Lecture Notes in Electrical Engineering 665

Gayadhar Pradhan
Stella Morris
Niranjan Nayak Editors

Advances
in Electrical
Control and
Signal Systems
Select Proceedings of AECSS 2019
Lecture Notes in Electrical Engineering

Volume 665

Series Editors

Leopoldo Angrisani, Department of Electrical and Information Technologies Engineering, University of Napoli
Federico II, Naples, Italy
Marco Arteaga, Departament de Control y Robótica, Universidad Nacional Autónoma de México, Coyoacán,
Mexico
Bijaya Ketan Panigrahi, Electrical Engineering, Indian Institute of Technology Delhi, New Delhi, Delhi, India
Samarjit Chakraborty, Fakultät für Elektrotechnik und Informationstechnik, TU München, Munich, Germany
Jiming Chen, Zhejiang University, Hangzhou, Zhejiang, China
Shanben Chen, Materials Science and Engineering, Shanghai Jiao Tong University, Shanghai, China
Tan Kay Chen, Department of Electrical and Computer Engineering, National University of Singapore,
Singapore, Singapore
Rüdiger Dillmann, Humanoids and Intelligent Systems Laboratory, Karlsruhe Institute for Technology,
Karlsruhe, Germany
Haibin Duan, Beijing University of Aeronautics and Astronautics, Beijing, China
Gianluigi Ferrari, Università di Parma, Parma, Italy
Manuel Ferre, Centre for Automation and Robotics CAR (UPM-CSIC), Universidad Politécnica de Madrid,
Madrid, Spain
Sandra Hirche, Department of Electrical Engineering and Information Science, Technische Universität
München, Munich, Germany
Faryar Jabbari, Department of Mechanical and Aerospace Engineering, University of California, Irvine, CA,
USA
Limin Jia, State Key Laboratory of Rail Traffic Control and Safety, Beijing Jiaotong University, Beijing, China
Janusz Kacprzyk, Systems Research Institute, Polish Academy of Sciences, Warsaw, Poland
Alaa Khamis, German University in Egypt El Tagamoa El Khames, New Cairo City, Egypt
Torsten Kroeger, Stanford University, Stanford, CA, USA
Qilian Liang, Department of Electrical Engineering, University of Texas at Arlington, Arlington, TX, USA
Ferran Martín, Departament d’Enginyeria Electrònica, Universitat Autònoma de Barcelona, Bellaterra,
Barcelona, Spain
Tan Cher Ming, College of Engineering, Nanyang Technological University, Singapore, Singapore
Wolfgang Minker, Institute of Information Technology, University of Ulm, Ulm, Germany
Pradeep Misra, Department of Electrical Engineering, Wright State University, Dayton, OH, USA
Sebastian Möller, Quality and Usability Laboratory, TU Berlin, Berlin, Germany
Subhas Mukhopadhyay, School of Engineering & Advanced Technology, Massey University,
Palmerston North, Manawatu-Wanganui, New Zealand
Cun-Zheng Ning, Electrical Engineering, Arizona State University, Tempe, AZ, USA
Toyoaki Nishida, Graduate School of Informatics, Kyoto University, Kyoto, Japan
Federica Pascucci, Dipartimento di Ingegneria, Università degli Studi “Roma Tre”, Rome, Italy
Yong Qin, State Key Laboratory of Rail Traffic Control and Safety, Beijing Jiaotong University, Beijing, China
Gan Woon Seng, School of Electrical & Electronic Engineering, Nanyang Technological University,
Singapore, Singapore
Joachim Speidel, Institute of Telecommunications, Universität Stuttgart, Stuttgart, Germany
Germano Veiga, Campus da FEUP, INESC Porto, Porto, Portugal
Haitao Wu, Academy of Opto-electronics, Chinese Academy of Sciences, Beijing, China
Junjie James Zhang, Charlotte, NC, USA
The book series Lecture Notes in Electrical Engineering (LNEE) publishes the
latest developments in Electrical Engineering—quickly, informally and in high
quality. While original research reported in proceedings and monographs has
traditionally formed the core of LNEE, we also encourage authors to submit books
devoted to supporting student education and professional training in the various
fields and applications areas of electrical engineering. The series cover classical and
emerging topics concerning:
• Communication Engineering, Information Theory and Networks
• Electronics Engineering and Microelectronics
• Signal, Image and Speech Processing
• Wireless and Mobile Communication
• Circuits and Systems
• Energy Systems, Power Electronics and Electrical Machines
• Electro-optical Engineering
• Instrumentation Engineering
• Avionics Engineering
• Control Systems
• Internet-of-Things and Cybersecurity
• Biomedical Devices, MEMS and NEMS
For general information about this book series, comments or suggestions, please
contact [email protected].
To submit a proposal or request further information, please contact the
Publishing Editor in your country:
China
Jasmine Dou, Associate Editor ([email protected])
India, Japan, Rest of Asia
Swati Meherishi, Executive Editor ([email protected])
Southeast Asia, Australia, New Zealand
Ramesh Nath Premnath, Editor ([email protected])
USA, Canada:
Michael Luby, Senior Editor ([email protected])
All other Countries:
Leontina Di Cecco, Senior Editor ([email protected])
** Indexing: The books of this series are submitted to ISI Proceedings,
EI-Compendex, SCOPUS, MetaPress, Web of Science and Springerlink **

More information about this series at http://www.springer.com/series/7818

Gayadhar Pradhan Stella Morris
• •

Niranjan Nayak
Editors

Advances in Electrical
Control and Signal Systems
Select Proceedings of AECSS 2019

123
Editors
Gayadhar Pradhan Stella Morris
Department of Electronics Universiti Tunku Abdul Rahman (UTAR)
and Communication Engineering Kajang, Malaysia
National Institute of Technology Patna
Patna, India

Niranjan Nayak
Department of Electrical
and Electronics Engineering
Sikshya ‘O’ Anusandhan University
Bhubaneswar, India

ISSN 1876-1100 ISSN 1876-1119 (electronic)

Lecture Notes in Electrical Engineering
ISBN 978-981-15-5261-8 ISBN 978-981-15-5262-5 (eBook)
https://doi.org/10.1007/978-981-15-5262-5
© Springer Nature Singapore Pte Ltd. 2020
This work is subject to copyright. All rights are reserved by the Publisher, whether the whole or part
of the material is concerned, specifically the rights of translation, reprinting, reuse of illustrations,
recitation, broadcasting, reproduction on microfilms or in any other physical way, and transmission
or information storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar
methodology now known or hereafter developed.
The use of general descriptive names, registered names, trademarks, service marks, etc. in this
publication does not imply, even in the absence of a specific statement, that such names are exempt from
the relevant protective laws and regulations and therefore free for general use.
The publisher, the authors and the editors are safe to assume that the advice and information in this
book are believed to be true and accurate at the date of publication. Neither the publisher nor the
authors or the editors give a warranty, expressed or implied, with respect to the material contained
herein or for any errors or omissions that may have been made. The publisher remains neutral with regard
to jurisdictional claims in published maps and institutional affiliations.

This Springer imprint is published by the registered company Springer Nature Singapore Pte Ltd.
The registered company address is: 152 Beach Road, #21-01/04 Gateway East, Singapore 189721,
Singapore
Organization

Chief Patron
Prof. (Dr.) Manojranjan Nayak, Hon’able President, SOA, India
Patron
Dr. A. Banerjee, Hon’able Vice-Chancellor, SOA, India
Finance Chair
Dr. M. K. Mallick, Director, FET (ITER), SOA, India
General Chair
Dr. P. K. Dash, SMIEEE, FINAE, FIE, SOA, India
Publication Chair
Dr. Gayadhar Pradhan, NIT Patna, India
Dr. S. Moriss, University of Tunku, Malaysia
Dr. H. S. Behera, VSSUT, India
Dr. P. K. Rout, SOA, India
Dr. R. K. Mallick, SOA, India
Dr. S. K. Kamilla, SOA, India
Dr. S. Dhar, SOA, India
Convener
Dr. Niranjan Nayak, SOA, India
Co-convener
Prof. L. M. Satpathy, SOA, India
Dr. T. P. Dash, SOA, India

v
vi Organization

Conference Committees

Advisory Committee
Dr. D. Acharya, SOA, India
Dr. R. P. Mohanty, SOA, India
Dr. P. K. Nanda, Dean Research, SOA, India
Dr. P. K. Sahoo, Dean, FET (ITER), SOA, India
Dr. R. Gokaraju, GUS, Canada
Dr. A. K. Moharana, Power Tech, Canada
Dr. N. P. Mahalik, California State University, Fresno
Dr. S. Majumdar, Carleton University, Canada
Dr. S. Dhar, San José State University, USA
Dr. M. Biswal, New Mexico State University, USA
Dr. A. Sahu, Coppin State University, USA
Dr. R. C. Bansal, University of Pretoria, South Africa
Dr. A. K. Pradhan, IIT Kharagpur, India
Dr. S. Mishra, IIT Delhi, India
Dr. B. Singh, IIT Delhi, India
Dr. B. K. Panigrahi, IIT Delhi, India
Dr. S. Samantray, IIT Bhubaneswar, India
Dr. D. Jena, NIT Surathkal, India
Organizing Committee
Dr. S. K. Dash, SOA, India
Prof. A. Gantayet, SOA, India
Dr. S. K. Routray, SOA, India
Prof. T. Parida, SOA, India
Dr. A. K. Patra, SOA, India
Prof. J. R. Panda, SOA, India
Dr. M. Mishra, SOA, India
Prof. S. K. Rout, SOA, India
Dr. S. Choudhury, SOA, India
Prof. A. Nanda, SOA, India
Mr. Debashis Sitikantha, SOA, India
Mrs. Sangita Rani Kar, SOA, India
Dr. M. Sahani, SOA, India
Mrs. Sunita S. Biswal, SOA, India
Dr. Snehamoy Dhar, SOA, India
Prof. R. Nanda, SOA, India
Prof. D. A. Gadnayak, SOA, India
Ms. Sandhyalati Behera, SOA, India
Prof. J. K. Rout, SOA, India
Organization vii

Ms. Archana Panda, SOA, India

Prof. A. Mishra, SOA, India
Ms. Sabita Pal, SOA, India
Prof. A. Pani, SOA, India
Ms. Jhansirani Jena, SOA, India
Preface

This volume LNEE contains the papers presented in the Advances in Electrical
Control and Signal System, AECSS 2019, held from November 8 to 9, 2019, at the
Institute of Technical Education and Research, Siksha ‘O’ Anusandhan Deemed to
be University, Bhubaneswar, Odisha, India. AECSS 2019 is treated as one of the
prestigious international conference series that aims at bringing together research-
ers, academicians, scholars and industry persons to report and review the latest
progress in cutting-edge research, innovation, sharing of new ideas on application
of control techniques to various areas like power system, MPPT control of a PV
power plant, energy, etc., along with implementation of signal processing for
solving specific problems, and finally to create awareness of these domains to a big
audience and participants.
AECSS 2019 received 165 paper submissions in total from various parts of the
globe out of which 81 papers were accepted for oral presentation to maintain the
standard of the conference proceedings. The accepted papers were reviewed
strongly by senior professionals of different reputed academic institutions. The
papers included in this LNEE volume cover a broad range of topics in control,
optimization, energy, machine learning, image and signal processing application
and problems selected from diverse domains of science and engineering.
The conference featured four distinguished keynote speakers. Professor
M. H. Naeem Hanoon, from Malaysia, talks on micro-grid control, which is an
emerging area and attracts many researchers in the present scenario. Professor
Subhransu Samantray, Indian Institute of Technology Bhubaneswar, delivered his
talk on “phasor measurement unit (PMU), design, testing and wide-area applica-
tions” which is an interesting topic in electrical power for a new researcher. The
phasor measurement unit with satellite communication was very interesting for the
audiences. The third keynote speaker was Prof. Gayadhar Pradhan, National
Institute of Technology Patna, India. He spoke on the topic of “front-end analysis
through single pole filter for development of robust speech-based applications.”
Some portion of his keynote speech was based on real-time work. Dr. Akash
Chakrabarty was one of the important keynote speakers in AECSS 2019, on the
topic “scientific writing and journal publishing, which is primarily addressed to the

ix
x Preface

young researchers.” His keynote speech was very informative and attractive and
motivated the young researchers.
We take this opportunity to thank the authors of all submitted papers in AECSS
2019, for their hard work, devotion to the deadlines and patience with the review
process. The quality of a refereed volume depends mainly on the expertise and
dedication of the reviewers. We are thankful to the reviewers for their timely effort
and help rendered to make this conference successful. We are indebted to the
program committee members who not only produced excellent reviews but also
constantly encouraged us during the short time frames to make the international
conference repute.
We would also like to thank our sponsors like OPTCL and LIT for providing all
the support and financial assistance. First, we are indebted to SOA Management,
vice-chancellor, director, deans and faculties and staff of the Department of
Electrical and Electronics Engineering for supporting and encouraging us to
organize the conference at ITER, SOA, Bhubaneswar. In particular, we would like
to express our heartfelt thanks and gratefulness to Prof. (Dr.) Manoj Ranjan Nayak,
Founder President, Siksha ‘O’ Anusandhan Deemed to be University, for providing
us the necessary financial support and infrastructural assistance to conduct the
conference. We thank Prof. (Dr.) Pradipta Kishore Dash, General Chair, AECSS
2019, for his continuous guidance. Our sincere thanks to Prof. Amit Banerjee,
honorable Vice-chancellor, SOA, for his continuous inspiring support. We convey
our thanks to Dr. Manas Ranjan Mallick, Director, ITER; Prof. P. K. Nanda, Dean
Research, SOA; and Prof. P. K. Sahoo, Dean, for their constant support and advice.
We would also like to thank the participants of this conference, who have
considered the conference above all hardships. Finally, we would like to thank all
the volunteers whose restless efforts in this conference and arranging every detail
ensured that the conference completed smoothly.

Kajang, Malaysia Stella Morris

Patna, India Gayadhar Pradhan
Bhubaneswar, India Niranjan Nayak
November 2019
Contents

Optimal Distribution Network Planning with Maximal Benefits

of Loss Reduction and Voltage Stability Development . . . . . . . . . . . . . 1
Amaresh Gantayet, Sudipta Mohanty, Aliva Arati Samantaray,
Lalit Mohon Satapathy, Akshaya Kumar Patra, and Alok Kumar Mishra
Novel Region Growing Mechanism for Object Detection
in a Complex Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
Tamanna Sahoo and Bibhuprasad Mohanty
A New Active Islanding Detection Technique Using Superimposed
Power Angle Disturbance of IBDER . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
Harikrishna Muda and Premalata Jena
MPPT-Based Inverter Control of Grid-Connected PV–Wind Hybrid
Power System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
A. V. Pavan Kumar and Y. Sai Varun
The Linear Quadratic Regulator Design for BG Control in Type-I
Diabetes Patient . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
Akshaya Kumar Patra, Anuja Nanda, Alok Kumar Mishra,
and Lalit Mohan Satapathy
ANN-Based Controllers for Improved Performance of BLDC
Motor Drives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
R. Shanmugasundaram, C. Ganesh, and A. Singaravelan
Frequency Regulation of Electric Vehicles Integrated Multi-area
Power System Under Deregulated Scenario Using Novel Dual
Mode Controller . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89
Sayantan Sinha and Ranjan Kumar Mallick
Photovoltaic Power Forecasting by Evolutionary Algorithm-Based
Improved Extreme Learning Machine . . . . . . . . . . . . . . . . . . . . . . . . . 109
Alok Kumar Pani and Niranjan Nayak

xi
xii Contents

Distribution System Optimization by Circular Reconfiguration

Technique . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131
Smrutirekha Mohapatra, Satwik Behera, and Subrat Kumar Dash
Novel Collecting Decision Optimization Algorithm for Enhanced
Dynamic Performance of Hybrid Power Source-Based SOFC
and Supercapacitor for Grid Integration . . . . . . . . . . . . . . . . . . . . . . . 147
Subhashree Choudhury, Bimal Sen, Nikhil Khandelwal,
and Anshuman Satpathy
Robust Controller Application to PV Fed CUK Converter
for Constant Voltage Regulation: “A Review” . . . . . . . . . . . . . . . . . . . 167
Nibedita Swain, Sadhna Mallik, and Gyana Ranjan Biswal
Crow Search Algorithm Optimized 3DOF-PID Controller for AGC
of Multi-area Power System Using RFB . . . . . . . . . . . . . . . . . . . . . . . . 183
Sk Alle Rasul Tarafdar, Digant Mishra, and D. P. Bagarty
Improvement of Transient Stability of AC-DC Power System
Using RPSO Based Sliding Mode Controller . . . . . . . . . . . . . . . . . . . . 199
Tanmoy Parida and Niranjan Nayak
Adaptive Resource Allocation for Cognitive Radio-Enabled Smart
Grid Network . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 213
Deepa Das and Niranjan Behera
Power Quality Improvement of a Fuel Cell-Based Distributed
Generation System Using Unified Power Quality Conditioner . . . . . . . 227
Sarita Samal, Tarakanta Jena, and Prasanta Kumar Barik
Numerical Approach to the Fractional Optimal Control Problem
of Continuous-Time Singular System . . . . . . . . . . . . . . . . . . . . . . . . . . 239
Tirumalasetty Chiranjeevi and Raj Kumar Biswas
Direct Torque Controlled Induction Motor Drive Using Super-Lift
Converter for Performance Improvement . . . . . . . . . . . . . . . . . . . . . . . 249
P. Elangovan, V. Maheswari, G. Nithiyanandham, and S. Prabhu
Design of Synchronous Reluctance Motor for Ceiling Fan
Application . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 263
Pratik Kochgabay, P. Ramesh, and N. C. Lenin
Modified Empirical Mode Decomposition and Teager–Kaiser Energy
Operator-Based Phasor Estimation in Presence of DC Offset
for Digital Relaying Application . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 273
Debadatta Amaresh Gadanayak and Ranjan Kumar Mallick
Improvement of Power System Stability Using Optimized Modified
Robust Controller Based STATCOM . . . . . . . . . . . . . . . . . . . . . . . . . . 285
Niranjan Nayak
Contents xiii

Comparative Analysis of Different Problem Formulation for Optimal

Capacitor Allocation in Distribution Systems . . . . . . . . . . . . . . . . . . . . 301
Shivangi Upadhyay and Sachin Singh
Self-tuned PI Controller Based Hybrid Shunt Active Power Filter
for Power Quality Enhancement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 315
Alok Kumar Mishra, Prakash Kumar Ray, Akshaya Kumar Patra,
Ranjan Kumar Mallick, and Soumya Ranjan Das
Performance Analysis of Si-Channel Nanosheet FETs with Strained
SiGe Source/Drain Stressors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 329
E. Mohapatra, T. P. Dash, J. Jena, S. Das, J. Nanda, and C. K. Maiti
Digital Implementation of OS-ELM for Data Classification
in Real-Time . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 339
Susanta Kumar Rout and Pradyut Kumar Biswal
Comparative Study of System Performances Using Integral Type
LQR with DE and Z-N Optimized PID Controller in AVR System . . . 349
Ashribad Pattnaik, Bidyadhar Rout, and Akshaya Kumar Patra
Voltage Control Method of Isolated Wind Power System . . . . . . . . . . . 361
Gyana Ranjan Biswal and Banaja Mohanty
Hybrid Sliding Mode Based Simplified NFC for Fuel Cell-Powered
Linearized IM Drive . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 373
Rabi Narayan Mishra, Kanungo Barada Mohanty, Abhimanyu Sahu,
and Partha Sarathi Behera
Kalman Filtering Sliding Mode Controller Design for Stabilizing
and Trajectory Tracking of Inverted Pendulum . . . . . . . . . . . . . . . . . . 387
Akshaya Kumar Patra, Alok Kumar Mishra, Anuja Nanda,
Lalit Mohan Satapathy, Amaresh Gantayet, Ramachandra Agrawal,
and Abhishek Patra
Electromagnetic and Thermal Analysis of Permanent Magnet BLDC
Wiper Motor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 405
Aneesh Jategaonkar, P. Ramesh, Pratik Kochgabay, and N. C. Lenin
Improved Sector-Based DTC-SVM for Induction Motor Drive
Using Hybrid Fuzzy-PI Controller . . . . . . . . . . . . . . . . . . . . . . . . . . . . 415
Abhimanyu Sahu, Kanungo Barada Mohanty, and Rabi Narayan Mishra
Detection and Classification of Brain Tumor Using Magnetic
Resonance Images . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 429
Limali Sahoo, Lokanath Sarangi, Bidyut Ranjan Dash,
and Hemanta Kumar Palo
xiv Contents

Real-Time Transmission Line Situational Awareness Using NI

Phasor Measurement Unit . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 443
Kunja Bihari Swain, Satya Sopan Mahato, Sushant K. Mandal,
and Murthy Cherukuri
Photovoltaic Cell with Shunt Active Power Filter for Harmonic
Cancelation Using Modified PSO-Based PI Controller . . . . . . . . . . . . . 455
Amba Subhadarshini Nayak, Devi Prasad Acharya,
and Subhashree Choudhury
Comparative Study of Segmentation of Noisy Color Synthetic Image
Using FCM and PFCM . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 469
P. Ganesan, B. S. Sathish, and L. M. I. Leo Joseph
FPGA Implementation of 16-Bit and 32-Bit Heterogeneous Adders . . . 479
Salony Mahapatro and Shasanka Sekhar Rout
Load Frequency Control of Multi-source Interconnected Power
System Controlled by Cascaded Controller . . . . . . . . . . . . . . . . . . . . . 491
Subhranshu Sekhar Pati and Saroj Kumar Mishra
A New Active Power Loss Allocation Method for Radial Distribution
Networks with DGs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 501
Ambika Prasad Hota, Sivkumar Mishra, and Debani Prasad Mishra
A GPS-Gradient Mapped Database-Based Fuzzy Energy
Management System for a Series—Parallel Hybrid
Electric Vehicle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 515
Jose Thomas, Allen Thomas, Akhil Biju, Aswin Mathew, C. Parag Jose,
and K. M. Haneesh
Optimal Placement of PMU: A Brief Review . . . . . . . . . . . . . . . . . . . . 529
Sandeep Kumar Mishra and Murthy Cherukuri
Design and Analysis of a PV-Based Cascaded H-Bridge Seven-Level
Inverter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 541
Parul Gaur, Yajvender Pal Verma, and Preeti Singh
Accurate Maximum Power Point Tracking with New Modified
Water Cycle-Based Ridge Extreme Learning Machine
for Photovoltaic-Based Microgrids . . . . . . . . . . . . . . . . . . . . . . . . . . . . 553
Sangram Keshari Pattnayak, Snehamoy Dhar, Durgesh Prasad Bagarty,
Meera Viswavandya, and Niranjan Nayak
Sensitivity Analysis of PI and Youla Controllers for a PV-Fed
Boost Converter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 567
Nivedita Pati, Babita Panda, and Bhagabat Panda
Contents xv

Improvement of Voltage Profile by Series Active Filter

in Grid-Connected PV System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 581
Rudranarayan Senapati, Sthita Prajna Mishra, Priyansha Sharma,
Rajendra Narayan Senapati, and Byomakesh Dash
Performance Improvement of Solar Photovoltaic Array by Modified
Bridge Link Array Configuration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 595
Chidurala Saiprakash, Alivarani Mohapatra, and Byamakesh Nayak
A Simple Decision Tree-Based Efficient Fault Classifier for PV-Based
AC Microgrid . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 605
Kanche Anjaiah and Rajesh Kumar Patnaik
Application of Salp Swarm Optimization for PI Controller
to Mitigate Transients in a Three-Phase Soft Starter-Based
Induction Motor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 619
Monalisa Mohanty, Sanjay Kumar Sahu, Manas Ranjan Nayak,
Anshuman Satpathy, and Subhashree Choudhury
An Improved MPPT Technique for Increasing Efficiency
of PV Module . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 633
Arjyadhara Pradhan, Babita Panda, Bhagabat Panda, and Aradhana Khillo
A Review of Advancements in DC Grid Technology . . . . . . . . . . . . . . 645
Jagadish Kumar Patra, Ramaprasad Panda, Smruti Pragnya Dash,
and Akanksha Rani
A Novel Approach for RFID-Based Smart EVM System . . . . . . . . . . . 661
Dipanjan Sen, Savio Jay Sengupta, Abhishek Sharma, Wasim Reja,
Bijoy Goswami, and Subir Kumar Sarkar
Design of a Low Power and Low Phase Noise VCO Using Active
Resistor and DTMOS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 671
Shasanka Sekhar Rout, Satabdi Acharya, and Kabiraj Sethi
A Stand-Alone Sine-Wave Inverter with Reduced Switching Loss . . . . 681
Nanda Kishor Panda, Ramprasad Panda, and Jagadish Kumar Patra
Mitigation of Power Quality Issues Using DSTATCOM Through
the Theory of Direct-Detection Light Wave Receivers Using
Optical Amplifiers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 691
S. M. Sanjay, M. Saritha, and M. H. Sidram
A Brief Review on Synchro Phasor Technology and Phasor
Measurement Unit . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 705
Monalisa Mohanty, Ravi Kant, Asit Kumar, Debasis Sahu,
and Subhashree Choudhury
xvi Contents

Evaluation of the Standard Procedure to Assess the Sphericity

and Size of Pellets Using Image Processing Techniques . . . . . . . . . . . . 723
Anuja Nanda, Akshaya Kumar Patra, Santisudha Panigrahi,
Alok Kumar Mishra, and Lalit Mohan Satapathy
Enhanced and Energy-Efficient Program Scheduling
for Heterogeneous Multi-Core Processors System . . . . . . . . . . . . . . . . . 737
Lavanya Dhanesh, S. Deepa, P. Elangovan, and S. Prabhu
A Prototype of Density-Based Intelligent Traffic Light Control
System Using Image Processing Technique and Arduino
Microcontroller in Lab VIEW Environment . . . . . . . . . . . . . . . . . . . . 749
Anita Mohanty, Subrat Kumar Mohanty, and Jitesh Kumar
Modeling and Performance Analysis of n-FinFETs:
A Comparative Study . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 765
J. Jena, T. P. Dash, E. Mohapatra, S. Das, J. Nanda, and C. K. Maiti
DOA Estimation on Fractal-Based Array . . . . . . . . . . . . . . . . . . . . . . . 777
P. Raiguru and R. K. Mishra
Output Voltage Regulated CUK and SEPIC Converter with High
Input Power Factor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 785
Alok Kumar Mishra, Akshaya Kumar Patra, Ramachandra Agrawal,
Nabajyoti Swain, Debadutta Dash, Shahil Sharma,
and Pranav Mohapatra
Energy Audit of Auxiliary Power Consumption: A Case Study . . . . . . 801
Avijit Nayak, Lalit Mohan Satapathy, Niranjan Nayak,
and Sudhansusekhar Dash
Atomistic Level Process to Device Simulation of GaNFET Using TNL
TCAD Tools . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 815
R. K. Nanda, E. Mohapatra, T. P. Dash, P. Saxena, P. Srivastava,
R. Trigutnayat, and C. K. Maiti
Modeling, Analysis, and Control of Vehicle Suspension System
Based on Self-Tuned PI Control . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 827
Akshaya Kumar Patra, Alok Kumar Mishra, Anuja Nanda,
Lalit Mohan Satapathy, Ramachandra Agrawal, and Narayan Nahak
IoT-Based Automatic Irrigation Control . . . . . . . . . . . . . . . . . . . . . . . 841
Akkenaguntla Karthik, Anumula Amarnath, T. M. Manohar Reddy,
Veldhanda Tulasi Krishna, and A. V. Pavan Kumar
Adaptive Controller Design for SMIB System Using Sliding
Mode Control . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 851
Subhranshu Sekhar Puhan, Sobhit Panda, and Animesh Kumar
Contents xvii

Classification of EEG Signal Using SVM . . . . . . . . . . . . . . . . . . . . . . . 859

Sandhyalati Behera and Mihir Narayan Mohanty
Dynamic Selection of the Best Server Out of Multiple Servers
Available Using Bully Algorithm in Any Geographical Area . . . . . . . . 871
Kartikeya Shukla and Bhaswati Sahoo
Heart Disease Prediction Using Machine Learning Techniques . . . . . . 879
Shekharesh Barik, Sambit Mohanty, Deepankar Rout, Subhra Mohanty,
Akshaya Kumar Patra, and Alok Kumar Mishra
Loss Allocation Strategies in Active Power Distribution Networks:
A Review . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 889
Ambika Prasad Hota, Sivkumar Mishra, and Debani Prasad Mishra
HSL Color Space Based Skin Lesion Segmentation Using
Fuzzy-Based Techniques . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 903
P. Ganesan, B. S. Sathish, and L. M. I. Leo Joseph
Industrial IoT-Based Gas Pipe Leakage Detector Robot . . . . . . . . . . . . 913
Henil Goswami, Udhav Goyal, and John Sahaya Rani Alex
Damping Control of a Multi—Area Thermal System Using SSO
Optimized 2DOF-PID-UPFC Based Controller . . . . . . . . . . . . . . . . . . . 923
Puja Dash and Lalit Chandra Saikia
Integration of Electric Vehicles in Distribution Grid for Maximizing
Profits Using MOGOA . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 933
Kumari Kasturi, Abhimanyu Behera, and Manas Ranjan Nayak
Video Object-Tracking Using Particle Filtering
and Feature Fusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 945
Jyotiranjan Panda and Pradipta Kumar Nanda
Voltage and Frequency Stabilization of a Hybrid Autonomous
System Using SMC-Based STATCOM Equipped with BESS . . . . . . . . 959
Abhijeet Choudhury, Soumya Mohanty, Swagat Pati, Amar Bijay Nanda,
Amiya Kumar Naik, and Sanjeeb Kumar Kar
Characterization of Power Quality Disturbances and Their Efficient
Classification . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 969
Laxmipriya Samal, Hemanta Kumar Palo, Badri Narayan Sahu,
and Debashisa Samal
A Novel Control Approach for Multi-level Inverter-Based
Microgrid . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 983
Sangram Keshari Routray, Buddhadeva Sahoo, and Sudhansu Sekhar Dash
An AIA-Based Parameter Extraction Method for PV System . . . . . . . 997
Swati Sucharita Pradhan and Raseswari Pradhan
xviii Contents

Smart Solar Electromagnetic Energy Harvesting Footware . . . . . . . . . 1009

Aneesh R. Nair and Vinit Kotak
A Novel Low Contrast Image Enhancement Using Adaptive
Multi-Resolution Technique and SVD . . . . . . . . . . . . . . . . . . . . . . . . . 1021
Lalit Mohan Satapathy and Pranati Das
Power Factor Corrected Cuk Converter with PI and Fuzzy
Logic Controller . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1035
Alok Kumar Mishra, Akshaya Kumar Patra, Ramachandra Agrawal,
Narayan Nahak, Amaresh Gantayet, Jnana Ranjan Swain,
and Samarjeet Satapathy
A Robust Competitive Optimization Algorithm Based Energy
Management Control Strategy in a Battery and Ultracapacitor
Based Hybrid Energy Storage System . . . . . . . . . . . . . . . . . . . . . . . . . 1049
Subhashree Choudhury, Nikhil Khandelwal, and Anshuman Satpathy
About the Editors

Dr. Gayadhar Pradhan is currently working as Associate professor in the

Department of ECE at NIT, Patna, India. He has received his M.Tech and Ph.D.
degrees in Electronics and Electrical Engineering from IIT, Guwahati, India, in 2009
and 2013, respectively. He has more than 12 years of teaching and research experi-
ence. His research interests are speech signal processing, speaker recognition, speech
recognition, keyword spotting, and VLSI signal processing. He has published more
than 50 papers in various reputed International Journals and Conference Proceedings.
He is also carrying two sponsored projects of worth nearly Rs. 1.5 crores.

Dr. Stella Morris is an Associate Professor in the Department of EEE, Universiti

Turku Abdul Rahman (UTAR), Malaysia. Stella Morris received her early educa-
tion in Tamilnadu, India before she obtained her B.E degree in EEE at the Madurai
Kamaraj University, Tamilnadu, in 1990. Subsequently, she obtained her M.E
degree at Anna University, Tamilnadu, in the area of the power system in 1992. In
2005, she obtained her Ph.D. degree from Multimedia University, Malaysia, in the
area of Engineering. Over the past 25 years in the academia, she has contributed her
expertise in a number of external/internal research grants and award-winning pro-
jects. She has over 50 research publications. She is a Professional Technologist
of the MBOT, BEM and ISTE.

Dr. Niranjan Nayak is an Associate Professor in the Department of EEE, SOA

deemed to be University, Odisha. He has received his Engineering degree from
Institution of Engineers, Kolkata, INDIA, in 1991. Subsequently, he has obtained
his M.Tech degree from UCE, Burla, Odisha, in 2003 and received his PhD degree
from SOA deemed to be university, Odisha In the year of 2015. Over the past 27
years in the industry and academia he has contributed his expertise in power system
engineering, Soft Computing, HVDC & FACTS, Renewable Energy, Power
System relaying, Evolutionary computing, Robust Controls and Machine Learning
Techniques to both UG and PG Students. He has published more than 35 papers in
various reputed International Journals and Conference Proceedings. Now he is the
member of different professional bodies like IEEE, IET, ISTE and IE.

xix
Optimal Distribution Network Planning
with Maximal Benefits of Loss Reduction
and Voltage Stability Development

Amaresh Gantayet , Sudipta Mohanty , Aliva Arati Samantaray,

Lalit Mohon Satapathy , Akshaya Kumar Patra, and Alok Kumar Mishra

Abstract Current electrical distribution and transmission infrastructure have wit-

nessed an upsurge for electrical energy requirements and a viable alternative is a
need for the hour, keeping in view the unbalance and deficit created in the generation
side of the system. The integration of distributed generation (DG) in distribution side
of the network, has thus been an emerging area of research. The placing and sizing
of distributed energy resources (DERs) have a significant impact on system behav-
ior. For the presented DG allocation problem, a multiobjective function is created
by fusing in two single objective functions of loss reduction and voltage stability
development. Genetic algorithm (GA) is employed to solve the multiobjective opti-
mization problem. The simulation results validate the presented methodology on 10,
12, 33, and 69 node radial distribution network (RDN).

Keywords Distributed energy resources · Distributed generation · Genetic

algorithm · Radial distribution network

A. Gantayet · A. A. Samantaray · L. M. Satapathy · A. K. Patra · A. K. Mishra

Institute of Technical Education and Research, Siksha ‘O’ Anusandhan (Deemed to be
University), Bhubaneswar 751030, India
e-mail: [email protected]
A. A. Samantaray
e-mail: [email protected]
L. M. Satapathy
e-mail: [email protected]
A. K. Patra
e-mail: [email protected]
A. K. Mishra
e-mail: [email protected]
S. Mohanty (B)
College of Engineering and Technology, Biju Pattnaik University of Technology, Bhubaneswar,
India
e-mail: [email protected]

© Springer Nature Singapore Pte Ltd. 2020 1

G. Pradhan et al. (eds.), Advances in Electrical Control and Signal Systems,
Lecture Notes in Electrical Engineering 665,
https://doi.org/10.1007/978-981-15-5262-5_1
2 A. Gantayet et al.

1 Introduction

For long, the ever-growing energy demand has been significantly analyzed by prac-
ticing engineers and system planning personnel. The onus is primarily of the existing
distribution systems to accommodate majority of the energy needs. A much of sup-
port can be obtained from decentralized generating units integrated into the existing
network. Various literature have suggested and discussed the role of renewable inte-
gration in electrical distribution system planning. The authors in [1] have suggested
various definitions of DG based on the competitive electricity market. [2] further
defines various aspects of distributed generation as a small-scale generation. In [3],
the authors have demonstrated a multiobjective process for optimum DG allocation
based on enhancement of voltage stability and reduction of power losses. The authors
in [4] have used a water cycle algorithm for integrating distributed energy units with
controllable power factors together with capacitor banks for multiobjective system
benefits. The authors in [5] have employed particle swarm optimization (PSO) for
multiple DG placement in the distribution network. Abri et al. in [6] have used a
voltage stability centric planning problem for renewable integration while including
the probabilistic nature of load and source. The authors in [7] have used the PSO
technique to get the sizing and locating parameters of DG and studied the operational
aspects together with the economic aspects of renewable integration in distribution
network. Bohre et al., in [8] used GA and PSO for optimal allocation of renewables in
distribution networks with load models. A multiobjective function with load models
is used in [9] to integrate multiple DG units in distribution network. Barik et al., in
[10] have obtained the location and sizing parameter by considering a seasonal vari-
ation of load and generation through a sequential optimization technique. Integration
of DERs in distribution network has been carried out in [11] to obtain maximum loss
reduction and voltage regulation through on-load tap changers. In [12], the authors
have solved the DG allocation problem to maximize distribution system operator’s
loss savings for a future planning duration. A random uncertainty based renewable
power generation and load variation are considered in [13] for optimal allocation set-
ting of renewable units through an analytical procedure. Sarkar et al., in [14], have
considered an average hourly load variation for integrating DG in the distribution
network through a network reconfiguration process. A combined voltage stability
index has been employed for DG site selection in [15] and an analytical procedure
is adopted to get the location setting for renewable integration. It can be well under-
stood from [16–20] that while considering renewable allocation problem location
and capacity of renewable DG units plays a vital role in getting maximum economic
and operational benefits.
The current work deals with a unified objective function which has been fused
from two individual objectives of voltage stability improvement of system buses and
line loss reductions of system branches. The Newton–Raphson load flow technique
adopted in the current work and genetic algorithm is employed for optimizing the
formulated fused objective function. Simulations are conducted over four different
test systems in order to test the efficacy and consistency of the presented methodology.
Optimal Distribution Network Planning with Maximal Benefits … 3

2 Renewable Integration Methodology

This section shall discuss about the process of obtaining the optimal position and
capacity of renewable units.

2.1 DG Integration Centered on a Unified Voltage Stability

Indicator

The site and capacity allocation of various DG units is a very crucial factor while con-
sidering DG penetration into a distribution grid. The node voltage stability scenarios
before and after DG penetration plays a vital role while integrating these renewables
into the distribution network. The prime objective of the work is to achieve dual ben-
efits of real power loss minimization and voltage stability improvement. A unified
stability index is presented in this paper and a process of normalization is adopted to
fuse both the objective into one. In [21, 22] two voltage collapse proximity indicators
are proposed which has been extensively used in various literature for locating most
voltage prone node of the distribution network. While [21] indicates a system node
centric indicator as shown in (1), [22] indicates a branch or line centric indicator (2).

SI(n2) =|V (n1)|4 − 4.0{P(n2)x(k) − Q(n2)r (k)}2

− 4.0{P(n2)r (k) + Q(n2)x(k)}|V (n1)|2 (1)

SI (n2) = Voltage stability indicator of bus n2 (n2 = 2, 3, …, NB)

For sturdy operation of the RDN, SI(n2) ≥ 0, n2 = 2, 3, …, NB. By using this
indicator, the proximity of voltage collapse associated with any bus can be estimated.
The bus with lowest indicator value is more reactive towards voltage collapse.

4x Q(n2)
LI(n2) = (2)
[V (n1) sin(θ − δ)]2

LI(n2) is termed as the stability index of that line with receiving end-node n2. For
desired operation of the RDN to be sturdy, LI (n2) ≤ 1.0, n2 = 2, 3, …, NB. The
above criterion (2) is used to find out the voltage collapse proximity indicator for
each line connected between two nodes of an interconnected network. More close
the indicator stays near one, more it shall be in proximity to voltage collapse and
vice versa.
Both indicators show the node prone towards voltage instability, (1) requires the
indicator to be away from zero, i.e., towards 1.0 pu, while (2) requires the opposite,
i.e., the indicator should be as low as possible. Both indicators are unified together
as shown in (3), so as to get a unified result out of it. A penalty factor is added to
the unified indicator in (3) so as to keep the values to a particular extent. In the work
lot of trial and run has been employed and finally, penalty factor (μ) of 5% has been
4 A. Gantayet et al.

selected.
 
1
USI = μ × min{SI(n2)} + (3)
max{LI(n2)}

2.2 Problem Formulation

In this section, the main problem is designed for the most favorable location and
capacity allocation of renewable DG keeping in view maximum real power loss
(RPL) reduction and voltage stability improvement. USI is considered as the first
objective (4) and δ1 need to be as small as possible, so as to increase the node voltage
stability. Real power loss is the second objective function (5), where δ2 needs to
be minimized. As the range of variation of real power loss and USI are different,
so to bring both objectives into one scaling platform a normalization function (6)
is used and finally the multiple objectives defined in (4) and (5) are fused into one
normalized objective function (7), which needs to be minimized.

δ1 = min{USI} (4)
 nbr 

δ2 = min |Ik | Rk
2
(5)
k=1

x(i) − xmin × ∅
Norm(x) = (6)
xmax − xmin × ∅
 nbr 

min(δ N ) = w1 × Norm{USI(i)} + w2 × Norm |Ik | Rk
2
(7)
k=1

While minimizing the required objective function the inequality constraints (8, 9)
have been considered. Equation (8) suggests that the minimum and maximum limits
of real power DG size is zero to total real power load and from (9) it is considered that
the search space for DG location has been taken as the entire system nodes excluding
the substation node. The complete procedural algorithm is shown in Sect. 3.

0 ≤ size of DG(PDG ) ≤ Pload (8)

2 ≤ location of DG ≤ Nbus (9)

Optimal Distribution Network Planning with Maximal Benefits … 5

2.3 Implementation of Multiobjective Function Through GA

The GA technique is a very profound iterative technique that has been used in a variety
of literature for optimizing the multidisciplinary and multidimensional optimization
problem [23]. Initially, a population is created either empirically or at random rep-
resented by strings of bits. In each iteration, the population pool is evaluated and
updated through a particular selection method to give rise to a new population pool.
The overall idea is inspired from the population genetics, and based upon “survival
of the best and discard the rest”. The selection technique assures the new pool to be
better at an average than the previous. Genetic operators like crossover and mutation
are used in order to have a slight variation in new population pool so that the complete
search space can be explored. The transition from one generation to another is guided
by the three processes—selection, crossover, and mutation—and are repeated until
some presumed stopping criterion is met. The stopping criteria can be maximum
number of generations or offspring new population not improving at a faster rate.
Step 1 Create population
Step 2 Evaluate each candidate solution through the fitness function
Step 3 Check Stopping criteria
Step 4 Selection
Step 5 Mating
Step 6 Crossover
Step 7 Mutation and back to step 2

3 Procedural Algorithm and Performance Criterion

3.1 Procedural Algorithm

Step 1 Read network and load data.

Step 2 Run power flow for base case and store results.
Step 3 Determine maximal and minimal limits of USI as per system constraints
mentioned in (8, 9) through GA
• Define GA parameters and create population of various UPF DG sizes
and location.
• Execute the objective function (5) with each generation.
• Run load flow for getting USI and RPL.
• Obtain the highest and lowest values of USI and RPL.

Step 4 Re-Run GA for optimal DG siting and sizing settings

• Define GA parameters and create population of various UPF DG sizes
and location.
6 A. Gantayet et al.

• Execute the objective function (7) by normalizing USI and RPL with each
generation.
• Store results

Step 5 Compare the various performance criterion of Sect. 3.2 for various test
systems.
Step 6 Store results

3.2 Performance Criterion

The robustness and efficacy of the presented methodology are tested via various
performance criterion [3] described in this section.

Voltage profile enhancement (VPE)

 
nbus nbus
i=1 Vi PLi − i=1 Vi PLi
w/DG Wo /DG
VPE% =  (10)
nbus
i=1 Vi PLi
Wo /DG

Voltage profile enhancement (VPE) is calculated using (10), where Vi represents

voltage magnitude at bus i, PLi represents the real power load at bus i, nbus represents
the maximum number of buses in the network.

Real Power Loss Reduction (RPLR) Real power loss reduction (RPLR) is
calculated using (11)
nbr  nbr
k=1 Ik,wo/DG Rk −
2 2
k=1 Ik,w/DG Rk
RPLR% = nbr 2 × 100 (11)
k=1 Ik,wo/DG Rk

where Rk is the branch resistance of kth branch in pu. Ik,w/DG is the branch current
value of kth branch with DG integration and Ik,wo/DG , branch current value of kth
branch without considering the DG integration and nbr represents the number of
branches.

Network loadability upliftment (NLU) NLU can be hypothetically defined as a

process to visualize the after-effects of DG integration, where the maximum system
loadability (λmax ) is obtained by increasing active and reactive power load all buses,
till the voltage collapse is observed. NLU is given by (12).

λmax(w/DG) − λmax(wo/DG)
NLU% = × 100 (12)
λmax(wo/DG)
Optimal Distribution Network Planning with Maximal Benefits … 7

4 Results and Discussion

Four different radial distribution systems (10 bus [24], modified 12 bus [17], 33
bus [25], and 69 bus [21]) are considered in the work to determine the robustness
and efficacy of the presented methodology. The system modeling and program was
carried out on Intel Core™ i5-4200 M CPU 2.5 GHz processor in MATLAB/Simulink
R2016a environment. The values of GA parameters used are as follows, Crossover
Probability PC = 1.0, Mutation Probability PM = 0.006, No. of Generations = 1000
(For Step-3 of Procedural Algorithm) and 100 (For Step-4 of Procedural Algorithm),
No of Bits = 12, No. of Population = 61, Selection Method: Roulette Wheel. The
USI and RPL Data of 69 Bus RDS is given for reference. MaxUSI = 0.4014 (pu),
MinUSI = 0.3375 (pu), MaxRPL = 614.1177 (kW), and MinRPL = 83.1799 (kW).
Table 1 shows the respective real power DG sizes for all the test systems under
consideration. Table 2 reflects compare and contrast of the system performance of
the presented GA method. The method is also compared with its base case and an
analytical method [17].
Figure 1 shows voltage profile of all the test systems under study and compares it
with base case and the analytical method [17]. With the presented methodology, the
voltage profile of all test systems has shown significant advancement and found to
be within acceptable limits (0.95 < V bus < 1.05).
Figure 2 shows the VPE with a radar plot for the test systems under study and, it
can be observed that the GA method is found better in performance than the analytical
method.
Figure 3 compares the loadability upliftment between the GA and analytical algo-
rithms. It is observed that both the methods share similar scope in system loadability
for all the systems under study, excluding 33 bus systems where GA method is
marginally better.
Figure 4 shows the effect of real power DG injection on RPLR using both the
methods. It is observed that using the present methodology, the power loss witnesses
a significant reduction. The results of the analytical method are also found equivalent
to the GA algorithm, with an exclusion in case of 33 bus systems, where GA method
gives around 32% more loss reduction. Thus, the presented methodology through
the application of GA serves as an efficient method of DG allocation problem.

Table 1 Application of the GA method on various distribution network

Test system GA method
Best node for DG installation Size (kW)
10-Bus 10 4424.7
12-Bus 9 235.5
33-Bus 7 2390.5
69-Bus 61 1872.3
8 A. Gantayet et al.

Table 2 System performance with DG integration using GA method

Test system System loadability Vmax —node Vmin —node System Losses (kW)
Without DG (base case)
10-Bus 2.0 1—1 0.8375—10 783.77
12-Bus 5.3 1—1 0.9434—12 20.71
33-Bus 3.6 1—1 0.9131—17 202.51
69-Bus 3.2 1—1 0.9092—65 224.95
With DG (analytical method)
10-Bus 3.0 1—1 0.9473—8 226.8
12-Bus 5.9 1—1 0.9838—7 10.777
33-Bus 3.66 1—1 0.9185—18 171.28
69-Bus 3.9 1—1 0.9685—27 83.2
With DG (GA method)
10-Bus 3.1 1–1 0.9531—7 243.509
12-Bus 5.9 1–1 0.9835—7 10.774
33-Bus 3.9 1–1 0.9514—18 105.02
69-Bus 3.9 1–1 0.9633—65 83.179

a. 10 Node RDS b. 12 Node RDS

c. 33 Node RDS d. 69 Node RDS

Fig. 1 Voltage Profile of various RDS using different DG Placement Algorithm

This is further justified in Table 3, where the current work (GA method) is also
compared with [3] where PSO is employed and [26] where a grid search algorithm
is employed. Table 3 proves the consistency and efficacy of the presented work.
Optimal Distribution Network Planning with Maximal Benefits … 9

Fig. 2 Comparison of VPE

with different DG integration
methodologies

Fig. 3 Comparison of NLU

with different DG integration
methodologies

Fig. 4 Comparison of
RPLR using different DG
placement algorithms
10 A. Gantayet et al.

Table 3 Comparison with existing references

System type Parameters Presented work [3] PSO method [26] Grid search
method
12 bus DG location 9 9 9
DG size (kW) 235.5 253.9 234.90
V max (pu) 1 1 1
V min (pu) 0.9835 0.9850 0.9835
System RPL (kW) 10.774 10.82 10.76
System loading 5.9 6.03 5.98
33 bus DG location 7 7 6
DG size (kW) 2390.5 2895.1 2600.5
V max (pu) 1 1 1
V min (pu) 0.9514 0.9501 0.9425
System RPL (kW) 105.02 114.89 111.03
System loading 3.9 3.78 3.73
69 bus DG location 61 61 61
DG size (kW) 1872.3 2026.4 1863.03
V max (pu) 1 1 1
V min (pu) 0.9633 0.9692 0.9683
System RPL (kW) 83.179 84.04 83.22
System loading 3.9 4.03 3.93

5 Conclusions

In the present work, a GA-based multiobjective approach is fused into a single objec-
tive approach for optimal allocation of real power DERs. Four types of systems are
simulated in the work to prove the consistency and efficacy of the methodology.
The simulation results verify the use of the proposed methodology in obtaining the
required objectives, like real power loss reduction and voltage profile advancement.
However, the work considers a fixed constant power loading pattern, whereas in
place the uncertainty associated with renewable sources and load can be simulated
as future scope of research.

References

1. Ackermann, T., Andersson, G., Söder, L.: Distributed generation: a definition. Electr. Power
Syst. Res. 57, 195–204 (2001). https://doi.org/10.1016/S0378-7796(01)00101-8
2. Pepermans, G., Driesen, J., Haeseldonckx, D., Belmans, R., D’haeseleer, W.: Distributed gen-
eration: definition, benefits and issues. Energy Policy 33, 787–798 (2005). https://doi.org/10.
1016/j.enpol.2003.10.004
Optimal Distribution Network Planning with Maximal Benefits … 11

3. Aman, M.M., Jasmon, G.B., Bakar, A.H.A., Mokhlis, H.: A new approach for optimum DG
placement and sizing based on voltage stability maximization and minimization of power
losses. Energy Convers. Manage. 70, 202–210 (2013). https://doi.org/10.1016/j.enconman.
2013.02.015
4. El-Ela, A.A.A., El-Sehiemy, R.A., Abbas, A.S.: Optimal placement and sizing of distributed
generation and capacitor banks in distribution systems using water cycle algorithm. IEEE Syst.
J. 12, 3629–3636 (2018). https://doi.org/10.1109/JSYST.2018.2796847
5. AlRashidi, M.R., AlHajri, M.F.: Optimal planning of multiple distributed generation sources
in distribution networks: a new approach. Energy Convers. Manage. 52, 3301–3308 (2011).
https://doi.org/10.1016/j.enconman.2011.06.001
6. Al, A.R., El-Saadany, E.F., Atwa, Y.M.: Optimal placement and sizing method to improve
the voltage stability margin in a distribution system using distributed generation. IEEE Trans.
Power Syst. 28, 326–334 (2013). https://doi.org/10.1109/TPWRS.2012.2200049
7. Ameli, A., Bahrami, S., Khazaeli, F., Haghifam, M.: A multiobjective particle swarm opti-
mization for sizing and placement of DGs from DG owner’s and distribution company’s view-
points. IEEE Trans. Power Deliv. 29, 1831–1840 (2014). https://doi.org/10.1109/TPWRD.
2014.2300845
8. Bohre, A.K., Agnihotri, G., Dubey, M.: Optimal sizing and sitting of DG with load models
using soft computing techniques in practical distribution system. IET Gener. Transm. Distrib.
10, 2606–2621 (2016). https://doi.org/10.1049/iet-gtd.2015.1034
9. Gampa, S.R., Das, D.: Optimum placement and sizing of DGs considering average hourly
variations of load. Int. J. Electr. Power Energy Syst. 66, 25–40 (2015). https://doi.org/10.1016/
J.IJEPES.2014.10.047
10. Barik, S., Das, D.: Determining the sizes of renewable DGs considering seasonal variation of
generation and load and their impact on system load growth. IET Renew. Power Gener. 12,
1101–1110 (2018). https://doi.org/10.1049/iet-rpg.2017.0528
11. Meena, N.K., Swarnkar, A., Gupta, N., Niazi, K.R.: Optimal integration of DERs in coordina-
tion with existing VRs in distribution networks. IET Gener. Transm. Distrib. 12, 2520–2529
(2018). https://doi.org/10.1049/iet-gtd.2017.1403
12. Prenc, R., Cuculiò, A., Capuder, T., Guerrero, J.M.: Optimal siting and sizing of DG units for a
MV network going through voltage transition. In: 2018 IEEE International Energy Conference
(ENERGYCON), pp. 1–6 (2018)
13. Zhang, C., Li, J., Zhang, Y.A., Xu, Z.: Optimal location planning of renewable distributed
generation units in distribution networks: an analytical approach. In: 2018 IEEE Power &
Energy Society General Meeting (PESGM), p. 1 (2018)
14. Sarkar, S., Das, D.: Effect of distributed generations on reconfiguration and voltage stability
of distribution networks considering average hourly load data. Int. J. Power Energy Convers.
7, 313–335 (2016). https://doi.org/10.1504/IJPEC.2016.079885
15. Gantayet, A., Mohanty, S.: An analytical approach for optimal placement and sizing of
Distributed Generation based on a combined voltage stability index. In: 2015 IEEE Power,
Communication and Information Technology Conference, PCITC 2015—Proceedings (2016)
16. Dixit, M., Kundu, P., Jariwala, H.R.: Incorporation of distributed generation and shunt capacitor
in radial distribution system for techno-economic benefits. Eng. Sci. Technol. Int. J. 20, 482–493
(2017). https://doi.org/10.1016/j.jestch.2017.01.003
17. Aman, M.M., Jasmon, G.B., Mokhlis, H., Bakar, A.H.A.: Optimal placement and sizing of a
DG based on a new power stability index and line losses. Int. J. Electr. Power Energy Syst. 43,
1296–1304 (2012). https://doi.org/10.1016/j.ijepes.2012.05.053
18. Celli, G., Ghiani, E., Mocci, S., Pilo, F.: A multiobjective evolutionary algorithm for the sizing
and siting of distributed generation. IEEE Trans. Power Syst. 20, 750–757 (2005). https://doi.
org/10.1109/TPWRS.2005.846219
19. Pereira, B.R., da Costa, G.R.M., Contreras, J., Mantovani, J.R.S.: Optimal distributed gen-
eration and reactive power allocation in electrical distribution systems. IEEE Trans. Sustain.
Energy 7, 975–984 (2016). https://doi.org/10.1109/TSTE.2015.2512819
12 A. Gantayet et al.

20. Bhattacharya, M., Das, D.: Multi-objective placement and sizing of DGs in distribution network
using genetic algorithm. In: 2016 National Power Systems Conference (NPSC), pp. 1–6 (2016)
21. Chakravorty, M., Das, D.: Voltage stability analysis of radial distribution networks. Int. J. Electr.
Power Energy Syst. 23, 129–135 (2001). https://doi.org/10.1016/S0142-0615(00)00040-5
22. Moghavvemi, M., Faruque, M.O.: Power system security and voltage collapse: a line outage
based indicator for prediction. Int. J. Electr. Power Energy Syst. 21, 455–461 (1999). https://
doi.org/10.1016/S0142-0615(99)00007-1
23. Das, D.: Reactive power compensation for radial distribution networks using genetic algo-
rithm. Int. J. Electr. Power Energy Syst. 24, 573–581 (2002). https://doi.org/10.1016/S0142-
0615(01)00068-0
24. Baghzouz, Y., Ertem, S.: Shunt capacitor sizing for radial distribution feeders with distorted
substation voltages. IEEE Trans. Power Deliv. 5, 650–657 (1990). https://doi.org/10.1109/61.
53067
25. Baran, M.E., Wu, F.F.: Network reconfiguration in distribution systems for loss reduction and
load balancing. IEEE Power Eng. Rev. 9, 101–102 (1989). https://doi.org/10.1109/MPER.
1989.4310642
26. Gözel, T., Eminoglu, U., Hocaoglu, M.H.: A tool for voltage stability and optimization
(VS&OP) in radial distribution systems using matlab graphical user interface (GUI). Simul.
Model. Pract. Theory 16, 505–518 (2008). https://doi.org/10.1016/j.simpat.2008.02.003
Novel Region Growing Mechanism
for Object Detection in a Complex
Background

Tamanna Sahoo and Bibhuprasad Mohanty

Abstract Object detection is vital for visual processing applications. In this work,
the desired object in an image is detected by the help of the wavelet coefficient fea-
ture (WCF) extraction and region growing technique. The region growing technique
is based upon the appropriate selection of seed block computation and adjacency
thresholding technique. The novelty of the proposed work is based on computation
of seed block using WCF from the dynamics of the image instead of an image itself.
Haar filter has been applied to transform the image after two level of decomposition
for WCF extraction and to take care of the reduction in time complexity of the system.
The extensive simulation-based experiment demonstrates the proposed methodology
efficiently detects the object even in the presence of complex or cluttered (dynamic)
background.

Keywords Object detection · Wavelet coefficient feature · Seed block · Region

growing

1 Introduction

The basic characteristics of human vision system are to rapidly select, separate and
localize the objects [1] as specified by the user [2]. In fact, the detection of objects
automatically [3] in a particular scene mostly depends on the focus of the vision that is
the attention [4], for which it is one of the basic steps of many high level algorithms
(such as moving object detection, tracking, classification and so on) of computer
vision area. Further, this has led to many applications like video surveillance of
traffic and border where detection of suspicious objects is very much necessary. The
main aim of this paper is to detect and identify the single object in a natural scene

T. Sahoo (B) · B. Mohanty

Department of Electronics and Communication Engineering, Institute of Technical Education and
Research, Siksha ‘O’ Anusandhan (Deemed to be University), Bhubaneswar 751030, India
e-mail: [email protected]
B. Mohanty
e-mail: [email protected]

© Springer Nature Singapore Pte Ltd. 2020 13

G. Pradhan et al. (eds.), Advances in Electrical Control and Signal Systems,
Lecture Notes in Electrical Engineering 665,
https://doi.org/10.1007/978-981-15-5262-5_2
14 T. Sahoo and B. Mohanty

using wavelet coefficient feature-based approaches. Usually, the identification of

objects under different environmental condition like moving clouds, heavy rainfall or
snowfall [1] and different categories of objects (either single or multiple, stationary
or moving) [5] is a difficult task, and it has led many researchers to work upon
this area. Many literatures have been provided regarding the approaches used by
the researchers such as thresholding methods, pattern recognition algorithms and
texture-based techniques. Most of such approaches are in spatial domain. Taking
the advantage of multiresolution analysis of wavelet transform [6], the two proposed
methods are developed in this work. The computation of seed block from the wavelet
coefficient features for each sub-block of an image is the main contribution of this
paper.
The flow of this paper is prepared accordingly, wherein Sect. 2 presents the liter-
ature survey related to discrete wavelet transform and target detection, in Sect. 3, the
proposed methods to obtain WCFs of each sub-block, selection of seed block and
region growing processes using seed block and seed point have been described and
discussed. Section 4 presents the experimental results of the proposed methodology
as well as discussion of the results and issues related to it. In Sect. 5, conclusion
and further extension of this work like detection of multiple objects related to this
technique have been discussed.

2 Literature Survey

The computation of wavelet coefficient features is basically the texture-based statis-

tical calculation. Usually, most of the traditional texture-based statistical approaches
(such as concurrence matrix [7], second order statistics [8], local linear transform [9,
10], Gaussian Markov Random Fields (GMRF) [11], etc.) are restricted to only single
scale when applied in spatial domain, whereas in wavelet domain, it can be applied
to multiple scales. This advantage of wavelet domain has gained lot of attention
recently. Further, the properties of wavelet domain such as multiresolution represen-
tation, orthogonality, fast processing and so on [12] make it suitable for this purpose.
In fact, the use of low pass and high pass filters in wavelet domain remains same for
successive scales unlike Gabor filter [13], which is another advantage of this domain.
Even in paper [14], for texture characterization and classification, wavelet statistical
feature (WSF) and WCF were proposed and calculated effectively.
The advantages of wavelet transform lead to its usage in detecting the objects. For
which in paper [15], the author made efforts to design a real-time vision system which
has a feature extractor and relaxation network using multiresolution decomposition.
In paper [16, 17], the wavelet transform is used for recognizing and detecting objects
in antisubmarine warfare. Wavelet-based algorithm is applied on Logan radar data
for detection of submarine wakes as well as wavelet packet methods for analyzing
large formats of sonar images. According to paper [18], Tial et al. used a spectral
statistics and wavelet coefficient characterization (SSWCC) for target detection and
classification. In this process, a feature vector has been extracted in both frequency
Novel Region Growing Mechanism for Object Detection … 15

and time–frequency domain, and further analysis of the seismic signals includes the
collaborative work of the spectrum, the power spectral density (PSD) and the wavelet
coefficients to obtain the classification. Even using Renyi’s information and wavelet
transform, the detection of objects in mammographic images has been proposed in
one of the paper [19]. This method works by maximizing the information available
at each resolution level of the representation. Although the above-discussed papers
involve detection of targets in wavelet domain, the approaches proposed in this paper
effectively exploit the co-occurrence features in wavelet domain from each detailed
subband of transformed image for the detection of specific single objects in clear
and cluttered backgrounds.

3 Proposed Methodology

Identification of target objects in a natural scene is one of the difficult tasks in

computer vision area. The proposed methodology has applied methods such as image
resizing, division of image into sub-blocks, 2D-discrete wavelet transform, wavelet
coefficient feature extraction (main contribution of this paper) and region growing
process to obtain the desired target, and it is presented in the below Fig. 1.

Fig. 1 Basic building block of the proposed target detected system

16 T. Sahoo and B. Mohanty

In both the proposed systems, the division of input image of size N × M into
sub-blocks of 32 × 32 which is an important task for which the input image is first
resized into 512 × 512. And for better processing, if the input image is color image,
it is then first transformed to gray-scale image.
The partition of sub-block images is carried out in such a way that it is non-
overlapping and adjacent blocks using raster scan technique. The process of division
of distinct sub-block images starts from the top-left corner of original image, and
further, each sub-block image is decomposed into two-level two-dimensional discrete
wavelet transform.
From the transformed coefficients, the gray level co-occurrence matrix and WCF
are calculated for the sub-block image using both the techniques. In the next step, eval-
uation of wavelet coefficient feature values using the statistical formula is calculated
for each sub-block.
The next step is the selection of seed block which is the sub-block image having
maximum of normalized feature values. The concept behind the selection of seed
block is that the high value of wavelet coefficient feature represents that it is a part of
the target. The center point of seed block is obtained, and it is referred as seed point
which is used along with region growing algorithm (adjacency of neighboring pixel
and threshold) for the identification of specific object in the image.

3.1 Two-Dimensional Discrete Wavelet Transform (2D-DWT)

Discrete wavelet transform is a hierarchical subband system where the spacing of

subbands is logarithmic in frequency domain. Further, the flexible nature of DWT
[20] that is multi-resolution or multichannel analysis [21] characteristics is one of
the major advantages which help researchers to work on each individual subband of
transformed image. It also can be implemented in pyramidal structure (dyadic) or
in tree structure (packet) which adds another feather to its wing. Other than that the
object information can be obtained in both spatial and wavelet domain making it one
of the convenient approach.
Basically, the DWT is performed by passing the input image through low pass
and high pass filter at each resolution level, and equation of 2D-DWT is presented
below:
   
−( j+1) −( j)
FIA ( p, q) = h ∗ (− p) ∗ h ∗ (−q) ∗ FIA ( p, q) ↓ 2q ↓ 2 p (1)
   
−( j+1) −( j)
FIH ( p, q) = h ∗ (− p) ∗ g ∗ (−q) ∗ FIH ( p, q) ↓ 2q ↓ 2 p (2)
   
−( j+1) −( j)
FID ( p, q) = g ∗ (− p) ∗ h ∗ (−q) ∗ FID ( p, q) ↓ 2q ↓ 2 p (3)
   
−( j+1) −( j)
FIV ( p, q) = g ∗ (− p) ∗ g ∗ (−q) ∗ FIV ( p, q) ↓ 2q ↓ 2 p (4)
Novel Region Growing Mechanism for Object Detection … 17

[W ] = FIA + FID + FIV + FIH (5)

where the F I represents the input image and F IA , F ID , F IV and F IH are the
wavelet decomposition subbands (approximation, diagonal, vertical and horizontal
coefficient) at second-level decomposition and W is the wavelet transform.

3.2 Co-occurrence Matrix

The significant wavelet coefficient features are computed by first calculating the
gray level co-occurrence matrix (GLCM) of the image. GLCM may be defined as
the distribution of gray pixels of an image in a matrix having certain relation. The
GLCM element Co(p, q, d, θ ) represents the probability of the pair of pixels (p, q)
which are located at a certain distance d and angle θ .
For instance, if the displacement vector is (0, 1), then it is interpreted as one pixel
below and one pixel to the right in the direction of 0°. Similarly, if the displacement
vector is (1, −1), then it is interpreted as one pixel below and one pixel to the left
in the direction of 135°. Figure 2 demonstrates the co-occurrence matrix in four
different orientations.
Using this information, co-occurrence matrix can be created at different orienta-
tions and different statistical measures having textural characteristics as defined by
Harllick [7] are extracted. These features are contrast, entropy, energy, dissimilarity,
cluster prominence, cluster shade, variance, autocorrelation which is given by
1. Autocorrelation: It is the measure of the magnitude of the fineness and
coarseness of texture.


N
Autocorrelation = pqCo( p, q) (6)
p,q=1

2. Cluster Prominence: It measures the asymmetry of the GLCM, and lower

value indicates peaks of the distribution are centered on the mean.

Fig. 2 Co-occurrence matrix orientation. a d = (0, 1), θ = 0°. b d = (1, 1), θ = 45°. c d = (1, 0),
θ = 90°. d d = (1, −1), θ = 135°
18 T. Sahoo and B. Mohanty


N
 4
Cluster prominence = p − Mx + q − M y Co( p, q) (7)
p,q=1

3. Cluster Shade: It measures skewness of the co-occurrence matrix with a

perceptual concept of uniformity.


N
 3
Cluster shade = p − Mx + q − M y Co( p, q) (8)
p,q=1

4. Contrast: The measurement of the local variations in an image and is very

much connected with the variance between the uppermost and the lowermost
values of a continuous set of pixels.


N
Contrast = ( p − q)2 Co( p, q) (9)
p,q=1

5. Energy: It represents the homogeneity of an image with the higher value

indicating uniformity in texture.


N
Energy = Co( p, q)2 (10)
p,q=1

6. Entropy: The uncertainty of the image intensity distribution is measured using

entropy, and it is high when all the possibilities Co( p, q) are identical and low
when the possibilities Co( p, q) are different.


N
Entropy = − Co( p, q) log2 Co( p, q) (11)
p,q=1

7. Sum of square (variance): It is the measurement of the gray level unevenness of

the pixel pairs indicating the measurement of heterogeneity in an image which
increases when the gray-scale values vary from their means.


N
Variance = ( p − μ)2 Co( p, q) (12)
p,q=1

where μ = μ p = μq is the mean gray level intensity.

8. Dissimilarity: It measures the mean of the gray level difference distribution of
the image. A larger value implies greater disparity in intensity values among
neighboring pixels.
Novel Region Growing Mechanism for Object Detection … 19


N
Dissimilarity = pCo( p, q) (13)
p=0

9. Inverse Difference Normalized: Here, the variance of the neighboring intensity

values is normalized by the total number of discrete intensity values.


N
Co( p, q)
Inverse Difference Normalized =   (14)
| p−q|2
p,q=1 1 + N

10. Maximum probability: It is the maximum of Co( p, q).


N
Maximum probability = max(Co( p, q)) (15)
p=0


N 
N
where Mx = pCo( p, q), My = qCo( p, q) (16)
p,q=1 p,q=1

3.3 Region Growing Algorithm

Region growing process [22] is one of the basic region-based image techniques in
which the regions are grown based on some criteria. The most convenient and basic
step of this technique to grow a region is by taking seed points. The selection of seed
point [1] is one of the important tasks of this method, and it is generally user defined
for which it is quite advantageous for researchers to formulate the seed point as per
required condition. After seed point selection, adjacent neighboring pixels are taken
into account which are further considering some threshold the neighboring pixels
are added to the region. The basic formulation of this technique is given below.
Other than seed point selection, the correct separation of regions having similar
properties leading to region having clear edges with better segmentation results is
another advantage of this process.

3.4 Dynamics Extraction

The process of extracting the dynamics (motion) [23] of an object, in a scene, by

using only the high frequency component provides the necessary edge information
and high contrast value information of the moving objects.
20 T. Sahoo and B. Mohanty

Fig. 3 Dynamics in wavelet transform

The extraction of dynamics in each sub-block of an image using 2D-DWT has

been shown in Fig. 3. For each sub-block, two-level 2D-DWT has been applied, the
low frequency component of both the levels is then masked to zero, and their inverse
transform is applied only to high frequency components(which contains the detail
information of the image) generating the dynamics of that sub-block which is given
by the Formula (19).
−j
W ∗−1 = [W ] − FIA (x, y) (19)

3.5 Algorithm I

Step-1 Image Preprocessing

For input image
• Resize it to 512 × 512.
• Divide it to 32 × 32 sub-block using raster scan.
Step-2 Wavelet Coefficient Feature Extraction
For each sub-block
Do
• Apply two-level 2D-DWT.
• Mask LL subband.
• Apply GLCM to HH, HL, LH subband
• Extract features of all subbands.
• Combine the individual feature values for all subbands.
 6
K 
SBB = (feature)i (17)
i=1 Sb=1
For k = 1, 2, …, 256.
• Repeat for all sub-blocks.
end
Step-3 Seed Block Computation
Do
• Maximum of feature values of all sub-blocks.
end
Step-4 Region Growing Algorithm
(continued)
Novel Region Growing Mechanism for Object Detection … 21

(continued)
Do
• Choose the seed block.
• Compute the seed point as the center of the sub-block.
• Compute adjacency in four-neighborhood pixels, such that,
PK = ds − di > T (18)
For i = 1, …, 4;
where ds is seed point pixels and di is the neighboring pixels.
• Merge PK with ds
• Calculate the mean PK S to find the next seed point and repeat the steps.
end

3.6 Algorithm II

Step-1 Image Preprocessing

Same as above Algorithm I.
Step-2 Wavelet Coefficient Feature Extraction using Dynamic Extraction
Do
• Apply two-level 2D-DWT.
• Mask LL subband to zero.
• Apply two-level inverse 2D-DWT
• Apply GLCM to the dynamics of sub-block
• Repeat for all sub-blocks.
End
Step-3 Seed Block Computation
Same as above Algorithm I.
Step-4 Region Growing Algorithm
Same as above Algorithm I.

4 Result and Discussion

4.1 Experimental Setup

To be specific in the experiment, the proposed approaches in the process of detecting

objects in the images are simulated in MATLAB environment (R2016a) with Intel
Core i5 processor (CPU-2.20 GHz). For the proposed detection system, four different
types of images having different types of object (such as bird, aeroplane, ship and a
monkey playing xylophone) have been taken into account. The basic reason of opting
22 T. Sahoo and B. Mohanty

Fig. 4 Input images for the proposed system a humming bird image, b xylophone image,
c aeroplane image, d ship image

different images in different environment is to verify the efficiency of the proposed

algorithm. Since the images are color and of different size, it is first converted to
gray-scale image and resized to 512 × 512 to get equal-sized sub-blocks (32 × 32
in this paper) which gives 256 blocks. The number of decomposition level adapted
for wavelet decomposition is second level using haar filter. And the region growing
process used for the purpose is the mean distance method (adjacency and threshold)
on the seed point the target region is obtained.

4.2 Test Images

Figure 4a, b represents two test sequences having single objects such as humming
bird image having size 960 × 632 and a monkey playing the xylophone image having
size 320 × 240. In the xylophone image, monkey playing the xylophone is taken
as single object. The motive of choosing these images is that the background is
clear for which the objects are very specific. The results obtained using the proposed
algorithms are shown in Figs. 5 and 6.
Figure 4c, d represents two test sequences having single objects such as aeroplane
image having size 1000 × 655 and ship image having size 780 × 371. The motive
for choosing these images is to detect the moving objects, while the background is
complex (presence of sky and water which are moving along with the objects).

4.2.1 Experimental Results for Algorithm I

Figure 5 shows the results of Algorithm I for the test images humming bird, xylo-
phone, aeroplane and ship image. For each test image, WCF extraction of each feature
is applied to obtain the seed block, and using the seed block, the region growing is
applied to obtain the detected object. For each WCF in the result, two rows consisting
of seed block number and detected object after region growing process have been
presented to show from which seed block the result has been obtained.
In humming bird image, the features like energy, entropy, maximum probability
and inverse difference normalized show only the detected object image, whereas
Novel Region Growing Mechanism for Object Detection … 23

Algorithm-I
Original Test Images

WCF:Autocorrelation

107 231 29 235

WCF: Cluster Prominence

76 105 135 208

WCF: Cluster Shade

92 105 101 208

WCF: Contrast And Dissimilarity

152 123 138 238

Fig. 5 Results of proposed Algorithm I for test images; a humming bird image for each WCFs
(seed block and detected object after region growing process), b xylophone image for each WCFs
(seed block and detected object after region growing process), c aeroplane for each WCFs (seed
block and detected object after region growing process), d ship image for each WCFs (seed block
and detected object after region growing process)
24 T. Sahoo and B. Mohanty

WCF: Energy, Maximum probability and

Inverse Difference Normalized

100 1 24 1

WCF: Entropy

105 153 83 145

WCF: Sum of squares(Variance)

120 186 29 235

(a) (b) (c) (d)

Fig. 5 (continued)

in xylophone image, features (energy, autocorrelation, sum of variance, maximum

probability and inverse difference normalized) have shown detected object only, but
results of features (cluster shade and cluster prominence) have shown the outline of
the detected object. For both aeroplane and ship image, in all features along with the
detected object, the background is also present as the background is also complex.

4.2.2 Experimental Results for Algorithm II

Figure 6 presents the results of Algorithm II, for the test images humming bird,
xylophone, aeroplane and ship image. For each test image, dynamic extraction is
used along with WCF extraction for each feature has been applied to obtain the seed
block, and further using region growing, the detected objects are obtained. In figure
Novel Region Growing Mechanism for Object Detection … 25

Algorithm-II
Original Test Images

WCF:Autocorrelation

76 230 135 173

WCF: Cluster Prominence

91 89 135 173

WCF: Cluster Shade

45 89 73 66

WCF: Contrast And Dissimilarity

103 139 150 152

Fig. 6 Results of proposed Algorithm II for test images; a humming bird image for each WCFs
(seed block and detected object after region growing process), b xylophone image for each WCFs
(seed block and detected object after region growing process), c aeroplane for each WCFs (seed
block and detected object after region growing process), d ship image for each WCFs (seed block
and detected object after region growing process)
26 T. Sahoo and B. Mohanty

WCF: Energy, Maximum probability and

Inverse Difference Normalized

251 1 246 13

WCF: Entropy

77 229 106 107

WCF: Sum of squares(Variance)

169 230 119 169

(a) (b) (c) (d)

Fig. 6 (continued)

for each wavelet coefficient feature, two rows consisting of seed block result obtained
after extraction and detected object after region growing process have been presented
to show from which seed block the detected object result has been obtained.
In humming bird image, the features like contrast and dissimilarity show only
the detected object image, whereas in xylophone image, features (energy, autocor-
relation, sum of variance, cluster shade, cluster prominence, maximum probability
and inverse difference normalized) have shown detected object only, but results of
features (contrast and dissimilarity) have shown the outline of the detected object.
Due to complexity in the background, (aeroplane and ship image), in all features
along with the detected object, the background is also present. For entropy feature,
the object is also not detected for ship image.
Novel Region Growing Mechanism for Object Detection … 27

5 Conclusion

In this work, the wavelet coefficient features (WCF) are being extracted from the
transformed image to detect the object present in that image using two techniques.
First method calculates the co-occurrence matrix of each detailed subband is applied,
and then, they are combined together to provide the WCF of image. Second approach
is the calculation of co-occurrence matrix by creating the dynamics of image and
then calculating WCF. The algorithm works efficiently in the single object with and
without background complexity. Even in the presence of the cluttered background,
the target image is detected for the wavelet features like entropy, cluster shade and
cluster prominence. The entire detection methodologies depend upon selecting the
correct seed block and then growing the region entirely from the seed point computed
from the chosen block. Results of the experiments for different types of images
with varying attributes are presented to demonstrate the potential of the proposed
methodology.
The first method gives satisfactory results for the static image. However, for the
cluttered background, the second method utilizes the dynamics of the scene in the
wavelet domain for the purpose of detection of the object. Although, the result of
detection is acceptable at par, the proposed methodology in the presence of dynamics
is not robust. Hence, for such images with complex background, the wavelet fea-
tures of similar characteristics may be integrated together to find out a robust algo-
rithm. This, we propose, may be extended to multiple object scenario with complex
background.

References

1. Pulla Rao, C., Guruva Reddy, A., Rama Rao, C.B.: Target detection using multi resolution
analysis for camouflaged images. Int. J. Cybern. Inf. 5(4), 135–147 (2016)
2. Li, Z.Q., Fang, T., Huo, H.: A saliency model based on wavelet transform and visual attention.
Sci. China Inf. Sci. 53(4), 738–751 (2010)
3. Arivazhagan, S., Ganesan, L.: Automatic target detection using wavelet transform. EURASIP
J. Appl. Sig. Process. 2004(17), 2663–2674 (2004)
4. Sahoo, T., Mohanty, B.: A systematic review on visual attention and its application. Indian J.
Public Health Res. Dev. 9(11), 2278–2286 (2018)
5. Pan, Y., Chen, Y., Fu, Q., Zhang, P., Xu, X.: Study on the camouflaged target detection method
based on 3D convexity. Proc. Mod. Appl. Sci. 5, 152–157 (2011)
6. Tsapatsoulis, N., Rapantzikos, K.: Wavelet based estimation of saliency maps in visual attention
algorithms. In: LNCS, vol. 4132, pp. 538–547. Springer, Berlin (2006)
7. Haralick, R.M., Shanmugan, K., Dinstein, I.: Textural features for image classification. IEEE
Trans. Syst. Man Cybern. SMC-8(6), 610–621 (1973)
8. Chen, P.C., Pavlidis, T.: Segmentation by texture using correlation. IEEE Trans. Pattern Anal.
Mach. Intell. PAMI-5, 64–69 (1983)
9. Laws, K.I.: Textured image segmentation. Ph.D. dissertation, Rept. 940, Image Processing
Institute, University of Southern California (1980)
10. Unser, M.: Local linear transforms for texture measurements. Sig. Process. 11(1), 61–79 (1986)
28 T. Sahoo and B. Mohanty

11. Kashyap, R.L., Chellappa, R., Khotanzad, A.: Texture classification using features derived
from random field models. Pattern Recogn. Lett. 1, 43–50 (1982)
12. Unser, M.: Texture classification and segmentation using wavelet frames. IEEE Trans. Image
Process. 4(11), 1549–1560 (1995)
13. Changand, T., Kuo, C.-C.J.: Texture analysis and classification with tree-structured wavelet
transform. IEEE Trans. Image Process. 2(4), 429–441 (1993)
14. Arivazhagan, S., Ganesan, L.: Texture classification using wavelet transform. Pattern Recogn.
Lett. 24(9–10), 1513–1521 (2003)
15. Kubota, T., Huntsberger, T.L., Alford, C.O.: A vision system with real-time feature extractor
and relaxation network. Int. J. Pattern Recogn. Artif. Intell. 12(3), 335–354 (1998)
16. Huntsberger, T.L., Jawerth, B.D.: Wavelet based automatic target detection and recognition.
Annual Technical Report, University Research Initiative Program for Combat Readiness,
University of South Carolina, Columbia, SC, USA (1998)
17. Huntsberger, T.L., Jawerth, B.D.: Wavelet based algorithms for acoustic and non-acoustic
antisubmarine warfare. Annual Technical Report, University Research Initiative Program for
Combat Readiness, University of South Carolina, Columbia, SC, USA (1998)
18. Tian, Y., Qi, H., Wang, X.: Target detection and classification using seismic signal process-
ing in unattended ground sensor systems. In: Procedding IEEE International Conference on
Acoustics, Speech, Signal Processing (ICASSP’02), vol. 4, p. 4172, Orlando, FL, USA, May
2002
19. Boccignone, G., Chianese, A., Picariello, A.: Using Renyi’s information and wavelets for target
detection: an application to mammograms. Pattern Anal. Appl. 3(4), 303–313 (2000)
20. Espinal, F., Huntsberger, T.L., Jawerth, B.D., Kubota, T.: Wavelet-based fractal signature
analysis for automatic target recognition. Opt. Eng. 37(1), 166–174 (1998)
21. Sastry, C.S., Pujari, A.K., Deekshatulu, B.L., Bhagvati, C.: A Wavelet based Multiresolution
algorithm for rotation invariant feature extraction. Proc. Pattern Recogn. Lett. 25, 1845–1855
(2004)
22. Arivazhagan, S.: Automatic target detection using wavelet transform. EURASIP J. Appl. Sig.
Process. (2004)
23. Sahoo, T., Mohanty, B.: Moving object detection using background subtraction in wavelet
domain. In: 2nd International Conference on Data Science and Business Analytics (ICDSBA),
Sept 2018
A New Active Islanding Detection
Technique Using Superimposed Power
Angle Disturbance of IBDER

Harikrishna Muda and Premalata Jena

Abstract In this paper, a new active islanding detection technique (AIDT) based
on superimposed positive-sequence (p-s) power angle disturbance is proposed. The
injection of disturbance signal leads to a superimposed positive-sequence phase
angle (SPPA) deviation at the inverter-based distributed energy resource (IBDER)
terminal during islanding event. It is found that the variation in SPPA is capable
of detecting unintentional islanding conditions. Since the sequence elements of
voltage and currents during islanding event are different from that of a pre-islanding
event, superimposed elements are derived at the IBDER terminal. Furthermore, the
proposed technique is capable of identifying nonislanding events such as voltage sag,
three-phase faults, load switching, and capacitor switching. The proposed technique
along with the control schemes of IBDER units is developed using real-time digital
simulator (RTDS). The performance of the proposed technique is compared with the
conventional techniques. The detection time for the proposed method is within the
specified range in IEEE Std. 1547, and is around 20 ms.

Keywords Active islanding detection technique · Distributed energy resources ·

Phase angle · Superimposed positive-sequence elements

1 Introduction

The islanded mode of operation of inverter-based distributed energy resources

(IBDERs) is evolving to improve electricity reliability. It provides a way to supply
local loads from IBDERs when the main grid is disconnected [1]. However, the inte-
gration of IBDERs such as photovoltaic systems and fuel cell results in control and
protection issues such as unintentional islanding operation of IBDERs.

H. Muda (B)
National Institute of Advanced Studies, IISc Bangalore, Bangalore, India
e-mail: [email protected]
P. Jena
Indian Institute of Technology Roorkee, Roorkee, India

© Springer Nature Singapore Pte Ltd. 2020 29

G. Pradhan et al. (eds.), Advances in Electrical Control and Signal Systems,
Lecture Notes in Electrical Engineering 665,
https://doi.org/10.1007/978-981-15-5262-5_3
30 H. Muda and P. Jena

Failure to detect unintentional islanding operation of power system imposes

issues like interference with grid protection, out of phase reclosing, inverter control
switching mode, power-quality degradation, malfunction of protective devices, and
equipment damage [2]. During the islanding event, IBDERs continue to energize a
portion of the system which may cause threats to personal or crewman safety [3]. The
new grid standard codes governing the grid integration of IBDERs are defined such
that tripping of IBDER cannot exceed a clearing time of 0.15 s after the occurrence
of unintentional islanding condition to avoid any risk in the maintenance operation
as well as protecting the electrical system against potential damages [4].
A comprehensive survey of existing literature work on passive, active, and
communication-based islanding detection techniques can be found in [1–3]. An active
islanding detection technique (AIDT) creates a disturbance in the IBDER output by
disturbing signal parameters such as magnitude [5], phase [6], and frequency [7–10].
The dynamic equations for control scheme of voltage source inverter are provided
in [10]. The variations in the output signal of the IBDER terminal are measured to
determine an islanding condition in a system. Active detection techniques are more
effective and able to provide a small nondetection zone (NDZ) [11].
In [12], an impedance-based AIDT is presented to switch the inverter from
current-controlled mode to a voltage-controlled mode of operation by using a nonhar-
monic frequency signal injection. In [13], an AIDT based on estimation of negative-
sequence impedance by injecting small negative-sequence current in a unbalanced
system at different frequency levels is proposed. However, variations in power
system parameters are eminent due to the addition of external disturbances at IBDER
terminal [14]. The slip mode-frequency shift method has small NDZ between 59.3
and 60.5 Hz for greater then load quality factor (Qf ) of 1.5. In [9], an active frequency
drift technique is proposed for islanding detection.
This paper introduces a new AIDT based on the injection of superimposed
positive-sequence (p-s) power angle in which a superimposed p-s phase angle (SPPA)
is found to be varied during an islanding condition. To test the proposed method,
converter control schemes of IBDERs are developed using real-time digital simu-
lator (RTDS). The proposed technique ensures very low degradation to IBDER output
power quality and is considered negligible when the IBDERs are interconnected to
the grid.

2 Analysis of Superimposed Sequence Elements

Figure 1 depicts a distribution system with local three-phase unbalanced loads where
the islanding mode of operation can be implemented by opening the switch S. Since
the sequence elements of voltage and currents during islanding event are different
from that of a pre-islanding event, superimposed elements are derived at the IBDER
terminal. The voltage source inverter is required to push apparent power from the
A New Active Islanding Detection Technique Using Superimposed … 31

E G H
AC F S
Grid
DC
IBDER
Loads

Fig. 1 Single line diagram of a distribution network with IBDER and unbalanced loads

IBDER to the grid because it is designed to provide a constant output current. There-
fore, p-s elements for IBDER are implemented using Thévenin’s representation of a
dependent voltage source in the sequence diagram of the network [15].

2.1 p-s Elements at the IBDER Location for Pre-islanding

Condition

For the case of grid-connected mode, the p-s diagram is represented in Fig. 2 [16].
The p-s equivalent representation for a multiple inverters is referred from [17]. Next,
IBDER is represented by an equivalent current-controlled voltage dependent source
in the sequence network. The power produced by IBDER is dependent on the phase
current reference. The pre-islanding p-s current (I1E) flowing through the IBDER
terminal is obtained using p-s equivalent circuit depicted in Fig. 2a. The voltage
Eqs. (1) and (2) by employing Kirchhoff’s voltage law to the left-hand and right-hand
side loops, respectively.
 
VE − I E Z 1 p + Z 1L + I H Z 1L = 0 (1)

 
−I H Z 1q + Z 1L − VH + I E Z 1L = 0 (2)

Here, Z 1q = Z 1H + Z 1G H and Z 1 p = Z 1E + Z 1E G . Z 1L is the p-s load impedance.

Z 1EG and Z 1GH are the p-s line impedances [18]. V E and V H are the voltages at
IBDER and grid terminals, respectively. The subscript 1 in the text refers to the p-s

E G H
Z1E Z1EG Z1GH Z1H

IH VH
V1E IE

Fig. 2 p-s equivalent circuit diagram of Fig. 1. a For pre-islanding event. b For post-islanding event
32 H. Muda and P. Jena

element. Z 1E and Z 1H are the p-s source impedances. The voltage Eqs. (1) and (2)
can be solved to obtain the loop currents (I E and I H ). Further, the I 1E flowing through
the relay at Bus-E can be written as

Z 1q + Z 1L Z 1L
I1E = I E = VE − VH (3)
Z 1 pq Z 1 pq

where Z 1 pq = (Z 1 p + Z 1L )(Z 1q + Z 1L ) − Z 1L Z 1L . It should be noted that Z 1p is

negligible in comparison to Z 1q since the main grid does not exchange power to
the load and IBDER during the grid-connected mode of operation. Thus, (3) can be
rewritten as
kq + 1 1
I1E = VE − VH (4)
Z 1q Z 1q

where kq = Z 1q Z 1L . By applying KVL to the p-s diagram shown in Fig. 2, the
pre-islanding p-s voltage (V 1E ) is obtained as
 
V1E = Z 1E G + Z 1L ||Z 1q I1E (5)

By substituting the value of I1E from (4) in (5), V 1E is expressed as

 
V1E = kd kq + 1 VE − kd VH (6)
 
where kd = Z 1E G + Z 1L ||Z 1q Z 1q .

2.2 p-s Elements at the IBDER Location for Post-islanding

Condition

The grid is disconnected with the help of the switch S as shown in Fig. 1. The
corresponding p-s circuit diagram is depicted in Fig. 2b during islanding condition.
Similarly, applying KVL to the left loop gives, the following equation:

VE − I E Z 1m = 0 (7)

where Z 1m = Z 1 p + Z 1L . Here, the p-s voltage and currents are affected by Z 1p , after

the occurrence of islanding condition. Furthermore, the p-s current (I1E ) is deduced
using (7) as

 1
I1E = IE = VE (8)
Z 1m

The p-s voltage (V1E ) at Bus-E in Fig. 2b is expressed as
A New Active Islanding Detection Technique Using Superimposed … 33

 
V1E = (Z 1E G + Z 1L )I1E (9)

 
By replacing I1E from (8) in (9), V1E can be expressed as

V1E = kd VE (10)

+Z 1L
where kd = Z 1EZG 1m . For an islanding event, the deviation between the post and
the pre-islanding elements is defined as superimposed component [19]. Using p-s
pre-islanding and post-islanding voltage phasors, the mathematical expression for
the superimposed p-s voltage is provided as [13]

V1E = V1E − V1E (11)

Similarly, the pre and post-islanding p-s current phasors are used to calculate the
superimposed p-s current phasor (I 1E ). Mathematically, I 1E is given by

I1E = I1E − I1E (12)

In order to detect an islanding condition in a system, the phase angle variations

obtained by superimposed p-s elements are used. Equations (5) and (9) show that the

phasor positionof V1E is leading the
 pre-islanding p-s voltage (V 1E ), as the equivalent
p-s impedance Z 1E G + Z 1L ||Z 1q is less than (Z 1E G + Z 1L ). Thus, the SPPA, φ 1E ,
between V 1E and I 1E is found to be positive during islanding event. It should
be noted that the proposed technique uses superimposed p-s power angle of IBDER
as a feedback signal to change phase current references. Since IBDER produces a
constant current in grid-connected mode, the change in the superimposed p-s power
angle is negligible. Therefore, the injected disturbances are insignificant during a
nonislanding case in the network to avoid maloperation of islanding condition. This
proposed active technique based on superimposed p-s power angle disturbance is
introduced in the next section.

3 Proposed Active Islanding Detection Technique

The p-s current (I 1E ) is injected from IBDER to the grid varies according to the
distribution network operating conditions to meet the required p-s apparent power
phasor (S 1E ). Thus, S 1E is a nonlinear function of either p-s voltage or p-s current
which can be written as follows:
∗
S1E = V1E I1E (13)

∗
where I1E is the complex conjugate p-s current. The following equation can be
obtained by substituting (4) in (13).
34 H. Muda and P. Jena

kq + 1 2 1
S1E = V1E − V1E V1H (14)
Z 1q Z 1q

Let us consider σ is the power angle of the IBDER. P1E and Q1E are assumed to
be the p-s active and reactive powers, respectively, which are extracted from the p-s
apparent power, S 1E . By applying the power triangle trigonometric identity to (14),
P1E and Q1E can be written as

1
P1E = V1E V1H sin σ (15)
Z 1q
1 kq + 1 2
Q 1E = V1E V1H cos σ − V (16)
Z 1q Z 1q 1E

The load p-s apparent power (S1L ) measured at Bus-E during islanding mode of
operation is represented as
 ∗
S1L = V1E I1E (17)


Substituting the value of I1E from (8) in (17), S1L can be rewritten as

1 2
S1L = V (18)
Z 1m E

where Z 1m = R1m + j X 1m . Then, Eq. (18) becomes

R1m 2
P1L = V (19)
|Z 1m | E
X 1m 2
Q 1L = V (20)
|Z 1m | E

where P1L is the load p-s active power which depends on the p-s resistance, R1m . Q1L
depends on the load’s p-s reactance, X 1L , and is called the load p-s reactive power.
The power supplied by the IBDER terminal is given by

P1L = P1E + P1 (21)

Q 1L = Q 1E + Q 1 (22)

where P1 and Q1 are equal to the differences in p-s active and reactive powers
within the load and IBDER that shared with the utility, respectively. Substituting (15)
and (19) in (21), one can obtain

R1m 2 1
V = V1E V1H sin σ + P1 (23)
|Z 1m | E Z 1q
A New Active Islanding Detection Technique Using Superimposed … 35

Similarly, using (16) and (20), Eq. (22) can be rewritten as

X 1m 2 1 kq + 1 2
V = V1E V1H cos σ − V + Q 1 (24)
|Z 1m | E Z 1q Z 1q 1E

Based on (23) and (24), the p-s power angle variation can be expressed by
 
Z 1q R1m V1E2
− P1 |Z 1m |
tan σ =   (25)
|Z 1m | k1m V1E
2
− Q 1 Z 1q

where k1m = 1m 1q |Z 1m1m| ( q ) .

X Z +|Z | k +1

Since the utility does not exchange the required P1 and Q1 with the island after
islanding occurrence, the variation in σ is very small. In the proposed method, the
variation of σ is taken as a feedback. It is observed there are no significant changes
in σ for the islanded mode. After the islanding occurrence, the variation of σ can
change the references of direct and quadrature axis currents (I d and I q ) with a suitable
feedback. The proposed injection method is based on (26) and (27) which are shown
below.
V1E
Id_ dis = sin σ = K sin σ (26)
Z 1q
V1E
Iq_ dis = cos σ = K cos σ (27)
Z 1q

Figure 3 shows two additional blocks that detect variations of cos σ and sin σ as
an error. The gain turns the variation of cosσ and sinσ to a current variation and is
equal to the maximum value of (26) or (27). A low pass filter (LPF) is introduced in
AIDT to filter out the distortions of power angle. The output signal is added to the
characteristics of the d- and q-axis. After the occurrence of islanding situation, the
change in σ causes a suitable variation in the SPPA during islanding condition.

(a) (b)

Fig. 3 Modified current controller of the proposed active islanding detection technique
36 H. Muda and P. Jena

4 Control and Simulation of IBDER in Real Time

The required steps for the proposed technique implementation are provided in this
section. The power system signals at the IBDER terminal are stored at a sampling
frequency of 1.0 kHz. Phasor values of the voltage and current signals are esti-
mated using a one-cycle-weighted least square-based phasor estimation technique
as described in [19]. The phasors are transformed into sequence elements using
symmetrical analysis. It should be noted that ten cycles of the pre-event voltages
and currents are stored in the memory [20]. p-s elements of power signals are calcu-
lated to obtain the p-s elements of powers for further process. The power angle σ is
computed using p-s active and reactive powers. Equations (15) and (16) show that
Z 1q can control p-s active and reactive powers, respectively. The IBDER output reac-
tive power is compared with a reference to the reactive power (Qref ). In this paper,
Qref is set to 0.365 MVAr. To demonstrate the effectiveness of the proposed AIDT,
a 4.16 kV 50 Hz distribution network with IBDER, shown in Fig. 1, is considered
[15] (Tables 1 and 2).

Table 1 Specifications of a
Parameter Value
single voltage source inverter
[19] Rated voltage 230 V
Rated current 125 A
Filter inductance 0.52 mH
Filter resistance 1.6 m
Filter inductance 137.83 μF
Switching frequency 2.0 kHz
System frequency 50 Hz
Transformer rating 0.23/4.16 kV

Table 2 Specification of
Parameter Value
single S230 MBT Tata Power
Solar panel at standard test Peak power (Pmax ) 230 W
conditions (25 °C and Max. voltage (V mpp ) 29.1 V
1000 W/m2 ) [19]
Max. current (I mpp ) 7.9 A
Open circuit voltage (V oc ) 36.7 V
Short-circuit current (I sc ) 8.4 A
Max. series fuse 20 A
A New Active Islanding Detection Technique Using Superimposed … 37

5 Simulation Results

5.1 Zero Power Mismatch Islanding Condition

The proposed islanding detection technique is tested on the IBDER under perfect
power match situation. The islanding event (by opening the switch, S, in Fig. 1)
and different nonislanding events are initiated at t = 1.0 s in the system. The load
is connected at bus-G as shown in Fig. 1. The parameters of the VSI and the PV
module (Tata power TS230 MBT) are provided in Appendix A. In this case of study,
the IBDER produces 2 MW and 0.365 MVAr, which is almost the same as the load
demand of 1.989 MW and 0.354 MVAr. Figure 4a, b provides the corresponding
voltage and current signals measured at the IBDER terminal. The perfect power

10 a
Voltage (kV)

b
0 c

-10
0.95 1 1.05 1.1 1.15
(a) Time (s)
0.5
Current (kA)

-0.5
0.95 1 1.05 1.1 1.15
(b) Time (s)
SPPA (rad.)

0.2

-0.2
0.95 1 1.05 1.1 1.15
(c) Time (s)
1
Trip

0.5

0
0.95 1 1.05 1.1 1.15
(d) Time (s)

Fig. 4 Responses of the proposed technique at the IBDER terminal for zero power mismatch
condition. a Three-phase voltages. b Three-phase currents. c Superimposed p-s phase angle. d Trip
signal
38 H. Muda and P. Jena

match islanding operation is conducted at t = 1.0 s. The SPPA for the case of
islanding is computed using the proposed technique and depicted in Fig. 4c. The
SPPA plot during an islanding condition increased in a positive rate and crossed the
predefined threshold of 0.2 rad at t = 1.016 s. Thus, a trip signal is generated within
20 ms, as shown in Fig. 4d, to disconnect the IBDER operation under zero power
mismatch islanding condition. It can be observed that the detection time is less than
20 ms.

5.2 Impact of Voltage Sag Condition

One of the power-quality issues is voltage sag which influences the current level
flowing through the distribution lines. To avoid false detection of islanding condition,
the IBDER units must be equipped with a reliable islanding detection scheme that
distinguishes islanding from nonislanding conditions. With the arrangement of the
voltage source regulator at Bus-H, the phase voltage is reduced by 10% between the
time period of t = 1.0 and 1.2 s. Voltages and currents at the terminal of IBDER
are varied as shown in Fig. 5a, b, respectively. Thus, SPPA element at the IBDER
terminal is varied under such a situation. Results for SPPA for such condition are
shown in Fig. 5c. It is found that the magnitude of SPPA is less than the threshold
value and consistent under voltage sag condition.

5
Voltage (kV)

a
b
0
c
-5
0.95 1 1.05 1.1 1.15 1.2 1.25 1.3 1.35
(a) Time (s)
1
Current (kA)

-1
0.95 1 1.05 1.1 1.15 1.2 1.25 1.3 1.35
(b) Time (s)
SPPA (rad.)

0.2
0
-0.2
0.95 1 1.05 1.1 1.15 1.2 1.25 1.3 1.35
(c) Time (s)

Fig. 5 Responses of the proposed technique at the IBDER terminal for voltage sag condition.
a Three-phase voltages. b Three-phase currents. c Superimposed p-s phase angle
A New Active Islanding Detection Technique Using Superimposed … 39

Voltage (kV) 5 a
b
0 c

-5
0.95 1 1.05 1.1 1.15
(a) Time (s)
Current (kA)

1
0
-1
0.95 1 1.05 1.1 1.15
(b) Time (s)
SPPA (rad.)

0.2
0
-0.2
0.95 1 1.05 1.1 1.15
(c) Time (s)

Fig. 6 Responses of the proposed technique at the IBDER terminal for abc fault. a Three-phase
voltages. b Three-phase currents. c Superimposed p-s phase angle

5.3 Results for Three-Phase Fault Conditions

The accuracy is depending on discrimination between nonislanding and islanding

situations. To test the performance of the proposed AIDT, a three-phase fault is
simulated at location F as shown in Fig. 1. The impact of the nonislanding situation
is shown in Fig. 6. It is noticed from the responses of Fig. 6a that the voltage reduces
significantly during fault condition. However, the SPPA curve as shown in Fig. 6c
did not cross the predefined threshold of 0.2 rad and affirms a nonislanding situation.

6 Conclusion

This paper demonstrates a novel active islanding detection technique based on super-
imposed p-s power angle disturbance in IBDERs. “Discrimination between islanding
and nonislanding conditions has been assessed utilising the most significant changes
obtained from superimposed p-s power angle. It is found that the proposed scheme
work effectively under a wide range of real-time simulation events.” The super-
imposed p-s power angle is stable during normal operating condition and hence,
the phase angle between superimposed p-s voltage and currents is negligible during
40 H. Muda and P. Jena

nonislanding situations. Furthermore, the proposed technique is shown to have negli-

gible effect on power quality during normal operation. The time of response is within
the specified range of IEEE Std. 1547, and is around 20 ms.

References

1. Ku Ahmad, K.N.E., Selvaraj, J., Rahim, N.A.: A review of the islanding detection methods in
grid-connected PV inverters. Renew. Sustain. Energy Rev. 21, 756–766 (2013)
2. Trujillo, L.C., Velasco, D., Figueres, E., Garcerá, G.: Analysis of active islanding detection
methods for grid-connected microinverters for renewable energy processing. Appl. Energy 87,
3591–3605 (2010)
3. Li, C., Cao, C., Cao, Y., Kuang, Y., Zeng, L., Fang, B.: A review of islanding detection methods
for microgrid. Renew. Sustain. Energy Rev. 35, 211–220 (2014)
4. IEEE Standards Coordinating Committee 21: IEEE Guide for Design, Operation, and Inte-
gration of Distributed Resource Island Systems with Electric Power Systems. IEEE Std
1547.4-2011, pp. 1–54 (2011)
5. Yu, B., Matsui, M., So, J., Yu, G.: A high power quality anti-islanding method using effective
power variation. Sol. Energy 82(4), 368–378 (2008)
6. Hung, G.K., Chang, C.C., Chen, C.L.: Automatic phase-shift method for islanding detection
of grid-connected photovoltaic inverters. IEEE Trans. Energy Convers. 18(1), 169–173 (2003)
7. Zeineldin, H.H., Salama, M.M.A.: Impact of load frequency dependence on the NDZ and
performance of the SFS islanding detection method. Ind. Electron. IEEE Trans. 58(1), 139–146
(2011)
8. Ropp, E.M., Begovic, M., Rohatgi, A.: Analysis and performance assessment of the active
frequency drift method of islanding prevention. IEEE Trans. Energy Convers. 14 (1999)
9. Lopes, L.A.C., Sun, H.: Performance assessment of active frequency drifting islanding
detection methods. IEEE Trans. Energy Convers. 21(1), 171–180 (2006)
10. Al Hosani, M., Qu, Z., Zeineldin, H.H.: Scheduled perturbation to reduce nondetection zone
for low gain sandia frequency shift method. IEEE Trans. Smart Grid 6(6), 3095–3103 (2015)
11. Zhang, J., Xu, D., Shen, G., Zhu, Y., He, N., Ma, J.: An improved islanding detection method
for a grid-connected inverter with intermittent bilateral reactive power variation. IEEE Trans.
Power Electron. 28, 268–278 (2013)
12. Asiminoaei, L., Teodorescu, R., Blaabjerg, F., Borup, U.: A digital controlled PV-inverter with
grid impedance estimation for ENS detection. IEEE Trans. Power Electron. 20(6), 1480–1490
(2005)
13. Wrinch, M., Martí, J., Nagpal, M.: Negative sequence impedance based islanding detection
for distributed generation (NSIID). In: 2008 IEEE Electrical Power and Energy Conference—
Energy Innovation, pp. 1–6 (2002)
14. Ye, Z., Kolwalkar, A., Zhang, Y., Du, P., Walling, R.: Evaluation of anti-islanding schemes
based on nondetection zone concept. IEEE Trans. Power Electron. 19(5), 1171–1176 (2004)
15. Wieserman, L., Mcdermott, T.E.: Fault current and overvoltage calculations for inverter-
based generation using symmetrical components. In: IEEE Energy Conversion Congress and
Exposition, pp. 2619–2624 (2014)
16. Blackburn, J.L.: Symmetrical Components for Power Systems Engineering. CRC Press (1993)
17. Liu, N., Aljankaway, A., Diduch, C., Chang, L., Su, J.: Passive islanding detection approach
based on tracking the frequency dependent impedance change. In: IEEE International
Symposium on Power Electrical for DG Systems (PEDG), pp. 364–367 (2012)
18. IEEE SA—1547.6-2011—IEEE Recommended practice for interconnecting distributed
resources with electric power systems distribution secondary networks (2014)
A New Active Islanding Detection Technique Using Superimposed … 41

19. Muda, H., Jena, P.: Rate of change of superimposed negative sequence impedance based
islanding detection technique for distributed generations. IET Gener. Transm. Distrib. 10(13),
3170–3182 (2016)
20. Garmrudi, M., Nafisi, H., Fereidouni, A., Hashemi, H.: A novel hybrid islanding detection
technique using rate of voltage change and capacitor tap switching. Electr. Power Compon.
Syst. 40(10), 1149–1159 (2012)
MPPT-Based Inverter Control
of Grid-Connected PV–Wind Hybrid
Power System

A. V. Pavan Kumar and Y. Sai Varun

Abstract This study presents a two-stage grid-tied three-phase inverter control

topology capable of performing maximum power point tracking (MPPT) and power
flow control. This topology is derived from the single-stage grid-tied system. The
grid-tied inverter requires a minimum value of DC input voltage for grid synchroniza-
tion. So, the amount of power available at low or minimal environmental conditions
cannot be utilized by single-stage topology. A two-stage grid-tied topology with PV–
wind-based generation, MPPT-controlled boost converter, and three-phase inverter
are implemented in MATLAB, Simulink, and the control logic is tested under vary-
ing environmental conditions. The simulation results validate the effectiveness of the
topology.

Keywords Hybrid power system · PMSG · Wind power · MPPT-based inverter ·

Grid tie · Load frequency control

1 Introduction

Photovoltaic (PV) and wind-based generation have experienced tremendous develop-

ment in a recent decade, mainly because of increasing concern for climatic changes
and oil prices, which has driven numerous nations to review new technologies to
advance renewable energy sources (RES) technology [1].
The RES power extracted beginning uncertain as the power is directly related to
change in climatic conditions, thereby the reliable and uninterrupted power supply
cannot be extracted from one type RES. However, this can be achieved by the com-
bination of two or more RES selected on the basis of geographical conditions to
from hybrid power generation. Solar PV and wind-based power generation are the

A. V. Pavan Kumar (B) · Y. Sai Varun

Department of EEE, Madanapalle Institute of Technology & Science, Madanapalle,
AP 517325, India
e-mail: [email protected]
Y. Sai Varun
e-mail: [email protected]

© Springer Nature Singapore Pte Ltd. 2020 43

G. Pradhan et al. (eds.), Advances in Electrical Control and Signal Systems,
Lecture Notes in Electrical Engineering 665,
https://doi.org/10.1007/978-981-15-5262-5_4
44 A. V. Pavan Kumar and Y. Sai Varun

most reliable, promising sources as the solar and wind energy resources are available
daily and seasonally. The hybrid PV–wind generation is gaining importance, capable
sources as compared to conventional generation [2]. The power electronic technology
plays a vital role in grid-connected distribution generation. During the last few years,
semiconductor technology has gone tremendous growth with respect to fast oper-
ating switches and high power handling capability. The real-time computer-based
controller technology which can implement complex control logic in coordination
with the power electronic technology has led to the development of efficient and
grid responsive converters [3]. Two different topologies of grid-connected RES are
available in the literature.
(1) Single-stage topology: where the output from RES generation is directly
connected to the grid with the help of voltage source inverter (VSI).
(2) Two-stage topology: Where the output from the RES is converted into a constant
DC voltage and then synchronized to the grid with the help of VSI.
A single-stage grid-integrated topology has more advantages, reliability, and
enhanced performance overt two-stage topology for a single-source PV generation
system [4–6], whereas a two-stage topology is advantageous for wind power gen-
eration and PV generation under low or varying climatic conditions. The two-stage
topology provides better frequency and power flow control as the frequency of the
system is independent of wind speed and is totally controlled by the inverter [7–9].
A hybrid system is formed by combining two RES generations, i.e., PV and wind
generation. The combined power generated from both the RES to be synchronized
and injected into the grid. A combined two-stage topology has more advantages and
fewer components over the two different single-stage grid-connected systems [10].
The number of voltage and current sensors required is reduced as only one VSI is
required for grid synchronization and power injection. In two-stage topology, first is
being a constant DC stage. The DC voltage at this stage is sustained constant never-
theless of fluctuations in climatic conditions, which can be used for supplying local
loads such as battery charging. In this paper, a two-stage grid-connected PV–wind
hybrid power system with an improved inverter control as shown in Fig. 1.
A maximum power point tracking (MPPT)-based inverter control is implemented
in the centralized controller as shown in Fig. 1 to enhance the maximum power point
(MPP) tracking and injecting maximum power harnessed into the grid.
A 300 kW PV, 300 kW wind-based generation is implemented in the MATLAB,
Simulink. The output voltage from RES generation is connected to a boost converter
forming stage-1. The purpose of the stage-1 is to maintain a constant DC voltage irre-
spective of change in environmental conditions and can be used to harness power at
lower solar illumination and wind speeds. The constant DC from stage-1 is connected
as an input to voltage source inverter (VSI) which forms stage-2. The stage-2 has
to perform DC-to-AC conversion, grid synchronizing operation, and power injec-
tion into the grid. The detailed modeling of the hybrid system and control logic
implementation will be discussed in detail in the following sections.
MPPT-Based Inverter Control of Grid-Connected PV–Wind Hybrid … 45

Fig. 1 Proposed two-stage topology

2 Modeling of Stage-1

Stage-1 consists of PV, wind generation, a boost converter, MPPT algorithm as shown
in Fig. 2. The MATLAB, Simulink implementation and mathematical modeling of
PV generation are taken from [11–13], and the equation-based modeling of wind
generation consisting of a wind turbine, permanent magnet synchronous generator
(PMSG), diode rectifier, and MPPT-controlled boost converter is taken from [14,
15].

Fig. 2 Proposed two-stage topology

46 A. V. Pavan Kumar and Y. Sai Varun

Fig. 3 I-V and P-V Characteristics of PV generation

Table 1 Electrical
Maximum Power (W) = 414 Cells per module = 128
parameters of PV panel
Open Circuit Voltage V _oc Short Circuit Current I_sc
(V) = 85.3 (A) = 6.09
Voltage at MPP V _mp (V) = Current at MPP I_mp (A) =
72.9 5.69

The Current (I)-Voltage (V) and Power (P)-Voltage (V) characteristics of PV

generation at different solar illumination levels at 25 °C operating temperature are
graphically represented in Fig. 3. A Sun Power SPR-415e-WHT-D module datasheet
is utilized to simulate one PV panel. Such modules are connected in series and
parallel combinations to achieve desired output power. The electrical parameters of
one module are given in Table 1. The simulated power characteristics of the wind
turbine at wind speeds are shown in Fig. 4. Modeling of the boost converter is taken
from [16].

3 Modeling of Stage-2

Stage-2 consists of voltage source inverter (VSI), inverter, measurement, grid syn-
chronizing control logic, and PWM generator. The graphical representation of pro-
posed MPPT-based inverter control technique is shown in Fig. 5. The inverter has
two different control strategies: A voltage regulator and current regulatory control.
The control technique is implemented in the d-q reference frame. The measurement
plays an important role in the development of control logic of grid-connected inverter
as shown in Fig. 6. A PLL is used to generate reference ωt control signal for grid
synchronization.
MPPT-Based Inverter Control of Grid-Connected PV–Wind Hybrid … 47

Fig. 4 Simulated power characteristics of wind turbine

Fig. 5 Block diagram of MPPT based inverter control technique

Fig. 6 Measurement and

conversion technique
48 A. V. Pavan Kumar and Y. Sai Varun
⎡ ⎤ ⎡ ⎤⎡ ⎤ ⎡ ⎤⎡ ⎤ ⎡ ⎤
Vi_RY R −R 0 IR L −L 0 P IR Vs_RY
⎣ Vi_Y B ⎦ = ⎣ 0 R −R ⎦⎣ IY ⎦ + ⎣ 0 L −L ⎦⎣ P IY ⎦ + ⎣ Vs_Y B ⎦
Vi_B R −R 0 R IB −L 0 L P IB Vs_B R
(1)
⎡ ⎤ ⎡ ⎤⎡ ⎤
I RY 1 −1 0 IR
⎣ IY B ⎦ = ⎣ 0 1 −1 ⎦⎣ IY ⎦ (2)
IB R −1 0 1 IB

In grid synchronized mode, the inverter acts as a voltage-controlled current source.

The power flow control of the inverter is achieved by controlling V d_meas and V q_meas
under synchronous reference frame. V i , V s represents the inverter voltage and grid
voltage, respectively. The power control can be mathematically derived, where P in
Eq. (1) is time derivative operator. The line to line can be expressed as Eq. (2) [17].
Substituting (2) in (1), we have
⎡ ⎤ ⎡ ⎤ ⎡ ⎤ ⎡ ⎤
Vi_RY I RY P I RY Vs_RY
⎣ Vi_Y B ⎦ = Ri ⎣ IY B ⎦ + Li ⎣ P IY B ⎦ + ⎣ Vs_Y B ⎦ (3)
Vi_B R IB R P IB R Vs_B R

where i is the identity matrix and Eq. (3) can be referred to in synchronous reference
frame as Eq. (4).
      
Vid R + L −ωL i d_ref Vsd
= + (4)
Viq ωL R + L i q_ref Vsq

The angular frequency of the grid is ω in Eq. (4). The active and reactive power
injection into the grid can be controlled by obtaining id_ref and iq_ref . The values
obtained can be submitted in Eq. (4) to find V id , V iq which can be implemented
using PWM techniques. In order to have a robust and accurate control, PI controller
is implemented to obtain the V *id , V *iq based on the error between the desired and
computed active and reactive power. The block diagram realization of mathematical
analysis of current regulator, active and reactive power flow control technique is as
shown in Fig. 7. The values of R = 0.2100 , L = 0.0039 H and PI controller gains
for active and reactive power control are K p = 0.30, Ki = 20. The block diagram
representation of the V dc regulator and voltage reference computation are shown in
Figs. 8 and 9, respectively.
The duty cycle computed by the MPPT algorithm is utilized to control the duty
cycle of the boost converter, and the same is utilized to compute Idc_ref as shown
in Fig. 8. The I dc_ref computed is utilized in the modeling of the current regulator,
active, and reactive power flow control technique as shown in Fig. 7. The control
signal computed from current regulator circuit is then utilized to generate voltage
reference signal which is utilized for grid synchronization, and the current regulator is
MPPT-Based Inverter Control of Grid-Connected PV–Wind Hybrid … 49

Fig. 7 Current regulator with active and reactive power flow control

Fig. 8 Block diagram

illustration of V dc regulator

utilized for current injection into the grid [17, 18]. The digital simulation of the grid-
connected PV–wind hybrid power system and simulation results will be discussed
in the next section.
50 A. V. Pavan Kumar and Y. Sai Varun

Fig. 9 Block diagram illustration of voltage reference computation

4 System Description and Simulation Results

A 600 kW hybrid system with 300 kW each generation is realized in MATLAB,

Simulink. The DC output from the RES generation is boosted up and maintained
constant at desired voltage using MPPT algorithm under varying solar irradiance
and wind speed. The real-time data of environmental conditions are captured at the
location [19, 20]. The voltage source inverter converts DC to AC with the help of
the proposed MPPT-based control technique, and grid synchronization is achieved.
The MPP tracking of PV and wind generation with the instantaneous data of envi-
ronmental conditions is shown in Figs. 10 and 11. The amount of power injected into
the grid by respective RES is plotted in the bottom plots in Figs. 10 and 11 and the
top plot being the real-time data. The DC bus-bar voltage is graphically represented
in Fig. 12; it is comprehended that the V dc measured tracks the V dc reference.

Fig. 10 PV generation with real-time data of solar illumination

MPPT-Based Inverter Control of Grid-Connected PV–Wind Hybrid … 51

Fig. 11 Wind generation with real-time data of wind speed

Fig. 12 DC bus-bar voltage

From Figs. 10, 11 and 12, it can be concluded that the MPPT algorithm operation is
as desired by tracking maximum power from PV and wind generation under varying
environmental conditions. The computation of direct axis current id and quadrature
axis current Iq play a vital role in power flow control of the inverter. The computed
values are compared with the reference values and are graphically represented in
Fig. 13.
It can be clearly comprehended that the measured values duplicate the reference
values. The initial portion of id current is magnified and shown in Fig. 13. The inverter
output voltage is shown in Fig. 14, and the per phase voltage measured at the 25 kV
AC bus-bar is graphically represented in Fig. 15; the load current is plotted in Fig. 16.
The ωt measured from the grid; the ωt generated for grid synchronization; and
voltage reference computation are shown in Fig. 17.
The computation of ωt plays a vital role in grid synchronization inverter. If the ωt
is not computed in the phase of ωt measured, then the generated inverter voltage and
frequency will not match that of the grid; the inverter will not be able to synchronize
to the grid. From Fig. 17, it can be observed that the measured and computed ωt
for grid synchronizing inverter are in phase and it can be comprehended from the
simulation results that the proposed controller was able to synchronize to the grid and
the power generated from the RES is injected into the grid. The total power injected
into the grid is graphically represented in Fig. 18.
52 A. V. Pavan Kumar and Y. Sai Varun

Fig. 13 Id current for inverter control

Fig. 14 Simulated inverter output voltage

Fig. 15 Simulated load voltage

MPPT-Based Inverter Control of Grid-Connected PV–Wind Hybrid … 53

Fig. 16 Simulated load current

Fig. 17 ωt measured and computed

Fig. 18 Total power injected into the grid

5 Conclusions

In this paper, an MPPT-based inverter control technique for grid-connected PV–

wind hybrid power system is proposed. Mathematical modeling of 300 kW each
PV and wind generation system is presented and detail analysis of the proposed
54 A. V. Pavan Kumar and Y. Sai Varun

controlled technique is represented. The real-time data of solar illumination and

wind speed is utilized to emphasize the control action of the controller under varying
environmental conditions. The model-based design of the system is implemented in
MATLAB, Simulink.
The simulation results demonstrate enhanced control of proposed controller under
varying environmental condition by maintaining DC bus-bar voltage at desired level.
The power generated by the PV and the wind under low solar illumination, and wind
speed respectively is fed into the grid by maintaining the required output voltage,
current, and frequency of the system.
The proposed control implementation was able to maintain the system in syn-
chronism with the grid under varying environmental conditions and injecting the
maximum power extracted from the RES generation into the grid. The future work
of the proposed system will be hardware implementation and testing of the system
under with practical grid and analyzing the performance of the controller under peak
load and for contingency conditions.

Acknowledgements I would like to thank the management of the Madanapalle Institute of tech-
nology and Science, Madanapalle, for encouraging UG students to carry research work through
SHARP program. This work was carried out with the UG student under SHARP program.

References

1. Kouro, S., Leon, J.I., Vinnikov, D., Franquelo, L.G.: Grid-connected photovoltaic systems: an
overview of recent research and emerging PV converter technology. IEEE Ind. Electron. Mag.
9(1), 47–61 (2015)
2. Mills-Price, M., et al.: Interconnection control of distributed generation with timesynchronized
phasors. In: 2011 IEEE/PES Power Systems Conference and Exposition, Phoenix, AZ, pp. 1–8
(2011)
3. Carrasco, J.M., et al.: Power-electronic systems for the grid integration of renewable energy
sources: a survey. IEEE Trans. Ind. Electron. 53(4), 1002–1016 (2006)
4. Dousoky, G.M., Ahmed, E.M., Shoyama, M.: MPPT schemes for single-stage three-phase
grid-connected photovoltaic voltage-source inverters. In: IEEE International Conference on
Industrial Technology (ICIT), Cape Town, pp. 600–605 (2013)
5. Bae, Y., Kim, R.Y.: Suppression of common-mode voltage using a multicentral photovoltaic
inverter topology with synchronized PWM. IEEE Trans. Ind. Electron. 61(9), 4722–4733
(2014)
6. Chen, H., Yang, Y.: Control scheme of grid-connected inverter based on virtual circuit. In:
Proceeding of the 11th World Congress on Intelligent Control and Automation, Shenyang,
pp. 5394–5398 (2014)
7. Manojkumar, M., Porkumaran, K., Kathirvel, C.: Power electronics interface for hybrid
renewable energy system—a survey. In: International Conference on Green Computing
Communication and Electrical Engineering (ICGCCEE), Coimbatore, pp. 1–9 (2014)
8. Wandhare, R.G., Agarwal, V.: Novel integration of a PV-wind energy system with enhanced
efficiency. IEEE Trans. Power Electron. 30(7), 3638–3649 (2015)
9. Soler-Bientz, R., Ricalde-Cab, L., Pérez, L.F.B., Baeza, J.G.C.: Study of a PV-wind system
in tropical conditions. In: 2011 37th IEEE Photovoltaic Specialists Conference, Seattle, WA,
pp. 001872–001875 (2011)
MPPT-Based Inverter Control of Grid-Connected PV–Wind Hybrid … 55

10. Sharma, R., Sathans: Survey on hybrid (Wind/Solar) renewable energy system and associated
control issues. In: 2014 IEEE 6th India International Conference on Power Electronics (IICPE),
Kurukshetra, pp. 1–6 (2014)
11. Pavan Kumar, A.V., Parimi, A.M., Uma Rao, K.: Performance analysis of FLC controlled
PV-wind hybrid power system for dc load with real-time data in Matlab, Simulink. J. Electr.
Electron. Eng. 10(1), 1520 (2017)
12. Kumar, A.V.P., Parimi, A.M., Rao, K.U.: Performance analysis of voltage regulated inverter for
FLC based PV-wind hybrid power system with real time data. In: 2016 International Conference
on Emerging Trends in Engineering, Technology and Science (ICETETS), Pudukkottai, pp. 1–8
(2016)
13. Pavan Kumar, A.V., Parimi, A.M., Uma Rao, K.: Tie-line frequency bias control of two-area
PV-wind hybrid power system. UPB Sci. Bull. Ser. C Electr. Eng. Comput. Sci. 80(1), 217–230
(2018). ISSN 2286-3540
14. Pavan Kumar, A.V., Parimi, A.M., Uma Rao, K.: Investigation of small PMSG based wind
turbine for variable wind speed. In: IEEE International Conference on Recent Developments
in Control, Automation and Power Engineering (RDCAPE), Mar 2015, pp. 107–112
15. Schmidlin Junior, C.R., Araujo Lima, F.K.: Wind turbine and PMSG dynamic modelling in
PSIM. IEEE Latin Am. Trans. 14(9), 4115–4120 (2016)
16. Pavan Kumar, A.V., Parimi, A.M., Uma Rao, K.: Implementation of MPPT control using
fuzzy logic in solar-wind hybrid power system. In: IEEE International Conference on Signal
Processing, Informatics, Communication and Energy Systems (SPICES), 19–21 Feb 2015,
pp. 1–5
17. Gan, L.K., Macpherson, D.E., Shek, J.K.H.: Synchronisation control and operation of micro-
grids for rural/island applications. In: 48th International Universities Power Engineering
Conference (UPEC), Dublin, pp. 1–6 (2013)
18. Bugade, V.S., Katti, P.K.: Dynamic modelling of microgrid with distributed generation for
grid integration. In: International Conference on Energy Systems and Applications, Pune,
pp. 103–107 (2015)
19. Davoodnezhad, R., Holmes, D.G., McGrath, B.P.: A three-level selfsynchronizing hystere-
sis current regulator with constant switching frequency. In: IEEE ECCE Asia Downunder,
Melbourne, VIC, pp. 38–44 (2013)
20. Pavan Kumar, A.V., Parimi, A.M., Uma Rao, K.: A discrete PLL based load frequency control
of FLC based PV-wind hybrid power system with real time data. Int. J. Power Electron. Drive
Syst. 7(4), 1320–1329 (2016)
The Linear Quadratic Regulator Design
for BG Control in Type-I Diabetes Patient

Akshaya Kumar Patra, Anuja Nanda, Alok Kumar Mishra,

and Lalit Mohan Satapathy

Abstract This manuscript presents a SIMULINK model of glucose metabolism

(GM) process and design of a linear quadratic regulator (LQR) to regulate the blood
glucose (BG) concentration in type-I diabetes mellitus (TIDM) patients. For design-
ing the LQR, a ninth-order-linearized model of the patient with micro-insulin dis-
penser (MID) is taken. The LQR is a novel approach whose gains dynamically vary
with respect to the error signal. The validation of improved control action of LQR
is established by comparative result investigation with other published control algo-
rithms. The comparative results clearly reveal the better performance of the proposed
approach to control the BG concentration (level) within the normoglycaemic range
in terms of accuracy, stability, and robustness.

Keywords Diabetes · Insulin dose · Glucose concentration · MID · LQR

1 Introduction

As per the World Health Organization (WHO) report, one among the widespread dis-
eases is diabetes mellitus and is resulted due to the malfunctioning of the pancreas.
This reduces insulin sensitivity affecting the normoglycaemic range of BG concentra-
tion (70–120 mg/dl) in a healthy human being. At present, numerous research projects
are undertaken by several researchers to get rid of this problem by devising advanced

A. K. Patra (B) · A. Nanda · A. K. Mishra · L. M. Satapathy

Department of EEE, Institute of Technical Education and Research, Siksha ‘O’ Anusandhan
(Deemed to be University), Bhubaneswar 751030, India
e-mail: [email protected]
A. Nanda
e-mail: [email protected]
A. K. Mishra
e-mail: [email protected]
L. M. Satapathy
e-mail: [email protected]

© Springer Nature Singapore Pte Ltd. 2020 57

G. Pradhan et al. (eds.), Advances in Electrical Control and Signal Systems,
Lecture Notes in Electrical Engineering 665,
https://doi.org/10.1007/978-981-15-5262-5_5
58 A. K. Patra et al.

medical equipment like automated MID. Till date, BG concentration is manually con-
trolled to adopt the open-loop control strategy. The hypoglycemic or hyperglycemic
conditions may evolve due to the difficulties in handling the internal system changes
and external disturbances by implementing the control loop technique. The develop-
ment of implanted artificial pancreases (AP) enabling the adequate dose of insulin
delivery proportionate to the sensor measurement in the patient’s body may pro-
vide the means to incorporate the closed-loop control strategy. Figure 1a describes a
closed-loop patient model with an AP. The AP consists of glucose sensor, MID, and
controller. The sensor measures the BG concentration of the human body continu-
ously and sends signal to the controller for generating the desired control actions.
The control signal generated by this controller also depicts the association of the
additional model uncertainties and disturbances. Thereafter, according to the control
signal u(t), the optimal insulin dose is infused into the patient’s venous blood by
MID to achieve the normoglycaemic range of BG concentration [1, 2].
To determine an optimal solution for the AP-like building, an appropriate model of
the complex BG regulatory system, a number of obstacles, and challenges such as the
effects of nonlinear behavior, time-dependent dynamics, presence of several sources
of disturbance, uncertainty, and lack of glucose sensing are to be faced. Additionally,
the challenges and constraints related to control of BG level are the glucose mea-
surement delay, insulin absorption delay, and random variations of model parameters
[1]. Despite of technical progress and considerable development on aforesaid issues,
substantial improvement is still required in the control algorithm. Assessment of the
glucose excursions following the insulin dose adjustment needs a controller for BG
regulation in AP and many authors suggested the PID controller as a viable solution
[3, 4]. However, due to glucose sensing time delay, insulin action, and non-variable
gain parameters, the desired performance, assuring high accuracy, reliability and
robustness could not be achieved. The fuzzy control [5, 6], LQG control [7], H ∞
control [8–10], sliding mode (SM) control [11–14], and model predictive (MP) con-
trol [15, 16] are some of the well-proven controllers to deal with the BG regulation
issues. The BG control in the diabetic patients within the normoglycaemia range
inculcating the above controllers enhanced the accuracy and robustness to some
extent compared to the PID controllers. However, these control algorithms are not
fully insensitive to the disturbances and the uncertainties of the model in spite of the
improved performance. Hence, optimal control parameters setting for better perfor-
mance and for avoiding slow response following meal disturbance, the current work
suggests an alternative novel technique implementing the linear quadratic regulator.
The LQR approach concept leads to an enhanced control performance with respect
to robustness and delay of time compensation characteristics to counteract the neg-
ative impact of associated errors, disturbances, and uncertainties. The stabilizing
control law u(t) is developed based on the LQR approach. Application of the LQR
to control the BG concentration in TIDM patient results to ensure a better robust
controller in comparison to other contemporary well-accepted methods under both
harmonized and incompatible uncertainties.
The highlights of this manuscript are as follows:
The Linear Quadratic Regulator Design for BG Control … 59

(a)

(b)

(c)

(d)

Fig. 1 a Overview of TIDM patient model with AP; b Compartmental model of TIDM patient
with AP; c SIMULINK model of TIDM patient with MID; d SIMULINK model of MID
60 A. K. Patra et al.

• Development of a SIMULINK model of the TIDM patient.

• Design of a novel LQR to control the BG concentration within the stable range.
• Evaluation of the control actions of the LQR under several abnormal conditions.
• Comparative investigation to certify the better response of the LQR.
This manuscript is organized as follows. Section 2 concisely illustrates the TIDM
patient model with mathematical details reflecting its dynamic characteristics of its
GM process. Also, it clearly demonstrates the simulation execution of the system
on MATLAB environment. A detailed presentation on how the control technique
is formulated and how it is implemented for this problem is presented in Sect. 3.
Comparative results of the proposed approach with other published control tech-
niques and the related analysis are provided in Sect. 4. The concluding comments
are summarized in Sect. 5.

2 Problem Formulation and Dynamic Modeling of TIDM

2.1 System Overview

The compartmental model schematic diagram as depicted in Fig. 1b reveals the GM

dynamics of the human body. The proposed controller is tested using the above
model. The human body has liver, kidney, periphery, heart/lungs, gut and brain as
different functional parts and is shown as six compartments separately. The blood flow
directions or circulations are indicated by the arrow marks. The controller receives
the measured arterial glucose value as input to provide the optimal insulin dose to
be injected as an output with the help of MID.

2.2 Clinical Background

A human body with a prolonged high arterial BG level that exceeds 144 mg/dl is
known to suffer from hyperglycemia disease. This type of the clinical disorder is
also termed as diabetes mellitus. The insulin deficiency along with its less resistive
or insensitiveness independently or combinedly cause hyperglycemia and are some
of the major reasons as discussed in [17, 18]. Diabetes mellitus may occur if the blood
glucose remains unutilized effectively under any normal human life cycle. Diabetes
is classified as type-I and type-II as per the rate of insulin creation in the patient’s
body. In the type-I diabetic patients, the insulin creation by the pancreatic cells is
completely impossible, but in case of the type-II is at a bit smaller rate, which also
gets inhibited in due course. Hence, in the present situation, an immediate insulin
injection to restore the normal BG level is essential that also prevents allied adverse
impacts on the normal operation of any human body. Subject to diverse fooding,
The Linear Quadratic Regulator Design for BG Control … 61

fasting, and exercise behaviors, the regulation of the BG levels are not very easy in
the real human life cycle.
The external sources like the carbohydrate foods are digested down into glucose
at different parts of the digestive system (gut), and then filtered. The filtered glucose
is added to the venous blood. Later on, the liver stores this glucose after being
transferred from the venous blood as glycogen. When the BG concentration in the
venous blood is less than normoglycaemic range of glucose concentration, glycogen
is re-injected as glucose by the liver into venous blood and is known as the internal
glucose source. The glucose utilization and production by the liver are known as the
net hepatic glucose balance (NHGB), which is the general phenomenon in the glucose
metabolism process. The total energy needed by human body organs irrespective of
insulin dependency derives from the degree of glucose utilization.
The creation of insulin is due to the β-cells present in the pancreas and thus, the
BG level is regulated. A normal human with high BG level has two major functions
of insulin; firstly, to allow the liver for the glucose absorption along with the storing
it in the form of glycogen, specifically under the meal intake conditions. As a result,
the production of excess ‘internal’ glucose by the liver and muscles is stopped.
Secondly, insulin speeds up the glucose absorption in the muscles and fulfills the
peripheral energy needs of the body. But, in case of diabetes patients, the abnormality
in the metabolic process dynamics, both the aforementioned functions are found to
impair partially or completely. In case of diabetes patient, the cells stop the glucose
utilization and internal glucose is produced by the liver, and then an uncontrolled
BG level is noticed. Later, as the BG level is more than the renal threshold glucose
(RTG) value of 162 mg/dl, venous BG in excess is extracted through the kidney.

2.3 Modeling of a TIDM Patient

In present scenario, with due consideration of the dynamics of glucose metabolism

process, several simulation models are proposed to control the BG concentration in
the TIDM patients [19–22]. Owing to the simplified structure and desired approx-
imation to the dynamics of human metabolism with reduced error, the model sug-
gested by Parker and Doyle [19, 20] and Lehmann and Deutsch [21, 22] is among
the widely accepted ones. The testing and verification of the proposed technique
using Lehmann and Deutsch [21, 22] model for BG control are considered in the
present study. Figure 1b describes the compartmental setup of diabetic patient with
the implanted AP. The controller computes the insulin injection amount to the venous
blood at 5 min interval using an integrated implanted insulin pump. Depending on
the supporting device and sensing technology, the sampling rate varies when is being
applied in the real-time domain [23, 24]. The gut compartment input is the meal and
the peripheral input is the exercise for integrating and executing the process dis-
turbances. Compartmental modeling method considering the equations of the fast
principle has been used to develop the GI interaction process with a MID model
62 A. K. Patra et al.

of the patient [21, 22]. Figure 1c demonstrates the simulated model diagram of the
patient with MID.

2.4 Micro-insulin Dispenser

In this study, a fifth-order model suggested by Cochin and Cadwallender [24] is

taken up to devise the MID model. Figure 1d represents all dynamics of the major
components of the MID model like the suitable control circuit, fluid circuit, insulin
pump with return valve, and the insulin storage capsule. The dynamics of its operation
are based on the principle of the variable pumping rate. The BG concentration is
regulated with the injection of the required insulin dose by the MID into venous
blood with respect to u(t).

2.5 Linearization

Figure 1c shows the nonlinear model of patient with the MID, and its simplified form
is as shown in Fig. 2a. The linearized equation of the patient with process disturbance
can be formulated as [25]:

ẋm (t) = Am xm (t) + Bm u(t) + Bd w(t)
(1)
y(t) = Cm xm (t) + Dm u(t)

where w(t) is the process disturbance, and it has two major components such as meal
disturbance w1 (t) and exercise w2 (t). The process disturbance w(t) is generally two
types in nature, first one is deterministic disturbance and the second one is stochastic
disturbance. xm (t) is the state variable of the NM of TIDM patient, and y(t) is the
measured output. Am , Bm , Cm , Dm , and Bd represent the state-space matrices of
the patient. The state-space matrices are found from the system dynamic equations
linearization surrounding the reference point (operating point). The detailed deriva-
tion is described in the literature [25]. In the current study, the two initial references
such as the BG concentration and the basal insulin dose are taken as 81 mg/dl and
22.3 mU/min, respectively. In MATLAB, the command ‘linmod’ is used to evalu-
ate the state-space matrices on the TIDM patient simulation model as depicted in
Fig. 1c. The nominal patient model is linearized surrounding the operating point for
the design of robust controllers to regulate the BG level within the normoglycaemic
range.
The Linear Quadratic Regulator Design for BG Control … 63

(a)

(b)

(c) (d)

Fig. 2 a Simplified block diagram of patient with MID; b BG concentration with respect to insulin
dose; c Glucose production and consumption rate by GM organs; d Glucose extraction rate of
kidney verses BG level

2.6 Analysis of Patient Dynamics

The patient model dynamic and characteristic operations are verified under several
operating constraints like the actuator and sensor noises, intake amount of carbohy-
drates, and varying exercises, etc. The BG concentration and the insulin dose of the
considered patient model with 60 gm meal at 600 min and the exercise for half-an-
hour at 1300 min are illustrated in Fig. 2b. The glucose production and consumption
rate of liver, and glucose consumption rate by the BG profile organs such as gut
compartment, CNS, and peripheral cells are shown in Fig. 2c.
64 A. K. Patra et al.

The overall BG regulation in the human body is carried out naturally through
glucose utilization by the organs such as peripheral cells (adipose tissues, muscle
cells) and the liver that exclusively depend on insulin. The peripheral cells and liver
consume the least amount of glucose under the condition of lack of insulin. Under this
situation, BG level rises abnormally and go beyond the glucose level of 144 mg/dl.
This results in the hyperglycemia problem. The kidney removes part of the glucose
from venous blood under the condition when the BG concentration rises up to RTG
value. The rate of glucose excretion of the kidney proportionate to the BG level is
represented by Fig. 2d. Figure 2b–d illustrates the unstable dynamics under various
model uncertainties and disturbances. These abnormal dynamics can be reduced by
applying the AP based on suitable control algorithm.

3 Control Algorithm

The LQR control algorithm is demonstrated in this section. The closed-loop sys-
tem response with respect to stability, accuracy, and robustness is analyzed. The
control specifications such as settling time (min), steady-state error ess (%), over-
shoot (mg/dl), and undershoot (mg/dl) are also evaluated and examined with proper
validation of the controller actions.

3.1 LQR Design

The linearized model of the TIDM patient as discussed in Sect. 2.5 has been taken for
the formulation of the LQR to regulate the BG concentration. For accomplishing an
upgraded performance and the adjustment of controller parameters of the suggested
LQR, it is integrated to the linearized model of the patient as illustrated in Fig. 3.
The linearized model of the patient with w(t) and ν(t) is formulated as follows:

dxm (t)
= Am xm (t) + Bm u(t) + Bd w(t) (2)
dt

y(t) = Cm xm (t) + ν(t) (3)

u(t) = −K c e(t) (4)

where e(t) is the error signal. Figure 3 shows the linearized model of the diabetes
patient with the LQR gain K c . The evaluation procedures of K c is mentioned in detail
in Sect. 3.1.1. The evaluation of the TF of the LQR is described in Sect. 3.1.2.
The Linear Quadratic Regulator Design for BG Control … 65

Fig. 3 Closed-loop TIDM patient with LQR in the state-space representation

3.1.1 LQR Gain K c

The K c is evaluated with respect to the patient dynamics for the minimum value of
the quadratic performance index j as specified in Eq. (5).

∞
 
j= xm (t)T Q 1 xm (t) + u(t)T R1 u(t) dt (5)
0

where Q 1 = CmT q1 Cm . The Q 1 , q1 , and R1 are denoted as the positive semi-definite

weighted matrix of state, the intensity of a weighted matrix of state, and the positive
definite weighted matrix of input, respectively. The K c is estimated for the minimum
value of the quadratic performance index as specified in Eq. (6).

K c = R1−1 BmT (6)
k

k is the solution of the Controller Algebraic Reccati Equation (CARE). The CARE
is specified as follows:
   
AmT + Am − Bm R1−1 BmT +CmT q1 Cm = 0 (7)
k k k k

3.1.2 TF of LQR

The TF of suggested controller K (S) is estimated with the use of K c . The K (s) of
LQR is specified as follows:
66 A. K. Patra et al.

Table 1 Optimal values of

Q1 R1
control parameters
1000 ∗ CmT ∗ Cm 1

K (s) = K c (s In − Am + Bm K c )−1 Bm (8)

The packed matrix notation of K (s) is specified as follows:


Am − Bm K c Bm
K (s) = (9)
−K c 0

For the design of suggested LQR, the optimal values of control parameters are
evaluated with help of MATLAB and represented as in Table 1.

4 Result and Discussions

Time domain response of glucose profiles, stability, and robustness of the closed-loop
model with proposed LQR is described in detail in this section. The proposed control
approach is compared with other popular control algorithms to justify its enhanced
performance.

4.1 Analysis of Patient Dynamics with LQR

In this section, all glucose profiles of the nonlinear patient model with LQR are
examined under different operating conditions such as variation of exercise, intake
amount of carbohydrate, and noise of sensors and actuators. All glucose profiles like
BG level, insulin dose, NHGB rate, gut rate, CNS, and peripheral glucose utilization
rate in TIDM patient model with proposed controller are illustrated in Fig. 4. Com-
pared to the condition of the uncontrolled process, the result obtained evidences a
higher utilization of the plasma glucose by the peripheral cells and liver, those are
dependent on the insulin availability and sensitivity. These bring back the BG level to
81 mg/dl with the quick settling time that results in the reduction or prevention of the
hyperglycemia occurrence probability as shown in Fig. 4a. The insulin-independent
organ like the CNS consumes the plasma glucose at a constant rate. Figure 4b illus-
trates the constant rate of glucose utilization by CNS at 84 mg/min. As the BG level
is below the RTG level, kidney does not extract any glucose and is depicted in Fig. 4c
by the constant glucose excretion rate at zero level. By implementing the proposed
control techniques to the MID, different operational parameters are evidenced and
have the enhanced performances.
The Linear Quadratic Regulator Design for BG Control … 67

(a)

(b) (c)

Fig. 4 Control action of LQR in glucose profiles of TIDM patient model: a BG concentration with
respect to insulin dose; b Glucose production and consumption by GM organs; c Glucose extraction
rate of kidney verses BG concentration (level)

4.2 Robustness of the LQR

Figure 5 illustrates the BG concentration of the patient model with suggested LQR
under the large variation of disturbance of the meal. The time domain outcomes
under different quantities of meal intake show the enhanced performance of the
patient model with suggested approaches. Overall, in each case, the patient model
achieves finally BG concentration of 81 mg/dl with less settling time. As indicated by
the results, the suggested controller’s robust performance under the huge deviation
of meal disturbances compared to LQR and other published control techniques is
much better.
68 A. K. Patra et al.

Fig. 5 BG concentration (level) with deviation of ±30% meal intake based on LQR

4.3 Stability Analysis

Figure 6a, b illustrates the magnitude plots result of the open-loop and closed-loop
patient model to verify and analyze the stability conditions. From the magnitude
plots, it is observed a better smoothness referring to the wider steady-state stability
of closed-loop system (Fig. 6b) than the open-loop system (Fig. 6a). In other words,
the bandwidth is increased in case of a closed-loop system with proposed controllers
than the open-loop system. This clearly indicates a faster stable dynamics, and also

(a) (b)

Fig. 6 a Magnitude plot of the patient model; b Magnitude plot of the patient model with LQR
The Linear Quadratic Regulator Design for BG Control … 69

Table 2 Comparative result analysis related to the BG concentration

Controller PID [4] Fuzzy [5] H ∞ [10] SM [14] LQR (Proposed)
Meal intake (mg) 60 60 60 60 60
Insulin infusion rate 59.6 59.2 59.1 59 59.1
(mU/min)
Settling time (min) 290 262 261 265 260
Peak overshoot (mg/dl) 6.5 6.5 6.6 6.5 6.4
Peak undershoot 3.1 2.1 1.4 1.5 1.3
(mg/dl)
Noise (%) 10 10 5 5 4
ess (%) 0 0 0 0 0

closed-loop patient model with LQR archives BG concentration of 81 mg/dl with

less settling time. This justifies better stability during system operation with LQR.

4.4 Comparative Study

The suggested LQR approach is compared with other popular control approaches
such as PID, fuzzy, H ∞, and SM to justify its enhanced performance as the con-
trollers. Figure 4 illustrates the effect of meal disturbance in the BG concentration of
the patient with the LQR approaches. Table 2 presents a comparative analysis with
respect to settling time (min), peak overshoot (mg/dl), peak undershoot (mg/dl),
noise (%), and steady-state error ess (%). The effect of meal disturbance in the BG
concentration of the patient model applying different control approaches such as
PID, fuzzy, H ∞ and SM is also presented in Table 2 based on the references [4, 5,
10, 14], respectively. Similar working conditions are followed with the same level
of actuator and sensor noise in all control techniques application for comparison.
The BG concentration in TIDM patient model ingested to 60 g meal under different
controllers is tested. The corresponding results are presented for the various control
approaches along with the proposed LQR with respect to control specifications such
as overshoots (mg/dl), undershoots (mg/dl), and settling time (min). The results
signify the better controllability of the LQR. The simulation results also demonstrate
the high noise elimination capability with high robustness for the proposed approach.
Overall, by looking at the above comparative analysis, the findings of suggested
approach advantages are the higher accuracy and stability, more robustness, high
noise elimination capability, and better capability to handle uncertainty under various
abnormal conditions and huge variation meal disturbance.
70 A. K. Patra et al.

5 Conclusions

In this paper, a novel control strategy LQR is proposed for BG control in TIDM
patient model. To justify its enhanced performance, it has been applied and tested to
control the BG concentration in patient within normoglycaemic range. Initially, the
patient is modeled as a ninth-order state-space representation. Then, the proposed
control approach LQR is designed. The comparative results clearly reflect that the
suggested LQR is arrived at better performance than the other control approaches
such as PID, fuzzy, H ∞, and SM, with respect to stability, reliability, and robust-
ness under various abnormal conditions and disturbances. The related better perfor-
mance of the suggested approach (LQR) in terms of improved accuracy and stabil-
ity, enhanced robustness, high noise elimination capability, and better capability to
handle uncertainty justify its real-time application.

References

1. Chee, F., Fernando, T.: Closed-loop glucose control in critically ill patients using continuous
glucose monitoring system (CGMS) in real time. IEEE Trans. Inf. Technol. Biomed. 7(4),
43–53 (2003)
2. Kamath, S., George, V.I., Vidyasagar, S.: Simulation study on closed loop control algorithm
of type 1 diabetes mellitus patients. IETE J. Res. 55, 230–235 (2009)
3. Frederick, C., Tyrone, L.: Expert PID control system for blood glucose control in critically ill
patients. IEEE Trans. Inf. Technol. Biomed. 7(4), 419–425 (2003)
4. Sutradhar, A., Chaudhuri, S.: Analysis and design of an optimal PID controller for insulin
dispenser system. J. Inst. Eng. (India) 82(2), 304–313 (2002)
5. Ibbini, M.: The PI- fuzzy logic controller for the regulation of blood glucose level in diabetic
Patients. J. Med. Eng. Technol. 30(2), 83–92 (2006)
6. Singh, M., Gupta, J.R.P.: A new algorithm-based type-2 fuzzy controller for diabetic Patient.
Int. J. Biomed. Eng. Technol. 1(1), 18–40 (2007)
7. Patra, A.K., Rout, P.K.: An automatic insulin infusion system based on LQG control technique.
Int. J. Biomed. Eng. Technol. 17(3), 252–275 (2015)
8. Chee, F., Andrey, V.: Optimal H∞ insulin injection control for blood glucose regulation in
diabetic patients. IEEE Trans. Biomed. Eng. 52(10), 1625–1631 (2005)
9. Yasini, S., Karimpour, A., Sistani, M.B.N.: Knowledge-based closed-loop control of blood
glucose concentration in diabetic patients and comparison with H∞ control technique. IETE
J. Res. 58, 328–336 (2012)
10. Patra, A.K., Rout, P.K.: Optimal H-infinity insulin injection control for blood glucose regulation
in IDDM patient using physiological model. Int. J. Autom. Control 8(4), 309–322 (2014)
11. Gallardo, H., Ana, G.: High-order sliding-mode control for blood glucose: practical relative
degree approach. Control Eng. Pract. 21(5), 747–758 (2013)
12. Rmileh, A., Gabin, W.: Wiener sliding-mode control for artificial pancreas: a new nonlinear
approach to glucose regulation. Comput. Methods Prog. Biomed. 107, 327–340 (2012)
13. Patra, A.K., Rout, P.K.: Backstepping sliding mode Gaussian insulin injection control for blood
glucose regulation in TIDM patient. J. Dyn. Syst. Meas. Control 140(9), 091006-091006-15
(2018)
14. Patra, A.K., Rout, P.K.: Adaptive sliding mode Gaussian controller for artificial pancreas in
TIDM patient. J. Process Control 58, 23–27 (2017)
The Linear Quadratic Regulator Design for BG Control … 71

15. Patra, A.K., Mishra, A.K., Rout, P.K.: Backstepping model predictive controller for blood
glucose regulation in type-i diabetes patient. IETE J. Res. (2018). https://doi.org/10.1080/
03772063.2018.1493404
16. Patra, A.K., Rout, P.K.: Adaptive continuous-time model predictive controller for implantable
insulin delivery system in Type I diabetic patient. Optimal Control Appl. Methods 38(2),
184–204 (2017)
17. Guyton, J.R., Foster, R.O.: A model of glucose-insulin homeostasis in man that incorporates
the heterogenous fast pool theory of pancreatic insulin release. Diab. Care 27(1), 1027–1042
(1978)
18. Barger, M., Rodbard, D.: Computer simulation of plasma insulin and glucose dynamics after
subcutaneous insulin injection. Diab. Care 12(1), 725–736 (1989)
19. Parker, R.S., Doyle III, F.J.: A model-based algorithm for BG control in type 1 diabetic patients.
IEEE Trans. Biomed. Eng. 46(2), 148–157 (1999)
20. Parker, R.S., Doyle III, F.J.: Variable-rate implantable insulin infusion pumps—closed loop
maintenance of normoglycaemia under patient variability for type 1 diabetes. In: Proceedings
of the 11th World Congress, International Society for Artificial Organs (1997)
21. Lehmann, E.D., Deutsch, T.: Physiological Model of glucose–insulin interaction in Type-1
diabetes mellitus. J. Biomed. Eng. 14(3), 235–242 (1992)
22. Lehmann, E.D., Deutsch, T.: Compartmental models for glycaemic prediction and decision
support in clinical diabetes care: promise and reality. Comput. Methods Prog. Biomed. 56(1),
193–204 (1998)
23. Sperr, G.: Biosensor research targets medical diagnostics. In: Medical Device and Diagnostic
Industry Magazine, Nov 1997
24. Cochin, L., Cadwallender, W.: Analysis and design of dynamic systems, 3rd edn. Addison-
Wesley, New York (1997)
25. Sutradhar, A., Chaudhuri, S.: Linear state space model of physiological process in a type-1
diabetic patient with closed loop glucose regulation. J. AMSE Fr. Adv. Model. Anal. Ser. C
66, 1–18 (2005)
ANN-Based Controllers for Improved
Performance of BLDC Motor Drives

R. Shanmugasundaram, C. Ganesh, and A. Singaravelan

Abstract This paper discusses the development and performance analysis of ANN-
based reference model controller and ANN-based self-tuned PID controller for
BLDC motor drives. As the BLDC motor drives are nonlinear due to its parameter
and load variations, there is a need to develop ANN-based controllers to overcome the
problems arising due to nonlinearity in BLDC motor drives. In this paper, ANN-based
self-tuned PID controller is developed for speed control of BLDC motor drives and its
performance is compared with the standard ANN-based reference model-controlled
BLDC motor drives. The unique feature of ANN-based self-tuned PID controller is
that it can dynamically change the PID controller gains to provide optimum perfor-
mance under changing dynamics of BLDC motor drive. Experimental results show
that ANN-based self-tuned PID-controlled BLDC motor drives can effectively deal
with speed tracking, load variations, and parameter variations.

Keywords ANN · Reference model · Self-tuned PID controller · BLDC motor ·

Parameter variations

1 Introduction

The BLDC motors are becoming increasingly popular in the electric vehicles, space
vehicles, military and medical equipment, etc., due to its excellent speed–torque
characteristics and less maintenance. But, the problems with conventional controllers
such as PI and PID are they cannot effectively track the speed under system parame-
ter and load variations. Moreover, the design of conventional controllers is based on
exact mathematical model obtained from the motor parameters. Under these oper-
ating conditions, conventional controllers fail to control the speed and yield desired

R. Shanmugasundaram (B)
Sri Ramakrishna Engineering College, Coimbatore 641022, India
e-mail: [email protected]
C. Ganesh · A. Singaravelan
New Horizon College of Engineering, Marathahalli, Bengaluru, Karnataka 560103, India

© Springer Nature Singapore Pte Ltd. 2020 73

G. Pradhan et al. (eds.), Advances in Electrical Control and Signal Systems,
Lecture Notes in Electrical Engineering 665,
https://doi.org/10.1007/978-981-15-5262-5_6
74 R. Shanmugasundaram et al.

performance. Artificial neural networks (ANNs) having non-linear mapping charac-

teristics are found to be a promising solution for controlling BLDC motor drives.
However, ANN required sufficient input–output data to obtain the exact input–output
mapping over the entire operating region. The objective of this paper is to develop
ANN-based self-tuned PID controller for speed tracking of BLDC motor drives
subjected to parameter and load variations, and step change in reference speed and
compare its performance with ANN-based reference model-controlled BLDC motor
drive and fuzzy controller-based BLDC motor drive [1].
In [2–12] parameter estimation and control techniques for BLDC motor drives
are investigated. In [6, 13] performance controllers for BLDC motor-driven electric
vehicle are analyzed. In [1, 5, 7] speed tracking performance of conventional PID
and fuzzy controllers is analyzed. In [1, 14–18] performance variations due to change
in parameters of the motor and sudden load disturbances of BLDC motor drive are
discussed. In [6, 16–22] adaptive control techniques to compensate for parameter
variations and load disturbances are discussed. In [1, 23–28] digital implementation
of controllers for improving the performance of BLDC motor drive is presented.
In [16, 20, 21, 29] performance of ANN-based control techniques is analyzed. The
following sections discuss the implementation and performance analysis of ANN-
based controllers for speed control of BLDC motor drives.

2 Inverse Dynamic Model of BLDC Motor with Propeller

Load

Based on the mathematical model of BLDC motor [1] and assuming that load con-
nected to the BLDC motor is a propeller type load, the dynamic model is obtained to
establish the relationship between the consecutive speed outputs and input terminal
voltage of motor. The load torque of the propeller load [20] is given by,
 
Tl = μωm2 (k) sign{ωm (k)} (1)

where ωm is the angular velocity (rad/s) and μ is a constant used for modeling the
nonlinear propeller load. The finite difference equation is obtained by combining (1)
and voltage, back-emf and torque equations of BLDC motor [20] and it is given by,
 
ωm (k + 1) = αωm (k) + βωm (k − 1) − γ sign{ωm (k)} ωm2 (k)
 
+ δ sign{ωm (k)} ωm2 (k − 1) + ς v(k) (2)

where ωm (k − 1), ωm (k), ωm (k + 1) are the three consecutive speeds; α, β, γ , δ and ζ

are the constants which are expressed in terms of the motor parameters and sampling
period [16]. The inverse dynamic model of the BLDC motor-driven propeller system
is obtained from (2) as,
ANN-Based Controllers for Improved Performance of BLDC … 75

v(k) = f (ωm (k + 1), ωm (k), ωm (k − 1)) (3)

Equation (3) clearly shows that there exist a nonlinear relationship between ter-
minal voltage ν(k) and the three consecutive speeds of BLDC motor [19–22, 30, 31].

2.1 ANN-Based Reference Model Controller for BLDC

Motor

In ANN-based reference model controller, ANN is trained to learn and mimic the
behavior of inverse dynamic model of BLDC motor. The block diagram of ANN-
based reference model controller and structure of ANN are shown in Figs. 1 and 2,
respectively. The inputs of ANN are three consecutive speeds [ωm (k + 1), ωm (k)
and ωm (k − 1)] and the output is terminal voltage v(k) of BLDC motor. The aim of
this controller is to control the BLDC motor such that its speed, ωm (k), follows the
specified speed trajectory, ωref (k), which is generated from the stable second-order
reference model [16, 19, 22] given by,

ωref (k + 1) = 0.6ωref (k) + 0.2ωref (k − 1) + r (k) (4)

where r(k) is the input to the reference model. For a desired speed trajectory ωref (k),
the input r(k) is generated from (4).

Fig. 1 Block diagram of ANN based reference model controller

76 R. Shanmugasundaram et al.

Fig. 2 Structure of ANN

The feed-forward ANN of size (3-5-1) is chosen for developing ANN-based ref-
erence model controller. The bipolar continuous activation function is used for all
neurons. The ANN is trained by error back-propagation training algorithm with
momentum factor of 0.1 and learning factor of 0.2. The final weights and biases of
trained ANN are initially used to start and run the motor. Later, during the normal
operation, weights and biases are updated on-line if the error [e(k) = v(k) − v * (k)]
exceeds a preset value. For the generated control sequence r(k), the motor speed
ω*ref (k + 1) is predicted from (4). The ANN outputs ‘terminal voltage v(k)’ and the
‘predicted voltage v*(k)’ are used to compute the error, e(k), and this error is used
to update the weights and biases of the ANN. The operating space is chosen based
on the specifications of BLDC motor (Appendix) is as follows:

−4000 ≤ ωm (k) ≤ 4000 rpm

|vt (k)| ≤ 36 V

The speed and control voltage, v(t), data are collected for the specified trajectory
[20] from (5) as shown in Fig. 3. The final weights and biases of the trained ANN
are given Table 1.

ωref (k) = 1333 sin(2π kT /4) + 2133 sin(2π kT /7) ∀ kT ∈ [0, 20] (5)

2.2 Hardware Implementation of ANN-Based Reference

Model Controller for BLDC Motor

The objective of the ANN-based reference model controller is to control the input
voltage of BLDC motor such that actual speed follows the set speed (or reference
ANN-Based Controllers for Improved Performance of BLDC … 77

Fig. 3 Speed and control voltage data for training ANN

Table 1 Final weights and biases of trained ANN

Input to hidden layer weights NW1
0.6550 −0.2574 0.8581 0.0052 −0.453
−0.9676 0.3696 −0.4738 0.1568 −0.113
0.5223 −0.3353 −0.3165 −0.4292 0.4549
Biases NB1
0 0.0001 0 0.0001 −0.0001
Hidden to output layer weights NW2
0.8094 −0.0583 0.7307 0.1984 −0.4075
Biases NB2
−0.0147

speed) under different operation conditions such as step change in speed, load distur-
bances, load variations and parameters variations of motor and load. The experimen-
tal setup is shown in Fig. 4. The IGBT power inverter (IC IMRAY20UP60A) is used
to control the input power supplied to the motor. The duty-cycle controlled PWM
gating signals at 20 kHz frequency is generated by the control algorithm implemented
in the DSP processor. The DSP processor reads reference speed, actual speed, hall
sensor signals, current, and terminal voltage. The rotor position is identified from the
78 R. Shanmugasundaram et al.

IGBT Power Inverter

T1 T3 T5

DC + BLDC
Load
Supply - MOTOR
T4 T6 T2
Speed
Sensor
Signal F/V
Conditioning
Unit
ia, ib, ic νa, νb, νc
Digital Signal A/D Converter Act. Speed
Processor Ref. Speed
TMS320LF2407A
Hall Sensors Signal
Digital Inputs

Fig. 4 Experimental set-up of ANN controller based BLDC motor drive

hall sensor signals and commutation signals are generated for IGBTs. The control
algorithm computes actual voltage v(t) and predicted terminal voltage v*(t), error
e(t), commutation signals and the duty-cycle for the gating signals of IGBTs. The
duty-cycle of PWM gating signals is varied based on the ANN output, v(t), to control
the terminal voltage, and hence the speed of the motor. The control action is applied
at fixed interval of one-time electrical constant of motor so as to allow current to
reach sufficient value to produce the required torque during the normal operation.
The weights and biases of ANN are updated when the error, e(t), exceeds the preset
value.

2.3 Results and Discussion of ANN-Based Reference Model

Controlled BLDC Motor Drive

The ANN-based reference model-controlled BLDC motor drive is tested under oper-
ating conditions such as step change in reference speed, change in inertia and phase
resistance of BLDC motor, and sudden load disturbance. The phase resistance of
BLDC motor is changed by adding external resistance in series with the winding,
and inertia is changed by attaching an iron piece to the motor shaft. The values of
inertia used for testing are J 1 = 23 × 10−6 kg m2 , J 2 = 550 × 10−6 kg m2 ; phase
resistances are R1 = 0.57 , R2 = 1.14 .
Figure 5 shows the speed response obtained for the step change in reference speed
(1500 rpm–2500 rpm–4000 rpm–2500 rpm) and sudden load disturbance at full load.
When the system is subjected to step change in reference speed and load disturbances,
ANN-Based Controllers for Improved Performance of BLDC … 79

Fig. 5 Speed response of the ANN based Reference model controlled based BLDC motor drive
for step change in reference speed with 100% load and load disturbance for different parameter
combinations of the drive. a Reference speed; b J 1 , R1 ; c J 1 , R2 ; d J 2 , R2 ; e J 2 , R1

it is observed from the results that controller tracks the speed and bring the actual
speed close to reference speed.
The speed response obtained for the parameter combinations (i) J 1 & R1 , (ii) J 1 &
R2 , (iii) J 2 & R1 , and (iv) J 2 & R2 is shown in Fig. 6. The speed error is the difference
between reference speed and actual speed, and duty-cycle implies the %duty-cycle
of gating signal applied to IGBTs for controlling the terminal voltage of motor. It is
observed from these figures that controller is able to track the reference speed and
bring actual speed closer to reference speed with a maximum error of ±60 rpm. The
rise time, settling time, and speed error are given in Table 2. The experimental results
of fuzzy controller-based BLDC motor drive from [1] are given in Table 3.
If both the resistance and inertia are increased, the rise time and settling time of
speed response are found to be increasing. From the results given in Tables 2 and
3, it is evident that rise time, settling time, and de-acceleration time of ANN-based
80 R. Shanmugasundaram et al.

a . Drive parameters J1, R1 b. Drive parameters J1, R 2

c. Drive parameters J2, R1 d. Drive parameters J2, R 2

Fig. 6 Speed response of the ANN based Reference model controlled BLDC motor drive for
step change in reference speed with 100% Load. a Reference speed; b Actual speed; c Error;
d %Duty-cycle

reference model controller is much higher than fuzzy controller-based BLDC motor
drive [1].
ANN-Based Controllers for Improved Performance of BLDC … 81

Table 2 Experimental results of ANN based reference model controlled BLDC motor drive
Parameters of Rise time t r Settling time t s Deceleration Steady-state
the drive (ms) (ms) time t d (ms) error
J 1 , R1 , Load: 250 300 200 ±60 rpm
100%
J 1 , R2 , Load: 300 400 200 ±60 rpm
100%
J 2 , R1 , Load: 250 350 200 ±60 rpm
100%
J 2 , R2 , Load: 400 500 200 ±60 rpm
100%

Table 3 Experimental results of Fuzzy controller based BLDC motor drive [1]
Parameters of Rise time, t r Settling time, t s Deceleration Steady-state
the tested (ms) (ms) time, t d (ms) error
system
J 1 , R1 , Load: 250 300 200 ±60 rpm
100%
J 1 , R2 , Load: 250 300 300 ±60 rpm
100%
J 2 , R1 , Load: 150 200 200 ±60 rpm
100%
J 2 , R2 , Load: 200 250 200 ±60 rpm
100%

3 ANN-Based Self-tuned PID Controller for BLDC Motor

Drive

The role of ANN in this controller is to learn the relationship between the controller
gains (k p , k i , k d ) of PID controller and the parameters that affect them most such as
system inertia (J) and phase resistance (R) of BLDC motor [32]. The block diagram
of ANN-based self-tuned PID controller and the structure of ANN is shown in Figs. 7
and 8, respectively. The ANN of size (2–5–3) has PID controller gains as outputs and
system inertia (J) and phase resistance (R) as inputs. The range of the parameters
considered for this work is as follows:
(i) J: 23 × 10−6 kg m2 –550 × 10−6 kg m2
(ii) R: 0.57–1.14
This neural network is trained using error back-propagation training algorithm
with a momentum factor of 0.1 and learning factor of 0.2. The final weights and
biases of the trained ANN are given in Table 4. The function of parameter estimator
is to compute the values of system inertia (J) and phase resistance (R). The output of
82 R. Shanmugasundaram et al.

Fig. 7 Block diagram of ANN based PID controller for BLDC motor drive

Fig. 8 Structure of ANN for mapping inputs (J & R) with the outputs (K p , K i & K d )

Table 4 Final weights and biases of the ANN trained to learn the relationship between inputs (J
& B) with the controller gains (K p , K i & K d )
Input to first hidden layer
Weights NW1 Biases NB1
0.1129 −6.3571 −6.1613
−1.0875 6.0905 3.2212;
4.9466 3.8374 0.0033
4.5462 4.3127 3.1272
−4.7593 −4.0432 −6.2723
First hidden layer to second hidden layer weights NW2
−0.9640 −1.1323 0.6696 2.9660 −1.4372
−0.3848 −1.2506 1.9528 −0.5512 2.3184
−0.2713 −2.4856 −1.1856 2.0284 0.5970
Second hidden layer biases NB2
−6.1613 3.2212 0.0033 3.1272 −6.2723
Second hidden layer to output layer weights NW3 Output layer biases NB3
1.0014 −0.2078 0.1855 −0.0211
0.7998 −0.4428 −0.0839 −0.3142
−1.0560 −0.5461 −0.3083 0.1296
ANN-Based Controllers for Improved Performance of BLDC … 83

ANN is used to change the controller gains of PID controller so as to achieve better
performance under system parameter variations.

3.1 Hardware Implementation of ANN-Based PID Controller

for BLDC Motor Drive

The experimental setup shown in Fig. 4 is used for the implementation of ANN-based
self-tuned PID controller. The DSP controller reads phase currents, phase voltages,
speed and hall sensor signals and computes system inertia (J) and phase resistance
(R) from the following equations [12, 32].
The phase resistance is given by,

V−K ω
R= (6)
i

where V, i, ω, and K denote phase voltage variation, phase current variation,

speed variation, and back-emf constant.
The inertia is given by,

τm (K 2 + R B)
J= (7)
R
The value of viscous friction coefficient, B, is determined from phase current
variation and speed variation, and the mechanical time constant (τ m ) is determined
by moment’s method [12].
The ANN gives the output as controller gains (k p , k i , k d ) based on the inputs J
and R. The PID controller gives output as control voltage, v(t), based on speed error
and controller gains. This control voltage is used to regulate the speed of the BLDC
motor. The performance of the ANN-based self-tuned PID-controlled BLDC motor
drive is tested with step change in reference speed for different combinations of J
and R and the speed responses are shown in Fig. 9. The rise time, settling time, and
speed error are given in Table 5.
It is observed from the results that ANN-based self-tuned PID controller is able
to track the change in reference speed and maintain actual speed close to reference
speed with a maximum speed error of ±20 rpm. It is evident from the results that
rise time, settling time, and speed error of ANN-based self-tuned PID controller are
better than ANN-based reference model controller and fuzzy controller-based BLDC
motor drives [1]. Hence, performance of ANN-based PID controller is found to be
better than ANN-based reference model controller and fuzzy controller-based BLDC
motor drives [1].
84 R. Shanmugasundaram et al.

a. Drive parameters J 1, R b. Drive parameters J1, R 2

c. Drive parameters J 2, R 1 d. Drive parameters J2, R 2

Fig. 9 Speed response of the ANN based self-tuned PID controlled BLDC motor drive for step
change in reference speed with 100% Load. a Reference speed; b Actual speed; c Error; d %Duty-
cycle

4 Conclusion

The ANN-based reference model controller and ANN-based self-tuned PID con-
troller have been developed and implemented successfully. The experimental results
indicate that both the controllers are able to track the reference speed under load dis-
turbances, load variations, parameter variations, and step change in reference speed.
However, the overall performance of ANN-based self-tuned PID controller is found
to better than ANN-based reference model controller and fuzzy controller-based
ANN-Based Controllers for Improved Performance of BLDC … 85

Table 5 Experimental results of ANN basedself-tuned PID Controller based BLDC motor drive
Parameters of Rise time, t r Settling time, t s Deceleration Steady-state
the drive (ms) (ms) time, t d (ms) error
J 1 , R1 , 200 290 200 ±20 rpm (±
Load:100% 0.5%)
J 1 , R2 , 200 270 200 ±20 rpm (±
Load:100% 0.5%)
J 2 , R1 , 190 260 220 ±20 rpm (±
Load:100% 0.5%)
J 2 , R2 , 180 240 200 ±20 rpm (±
Load:100% 0.5%)

BLDC motor drives [1] under parameter variations, load disturbances, and load vari-
ations. Hence, ANN-based self-tuned PID controller may be a better choice for speed
control of BLDC motor under load disturbances and parameter variations.

Acknowledgements The authors thank the management and principal of Sri Ramakrishna
Engineering College, Coimbatore, for providing support and facilities to carry out this work.

Appendix

Parameters of BLDC motor

Rated voltage 36 V
Rated current 5A
Number of poles 4
Number of phases 3
Rated speed 4000 rpm
Rated torque 0.42 N m
Torque constant 0.082 N m/A
Mass 1.25 kg
Inertia of motor 23e−06 kg m2
Resistance per phase 0.57
Inductance per phase 1.5 mH
86 R. Shanmugasundaram et al.

References

1. Shanmugasundram, R., Zakariah, K.M., Yadaiah, N.: Implementation and performance analysis
of digital controllers for brushless dc motor drives. IEEE/ASME Trans. Mechatron. 19(1),
213–224 (2014)
2. Krishnan, R.: Permanent Magnet Synchronous and Brushless DC Motor Drives: Theory, Oper-
ation, Performance, Modeling, Simulation, Analysis, and Design-Part 3. Permanent Magnet
Brushless DC Machines and their Control, pp. 451–563. CRC Press, Boca Raton, FL (2009)
3. Pillay, P., Krishnan, R.: Modeling, simulation, and analysis of permanent-magnet motor drives,
part ii: the brushless dc motor drive. IEEE Trans. Ind. Appl. 25(2), 274–279 (1989)
4. Shanmugasundram, R., Zakariah, K.M., Yadaiah, N.: Low-cost high performance brushless dc
motor drive for speed control applications. In: Proceedings of IEEE International Conference
on Advanced Recent Technology Communication Computer, Kottayam, Kerala, India, pp. 456–
460 (2009)
5. Shanmugasundram, R., Zakariah, K.M., Yadaiah, N.: Digital implementation of fuzzy logic
controller for wide range speed control of brushless dc motor. In: Proceedings of IEEE
International Conference Vehicle Electronics Safety, Pune, India, pp. 119–124 (2009)
6. Wu, H.-x., Cheng, S.-k., and Cui, S.-m.: A controller of brushless DC motor for electric vehicle.
IEEE Trans. Magn. 41(1), 509–513 (2005)
7. Shanmugasundram, R., Zakariah, K.M., Yadaiah, N.: Modelling, simulation and analysis of
controllers for brushless direct current motor drives. J. Vib. Control 19(8), 1250–1264 (2012)
8. Grabner, H., Amrhein, W., Silber, S., Gruber, W.: Nonlinear feedback control of a bearingless
brushless dc motor. IEEE/ASME Trans. Mechatron. 15(1), 40–47 (2010)
9. Oztk, S.B., Toliyat, H.A.: Direct torque and indirect flux control of brushless dc motor.
IEEE/ASME Trans. Mechatron. 16(2), 351–360 (2011)
10. Precup, R.-E., Preitl, S., Rudas, I.J., Tomescu, M.L.: Design and experiments for a class of
fuzzy controlled servo systems. IEEE/ASME Trans. Mechatron. 13(1), 22–35 (2008)
11. Tu, Y.-W., Ho, M.-T.: Robust second-order controller synthesis for model matching of interval
plants and its application to servo motor control. IEEE Trans. Control Syst. Technol. 20(2),
530–537 (2012)
12. Hadef, M., Bourouina, A., Mekideche, M.R.: Parameter identification of a dc motor via
moments method. Int. J. Electr. Electron. Eng. 7(2), 159–163 (2008)
13. Rodriguez, F., Emadi, A.: A novel digital control technique for brushless DC motor drives.
IEEE Trans. Ind. Electron. 54(5), 2365–2373 (2007)
14. Wallace, A.K., Spee, R.: The effects of motor parameters on the performance of brushless dc
drives. IEEE Trans. Power Electron. 5(1), 2–8 (1990)
15. Varatharaju, V.M., Mathur, B.L., Udhyakumar, K.: Speed control of PMBLDC motor using
MATLAB/Simulink and effects of load and inertia changes. In: Proceedings of 2nd International
Conference on Mechanical and Electrical Technology, Singapore, pp. 543–548 (2010)
16. Rahman, M.A., Hoque, M.A.: On-line self-tuning Ann-based speed control of a PM DC motor.
IEEE/ASME Trans. Mechatron. 2(3), 169–177 (1997)
17. Shanmugasundram, R., Zakariah, K.M., Yadaiah, N.: Effect of parameter variations on the
performance of direct current (dc) servomotor drives. J. Vib. Control 19(10), 1575–1586 (2012)
18. Ganesh, C., Patnaik, S.K.: Artificial neural network based proportional plus integral plus deriva-
tive controller for a brushless dc position control system. J. Vib. Control 18(14), 2164–2175
(2012)
19. Rubaai, A., Ricketts, D., Kankam, M.D.: Development and implementation of an adaptive
fuzzy-neural-network controller for brushless drives. IEEE Trans. Ind. Appl. 38(2), 441–447
(2002)
20. Weerasooriya, S., El-Sharkavi, M.A.: Identification and control of dc motor using back-
propagation neural networks. IEEE Trans. Energy Convers. 6(4), 663–669 (1991)
21. Melkote, H., Khorrami, F.: Nonlinear adaptive control of direct drive brushless dc motors and
applications to robotic manipulators. IEEE/ASME Trans. Mechatron. 4(1), 71–81 (1999)
ANN-Based Controllers for Improved Performance of BLDC … 87

22. Weerasooriya, S., EI-Sharkawi, M.A.: Adaptive tracking control high performance dc drives.
IEEE Trans. Energy Convers. 4(3), 502–508 (1989)
23. Lai, Y.-S., Shyu, F.-S., Chang, Y.-H.: Novel loss reduction pulse width modulation technique
for brushless dc motor drives fed by MOSFET inverter. IEEE Trans. Power Electron. 19(6),
1646–1652 (2004)
24. Sathyan, A., Milivojevic, N., Lee, Y.-J., Krishnamurthy, M., Emadi, A.: An FPGA-based novel
digital PWM control scheme for BLDC motor drives. IEEE Trans. Ind. Electron. 56(8), 3040–
3049 (2009)
25. Murphree, J., Brzezinski, B., Parker, J.: Using a fixed-point digital signal processor as a PID
controller. In: Proceedings of American Society for Engineering Education Annual Conference
Exposition, Montreal, QC, Canada, pp. 1–8 (2002)
26. Horvat, R., Jezernik, K., Curkovic, M.: An event-driven approach to the current control of a
BLDC motor using FPGA. IEEE Trans. Ind. Electron. 61(7), 3719–3726 (2014)
27. Kim, J.-W., Kim, S.W.: Design of incremental fuzzy PI controllers for a gas-turbine plant.
IEEE/ASME Trans. Mechatron. 8(3), 410–414 (2003)
28. Nigam, V., Hussain, S., Agarwal, S.N.: A hybrid fuzzy sliding mode controller for a BLDC
motor drive. In: 1st IEEE International Conference on Power Electronics, Intelligent Control
and Energy Systems, Delhi, India, pp. 1–4 (2016)
29. Narendra, K.S., Parthasarathy, K.: Identification and control of dynamical systems using neural
networks. IEEE Trans. Neural Netw. 1(1), 4–27 (1990)
30. El-Sharkawi, M.A., El-Samahy, A.A., El-Sayed, M.L.: High performance drive of dc brushless
motors using neural network. IEEE Trans. Energy Convers 9(2), 317–322 (1994)
31. Tipsuwanporn, V., Piyarat, W., Tarasantisuk, C.: Identification and control of brushless dc
motors using on-line trained artificial neural networks. In: IEEE International Conference on
Power Conversion, Osaka, pp. 1290–1294 (2002)
32. Vinodhini, R., Ganesh, C., Patnaik, S.K.: Genetic Algorithm optimized on-line Neuro-
tuned robust position control of BLDC motor. In: IEEE Students’ Conference on Electrical,
Electronics and Computer Science, Bhopal, India, pp. 1–4 (2012)
Frequency Regulation of Electric
Vehicles Integrated Multi-area Power
System Under Deregulated Scenario
Using Novel Dual Mode Controller

Sayantan Sinha and Ranjan Kumar Mallick

Abstract The research work focuses on the automatic generation control (AGC) of
a multi-source two area power system with vehicle-to-grid technology under deregu-
lated environment. The electric vehicles are the future for the clean automobile sector
and hence the charging and discharging process for the vehicles have an important
role in the power system stability. The main role of AGC is to minimize the system
frequency variations and hence attempt has been made to design novel DUAL mode
control scheme. The dual mode control basically takes the magnitude of the error
and effectively chooses between proportional and integral control for better action.
This paper has proposed a Dual mode PID controller whose performance is put to
comparison with PID and 2 DOF PID controller. The controller gains are tuned by
a newly proposed Dragonfly algorithm. The system is to be simulated under base
case and bilateral contract market scenario of deregulated environment. Analysis is
also made for both the charging and discharging modes of electric vehicles. The
robustness of the proposed controller is to be established by analyzing the dynamic
system performances in terms of settling time, maximum overshoot, and undershoot.

Keywords AGC · Deregulated · Dual mode controller · PID · Electric vehicles

1 Introduction

With the expansion of population all over the world, the need for power has increased
to a large extent. This rapid increase in power demand has resulted in interconnecting
more and more powerful systems all over the world. This huge interconnection has
made the power system much more vulnerable to frequency changes in case of

S. Sinha
Department of Electrical Engineering, Institute of Technical Education and Research, Siksha ‘O’
Anusandhan (Deemed to be University), Bhubaneswar 751030, India
R. K. Mallick (B)
Department of EEE, Institute of Technical Education and Research, Siksha ‘O’ Anusandhan
(Deemed to be University), Bhubaneswar 751030, India
e-mail: [email protected]

© Springer Nature Singapore Pte Ltd. 2020 89

G. Pradhan et al. (eds.), Advances in Electrical Control and Signal Systems,
Lecture Notes in Electrical Engineering 665,
https://doi.org/10.1007/978-981-15-5262-5_7
90 S. Sinha and R. K. Mallick

sudden load demand. The power systems should be able to mitigate the deviations
in frequency as fast as it can and keep it under nominal values for maintaining the
stability and the reliability of the power system. This is efficiently carried out by
the automatic generation control. Automatic generation control acts as an ancillary
service and ensures that the frequency deviations caused by sudden load disturbances
is contained within suitable values and the load demand is met so that the system
stability is restored. Extensive literature survey has been conducted to get a clear
conception of the role of AGC in maintaining power system reliability. The concept
of automatic generation control was first mentioned in [1, 2]. Discussions were made
about a suitable frequency control technique for conventional and distribution power
systems in [3].
Another growing concern with increasing population is the depletion of conven-
tional sources of power and an alarming rate of global warming posing an alarming
scenario to environmentalists and scientists all over the world. It is at this juncture
that people all over the world started harnessing alternative sources of power to
protect the conventional power sources and to save global warming and reduce it.
The automobile sector has also come forward in the use of renewable fuels and has
replaced conventional vehicles with electric cars. These cars draw power from the
grid in times of need and after a day’s trip, it sends the unused power back to the
grid. This integration of electric vehicles in the grid is of a delicate issue and is a lot
more dependent on system stability. Paper [4] proposes the conceptual framework
which explains the science behind V2G technology. In paper [5], effort has been
made to model the V2G successfully and to study the benefits it has on a power sys-
tem. In paper [6] this approach is extended to smart grids and various optimizations
techniques for obtaining the optimal performances are studied. Paper [7] reviews the
impact this V2G technology has on the power system.
The modern-day power sector has been decentralized, unlike the previous days
when each and every unit of the power system were to be connected to and con-
trolled by a single entity called Vertical Integrated Utility (VIU). Decentralization
or Deregulation of power system ensures that the distribution companies have the
liberty to come into contract with the generation companies or IPPs (Independent
power producers) of different areas thereby selling and buying power units at a more
economical price. Therefore, there is a huge decentralization of control in a dereg-
ulated environment and it is the ISO (Independent service operator) responsible for
monitoring the frequency disturbance in the system and minimizing the tie line power
flow.
It has been utmost sure from a literature survey that the performance of the system
in maintaining the frequency balance depends on the control technique and the opti-
mal gain values of the controller. A varied number of controllers and optimization
technique for tuning their gains has been proposed in the literature survey both for
conventional as well as the deregulated environment. Paper [3] consists of an exten-
sive literature survey on the load frequency control for conventional and distributed
generation sources. A detailed review of the varied performance of different classical
controllers employed in an automatic generation is conducted effectively in [8]. In
paper [9] a model predictive based load frequency control of an interconnected power
Frequency Regulation of Electric Vehicles Integrated … 91

system under a restructured scenario is proposed. Paper [10] proposed a reinforce-

ment type learning of the controllers for the automatic generation control of an inter-
connected power system. Use of neural network strategy for bringing about the load
frequency control is also studied in [11]. Application of firefly algorithm to tune the
controller gains for the LFC of an interconnected power system was effectively done
in [12]. Paper [13] effectively discusses the optimized multi-area AGC simulation
for a restructured environment. A similar scenario was also taken for consideration
and in paper [14] the authors showcased the load frequency control technique in an
interconnected power system under deregulated environment. In paper [15] use of
local controllers was taken to address the LFC issues in a deregulated environment. A
new type of control theory named the Disturbance Accommodation control theory is
proposed for the load frequency control under decentralized market scenario in [16].
Paper [17] put forward an optimal gravity search optimization scheme for the AGC
of an interconnected power system. The traditional controller suffered from control
constraints as they were predominantly one degree of freedom in nature. Paper [18]
put forward a 2 DOF PID controller for the LFC of an interconnected power system
inclusive of governor dead band and non-linearity. Hybridization of two different
optimization schemes was brought about in [19] and coordinated design of PSS and
SSSC based controller is proposed for the AGC of an interconnected power sys-
tem. Paper [20] basically gives us a detailed review of the various types of classical
controllers that are being put to use for the AGC of an interconnected power sys-
tem. Bacteria foraging algorithm of optimizing controller gains has found its maiden
application in the AGC of a hydrothermal power system in [21].
The main objective behind this research work is
• Design of a two area power system and with the integration of bidirectional electric
vehicles.
• Effort has been made to design dual mode Proportional-Integral-derivative
controller for minimizing the frequency deviations.
• The controller gains are tuned with the help of Dragonfly algorithm.
• The dynamic responses of the system are to be investigated and comparison has
to be done between PID and 2 DOF PID controller.
• The system performance is investigated under step load disturbances for both
charging and discharging case of the electric vehicles under a deregulated
environment.

2 System Modeling

The proposed research work considers a two area power system of a system rating of
2000 MW respectively. The linearized system model is described in Fig. 1. The two
area power system consists of a thermal reheat power system with 3% Generation
rate constraint and considerable governor dead band in both the areas. There is
also the inclusion of a gas turbine power plant in both areas. Due to depletion in
92 S. Sinha and R. K. Mallick

Fig. 1 Linearized model of the system considered

natural resources worldwide, a global attempt has been made to cut down the use of
petroleum and coal as primary fuels and replacing with renewables. The automobile
sector over the past few years has attempted to bring about a revolutionary change
and has been successful in implementing electric vehicles which now stands out as
a future to the automobile industry. The research work has therefore considered an
electric vehicle lumped model to study the effects on the power system frequency
during charging and discharging. The entire power system is to be analyzed under
a deregulated environment. In a deregulated environment, frequency stability is a
growing concern.
Owing to a large amount of interconnections and a huge rate of exchange of
power, a slight disruption in the operating frequency will make the DISCOs and
the GENCOs fall out of the market and finally lead to system instability. Automatic
Generation Control thus plays a vital role in minimizing frequency deviations. This
is done by minimizing the Area Control Error to zero in the shortest time possible.
Frequency Regulation of Electric Vehicles Integrated … 93

For a normal power system, ACE can be mathematically expressed as:

ACEi = βi  f i + Ptieerrori (1)

where i indicates the respective areas taken into consideration in a power system. As
visible from the above expression, the actual value of the ACE is dependent on the
tie line power error which can be mathematically expressed as:

Ptieerror = Ptieactual − Ptiescheduled (2)

where the actual tine power flow can be written as

2π T12
Ptieactual = ( f 1 −  f 2 ) (3)
s
and the scheduled tie line power flow is mathematically denoted as:


2 
4 
4 
2
Ptiescheduled = cpfi j PL j − cpfi j PL j (4)
i=1 j=3 i=3 j=1

where cpf stands out for the contract participation factor in deregulated market
scenario

3 Modeling of Electric Vehicle

The use of plug-in hybrid electric vehicles plays a considerable role in compensating
additional real power requirements where the traditional frequency control reaches
its limitation. Hybrid electric vehicles are characterized to have a much smaller time
constant than the turbine or the governor of a traditional thermal power plant and are
hence faster in response time. They are basically important in damping the power
oscillation and frequency deviation as fast as possible. Added to the hybrid vehicles,
the role of the governor and the turbine is to eliminate the steady-state frequency
error. Since in the future a large number of electric vehicles are expected to run on
streets the proposed research work has thereby considered a lumped electric vehicle
model for studying its role in maintaining the frequency under nominal limits. The
lumped model is depicted in Fig. 2. The lumped model takes the LFC signal as its
input and the discharging power of one EV is considered as the input. The battery
output capacity of the battery is considered as ±B KW. The charge present in the
battery at the initial instant is denoted by E. This battery charge can be controlled
within limits of E max and E min . K1 and K2 are the calculated energy differences
from the maximum and minimum levels respectively. The electric vehicles do not
take part in frequency regulation when the charge in the battery is above 90% and
94 S. Sinha and R. K. Mallick

B KW
B KW K2>0
K1>0
ΔUe 1 0
1 + sT 0

-B KW K1<0 K2<0

-B KW

Emax
-
K1 Emin _ K2

+ +

TOTAL ENERGY MODE

Energy of one EV

Fig. 2 Lumped model of electric vehicle system taking part in frequency regulation

below 80%. The total power output of the electric vehicles depends on the number
of electric vehicles participating in the frequency regulation.
The total power that a V2G can provide to the power system strictly depends upon
three factors namely (a) the current carrying capacity of the wires (b) the stored energy
in the vehicle and the rate at which the energy is used and (c) the rated maximum
power of the vehicles power electronics. The lowest values of the three mentioned
factors usually indicate the power capacity of the V2G. The mathematical formula
prepared for calculating the power limitation of an electric vehicle for on-board
energy storage is given below
 
dd +drb
Es − ηvehicle
ηinv
Pvehicle = (5)
tdis

where Pvehicle stands for the maximum power rating of the V2G. E s denotes the total
DC energy stored in batteries. d rb stands for the buffer range defined by the driver
for getting the idea for minimum reserve required for an unplanned trip. Db denotes
the total distance in miles driven by the car. ηvehicle stands for the vehicle efficiency
in miles per Kwh ηinv stands for the inverter electrical power conversion efficiency
from DC to AC t dis signifies the total time in which the stored energy of the electric
vehicle is dispatched in hours. The transfer function model of the hybrid electric
vehicle is given in Fig. 3.
The V2G power droop characteristic against frequency deviation can be mathe-
matically expressed as

Fig. 3 Transfer function output

input
model of the hybrid electric 1
vehicle Δf 2 T phev s + 1
Frequency Regulation of Electric Vehicles Integrated … 95
   
K v2g f  K v2g ≤ Pmax

Pv2g = (6)
Pmax Pmax <  K v2g 

where K v2g stands for the vehicle gain which is determined by taking a trade-off
between v2g and the battery state of charge (SOC) deviation into consideration.
Pmax determines the maximum power demanded or dissipated by a 200 V/25 A v2g
outlet.
n
SOC − SOClow(high)
K v2g = K max 1 − (7)
SOCmax(min) − SOClow(high)

where n stands for the design parameters and SOC (min) and SOC (max) stands for
the minimum and maximum state of charge of the v2g battery respectively. K max and
K min indicate the maximum and minimum gains of V2G electric vehicles.

4 Controller Proposed

The proportional-integral controller has served as the preferred controller throughout

industries and for engineers owing to its simplicity in structure and reliability. The
PID controller does not demand higher skill levels for its operation. Proportional-
integral-derivative controller consists of three basic modes, i.e., proportional, inte-
gral, and differential mode modes. The proportional controller improves the closed-
loop system gains, thereby reducing rise time and hence improving the transient
system performance of the system but fails in eliminating the steady-state errors.
The integral controller helps in reducing the steady-state error to zero and thereby
has a slower response time during the transient period. The differential controller
helps in improving the transient characteristics and minimized the overshoot and
undershoots and proves effective in bringing the system deviations to the steady-
state in the fastest time possible. Using the proportional and the integral controller
simultaneously fails to differentiate between static and dynamic accuracy. The use of
proportional reduces the rise time but fails to damp the system whereas the integral
controller reduces settling time but has no control over the rise time. Figure 4 gives
us the detailed aspect of a traditional PID controller.
This anomaly in the operation is solved by using the concept of dual mode con-
troller. The dual mode controller effectively switches between the proportional and
the integral controllers depending on the magnitude of the error. The magnitude of the
error controls the priority of the following control actions as mathematically stated
below. In case of a larger rate of change of error, the proportional control comes to
dominance and reduces the system oscillations to a large extent.

dACE
> |ε| (8)
dt
96 S. Sinha and R. K. Mallick

PROPORTIONAL
Kp

+
ACE 1 U(t)
Ki s +
INTEGRAL +

du
Kd dt

DERIVATIVE

Fig. 4 Traditional PID controller

Then the proportional controller becomes dominant as clear from the mathemat-
ical expression

d
Pc = K p (ACE) + K D (ACE) (9)
dt
And when the change in error is small than the threshold value, the proportional
controller need not be active and then the main concern is to minimize the steady-state
error. So the controller is shifted to an integral part as described below:

dACE
< |ε| (10)
dt
Then,

d
Pc = K i (ACE) + K D (ACE) (11)
dt

The differential controller is used in both cases to minimize the transient

oscillations of the system. Figure 5 explains the dual mode concept in detail.
The performance of the controller mainly depends on the values of the controller
gains. These values generally hint to a control action around a fixed operating point.
But due to sudden disturbances in the system, the operating points get changed and
it becomes a tedious task to reach the optimum gain values of the controller by a
traditional hit and trial method. In order to solve this problem, help has been taken
of optimization techniques. These techniques being explorative and exploitative in
nature provide optimum values to the controller gains with a change in operating
Frequency Regulation of Electric Vehicles Integrated … 97

INTEGRAL

d Ki
ACE
s
sK d
ACE dt
+ U(t)
d
dt
ACE
Kp sK d +

PROPORTIONAL

Fig. 5 Dual mode controller

points. In this work, an explorative technique called Dragonfly algorithm is applied

for optimal tuning of the controller gains

5 Optimization Scheme

Dragonfly algorithm is motivated by the movement and the behavior of the drag-
onflies [23]. They use their unique and superior swarming technique hunting and
migration. They are small predatory insects that hunt small insects. The hunting
behavior of the dragonflies is called static (feeding swarm), and the migrating behav-
ior is called dynamic (migratory) swarm. The difference between the static and the
dynamic swarm is that in the static swarm dragonflies in small groups and over a
small area hunt other flying rapine like mosquito and butterflies. The key characteris-
tics of a static swarm are the local movements and sudden changes in the flying path.
However, the dynamic swarms, the dragonflies in the large number make the swarm
one direction over the long distances for migration purposes. The main inspiration for
the algorithm is these two types of swarming behaviors depict the two main phases of
optimization: exploration and exploitation. The behavior of swarms mainly follows
fire basic rules: separation consistency, affinity, alignment, near a food source, and
distraction from the foe sources.
Isolation (S i ) is defined as the stationary clashing prevention of one form from
other forms in the vicinity.
Alignment (Ai ) is termed as the velocity equity of individuals to others in the
locality.
Cohesion (C i ) indicates an inclination of particles at the mass’ center of the region.
Appropriate weight is given to operators individually and is adjusted for turning
to make certain of the meeting of dragonflies towards best result
The mathematical formulation of DA can be described as:
Let us assume the population of dragonflies of magnitude N. The point of ith
dragonfly can be mentioned as:

X i = (X i1 , X di , . . . , X i N )
98 S. Sinha and R. K. Mallick

where i = 1, 2, 3, …, N, X id correlates to the instance of the ith dragon fly in dth

dimensional attribute of the search space and N is the number of causalities.
The separation (S i ) can be calculated as follows:


N
Si = − X − Xj (12)
j=1

where X is the position of the current individual, X j shows the position jth neighboring
individual, and N is the number of neighboring individuals.
Alignment is calculated as follows:
N
j=1 Vj
Ai = (13)
N
The velocity of jth neighboring individuals is shown by X j .
The cohesion is calculated as follows:
N
j=1 X j
Cj = −X (14)
N
where X is the state of the current individuals, N is the number of localities, and X j
shows the point jth neighboring individual.
Affinity towards the food’s origin is calculated as follows:

Fi = X + − X (15)

Here X is the position of the present individuals, and X+ conveys the original
place of the food source.
Deviation outwards an enemy is calculated as follows:

Ei = X − + X (16)

6 Use of Dragonfly Algorithm in the Proposed Problem

The proposed research work has made use of Dragonfly algorithm to optimize the
gain values of the controller for providing an optimal solution. The controller accepts
the Area Control Error (ACE) as its input and returns a suitable control action that
tries to minimize the ACE. An objective function is taken into consideration for
minimization purpose in order to return the minimum value of ACE. The below-
mentioned objective functions are incorporated in the proposed Dragonfly algorithm
and their performances is tabulated in Table 1. Table 2 lists the optimal gain values
Frequency Regulation of Electric Vehicles Integrated … 99

Table 1 Objective function values for base case and bilateral transactions
Market scenario Objective function
ITAE IAE ITSE ISE
Base case 1.9196 0.6657 0.0567 0.0145
Bilateral transaction 1.9867 0.7232 0.0645 0.0206

Table 2 Optimized controller parameters tuned by DA technique

Controller Gain values
Base case scenario
Kp1 Ki1 Kd1 Kp2 Ki2 Kd2
PID 1.1882 0.7416 0.3695 0.3202 0.8661 0.1099
DUAL PID 0.3555 1.5164 1.2684 1.7124 0.9748 1.9714
2 DOF PID 0.6817 0.3162 0.2310 0.2977 1.2268 1.2364
Bilateral scenario
Kp1 Ki1 Kd1 Kp2 Ki2 Kd2
PID 1.8956 1.1969 0.9592 0.7724 0.7911 1.8904
DUAL PID 0.6346 1.5571 1.5472 0.8450 0.8573 1.5293
2 DOF PID 0.2241 0.4452 1.7782 1.4129 1.3088 0.5927

of the controller attained by using the most appropriate objective function

t
IAE = (| f 1 | + | f 2 | + |ptie |)dt (19)
0

t
  
ITAE =  f f  + | f 2 | + |ptie |tdt (20)
0

t
 2 
ISE =  f 1 +  f 22 + ptie
2
dt (21)
0

t
 2 
ITSE =  f 1 +  f 22 + ptie
2
tdt (22)
0
100 S. Sinha and R. K. Mallick

7 Results and Analysis

The proposed two area system incorporated with lumped electric vehicles in area
2 is simulated under deregulation environment. The gains of the newly proposed
dual PID controller tuned with the help of Dragonfly algorithm using Integral Time
Absolute Error as the objective function as it gave the least value among the rest
as stated in Table 1. The system is put to simulation under two deregulated market
scenarios. In each case the performances of the dual mode controllers are compared
with the traditional controller and the dynamic system performance is studied for
establishing system supremacy.

7.1 Case 1: Base Case

The proposed system is simulated under base case market scenario. The analysis
is carried out by considering both charging and discharging cases for the electric
vehicle. For base case market scenario, it is necessary that the DISCOs of an area
enter in a contract with the GENCOs of the same area and is not entitled to enter into
inter-area contract. The distribution participation matrix DPM can be mathematically
expressed as:
⎡ ⎤
0.5 0.5 0 0
⎢ 0.5 0.5 0 0⎥
DPM = ⎢
⎣ 0
⎥ (23)
0 0 0⎦
0 0 0 0

Charging mode:
The analysis is first done taking into consideration that the electric vehicles accept
power from the grid for charging purposes. So, each electric vehicle can be interpreted
as a small p.u load demand which disturbs the standard working frequency of the
system. The role of the dual controller in maintaining the system frequency within
nominal values is then studied and the system dynamic performances are analyzed.
The system dynamic parameters like the settling time, maximum overshoot and
minimum undershoot are analyzed and noted down in Table 3a, b. The value clearly
explains the superiority of the proposed dual controller over their standard forms.
The plots Figs. 6, 7 and 8 clearly establish the efficacy of DUAL PID over traditional
PID form and 2 DOF PID controller.
Discharging mode:
In this case, the electric vehicles at the end of the day return the excess power to the
grid. In the discharging case, each and every electric vehicle acts as a small source of
electric power fed to the grid. They play a role in minimizing the system frequency
deviations during sudden load changes when applied to any area of a power system.
Frequency Regulation of Electric Vehicles Integrated … 101

Table 3 a Dynamic performance of the proposed system during charging for both bilateral and
base case scenario. b Dynamic performance of the proposed system during discharging mode in
deregulated scenario (bilateral market and base case)
System performance Settling time (T s ) Max. overshoot (U max ) Min undershoot (U min )
parameters
Bilateral (charging condition)
PID f 1 11.7443 0.4869 −0.0017
f 2 12.5410 0.2353 −0.0020
ptie 16.5951 0.0928 −0.0097
DUAL PID f 1 11.4003 0.4304 −0.0032
f 2 7.3710 0.2158 −0.0026
ptie 11.2300 0.0903 −0.0285
2 DOF PID f 1 13.8716 0.2007 −0.0277
f 2 19.6927 0.1068 −0.0098
ptie 20.8468 0.1208 −0.0163
Base case (charging condition)
PID f 1 11.5582 0.4869 −0.0012
f 2 12.9604 0.2353 −0.0014
ptie 17.3600 0.0928 −0.0073
DUAL PID f 1 11.7066 0.4304 −0.0203
f 2 7.7915 0.2158 −0.0021
ptie 11.6313 0.0903 −0.0235
2 DOF PID f 1 14.1838 0.2207 −0.0277
f 2 20.3038 0.0940 −0.0088
ptie 20.6690 0.1368 −0.0153
Bilateral market (discharging condition)
PID f 1 10.5615 0.4869 0.0099
f 2 11.3300 0.2816 −0.0075
ptie 17.9608 0.0908 −0.0021
DUAL PID f 1 9.3996 0.4304 −0.0292
f 2 12.1765 0.2726 −0.0331
ptie 15.2989 0.0850 −0.0071
2 DOF PID f 1 13.6655 0.2207 −0.0356
f 2 12.4948 0.1202 −0.0104
ptie 22.4314 0.1115 −0.0134
Bilateral market (discharging condition)
PID f 1 9.3279 0.4869 −0.0100
f 2 6.9047 0.2816 −0.0033
ptie 18.0962 0.0908 −0.0026
(continued)
102 S. Sinha and R. K. Mallick

Table 3 (continued)
System performance Settling time (T s ) Max. overshoot (U max ) Min undershoot (U min )
parameters
DUAL PID f 1 12.6534 0.4304 −0.0450
f 2 12.5189 0.2726 −0.0423
ptie 15.0104 0.0850 −0.0072
2 DOF PID f 1 14.0228 0.2207 −0.0352
f 2 18.1511 0.1005 −0.0092
ptie 21.4854 0.1295 −0.0123

Fig. 6 Frequency deviation in area 1 with electric vehicle charging mode

Fig. 7 Frequency deviation in area 1 with electric vehicle charging mode

The plots Figs. 9, 10, and 11 clearly establish the efficacy of DUAL PID over
traditional PID and standard 2 DOF PID controller The dynamic system parameters
like settling time, maximum overshoot, and minimum undershoot is also analyzed
Frequency Regulation of Electric Vehicles Integrated … 103

Fig. 8 Frequency deviation in area 1 with electric vehicle charging mode

Fig. 9 Frequency deviation in area 1 with electric vehicle discharging mode

Fig. 10 Frequency deviation in area 2 with electric vehicle discharging mode

104 S. Sinha and R. K. Mallick

Fig. 11 Tie line power deviation with electric vehicle discharging mode

and tabulated in Table 3a, b. The value clearly explains the superiority of the proposed
DUAL controller technique over its traditional forms.

7.2 Bilateral Market Scenario

A similar analysis is also executed in a bilateral market scenario. The term bilateral
scenario indicates that the DISCOs come in contract with the GENCOs not only in the
same area but also in the adjoining areas and the power exchange takes place through
tie lines. The contract statement for bilateral market is mathematically denoted as
⎡ ⎤
0.5 0.25 0 0.3
⎢ 0.2 0.25 0 0 ⎥
DPM = ⎢
⎣ 0
⎥ (24)
0.25 1 0.7 ⎦
0.3 0.25 0 0

The analysis, in this case, is done for both the charging and discharging conditions
of the electric vehicles. As already stated when the vehicle charges, it behaves as a
sudden load disturbance which gives rise to frequency deviations which are to be
minimized and maintained within nominal values. The Dragonfly algorithm tuned
DUAL controllers prove to be very effective in maintaining system stability as clearly
depicted in the following plots:
The plots Figs. 12, 13 and 14 clearly establish the efficacy of DUAL PID over
traditional PID and 2 DOF PID controller The dynamic system performance during
charging mode in terms of settling time, maximum overshoot, and minimum under-
shoot is analyzed and tabulated as Table 3a, b. It can be effectively inferred that the
DUAL mode controllers have a superior performance over the traditional forms.
Discharging mode:
As already stated above, the excess power from the vehicles is fed to the grid. This
extra power plays an important role to meet the load demands and help AGC in
maintaining system stability.
Frequency Regulation of Electric Vehicles Integrated … 105

Fig. 12 Frequency deviation in area 1 during charging condition

Fig. 13 Frequency deviation in area 2 during charging condition

Fig. 14 Tie line power deviation during charging conditions

The plots Figs. 15, 16 and 17 clearly establish the efficacy of DUAL PID over
traditional PID form and 2 DOF PID controller The system dynamic performances
are to be studied in terms of settling time, maximum overshoot minimum undershoot
and tabulated in Table 3a, b. The values clearly indicate the superiority of the dual
controllers in minimizing frequency deviations in a shorter time possible. It has been
clearly observed that the performance of the DUAL PID is better than that of PID
controller and 2 DOF PID controllers
106 S. Sinha and R. K. Mallick

Fig. 15 Frequency deviation in area 1 during discharging condition

Fig. 16 Frequency deviation in area 2 during discharging condition

Fig. 17 Tie line power deviation during discharging conditions

Frequency Regulation of Electric Vehicles Integrated … 107

8 Conclusion

In the proposed research work, effort has been taken to maintain the frequency devi-
ations and the tie line power variances within nominal limits with the inclusion of
electric vehicles which take power from the grid during charging and again reverts
back the excess power to the grid when not in use. This power exchange between
the vehicles and the grid may allow the system frequency to deviate and the role
of AGC in minimizing these deviations is therefore analyzed. Effort has been made
to incorporate the DUAL mode of controller by taking PID and 2 DOF PID con-
trollers into reference. The gains of the DUAL controllers are tuned with the help
of Dragonfly algorithm. The system performance has been studied under a step load
perturbations for both charging and discharging cases. The system is analyzed for
deregulated market scenario where frequency stability is an important issue. The
dynamic performance in terms of settling time, maximum overshoot and minimum
undershoot is also studied and analyzed to determine the superiority of the proposed
DUAL mode controllers over the traditional values.

References

1. Elgerd, O.: Electric Energy Systems Theory: An Introduction (1982)

2. Bevrani, H.: Robust Power System Frequency Control, vol. 85. Springer, New York (2009)
3. Pandey, S.K., Mohanty, S.R., Kishor, N.: A literature survey on load–frequency control for
conventional and distribution generation power systems. Renew. Sustain. Energy Rev. 25,
318–334 (2013)
4. Deivanayagam, R.: Vehicle-to-grid technology: concept, status, and challenges. J. Undergrad-
uate Res. Univ. Ill. Chicago 10(1) (2017)
5. Guille, C., Gross, G.: A conceptual framework for the vehicle-to-grid (V2G) implementation.
Energy Policy 37(11), 4379–4390 (2009)
6. Ma, Y., et al.: Modeling the benefits of vehicle-to-grid technology to a power system. IEEE
Trans. Power Syst. 27(2), 1012–1020 (2012)
7. Yilmaz, M., Krein, P.T.: Review of the impact of vehicle-to-grid technologies on distribution
systems and utility interfaces. IEEE Trans. Power Electron. 28(12), 5673–5689 (2013)
8. Saikia, L.C., Nanda, J., Mishra, S.: Performance comparison of several classical controllers in
AGC for multi-area interconnected thermal system. Int. J. Electr. Power Energy Syst. 33(3),
394–401 (2011)
9. Mohamed, T.H., et al.: Decentralized model predictive based load frequency control in an
interconnected power system. Energy Convers. Manage. 52(2), 1208–1214 (2011)
10. Ahamed, T.P.I., Nagendra Rao, P.S., Sastry, P.S.: A reinforcement learning approach to
automatic generation control. Electr. Power Syst. Res. 63(1), 9–26 (2002)
11. Chaturvedi, D.K., Satsangi, P.S., Kalra, P.K.: Load frequency control: a generalised neural
network approach. Int. J. Electr. Power Energy Syst. 21(6), 405–415 (1999)
12. Padhan, S., Sahu, R.K., Panda, S.: Application of firefly algorithm for load frequency control
of multi-area interconnected power system. Electr. Power Compon. Syst. 42(13), 1419–1430
(2014)
13. Bhatt, P., Roy, R., Ghoshal, S.P.: Optimized multi area AGC simulation in restructured power
systems. Int. J. Electr. Power Energy Syst. 32(4), 311–322 (2010)
108 S. Sinha and R. K. Mallick

14. Parmar, K.P.S., Majhi, S., Kothari, D.P.: LFC of an interconnected power system with multi-
source power generation in deregulated power environment. Int. J. Electr. Power Energy Syst.
57, 277–286 (2014)
15. Abraham, R.J., Das, D., Patra, A.: Load following in a bilateral market with local controllers.
Int. J. Electr. Power Energy Syst. 33(10), 1648–1657 (2011)
16. Sedghisigarchi, K., Feliachi, A., Davari, A.: Decentralized load frequency control in a dereg-
ulated environment using disturbance accommodation control theory. In: Proceedings of the
Thirty-Fourth Southeastern Symposium on System Theory, 2002. IEEE (2002)
17. Sahu, R.K., Panda, S., Padhan, S.: Optimal gravitational search algorithm for automatic
generation control of interconnected power systems. Ain Shams Eng. J. 5(3), 721–733 (2014)
18. Sahu, R.K., Panda, S., Rout, U.K.: DE optimized parallel 2-DOF PID controller for load
frequency control of power system with governor dead-band nonlinearity. Int. J. Electr. Power
Energy Syst. 49, 19–33 (2013)
19. Panda, S., Yegireddy, N.K., Mohapatra, S.K.: Hybrid BFOA–PSO approach for coordinated
design of PSS and SSSC-based controller considering time delays. Int. J. Electr. Power Energy
Syst. 49, 221–233 (2013)
20. Mohanty, B., Hota, P.K.: Comparative performance analysis of fruit fly optimisation algorithm
for multi-area multi-source automatic generation control under deregulated environment. IET
Gener. Transm. Distrib. 9(14), 1845–1855 (2015)
21. Saikia, L.C., Sinha, N., Nanda, J.: Maiden application of bacterial foraging based fuzzy IDD
controller in AGC of a multi-area hydrothermal system. Int. J. Electr. Power Energy Syst. 45(1),
98–106 (2013)
22. Shi, Y.: Particle swarm optimization: developments, applications and resources. In: Proceedings
of the 2001 Congress on Evolutionary Computation, 2001, vol. 1. IEEE (2001)
23. Mirjalili, S.: Dragonfly algorithm: a new meta-heuristic optimization technique for solving
single-objective, discrete, and multi-objective problems. Neural Comput. Appl. 27(4), 1053–
1073 (2016)
Photovoltaic Power Forecasting
by Evolutionary Algorithm-Based
Improved Extreme Learning Machine

Alok Kumar Pani and Niranjan Nayak

Abstract In order to improve the grid energy management capability in a smart grid
environment, this work aims to apply modified extreme learning machine (ELM)
forecasting method for short-term prediction of PV power and overcomes some of
the major challenges during on-grid mode of operation. Solar power generation is
irregular and changeable in nature. The solar source is highly reliant on temperature
and other atmospheric factors. Integration of solar plant with conventional grid brings
the smart grid concept to the power engineering space. Now the major challenge
to grid supervision is accurate forecasting of photovoltaic power in a smart grid
with less cost. Thus, a modified extreme learning machine forecasting model is
designed for short-term prediction of solar power. Further to improve the accuracy
and stability of the system, the weights of ELM technique are optimized by runner and
root optimization technique. Results from the current study depict that the proposed
method performs better by providing a smaller forecasting error than other methods.
The modified extreme learning machine is applied to a real-time data set of a real-
time solar power plant, located at the roof top of an academic building, situated at
Bhubaneswar, Odisha, India, whose geographical location is given in Table 1. Also a
new optimization technique is applied for its weight optimization which minimizes
the forecasting error.

Keywords PV array · Extreme learning machine · PSO-ELM · CPSO-ELM ·

RRA-ELM

1 Introduction

Use of fossil fuel in conventional power generation created serious ecological prob-
lems such as the production of huge carbon dioxide. Photovoltaic energy is a clean

A. K. Pani (B) · N. Nayak

Institute of Technical Education and Research, Siksha ‘O’ Anusandhan (Deemed to be
University), Bhubaneswar, Odisha 751030, India
e-mail: [email protected]

© Springer Nature Singapore Pte Ltd. 2020 109

G. Pradhan et al. (eds.), Advances in Electrical Control and Signal Systems,
Lecture Notes in Electrical Engineering 665,
https://doi.org/10.1007/978-981-15-5262-5_8
110 A. K. Pani and N. Nayak

renewable energy which can replace the fossil fuel in power generation. Neverthe-
less, the integration of renewable energy resource is most challenging such that its
reliability, to be a major part of national power generation, will be marginal. But irre-
spective of production cost, it has been increasing speedily in past two decades. The
PV power generation is stochastic because of unexpected weather conditions, which
affect the steadiness of the source. Power system supervision, thus short-length PV
power prediction is needed for assurance of grid quality and energy management
[1–5]. Generally, there are short-term (less than 48 h) and long-term forecasting in
solar power prediction depending on the forecasting horizon. There are many types
of forecasting techniques out of which, numerical approaches are most appropriate
methods for short-term forecasting. This approach is statistical connection of linked
past data or numerical weather prediction (NWP) data to the short future and is widely
used in PV power forecasting in short duration. This type of forecasting includes time
series-based artificial neural network (ANN) and many other such techniques [6–8].
The autoregressive integrated moving average (ARIMA) forecasting method takes
care of effect of weather change in the PV power generation which is discussed in the
literature [9, 10]. A partial functional linear regression approach, which is suitable
for PV power forecasting of time 1 day, is implemented in the study of the literature
[11]. Historical time series forecast precision in steady weather condition and its
accuracy reduce with change in weather condition. Artificial neural network (ANN)
is most suitable machine learning technique for short-term prediction of PV power
which takes care of the complex nonlinear relationship between input and output. A
hybrid data set composed of local geographical data and capacity of adjacent photo-
voltaic cell is considered as the input vectors to nonlinear autoregressive exogenous
(NARX) neural network [12]. A modified and improved back-propagation neural
network based on genetic algorithm was proposed for photovoltaic power prediction
connected to IEEE-30 bus system is proposed in [12–14]. In the recent past, most
of the studies in PV power prediction focused on forecasting horizon from 1 to 24 h
duration. In the reference [15, 16], the authors forecasted the cloud movement by
trapping the sky images. The forecasting technique uses the irradiance sensor array
in the region of the solar plant but they need extra equipment and have higher thresh-
old. In all above research works, the weights of the forecasting models are randomly
selected which affects the exactness of the model [17]. But in the above papers, the
weights of the ELM are selected randomly. The proper selection of weights also
affects the capability of ELM model. Thus in the literature [18, 19], the PSO and
sine–cosine algorithm are used toward weight optimization in the ELM technique.
In [20], genetic algorithm is used obtain optimized ELM weights. Further, the opti-
mization techniques do not provide the ultimate values. Still the selection of proper
weights is a challenging task.
In this paper, a much widely used forecasting model known as extreme learning
machine is used to predict the photovoltaic power generation for short time length like
15, 30 min and 1 h time horizons. The weights of ELM are optimized by the runner
root algorithm (RRA), and the results are matched with the PSO-ELM and CPSO-
ELM. Unsystematic selection of weights in the input layer and bias in the hidden
layer affects seriously the performance of ELM. The challenges like uncertainty in
Photovoltaic Power Forecasting by Evolutionary Algorithm-Based … 111

output and over fitting are still needed to be taken care. Thus, the input weights are
optimized by a modified PSO, and then the hidden layer weights are optimized, then
the system is simulated separately in each case, and the results are compared.
Regarding arrangement of this paper, the characteristics of solar forecasting, data
selection, mathematical analysis of photovoltaic model and the operating process are
explained in Sect. 2. Empirical mode decomposition is discussed in Sect. 3. ELM
and variants are discussed in Sect. 4. Different optimization techniques like PSO,
craziness PSO and runner root algorithm are explained in Sect. 5. The simulation
results and performance of projected model are presented in Sect. 6. At last, the
conclusion of the entire work described in Sect. 7.

2 Characteristics of Solar Forecasting and Data Selection

Taking many factors into account like intensity of solar radiation, weather condition,
ambient temperature, time, cloud and geographical location, the output of PV power
generation is fluctuating, and it is difficult to control. Also the PV output depends on
environmental conditions [11].
Because of meteorological reservations, the PV generation has tough cyclical
characteristics including the day time cycle and yearly cycle. The situation becomes
extremely challenging when PV generation is integrated with conventional grid for
solar power penetration. The duration of harvesting of solar energy is from 8.00 AM
to 5.00 PM. The working temperature of PV power plant also affects the conversion
efficiency. In general, the PV plant temperature is higher than the atmospheric tem-
perature. For better prediction, both the temperatures are assumed to be same. The
conversion efficiency of the PV cell is expressed as follows:
  
η = η0 1 − μ T p − Tμ (1)

where T p is the temperature corresponds to point P, Tμ is the reference temperature

(298 K), and η0 is the conversion efficiency at reference temperature. Considering
the above factors in the prediction horizons is strongly recommended for PV power
forecasting and energy management. Thus, out of various forecasting methods, the
extreme learning machine technique is used here for prediction of solar power.

2.1 PV Power Generation

The highest PV power yield is modeled by

Wv = η AI [1 − 0.05(t − 25)] (2)

112 A. K. Pani and N. Nayak

where
η conversion efficiency,
A area of solar panel in m2 ,
I solar radiation (kW/m2 ),
t outside air temperature (°C) [18].
Maximum value of power, voltage and current are Pmax = 0.106 W, Vmax =
18.453 V and Imax = 5.76 A, Isc = 6.11 A, Rsh = 1000  and Rs = 0.0001 ,
21.6 V, N s = 36 cells in series, N P = 1, cells in parallel.

2.2 PV Power Forecasting

The performance of the forecasting model may be influenced by the factors like
input parameters and the time horizon. There are a few major variables used as input
of prediction model but important ones are (1) past data of PV power production
and (2) previous descriptive variables, like meteorological factors, which consist of
horizontal irradiance, temperature, cloud, moisture content, wind speed and so on.
Different prediction horizons are selected as per the needs of decision making and
grid energy management, like short-term and long-term horizons.
To get accuracy, proper grid management and smooth functioning of power sys-
tem, these time horizons are suitable and helpful in PV power generating stations. The
main functions in generating stations are unit commitment, battery control and energy
trading etc. Thus, most of the research work focused on the recent development of
PV system and forecasting models.

2.3 Performance Estimation

In this work, various numerical measurements like RMSE, MAE and MAPE are
calculated using soft computing methods, and their mathematical expression are
given below.

1   
N
MAE = Pj − t j  (3)
N i=1
 
1   Pj − t j 
N
MAPE = × 100% (4)
N i=1 yj

1  2
N
RMSE = Pj − t j (5)
N i=1
Photovoltaic Power Forecasting by Evolutionary Algorithm-Based … 113

where Pj and t j represent present and predicted PV powers, respectively.

2.4 Data Processing

The input and output data are normalized by the use of following equation. The input
weights and bias are initially selected in the range of [0, 1], such that the training
speed and calculation overflow need not be considered.

x j − xmin
x∗ = ; 0 ≤ x∗ ≤ 1 (6)
xmax − xmin

Here, x max and x min are the upper and lower bound.

3 Empirical Mode Decomposition

Empirical mode decomposition (EMD) is a signal processing technique applied

to divide a signal into multiple component signals called intrinsic mode functions
(IMFs). The result of EMD also affects the stability and efficiency of the forecasting
model. Nevertheless, the presentation of the model improves by signal disintegra-
tion, and prediction accuracy becomes very high. The model asserts that the following
expression can describe any fraction of a fresh string of signals.


n
A(t) = βi(t) + remn (t) (7)
i=1

where β and remn (t) are IMFs and remainder of the signal in that order.
The decomposition process is a computational technique which followed iteration
method, and the splitted signal has a range of amplitudes and frequencies.
The constituent formation agrees to the conditions that the maxima or minima be
either same with the number of zero crossings or their difference is one, in the entire
data set. Other condition is that their average must be zero in the envelop.
The first term of IMF is described by

i 1 (t) = A(t) − γ1 (t) (8)

When A(t) is divided into many IMFs, IMF1 is the first term. If it is not, then it is
to be measured as the signal itself and split frequently. KIMF1 is taken to be an IMF
and is given as

σ1 = ki 1 (t) (9)
114 A. K. Pani and N. Nayak

This term (σ 1 ) is subtracted from A(t) by taking

A(t) − ρ1 (t) = λ1 (t) (10)

R is residue of the given signal and other IMFs are ρ1 , ρ2 , . . . , ρn .

Then

λ1 (t) − ρ2 (t) = γ (t) (11)

λn (t) − ρn (t) = γ (t) (12)

(11) and (12) are the altered version of Eq. (1)

The terminating state for this uneven procedure was recommended by Huag et al.,
and a normalized squared variation among the two consecutive changing operations
is essential.
T  2
 i n(k−1) (t) − i nk (t)
A 1 Dk = (13)
t=0
2
i n(k−1) (t)

This continuous method is ended by any pre-considered criteria, by adding all the
IMFs and the ending residue the signal An (t) can be restated as


m
An (t) = αni (t) + remnm (t) (14)
i=1

4 Extreme Learning Machine (ELM)

ELM is a machine learning procedure, to train a single layer feed forward network,
with arbitrary weights. The computational speed of ELM can be much higher than
conventional feed forward network learning algorithms like back-propagation (BP)
and some other neural networks. It is a high performed machine learning technique
can be applied to a system with small input data. Thus, the suggested learning algo-
rithm tends to have better presentation for feed forward neural networks. Due to
the advantages of such machine learning algorithm like easy implementation, least
training error, smallest weight norms and extremely faster speed over other con-
ventional neural networks, it is called ELM. In this algorithm, the input parameters
are selected randomly after computation, the system becomes linear, and the output
weights are computed analytically through Moore–Penrose generalized inverse of
the hidden layer output matrices. The structure of ELM is shown as in Fig. 1, and
the block diagram of PV system using ELM algorithm is shown in Fig. 2.
Photovoltaic Power Forecasting by Evolutionary Algorithm-Based … 115

Input Hidden Output

Layer Layer Layer

g(a,b,x)

x
t

Fig. 1 Structure of the SLFN model (ELM)

Irradiance
PV Power
Vpv Vdc Load
Array Converter
Temperature

Raw real time

PV deta Forecasted PV
power
Voltage & PWM Maximum
ELM
Current Sensing Generator PV power

Random weights
and bias of ELM Optimized
weights and bias
MPPT Algorithm + Duty Optimization to ELM
Cycle Adjustment technique

Fig. 2 Block diagram representation of proposed PV model using ELM

For m distinct sample, the parameters

 T
x j = x j1 , x j2 , . . . x jn ∈ R n (15)

and

yi = [yi1 , yi2 , . . . , yim ]T ∈ R m , (16)

usual SLFN with K hidden neurons and activation function, g, is expressed as

116 A. K. Pani and N. Nayak


K
 
α j · g a j xi + b j = ti , i = 1, . . . , N , (17)
j=1

where aj is the weight interconnecting inputs node with the jth hidden nodes, α j is
the weight interconnect output node with the jth hidden nodes, bj is the threshold of
the jth hidden nodes, and t i is the ELM yield for the ith data.
This network can estimate these N data samples with 0 error


N
Pi − ti  = 0. (18)
i=1

That is, bj , aj and α j exist such that


K
 
α j · g a j xi + b j = yi , i = 1, . . . , N , (19)
j=1

The above N equations can be expressed into a compact form:

Gβ = T (20)
 
where G = a j , . . . , a X , b j , . . . , b X , xi , . . . , x N
⎡ ⎤
g(a1 x1 + b1 ) . . g(a K x1 + b K )
⎢ . .. . ⎥
=⎢
⎣
⎥
⎦ (21)
. .. .
g(a1 x N + b1 ) . . g(a K x N + b K ) N ×K
⎡ ⎤ ⎡ ⎤
α1T t1T
⎢ . ⎥ ⎢ . ⎥
⎢ ⎥ ⎢ ⎥
⎢ ⎥ ⎢ ⎥
β=⎢ . ⎥ and T = ⎢ . ⎥ (22)
⎢ ⎥ ⎢ ⎥
⎣ . ⎦ ⎣ . ⎦
α KT K ×m t NT N ×m
 
In ELM, a j , b j remain unchanged once arbitrarily produced. To train the net-
work, it is simply corresponding to find a least square solution α̂ of the linear system
refer to Eq. (9), that is,
 
G α̂ − yi  = min Gα − yi  (23)
α

The smallest norm least square solution to Eq. (9) is

β̂ = G + T (24)
Photovoltaic Power Forecasting by Evolutionary Algorithm-Based … 117

where, H + is the Moore–Penrose generalized inverse of H, which can be computed

through orthogonal projection

G + = (G T G)−1 G T (25)

In general, the learning algorithms tend to attain minimum training error, but they
cannot, because of the local minimum or restriction of infinite training iterations.
The minimum norm least squares solution of Gβ = T is unique, which is

β̂ = G + T. (26)

4.1 Modified Extreme Learning Machine

In general, the resolution of a linear system is given

AZ = P (27)

where

A ∈ R m×n (28)

Z ∈ Rm (29)

In a Euclidean space, a SLFN with arbitrary weight may be treated as a linear

system, for ‘n’ samples (xi , pi )

where xi = [xi1 , xi2 , . . . , xm ]T ∈ R (30)

and pi = [ pi1 , pi2 , . . . , pm ]T ∈ R (31)

i.e., X i = [xi j ]T ∈ R (32)

and Pi = [ pi j ]T ∈ R (33)

The ELM can be modeled as


k
βi f (wti . x j + ci ) = y j , j = 1, 2, . . . , n (34)
i=1
118 A. K. Pani and N. Nayak

where wti = [wti1 , wti2 , . . . , wtin ]T is the input weight vector. It is input weight
vector, arbitrarily selected and coupled with ‘ith’ hidden neuron.
βi is represented as weight vector,
bi is sill assigned to ‘ith’ unseen neuron.
Equation (1) can be expressed as in the matrix form, as follows, where
⎡ ⎤
g(w1 x1 + b1 ) . . g(ak x1 + bk )
⎢ · . · . ⎥
Gn = ⎢
⎣
⎥
⎦ (35)
. . · ·
g(w1 xn + b1 ) . · g(ak xn + bk ) n×k

βn = [β1 , β2 , . . . , βk ] (36)

To keep ‘G’ as unchanged, the learning parameters are chosen randomly.

ELM can be trained as a least square solution. β̂ of the linear system.

G n βn = Pn (37)
 
 
G n (w1 . . . wm , b1 . . . bm )β̂n − Pn  = β̂n G n (w1 . . . wm , b1 . . . bm )βn − Pn  (38)

By the use of Moore–Penrose inverse matrix, theory output weight βn

β̂n = (G Tn G n )−1 G nT Pn (39)

Throughout the simulation work, the number of hidden nodes taken is 5.

4.2 Optimized Extreme Learning Machine (OELM)

In case of number of training samples and hidden neurons are same, then the network
can estimate error-free parameters. For a large data system with big number of hidden
neurons, it becomes extremely challenging for computation. Thus, it is needed to
select the approximate parameters to get outputs close to the pragmatic solution with
least error. So, it is advisable to train a SLFN with set input weights a j , bias b j along
with a single hidden layer, and the least square solution is calculated which helps to
reduce the error.
The optimized ELM algorithm improves the accuracy by applying optimization
algorithm like PSO, CPSO and runner root algorithm. Figure 3 shows flowchart of
the working of optimized ELM algorithm.
Photovoltaic Power Forecasting by Evolutionary Algorithm-Based … 119

Parameter
initialization

Initialize
population
Fitness calculation

ELM

Testing accuracy Update the

velocities
and positions
Update the
best position

Termination No
criterion

Yes

Output

Fig. 3 Flowchart of PSO optimizing ELM

5 Optimization Techniques

5.1 Particle Swarm Optimization

PSO is a sufficiently recognized sensible optimization method which has got very
adaptive features for its application and is capable of converging quickly to get
suitable solution. Also PSO is well competent to ensure working with huge search
space and an objective function which is non-differential.
PSO algorithm is developed to simulate assuming random movement of birds in
sky or fish in water. Velocity and position of each element are customized as per (40)
and (42), respectively.
 
velik+1 = w ∗ velik + η1 ∗ rand1 ∗ pbesti (k) − si (k)
 
+ η2 ∗ rand2 ∗ gbesti (k) − si (k) (40)

where

veli = velmax where veli > velmax

= velmin where veli < velmin (41)
120 A. K. Pani and N. Nayak

and

sik+1 = sik + vik+1 (42)

The step-by-step procedure of the PSO algorithm is presented as follows:

Step 1: The entire data sheet is normalized within the range [0, 1]
Step 2: The preliminary velocities of all particles are arbitrarily selected by
conveying the swarm dimension, final iterations and speed.
Step 3: The fitness value of the particle assigned as per the performance of ELM
and the best location of the particles are set according to the swarm’s maximum
fitness.
Step 4: Both the velocity and position of individual particle are updated as per
Eqs. (26) and (27).
Step 5: The termination conditions are verified, and if the utmost iterations are
not reached, then go back to step 3, or else move to the following step.
Step 6: The outcome is the best arrangement of (a, b) of ELM corresponding to
the highest fitness value.

5.2 Craziness PSO

There are certain short falls of the classical PSO which can be overcome by its mod-
ification. Craziness PSO is a modification the classical PSO particularly in velocity
expression. In classical PSO, the bird and fishes change direction suddenly. But that
is taken care by a craziness factor η in velocity equation of craziness PSO. The
velocity expression here in CPSO is given by

velik+1 = ran2 ∗ sign(ran3 ) ∗ velik

+ (1 − ran2 ) ∗ η1 ∗ ran1 ∗ {pbestik − sik }
+ (1 − ran2 ) ∗ η2 ∗ (1 − ran1 ) ∗ {gbestik − sik } (43)

where ran are the random parameter chosen whose values lie in [1, 0], sign(ran3 ) is
defined as

sign(ran3 ) = −1 ran3 ≤ 0.05

=1 ran3 > 0.05 (44)

ran1 , ran2 are two arbitrary parameters chosen separately. If both the values chosen
are big and have social experience, then the element will flutter from the neighboring
optimum. The optimization technique will have slow convergence for small values
of r1 , and r2 that will converge faster for big value of r 1 and the least value of (1 – r 1 ).
Photovoltaic Power Forecasting by Evolutionary Algorithm-Based … 121

The equilibrium of global and local search may be achieved by selection of another
random number r 2 . In some exceptional cases during changing the position for food
searching, a bird may not able to fly for food searching because of inertia. But the
element may fly to the probable region in opposite direction which is taken care by
sign (r 3 ).
The importance of CPSO is prior to updating this position the velocity of the
element is crazed by

velik+1 = velik+1 + p(ran4 )sign(ran4 ) ∗ velicraziness (45)

where is arandom variable in [0, 1], velicraziness is craziness velocity. velicraziness ∈

 minran4 max
veli , veli is the signum function and probability is defined as

p(ran4 ) = 1 ran4 ≤ pcr

(46)
= 0 ran4 > pcr

where pcr is probability craziness

sign(ran4 ) = 1 ran4 ≥ 0.5

(47)
= −1 ran4 < 0.5

Reverse flow of birds may not occur frequently for which r3 < 0.05 very small
values can be assigned and sign(ran3 ) = −1 is assumed in opposite direction.
Similarly, pcr ≤ 0.3 is chosen such that r4 will be assigned more chance, and lastly
it comes to more than pcr . p(ran4 ) which will be zero in majority cases. Otherwise,
there will be an unexpected oscillation in the convergence curve. velcraziness is selected
very minute values (=0.0001).

5.3 Runner and Root Algorithm

Plants like strawberry and spider elongate from one place to another, with the help
of that runner (also stolon). Plant generates a new daughter plant at the end of the
runner. This daughter plant generates further and gets converted to a new mother plant
after certain time. The series of growth events goes without hindrance. In this case,
activities of runner roots are modeled as the global and local search, respectively.
The plants grow the runners, distribute roots and pierce hairs in order to spread it to
find nutrients and water, which is equivalent to search techniques in an optimization
algorithm. Hence, it is derived that if a daughter plant traps a confined best possible
point, it grows new runners and roots and becomes an able parent.
This can be considered as an unobstructed optimization function as

min η(x), yl ≤ y ≤ yu (48)

122 A. K. Pani and N. Nayak

where η: R m ∈ R is ‘m’ cost function.

y ∗ = arg minη(x) ∈ R m is the most outstanding decision.
yl , yu ∈ R m are min and max values.
Each parent plant produces a daughter plant iteratively, similar to RRA. The
expression for the offspring is given as

1
ymother (i) k=1
k
ydaughter (i) = (49)
ymother (i) + drunner × rk k = 2, 3, . . . , P
k

where rk ∈ R m is a arbitrary vector within [−0.5, 0.5].

drunner is offspring and parents.
The cost function is tested at new daughter plant. At least one among these plant
grows considerably in the worth of the cost function matched to the best one in the
preceding iteration, i.e., mathematically if the expression
    
 
 min η ydaughter
k
(i) − min η ydaughter
k
(i − 1) 
 ≥l (50)
 k = 1, 2, . . . , P k = 1, 2, . . . , P 

suits, then the algorithm does not make the local search, i.e., the global search is in
progress. In this process of finding, the best new solution may be chosen as a parent
for the next generation, i.e.,
1
ymother (i + 1) ← ydaughter best (i) (51)

Pseudocode:
Initialize the distance of runner and root and, n pop , stall_ max, tol, a
k
xmother (1) ← xl + rand × (xu − xi ) for k = 1, …, npop
Assign initial values of parents.
stal_ count ← 0, i ← 1
Repeat until termination condition does not fulfill
 k
ymother (i)
ydaughter (i) =
k
for k = 1, . . . , n pop (52)
k
ymother (1) + drunner × r1

ydaughter,best (i) ← arg min f (x) (53)

//utilizes n pop function evaluations y = xdaughter k

(i)
IF i > 1 AND
    
 min f y k (i) − k
(i − 
 k=1,...,N daughter min f ydaughter 1) 
 k=1,...,Npop 
 
pop
  < tol (54)
 min f ydaughter (i − 1)
k 
 k=1,...,Npop


then
Photovoltaic Power Forecasting by Evolutionary Algorithm-Based … 123

for k = 1: npop
do//test locally at big steps
xperturbed,k ← diag{1, 1, . . . , 1, 1 + drunner n k , 1 . . . , 1} × ydaughter,best (i)

   
if f ypeturbed,k < f ydaughter,best (i) (55)

then//uses a function estimate

ydaughter,best (i) ← yperturbed,k

end
end
(k- Loop)
for k FROM 1 UNTIL Npop
do//test locally with small steps
yperturbed,k ← diag{1, 1, . . . , 1, 1 + droot rk , 1 . . . , 1} × ydaughter,best (i) (56)
   
If f ypeturbed,k < f ydaughter,best (i)
then//consumes a function estimation.
xnew,best (i) ← xperturbed,k

end
end (k- loop)
end (if)
1
ymother (i + 1) ← ydaughter,best (i)
Compute the eligibility of kth offspring from the following and
  1
k
fit ydaughter (i) ←     (57)
a + f ydaughter
k
(i)− − f ydaughter,best (i)

the chance of selecting it from the following

 
k
fit ydatgher (i)
pk =  N   for k = 1, . . . , Npop (58)
j
pq p
j=1 fit ydatgher (i)

for k from2 until npop

do //growing the mother plants of next generation
k
ymother (i + 1) ← ydaughter
ind
(i) (59)

here ‘ind’ is the index of the daughter plant chosen. This is selected from among
the present iteration through a roulette wheel selection.
end (k-Loop)
124 A. K. Pani and N. Nayak

if
     
f y
 daughter,best (i) − f ydaughter,best (i − 1) 
    < tol (60)
 f ydaughter,best (i − 1) 

then
stal count ← stal count + 1
else
stal_ count ← 0
end
if stall_ count > stall_ max
then resume the program i ← i + 1
end. (repeat)

6 Simulation Results

In this work first of all, ELM technique is used to forecast the PV power generation of
a real-time solar on-grid power plant, (located at the roof top of an academic block,
Bhubaneswar in Odisha, India) whose specification is given in Table 1. ELM is
applied to the same plant data (80% testing and 20% training), and its performance
was investigated. In further investigation, to improve the accuracy of prediction,
the signal is decomposed by empirical mode decomposition, (EMD) followed by
different optimization techniques like PSO, CPSO and RRA, which are applied and
the results are tabulated for a short-term forecasting (15, 30 min and 1 h). Here, IMFs
generated as a result of EMD are applied to the input layer nodes samplewise, i.e.,
one set of IMFs to one node, however small their amplitude may be. Sum total of
all node input IMF sets represents the signal. It is like Fourier series decomposition.
Figure 4 shows a part of it.
Sometimes if the IMF is very feeble, 0 value is considered for that automatically.
Here since the signal is random, so also is the IMFs. After this, various said optimiza-
tion techniques are applied to train the weights in the branches. The simulation result
depicts that the RRA-based EMD-ELM performs better than other ELM techniques.

Table 1 Real-time data

S. No. Design Scheme
requirement
1 Cell type Polycrystalline
2 Cell output 11.2 kW
3 Latitude 20° 25
4 Longitude 85° 80
5 Area covered 106.25 m2
Photovoltaic Power Forecasting by Evolutionary Algorithm-Based … 125

Fig. 4 IMFs of the data with EMD in 30 min interval

The result of simulation with above specifications and incremental conductance MPP,
P-V characteristics of the said photovoltaic cell is shown in Fig. 5.

Case-1: Prediction of PV power in 15 min Horizon

In this case, the PV power generated data between January 2010 and December
2010. The entire data are arranged in each 15 min time horizon. The data set is
subjected to empirical mode decomposition (EMD) and then trained by the ELM

120
Ir=1000W/m2 MPP
100 Ir=800W/m2
Ir=600W/m2
80 Ir=400W/m2
Power(Watt)

60

40

20

0
0 5 10 15 20 25
Voltage(Volt)

Fig. 5 P-V characteristics for irradiance from 400 to 1000 W/m2 at 25 °C

126 A. K. Pani and N. Nayak

techniques. Further different optimization techniques, like PSO, CPSO and runner
root algorithm, are applied to enhance the stability and forecasting capability of the
model. The prediction curves are shown in Fig. 6. Errors are shown in Table 2. It is
depicted from the graphs and the forecasting measuring indices that the RRA-based
ELM shows better performance than other models.

Case-2: Prediction of PV power in 30 min Horizon

In this study, the same PV power generated data of the same plant, between January
2010 and December 2010, were arranged in each 30 min interval of time. The data
set is decomposed into different IMFs by empirical mode decomposition method
(EMD). The data set was decomposed to enhance the strength of the ELM model.
Further different optimization techniques, like PSO, CPSO and runner root algo-
rithms, are applied to enhance the stability and forecasting capability of the model.
The prediction curves are shown in Fig. 7. Errors are shown in Table 3 that fol-
lows. It is depicted from the graphs and the forecasting measuring indices that the
RRA-based ELM shows better performance than other models.

Fig. 6 15 min prediction in various algorithms

Table 2 Forecasting errors in 15 min horizon

Forecasting models Horizon (in minute) MAPE (%) MAE RMSE
ELM 15 3.0091 0.0298 0.0298
PSO-ELM 15 2.1215 0.0301 0.0321
CPSO-ELM 15 2.1001 0.0199 0.0301
RRA-ELM 15 1.4448 0.0146 0.0179
Photovoltaic Power Forecasting by Evolutionary Algorithm-Based … 127

Fig. 7 30 min prediction in various algorithms

Table 3 Forecasting errors in 30 min horizon

Forecasting models Horizon (in minute) MAPE (%) MAE RMSE
ELM 30 4.6112 0.0431 0.0522
PSO-ELM 30 3.7911 0.0461 0.0531
CPSO-ELM 30 3.4371 0.0432 0.0502
RRA-ELM 30 3.1148 0.0406 0.0389

Case-3: Prediction of PV power in 60 min horizon

In this case, the total data set is arranged in 60 min time interval with same ratio of
80% testing and 20% training.
The data set is decomposed into different IMFs by empirical mode decomposition
method (EMD). The ELM based on different optimization techniques, like PSO,
CPSO and RRA, investigates the forecasting errors with 60 min time interval. The
prediction error curves are shown in Fig. 8. Errors are shown in Table 4. It is depicted
from the graphs and the forecasting measuring indices that the RRA-based ELM
shows better performance than other models.
128 A. K. Pani and N. Nayak

Fig. 8 60 min prediction in various algorithms

Table 4 Forecasting errors in 60 min horizon

Forecasting models Horizon (in minute) MAPE (%) MAE RMSE
ELM 60 6.0081 0.0724 0.0813
PSO-ELM 60 5.0912 0.0621 0.0661
CPSO-ELM 60 5.0821 0.0617 0.0803
RRA-ELM 60 4.8552 0.1032 0.0615

7 Conclusion

Here in this work, a real-time PV model historical data set is prepared with time
slots ranging 15, 30 and 60 min, and then simulated with application of a modified
extreme learning machine technique. Further to increase the accuracy in forecasting,
the arbitrary selected weights are optimized through PSO, CPSO and RRA and
simulated. It is depicted that RRA-ELM shows better performance than other models.
Thus, the use of various optimization algorithms is made, giving appropriate weights
in the model and bias to the nodes in the model, so that ELM representation can offer
a computed guess of PV power with high accuracy and low error. Here, the runner
root algorithm is rigorous optimization techniques which optimize the random input
weights of the ELM model. Hence, the RRA-ELM is superior to other models which
can be verified from the table.
Photovoltaic Power Forecasting by Evolutionary Algorithm-Based … 129

References

1. Kelleher, J., Ringwood, J.V.: ‘A computational tool for evaluating the economics of solar and
wind micro generation of electricity. Energy 34, 401–409 (2009)
2. Parida, B., Iniyan, S., Goic, R.: A review of solar photovoltaic technologies. Renew. Sustain.
Energy Rev. 15, 1625–1636 (2011)
3. Shiva Kumar, B., Sudhakar, K.: Performance evaluation of 10 MW grid connected solar
photovoltaic power plant in India. Energy Rep. 1, 184–192 (2015)
4. Pannase, V.R., Nanavala, H.B.: A review of PV technology power generation, PV material,
performance and its applications. In: IEEE Xplorer, ICISC-2017, pp. 1–6 (2017)
5. Akhter, M.N., Mekhilef, S., Mokhlis, H., Shah, N.M.: Review on forecasting of photovoltaic
power generation based on machine learning and metaheuristic techniques. IET Renew. Power
Gener. 13(7), 1009–1023 (2019)
6. Voyant, C., et al.: Numerical weather prediction (NWP) and hybrid ARMA/ANN model to
predict global radiation. Energy 39(1), 341–355 (2012)
7. Dahmani, K., et al.: Estimation of 5-min time-step data of tilted solar global irradiation using
ANN (artificial neural network) model. Energy 70, 374–381 (2014)
8. Mellit, A., Pavan, A.M.: A 24-h forecast of solar irradiance using artificial neural network:
application for performance prediction of a grid-connected PV plant at Trieste. Italy. Sol.
Energy 84(5), 807–821 (2010)
9. Alsharif, M.H., Younes, M.K., Kim, J.: Time series ARIMA model for prediction of daily and
monthly average global solar radiation: the case study of Seoul, South Korea. Symmetry 11,
240 (2019)
10. Singh, B., Pozo, D.: A guide to solar power forecasting using ARMA models. In: Statistic
Application, Cornel University, pp. 1–6 (2018)
11. Shin, H.: Partial functional linear regression. J. Stat. Plann. Inference 139, 3405–3418 (2009)
12. Boussaada, Z., Curea, O., Remaci, A., Camblong, H., Bellaaj, N.M.: A nonlinear autoregressive
exogenous (NARX) neural network model for the prediction of the daily direct solar radiation.
Energy 11, 620–642 (2018)
13. Wang, Guochang, Su, Y., Shu, Lianjie: One-day-ahead daily power forecasting of photovoltaic
systems based on partial functional linear regression models. Renew. Energy 96, 469–478
(2016)
14. Islam, Badar, Baharudin, Zuhairi, Nallagownden, Perumal: Modified meta heuristics and
improved back propagation neural network-based electrical load demand prediction technique
for smart grid. IEEJ 12(S1), S20–S32 (2017)
15. Chaturvedi, D.K., Isha, I.: Solar power forecasting: a review. Int. J. Comput. Appl. 145, 28–50
(2016)
16. Schmidt, T., Calais, M., Roy, E., Burton, A., Heinemann, D., Kilper, T., Carter, C.: Short-term
solar forecasting based on sky images to enable higher PV generation in remote electricity
networks. Renew. Energy Environ. Sustain. 2(23)
17. Ruf, H., Schroedter-Homscheidt, M., Heilscher, G., Beyer, H.G.: Quantifying residential PV
feed-in power in low voltage grids based on satellite-derived irradiance data with application
to power flow calculations. Sol. Energy 135, 692–702 (2016)
18. Behera, M.K., Nayak, N., Majumder, I.: Solar photovoltaic power forecasting using optimized
modified extreme learning machine technique. Int. J. Eng. Sci. Technol. 21(3), 428–438 (2018)
19. Behera, M.K., Nayak, N.: A comparative study on short-term PV power forecasting using
decomposition based optimized extreme learning machine algorithm. Int. J. Eng. Sci. Technol.
(in press)
20. Eniola, V., Suriwong, T., Sirisamphanwong, C., Ungchittrakool, K.: Hour-ahead forecasting
of photovoltaic power output based on hidden Markov model and genetic algorithm. Int. J.
Renew. Energy Res. IJRER 9(9) (2019)
Distribution System Optimization
by Circular Reconfiguration Technique

Smrutirekha Mohapatra, Satwik Behera, and Subrat Kumar Dash

Abstract This article presents a new fuzzy-based circular reconfiguration method-

ology for solving the reconfiguration problem of a radial distribution system. Mul-
tiple objectives are considered such as real power loss minimization, node voltage
deviation minimization and branch current loading minimization while subjected
to constraints of maintaining radial structure, all load energization and validation
of KCL and KVL laws in the network. Further fuzzy membership functions are
defined to normalize multiple objectives as well as to combine them to make a single
objective. Present technique is applied to IEEE 69 radial distribution system under
two loading conditions (constant power load, composite load), and the results are
encouraging.

Keywords Circular reconfiguration · Power distribution system · Fuzzy

multi-objective approach

S. Behera (B) · S. K. Dash

Goverment College of Engineering Kalahandi, Bhawanipatna, Odisha, India
e-mail: [email protected]
S. K. Dash
e-mail: [email protected]
S. Mohapatra
Institute of Technical Education and Research, Siksha ‘O’ Anusandhan (Deemed to be
University), Bhubaneswar, Odisha 751030, India
e-mail: [email protected]

© Springer Nature Singapore Pte Ltd. 2020 131

G. Pradhan et al. (eds.), Advances in Electrical Control and Signal Systems,
Lecture Notes in Electrical Engineering 665,
https://doi.org/10.1007/978-981-15-5262-5_9
132 S. Mohapatra et al.

1 Introduction

There are two types of switches in the radial distribution system. Network reconfig-
uration is the process of altering the status of these two types of switches in order to
achieve different objectives. It means the aim of network reconfiguration is to achieve
a configuration such that it can act as an efficient system which satisfies a number
of objectives, such as real power loss reduction, voltage deviation minimization and
load balancing.
Table 1 represents detailed literature review of wide variety of approaches
suggested by different researchers to solve the network reconfiguration problem.
After reviewing the above contributions, this paper presents a new circular recon-
figuration technique. To handle the combinatorial nature implicit to the network
reconfiguration problem, heuristic rules are used to guide the optimization in fuzzy

Table 1 Literature review of network reconfiguration approaches

Author(s), Reference Salient features/contributions
Marlin and Back [1] First applied branch-bound technique for network reconfiguration
to minimize real power loss
Shirmohammadi et al. [2] Suggested to open the switch having minimum branch current after
closing all normally open switches
Civanlar et al. [3] Used switch interchange technique to represent the variation of
loss due to load transfer
Baran and Wu [4] Proposed spanning structure for load balancing and loss reduction
Taylor et al. [5] Adopted heuristic approach for loss minimization, relieving
transformer overload and limiting feeder constraint violations
Goswami and Basu [6] Solved network reduction problem by applying Kirchoff’s
equations
Kashem et al. [7] Created switch pools that include tie switches and its neighboring
two switches to attain optimal network reconfiguration
enhancement
Kashem et al. [8] Used first stage that identifies the network structure for minimum
loss and second stage that determines the optimal switching
operation
Venkatesh et al. [9] Fuzzy adaptation of evolutionary programming is presented for
real power loss minimization and power quality improvement
Venkatesh et al. [10] Presented network reconfiguration to enhance loadability
Das [11] Adopted fuzzy max–min principle and heuristic rules for achieving
load balancing, loss minimization, node voltage deviation and
branch current constraint violation
Das [12] Formulated the fuzzy satisfaction objective function using
weighting factors to satisfy all the objectives and constraints
Lin [13] Developed a unique algorithm for balancing the loads on
transformers and feeders
(continued)
Distribution System Optimization by Circular Reconfiguration … 133

Table 1 (continued)
Author(s), Reference Salient features/contributions
Prasad et al. [14] Formulated a fuzzy mutated genetic algorithm for loss
minimization and power quality improvement
Ahuja et al. [15] Implemented an AIS–ACO hybrid approach for solving
reconfiguration problem for addressing loss reduction, transformer
load balancing and voltage deviation minimization in the
pareto-optimal framework
Zhang et al. [16] Adopted tabu search algorithm for minimizing loss in large-scale
distribution systems
Su et al. [17] Proposed a technique based on mixed integer hybrid differential
evolution technique to reduce power loss and improve voltage
profile
Gupta et al. [18] Presented a fuzzy multi-objective approach using an adaptive
genetic algorithm. Used ‘max-geometric mean’ operator for
finding the degree of overall satisfaction
[19–22] Genetic algorithm is implemented for obtaining the optimal
network by reducing the real power loss
Nikam et al. [23] Adopted MHBMO algorithm to solve the reconfiguration problem
by considering the effect of renewable energy sources
Narimani et al. [24] Obtained optimal configuration for loss minimization, operational
cost minimization and improvement of reliability using enhanced
gravitational search algorithm
Rani et al. [25] Employed HSA to solve the reconfiguration problem for both
active and reactive power loss reduction
Raju et al. [26] Addressed the reconfiguration of balanced and unbalanced
distribution systems
[27, 28] ACO search is adopted to solve the reconfiguration problem
Rao et al. [29] Used HSA for simultaneous reconfiguration and DG installation
Farahani et al. [30] Implemented network reconfiguration and capacitor placement
simultaneously to reduce power loss and to improve voltage
profiles

environment as well as to reduce number of tie switch operations. In order to check the
effectiveness of the proposed methodology, IEEE 69 bus radial distribution system
is taken in the presence of different loading conditions.

2 Network Reconfiguration Methodology

The proposed methodology implements a branch exchange technique (for every

closing operation of switch is followed by opening of another switch to restore radial
configuration of network) as explained in [14].
134 S. Mohapatra et al.

3 Problem Formulation

Here the problem is formulated by considering three different objectives, such as (a)
real power loss reduction (b) node voltage deviation minimization and (c) branch
current loading minimization subject to constraints of maintaining radiality of the
network, energization of all loads and validation of KCL and KVL. Here all the three
objectives are modeled in the forms of fuzzy membership function with restricted
lower and upper bounds as mentioned in [11].

(a) Membership function for real power loss reduction (µL i )

Let us define
PLOSS(i)
xi = , for i = 1, 2, . . . , Nk (1)
PLOSS0
Equation (1) indicates that if xi is high, power loss reduction is low and if xi low,
power loss reduction is high. Membership function for real power loss reduction is
given in Fig. 1 and µL i can be expressed as follows (Fig. 2):

Fig. 1 Membership function for power loss reduction

Fig. 2 Membership function for node voltage deviation

Distribution System Optimization by Circular Reconfiguration … 135
⎧ ⎫
⎪ (xmax −xi )
⎨ (xmax −xmin ) for xmin < xi < xmax ⎪
⎬
µL i = 1 for xi < xmin (2)
⎪
⎩0 ⎪
⎭
for xi ≥ xmax

(b) Membership function for node voltage deviation (µVi )

Basic purpose of this membership function is that the deviation of node voltages
should be less. Let us define

yi = max Vs − Vi, j , for i = 1, 2, . . . , Nk ; j = 1, 2, . . . , N B (3)

If the value of node voltage deviation is less, then a higher membership value is
assigned, and if the deviation is more, then a lower membership value is assigned.
⎧ ⎫
⎪ (ymax −yi )
⎨ (ymax −ymin ) for ymin < yi < ymax ⎪
⎬
µVi = 1 for yi < ymin (4)
⎪
⎩0 ⎪
⎭
for yi < ymax

(c) Membership function for maximum branch current loading index (µAi )

Basic purpose of this membership function is that to minimize the branch current
constraint violation.
Let us define
|I (i, m)|
Branch current loading index = , (5)
Ic (m)

For i = 1, 2,…, N k ; m = 1, 2, …, LN
When the maximum value of branch current loading index exceeds unity, mem-
bership value will be lower and as long as it is less than or equal to unity, membership
value will be maximum, i.e., unity. Let us define

|I (i, m)|
z i = max , for i = 1, 2, . . . , Nk ;
Ic (m)
m = 1, 2, . . . , L N (6)

Membership function for maximum branch current loading index is shown in

Fig. 3. From Fig. 3, we can write
⎧ ⎫
⎪ (z max −z i )
⎨ (zmax −zmin ) for z min < z i < z max ⎪
⎬
µAi = 1 for z i ≥ z max (7)
⎪
⎩0 ⎪
⎭
136 S. Mohapatra et al.

Fig. 3 Membership functions for maximum branch current loading index

3.1 Fuzzy Multi-objective Formulation [18]

Multi-objective reconfiguration is carried out to get an optimal configuration which

satisfies a number of objectives. A number of operators have been proposed by
the researchers such as ‘max–min,’ ‘max-weighted addition’ and ‘max-product’.
Drawbacks of these operators are described in [18]. In this paper, ‘max-geometric
mean’ operator has been implicated to find out the degree of fuzzy satisfaction (DFS)
value.

DFS = (µL i ∗ µVi ∗ µAi )1/ 3 (8)

For i = 1, 2, …, N k .
The network having maximum DFS will give the best compromising solution.

3.2 Circular Reconfiguration Methodology

The method adopted is based on circular updating mechanism. According to this

method after completion of the reconfiguration process once, the opened branches
in the newly configured network are considered as the tie branches for the next
reconfiguration process. This process goes on until we get the satisfactory objective
function value or the optimal solution. The detailed algorithm is described below.
Algorithm
Step1: Read the system data.
countb=1
Store all the tie branches in tie_br.
Step 2: Calculate the voltage differences across the open switches by running the
load flow.
Step 3: i={1,2,….,ntb}, Close the tie switch i across which voltage difference is
maximum and is greater than a prespecified value (0.01) [11]. This condition is
applied in order to minimize the switching operations.
Step 4: Run reconfiguration mechanism as described in previous algorithm.
Distribution System Optimization by Circular Reconfiguration … 137

Step 5: Evaluate all the membership function values by running load flow as
described in section [III] one by one opening all the loop forming branches. (In
this work load flow is done by forward and backward sweep method [31])
Step 6: Evaluate the overall degree of satisfaction by using equation no.
Step 7: Open that branch among the loop forming branches for which the value
of DFS is maximum.
Step 8: Check if length(ntb)=1 go to step 2 else go to step 9.
ntb=ntb-1
Step 9: Consider the opened switches in the previous configuration as the tie
switches in the current configuration.
tie_br=opened_branch
count=countb+1
Step 10: Run load flow and calculate the parameters like real power loss, node
voltage, branch current etc.
Check If PLoss (count)<PLoss (countb), go to step 11 else goto step 2.
countb=count
Step 11: Stop

3.3 Load Modeling [32]

The real and reactive power loads of node ‘n’ are given as follows:

Ppu (n) = Ppu (n) c1 + c2 |V (n)| + c3 |V (n)|2 (9)


Q pu (n) = Q pu (n) d1 + d2 |V (n)| + d3 |V (n)|2 (10)

The constants c1 , d 1 , c2 , d 2 , c3 and d 3 are the compositions of constant power,

constant current and constant impedance loads, respectively. For constant power
loads, c1 = d 1 = 1 and rest are zero. Similarly for composite load comprising 40% of
constant power loads, 30% of constant current loads and 30% of constant impedance
loads, the values of above constants are c1 = d 1 = 0.4 and c2 = d 2 = c3 = d 3 = 0.3.

4 Result Analysis

So as to test the effectiveness of proposed method, an IEEE 69 bus distribution

system is analyzed [33]. It is having 69 nodes, 68 sectionalizing switches and five
tie switches. Its single line diagram is depicted in Fig. 4. For the base configuration,
total real power load and total reactive power load are 3802.19 kW and 2694.06
kVAr, respectively. The system has a total real power loss of 224.9783 kW and total
reactive power loss of 102.1948 kVAr.
138 S. Mohapatra et al.

Fig. 4 Base configuration of 69 node system

The details of the comparisons of the two reconfiguration processes are mentioned
below in the form of tables and graphs. Figure 5 represents the optimal reconfiguration
of the network using proposed technique.

(a) Constant Power Load

Figures 5, 6, 7 and 8 depict the voltage profile, branch current variations, branch
real power loss and voltage stability index for the test system under base case, single
reconfiguration and circular reconfiguration, respectively (Fig. 9).
Table 2 compares the real power loss, minimum voltage and stability index for
base case, single reconfiguration and circular reconfiguration for constant power
loading. Switching sequence for the network reconfiguration is also mentioned in
Table 2. It is observed that in circular reconfiguration, better real power loss reduction

Fig. 5 69 Node distribution system after optimal reconfiguration

Distribution System Optimization by Circular Reconfiguration … 139

Fig. 6 Voltage profile for constant power load

Fig. 7 Branch current for constant power load

is achieved at same time maintaining better minimum voltage and stability margin
of the network.

(b) Composite Load

Table 3 compares the real power loss, minimum voltage and stability index for
base case, single reconfiguration and circular reconfiguration for composite loading.
Switching sequence for the network reconfiguration is also mentioned in Table 3. It is
observed that in circular reconfiguration, better real power loss reduction is achieved
at the same time maintaining better minimum voltage and stability margin of the
network.
140 S. Mohapatra et al.

Fig. 8 Branch real power loss for constant power load

Fig. 9 Voltage stability index for constant power load

Table 2 Configuration of 69 bus system for constant power load

Parameters Base case Single reconfiguration Circular
reconfiguration
Real power loss (kW) 224.9783 100.0235 98.6064
Minimum voltage @ 0.9092 @ 65 0.9483 @ 62 0.9495 @ 61
node (p.u.)
Stability index @ node 0.6833 @ 65 0.8087 @ 62 0.8126 @ 61
Objective function – 0.7784 0.7652
Switches opened 69, 70, 71, 72, 73 69, 55, 62, 14, 12 69, 14, 58, 61, 70
Distribution System Optimization by Circular Reconfiguration … 141

Table 3 Configuration of 69 bus system for composite load

Parameters Base case Single reconfiguration Circular
reconfiguration
Real power loss (kW) 171.8230 86.3310 85.4107
Minimum voltage @ 0.9214 @ 65 0.9524 @ 62 0.9534 @ 61
node (p.u.)
Stability index @ node 0.7207 @ 65 0.8229 @ 62 0.8263 @ 61
Objective function – 0.7224 0.7144
Switches opened 69, 70, 71, 72, 73 69, 55, 62, 14, 12 69, 14, 58, 61, 70

Fig. 10 Voltage profile of constant power load

Figures 10, 11, 12 and 13 show the voltage profile, branch current variation,
real power losses of each branch and stability index of each node for composite load,
respectively. As expected, it is clear from the above plots that circular reconfiguration
yields better result than the other scenarios. The points of minimum voltage and
stability index are improved, and overall branch current loading is reduced. Similarly,
overall real power loss is reduced to 49.76% for single reconfiguration and 50.29%
for circular reconfiguration.
Table 4 compares the value of real power loss and minimum voltage obtained by
different optimization methods of some researchers and our proposed method. In the
final solution obtained by the proposed method, circular reconfiguration achieved
the best result among the all other methods. In [34] although the real power loss is
approximately 0.1 kW less than that of proposed method, the system voltage profile
in the proposed method is better than [34].
142 S. Mohapatra et al.

Fig. 11 Current profile of constant power load

Fig. 12 Branch wise real power loss for composite load

5 Conclusion

Branch exchange-based technique of reconfiguration has been presented for network

reconfiguration with the objective of real power loss minimization, voltage profile
improvement and branch current restriction. In this regard, a standard 69 node IEEE
system has been studied under two loading conditions. Heuristic rules have been
suggested to minimize tie switch operation for obtaining optimal reconfiguration of
the network. Fuzzy multi-objective approach proved to be providing better optimal
reconfigured network. Stability analysis has also been carried out before and after
network reconfiguration. Stability index was found to be better for the reconfigured
network than the original network. Finally, the result of circular reconfiguration
methodology has been compared with different methods adopted by different authors
Distribution System Optimization by Circular Reconfiguration … 143

Fig. 13 Voltage stability index for composite

Table 4 Comparison of reconfiguration by different approaches

Methods Real power loss (kW) Minimum voltage Opened switches
(p.u.)
Initial configuration 224.9427 0.9092 {69, 70, 71, 72, 73}
[34] BPSO 105.1431 0.9239 {13, 20, 55, 61, 69}
[35] SAI 99.62 0.9428 _
[34] SPSO 98.59 0.9494 {14, 56, 61, 69, 70}
[29] HSA 99.35 0.9428 {69, 18, 13, 56, 61}
Single reconfiguration 100.02 0.9483 {69, 55, 62, 14, 12}
Circular 98.61 0.9495 {69, 14, 58, 61, 70}
reconfiguration

and single reconfiguration in order to validate the effectiveness of the proposed

method. The comparison result indicates the best efficiency among all the methods.

References

1. Merlin, A., Back, H.: Search for a minimum loss optimal spanning tree configuration for
an urban power distribution system. In: 5th International Proceedings on Power System
Computation Conference, Cambridge, U.K., pp. 1 2/6 (1975)
2. Shirmohammadi, D., Hong, H.W., Semlyen, A., Luo, G.X.: A compensation based power flow
method for weakly meshed distribution and transmission network. IEEE Trans. Power Syst. 3,
753–762 (1988)
3. Civanlar, S., Grainger, J.J., Yin, H., Lee, S.S.H.: Distribution feeder reconfiguration for loss
reduction. IEEE Trans. Power Deliv. 3, 1217–1223 (1988)
4. Baran, M.E., Wu, F.: Network reconfiguration in distribution system for loss reduction and
load balancing. IEEE Trans. Power Deliv. 4, 1401–1407 (1989)
144 S. Mohapatra et al.

5. Taylor, T., Lubkeman, D.: Implementation of heuristic search strategies for distribution feeder
reconfiguration. IEEE Trans. Power Deliv. 5, 239–246 (1990)
6. Goswami, S.K., Basu, S.K.: A new algorithm for the reconfiguration of distribution feeders for
loss minimization. IEEE Trans. Power Deliv. 7, 1484–1491 (1992)
7. Kashem, M.A., Ganapathy, V., Jasmon, G.B.: Network reconfiguration for enhancement of
voltage stability in distribution networks. IEEE Proc.-Gener. Transm. Distrib. 147, 171–175
(2000)
8. Kashem, M.A., Ganapathy, V., Jasmon, G.B.: A geometrical approach for network reconfigu-
ration based loss minimization in distribution systems. Electr. Power Energy Syst. 23, 295–304
(2001)
9. Venkatesh, B., Ranjan, R.: Optimal radial distribution system reconfiguration using Fuzzy
adaptation of evolutionary programming. Electr. Power Energy Syst. 25, 775–780 (2003)
10. Venkatesh, B., Ranjan, R., Gooi, H.B.: Optimal reconfiguration of radial distribution systems
to maximize loadability. IEEE Trans. Power Syst. 19 (2004)
11. Das, D.: A fuzzy multi-objective approach for network reconfiguration of distribution systems.
IEEE Trans. Power Deliv. 21, 202–209 (2006)
12. Das, D.: Reconfiguration of distribution system using Fuzzy multi-objective approach. Electr.
Power Energy Syst. 28, 331–338 (2006)
13. Lin, C.-H.: Distribution network reconfiguration for load Balancing with a coloured Petri
net algorithm. IEEE Proc. Gener. Transm. Distrib. 150 (2003)
14. Prasad, K., Sahoo, N.C.: A Simplified Approach for Computer-Aided Education of Network
reconfiguration in Radial Distribution Systems. Wiley Periodicals Inc. (2007)
15. Ahuja, A., Das, S., Pahwa, A.: An AIS-ACO hybrid approach for multi-objective distribution
system reconfiguration. IEEE Trans. Power Syst. 22 (2007)
16. Zhang, D., Zhengcai, F., Zhang, L.: An improved TS algorithm for loss-minimum reconfigu-
ration in large-scale distribution systems. Electr. Power Syst. Res. 77, 685–694 (2007)
17. Su, C.-T., Lee, C.-S.: Network reconfiguration of distribution systems using improved mixed-
integer hybrid differential evolution. IEEE Trans. Power Deliv. 18 (2003)
18. Gupta, N., Swarnkar, A., Niazi, K.R., Bansal, R.C.: Multi-objective reconfiguration of dis-
tribution Systems using adaptive genetic algorithm in fuzzy framework. IET Gener. Transm.
Distrib. 4(12), 1288–1298 (2010)
19. Zhu, J.Z.: Optimal reconfiguration of electrical distribution network using the Refined genetic
algorithm. Electr. Power Syst. Res. 62, 37–42 (2002)
20. Mendoza, J., López, R., Morales, D., López, E., Dessante, P., Moraga, R.: Minimal loss recon-
figuration using genetic algorithms with restricted population and addressed operators: real
application. IEEE Trans. Power Syst. 21 (2006)
21. Edgar Manuel Carreno, Rubén Romero, Antonio Padilha-Feltrin, “An Efficient Codification to
Solve Distribution Network Reconfiguration for Loss Reduction Problem” IEEE transactions
on power systems, vol. 23, November (2008)
22. Enacheanu, B., Raison, B., Caire, R., Devaux, O., Bienia, W., Hadjsaid, N.: Radial network
reconfiguration using genetic algorithm based on the Matroid theory. IEEE Trans. Power Syst.
23 (2008)
23. Nikam, T.: An efficient multi-objective HBMO algorithm for distribution feeder reconfigura-
tion. Expert Syst. Appl. 38, 2878–2887 (2011)
24. Narimani, M.R., Vahed, A.A., Azizipanah-Abarghooee, R., Javidsharifi, M.: Enhanced grav-
itational search algorithm for Multi-objective distribution feeder reconfiguration considering
reliability, loss and operational cost. IET Gener. Transm. Distrib. 8(1), 55–69 (2004)
25. Sudha Rani, D., Subrahmanyam, N., Sydulu, M.: Self adaptive harmony search algorithm for
optimal network reconfiguration. IEEE Trans. Power Deliv. (2014)
26. Viswanadha Raju, G.K., Bijwe, P.R.: Efficient reconfiguration of balanced and unbalanced
Distribution systems for loss minimisation. IET Gener. Transm. Distrib. 2(1), 7–12 (2008)
27. Su, C.-T., Chang, C.-F., Chiou, J.-P.: Distribution network reconfiguration for loss reduction
by ant colony search algorithm. Electr. Power Syst. Res. 75, 190–199 (2005)
Distribution System Optimization by Circular Reconfiguration … 145

28. Carpaneto, E., Chicco, G.: Distribution system minimum loss reconfiguration in the Hyper-
Cube Ant Colony Optimization Framework. Electr. Power Syst. Res. 78, 2037–2045 (2008)
29. Srinivasa Rao, R., Ravindra, K., Satish, K., Narasimham, S.V.L.: Power loss minimization in
distribution system using network reconfiguration in the presence of distributed generation.
IEEE Trans. Power Syst. 28 (2013)
30. Farahani, V., Vahidi, B., Askarian Abyaneh, H.: Reconfiguration and capacitor placement
simultaneously for energy loss reduction based on an improved reconfiguration method. IEEE
Trans. Power Syst. 27 (2012)
31. Ghosh, S., Das, D.: Method for load-flow solution of radial distribution networks. IEEE Proc.
Gener. Transm. Distrib. 146, 641–648 (1999)
32. Chakrovorty, M., Das, D.: Voltage stability analysis of radial distribution networks. Electr.
Power Energy Syst. 23, 129–135 (2003)
33. Savier, J.S., Das, D.: Impact of network reconfiguration on loss allocation of radial distribution
systems. IEEE Trans. Power Deliv. 2, 2473–2480 (2007)
34. Tandon, A., Saxena, D.: A comparative analysis of SPSO and BPSO for power loss mini-
mization in distribution system using network reconfiguration. In: International Conference on
Innovative Applications of Computational Intelligence on Power, Energy and Controls with
their Impact on Humanity (CIPECH14), 28 & 29 Nov 2014
35. Chena, J., Zhang, F., Zhang, Y.: Distribution network reconfiguration based on simulated
annealing immune algorithm. Energy Procedia 12, 271–277 (2001)
Novel Collecting Decision Optimization
Algorithm for Enhanced Dynamic
Performance of Hybrid Power
Source-Based SOFC and Supercapacitor
for Grid Integration

Subhashree Choudhury, Bimal Sen, Nikhil Khandelwal,

and Anshuman Satpathy

Abstract This paper describes a hybrid grid-connected microgrid system

comprising of a solid oxide fuel cell (SOFC) and a supercapacitor (SC). The SOFC
possesses many advantages over other fuel cells in having high dynamic perfor-
mance, high combined heat, long-term stability, fuel flexibility, low emissions and
relatively low cost. Nevertheless, the most significant disadvantage of SOFC is that
it alone fails to compensate the system because of its high operating temperature
which makes the system response sluggish in nature. To eradicate this problem, a
SC-based energy storage system (ESS) is coordinated with SOFC to enhance the
transient response of the system. SC has advantages over other ESSs with its high
capacitor capacity and faster charging and discharging capability. A hybrid system is
considered to be more efficient as it emits zero amount of carbon pollutants, thereby
making the system environment-friendly. The conventional PID controller fails to
respond to nonlinearities in the system. So for dynamic operation of the PID con-
troller, a novel collective decision optimization algorithm (CDOA) is proposed in
this paper. The hybrid system is designed in MATLAB/Simulink architecture, and
to verify the efficacy of the proposed controller, the system is led to a disturbance.
A detailed comparative analysis of SOFC with PID, hybrid system with PID and
hybrid system with CDOA-tuned PID has been done. In addition to this, harmonics
study of each case has been carried out. The proposed hybrid system with the CDOA
technique outperforms others in satisfying well the IEEE constraints.

Keywords Solid oxide fuel cell (SOFC) · Supercapacitor (SC) · Boost converter ·
Bidirectional buck–boost converter (BBC) · Total harmonic distortion (THD) ·
Collective decision optimization algorithm (CDOA)

S. Choudhury (B) · B. Sen · N. Khandelwal · A. Satpathy

Department of Electrical and Electronics Engineering, Institute of Technical Education and
Research, Siksha ‘O’ Anusandhan (Deemed to be University), Bhubaneswar 751030, India
e-mail: [email protected]

© Springer Nature Singapore Pte Ltd. 2020 147

G. Pradhan et al. (eds.), Advances in Electrical Control and Signal Systems,
Lecture Notes in Electrical Engineering 665,
https://doi.org/10.1007/978-981-15-5262-5_10
148 S. Choudhury et al.

1 Introduction

The rapid degradation of fossil fuel and the pollution caused by the consumption of
these fossil fuels are dragging the environment towards unhygienic conditions. This
affects public health, so for maintaining hygiene and public health more attention
is given towards sustainability for which distributed renewable energy production is
necessary. Solid oxide fuel cell (SOFC) gives neat and pollution-free automation to
an electrochemically created electricity at a higher efficiency. SOFC is the perfect
energy source with compatibility and portability ability, which enables the SOFC to
be used in a wide area of applications [1]. The SOFC offers significant advantages
over other fuel cells by showing high dynamic performance, high combined heat,
long-term stability, fuel flexibility, low emission and relatively low cost. Applicable
for long distribution (>250 km) transmission line and electrolyte used does not affect
the SOFC working process [2, 3].
Nevertheless, SOFC also possesses some major disadvantages such as: (1) it
has high operating temperature for which SOFC alone fails in giving a dynamic
performance during a fault, and (2) it works at high temperature so the material cost
will be increased which is uneconomical [4]. To improve the robustness of the SOFC,
this paper introduces a hybrid power source (HPS) approach. Hybrid power source
(HPS) provides higher peak power and good energy density and fulfils the power
demand. HPS can be utilized in many ways in dynamic loads [5]. It also reduces
the fuel expenditure by adding natural renewable sources which are benefited for
the consumer. It controls the transient current and energy density for many types of
the electrical circuit [6]. Environmental impact can be reduced by the help of hybrid
power sources. It is also used for system accuracy and increased purpose. Nowadays
as fuel is very expensive, it helps to reduce cost and maintenance cost [2].
Here, the SOFC is hybridized with supercapacitor (SC) to form a HPS. SC acts
as a support and accompaniment component which balances the load mismatch
of SOFC. The supercapacitor (SC) is synchronized through DC/DC bidirectional
buck/boost converter with the system. In case of power transmission, SC is superior
to other energy storage system (ESS) because of its high capacitor capacity, increased
lifespan, better energy density, faster rate of charging and discharging features and
quicker than other capacitors in current distribution and consumption [7, 8]. Further
SC working principle does not depend upon chemical reaction, and it cannot be
overcharged or discharged as it always works in a boundary limit [9].
PID controller is very easy for implementation and simpler in structure and pro-
vides immediate feedback [10]. PID controller is divided into three parts like the
series controller, lateral controller and combination of lateral and series controller.
In P mode, it reduces the system output variation. In I mode, system fault can be zero.
In D mode, it gives a quick response to change the controller input [11]. However,
the major shortcoming of the PID controller is that it is a linear controller and it
cannot respond to nonlinearity present in the power system. And hence, it fails to
give optimal solution.
Novel Collecting Decision Optimization Algorithm for Enhanced … 149

So to dynamically tune the PID control parameters, many optimization techniques

have been proposed by many authors in the literature such as brain storm optimiza-
tion (BSO) [12], ant colony optimization (ACO) algorithm [13], bacterial foraging
optimization (BFO) algorithm [14], genetic algorithm (GA) [15], particle swarm
optimization (PSO) [16] and collective decision optimization algorithm (CDOA)
[17]. Among all of them, CDOA has more advantages of faster execution and it
yields more precise result being a group-based algorithm [17].
In this paper, a SOFC- and SC-based HPS with grid tied system is modelled in
the MATLAB/Simulink software. To test the robustness of the proposed controller,
the system is led to addition of a rectifier load for time t = 0.1–0.2 s. From the
characteristics obtained, it can be seen that alone SOFC with classical PID controller
fails to meet the load demand and mitigate nonlinearity from the system, whereas it
can be observed that when SOFC and SC form a HPS with conventional PID, the
compensation of the load demand is little enhanced but fails to mitigate the nonlin-
earities of the system. So for the robust operation of the PID controller, a collective
decision optimization algorithm (CDOA) is proposed in this paper which dynami-
cally tunes the PID parameters. The HPS with CDOA-tuned PID proves its efficiency
by mitigating nonlinearities, compensating the load demand and improving the tran-
sient stability of the system. Further, this paper provides a detailed comparison of
various system parameter characteristics by taking into account SOFC with classical
PID and SOFC- and SC-based HPS with conventional PID and SOFC- and SC-
based HPS with CDOA-tuned PID controller. The total harmonic distortion (THD)
through FFT analysis for each case has been done, and SOFC- and SC-based HPS
with CDOA-tuned PID controller has the minimum THD value as per IEEE 1547
constraint.
The rest part of the paper is organized as follows. In Sect. 2, the detailed modelling
of the system configuration is mentioned. The conventional and proposed controller
structure is elaborated in Sect. 3. In Sect. 4, the MATLAB/Simulink model designed
is shown and the result obtained is analysed. Finally, in Sect. 5 the entire paper work
is concluded.

2 Mathematical Modelling of System Configuration

2.1 Modelling of a Solid Oxide Fuel Cell (SOFC)

The fuel cell (FC) is a device which develops electrical energy through a chemical
reaction between a fuel and an oxidant. There are various types of FCs like proton
exchange membrane fuel cell (PEMFC), molten carbonate fuel cell (MCFC), phos-
phoric acid fuel cell (PAFC), alkaline fuel cell (AFC), solid oxide fuel cell (SOFC),
etc. However, SOFC has advantages over others such as [18]: (1) operates at a high
temperature; (2) no need of maintenance for the electrolyte loss; (3) suitable for
long-distance transmission (>250 km); and (4) avoid expensive catalysts such as
150 S. Choudhury et al.

Fig. 1 A simple model of a

SOFC

ruthenium or platinum. SOFC can be otherwise termed as energy alternation devices

[19]. A simple model of a SOFC is shown in Fig. 1 which is constructed using resis-
tors ‘R1 ’, ‘R2 ’ and ‘R3 ’, capacitor ‘C’, voltage source ‘E’ and dependent voltage
sources ‘ f E 0 (I, T )’ and ‘Vact ’.
The output voltage of a SOFC can be written as Eq. (1):

V0 = E − V1 − V2 − V3 (1)

where ‘V0 ’ is the output voltage, ‘V1 ’ is activation overvoltage across ‘R1 ’, ‘V2 ’ is
the concentration overvoltage across ‘R2 ’, ‘V3 ’ is the voltage drop across ‘R3 ’ and
‘E’ is the open-circuit voltage.
Using Nernst equation, ‘E’ can be found as shown in Eq. (2):
  
RT PH2 (PO2 )0.5
E = N E0 + ln (2)
2F PH2 O

where ‘N ’ is the number of stacked cells, ‘E 0 ’ is the reversible cell voltage, ‘R’
represents the universal gas constant (JK/kmol), ‘T ’ denotes the temperature of the
stack in Kelvin, ‘PH2 ’, ‘PO2 ’ and ‘PH2 O’ are the mole fractions of the species ‘H2 ’,
‘O2 ’ and ‘H2 O’, respectively, and ‘F’ is the Faraday constant equal to 9.648 × 104
C mol−1 .
Voltage across the resistor occurs due to the resistance of the electrodes and can
be calculated through Ohm’s law as represented in Eq. (3):

V3 = i ∗ R3 (3)

where ‘i’ is the current flowing through the SOFC.

The activation voltage due to the hydrogen concentration and cell current can be
written as Eq. (4):

V1 = −A ∗ ln(i) (4)

where ‘A’ is a constant value which depends upon the type of FC used.
Novel Collecting Decision Optimization Algorithm for Enhanced … 151

Concentration voltage ‘V2 ’ can be ignored under consideration of the linear

operation region, hydrogen and oxygen flow constant.

2.2 Modelling of Supercapacitor (SC)

A supercapacitor, also known by the electrochemical double-layer capacitor, has a

high value of capacitance due to the electrolytic solution present between the two
conductors. Unlike normal capacitors, it can store a large amount of energy. But its
low internal resistance and high-power density give rise to its own unique charac-
teristics. Faster charging–discharging rate, cycle firmness, low service, robustness,
operation at higher temperatures and better lifespan are the expectancies which can
be fulfilled through an SC [1]. In most of the applications, an SC is used because of
its faster rate of charging and discharging. Due to having a high-power density, it can
be charged sooner and efficiently than other elements. It allows the power to flow at
its peak level to the load and controls the energy level if system is unstable. However,
its low energy density does not allow it to be an independent energy storage device
[2, 3]. A well-known model of SC is manifested in Fig. 2 where ‘U ’ represents the
voltage across the terminals of the SC, ‘i’ denotes the SC current, ‘R’ is the internal
resistance, ‘i 1 ’ and ‘i 2 ’ are the currents flowing through the capacitors ‘C1 ∗ V0 ’ and
‘C0 ’, respectively, and ‘V0 ’ shows the voltage across the parallel combination of the
two capacitors.
The voltage across the terminals of the SC is depicted in Eq. (5):

U = Ri + VO (5)

Current through the capacitors is shown in Eqs. (6) and (7), and total current
flowing in the SC is shown in Eq. (8):

dVO
i 1 = C 1 VO (6)
dt

Fig. 2 Simple model of a SC

152 S. Choudhury et al.

Fig. 3 Simple model of

a Boost converter

dVO
i 2 = C0 (7)
dt
dVO dVO
i = i 1 + i 2 = C 1 VO + C0 (8)
dt dt

2.3 Modelling of Boost Converter

When the output voltage ‘V0 ’ is greater than the input voltage ‘Vin ’, then the converter
is called a boost converter. In this converter, an inductor ‘L’ is connected in series
with a voltage source as shown in Fig. 4. It stores energy when switch ‘S’ is turned
on and dissipates it along with the source energy to provide an increased output
voltage across the resistive load ‘R L ’ when the same switch is turned off [20]. Here,
capacitor ‘C’ is used as an alternative of the source which provides power during the
switch on time ‘Ton ’ and diode ‘D’ is used to provide a path to the current during
the switch off time ‘Toff ’. The ‘Ton ’ and ‘Toff ’ of the switch are controlled by a pulse
width modulation (PWM) signal. The relation between the output and input voltage
is shown in Eq. (9):
 
1
V0 = ∗ Vin (9)
1 − DT

where ‘DT ’ is the duty cycle which can be formulated as [Ton /(Ton + Toff )] (Fig. 3).

2.4 Modelling of Boost Converter

BBC works as both buck and boost converters for the same system but not at the same
time. The operation is held by a PWM signal applied to the switches of the BBC.
Since the BBC is economical and smaller in size as compared to other converters, it
is used as a medium of charging and discharging of the storing components of the
system [21, 22]. In Fig. 4, all the elements of the circuit are in their ideal state and
Novel Collecting Decision Optimization Algorithm for Enhanced … 153

Fig. 4 Bidirectional
buck–boost converter

they are: inductor with inductance value ‘L’, capacitors used as storing device and
filter having capacitances ‘C 1 ’ and ‘C 2 ’, respectively, resistance ‘R1 ’ connected in
series with ‘C 1 ’, ‘RL ’ as load and ‘S 1 ’ and ‘S 2 ’ as switches.
In the above circuit, ‘I L ’ denotes the current flowing through the inductor, and
‘U 1 ’ and ‘U 2 ’ are the voltages across ‘C 1 ’ and ‘C 2 ’, respectively. When switch ‘S 1 ’ is
closed and switch ‘S 2 ’ is opened, then the BBC acts as a boost converter and elements
used to store energy transmit power. When ‘S 1 ’ is opened and ‘S 2 ’ is closed, then
the BBC behaves like a buck converter and the same elements start absorbing power
from the DC output voltage [23].

3 Control Structure

3.1 PID Controller

PID is one type of linear controller, and it is the combination of proportional (P),
integration (I) and derivative (D). The PID controller represents three parameters
‘K p ’, ‘K i ’ and ‘K d ’. ‘K p ’ represents the proportional gain which decreases the
rise time, ‘K i ’ represents the integral gain which helps in reducing the steady-state
error, and ‘K d ’ represents the derivative gain which reduces the peak overshoot,
increases the transient response and also creates a more stable system [24, 25]. All
the parameters are interdependent on each other. The time-domain function of the
PID controller is given in Eq. (10), and the transfer function is represented in Eq. (11)

t
de(t)
c(t) = K p e(t) + K i e(T ) + K d (10)
dt
0

Ki Kd s2 + K p s + Ki
C(s) = K p + + Kd s = (11)
s s
154 S. Choudhury et al.

3.2 Collective Decision Optimization Algorithm (CDOA)

In day-to-day life, the peoples are not satisfied with their own outcomes; rather, they
want higher gain in a minimum amount of investment. So, people having this type of
attribute need to optimize their problems. Till now, there are many natures inspired
optimized techniques (OTs) present in the literature such as brain storm optimization
(BSO) [26], ant colony optimization (ACO) algorithm [27], bacterial foraging opti-
mization (BFO) algorithm [28], genetic algorithm (GA) [29] and collective decision
optimization algorithm (CDOA) [30]. All these OTs can only be expressed in terms
of problems of the ant, bacteria, human being, etc. However, CDOA is a more con-
vincing OT because of its simpler design of solving the human-based problems and
faster convergence speed. It is generally required to enhance the problem-solving
skill of human beings. CDOA depends on the inspection of the people’s answerable
quality and how they create or select solutions for the problem. In CDOA, group of
peoples refers to the all applicant solutions. It shows an excellency in the performance
while solving the deep-down problems. This optimization needs equity between the
group members. All the candidates in a group are known as a resolver, and every
resolver has their own ideas. Thereby, every resolver speaks their own ideas and
shares their knowledge with the other members. The best plan is preferred from the
offered ideas. Here, individual’s special maturity is required to create unique ideas.
As it is a group-based process, it is easy to create influence towards the opinion as
controlled by the resolver. The basic working of CDOA is demonstrated through a
flow chart as shown in Fig. 5.
CDOA has two types of groups, namely ‘personal-based’ and ‘culture-based’.
Nevertheless, they depend upon applicant figures to solve the problems.
1. Personal-based: In this case, people individually select ideas as being self-
dependent. Personal-based peoples also need less computation charge.
2. Culture-based: The main target of this process is to gather a group of people and
discuss new innovative ideas to solve the problems. In this case, many groups
are required and a group leader. Here, the group leader plays a main role as
well as the resolver because resolver must deal with few factors which are most
important in the answerable procedure. This procedure inspires the resolver to
create new ideas or solutions for society. Here, many unique ideas are created by
any of the groups, so the number of solutions is more and improved from genesis
to genesis [30, 31]. The group leader is required to regulate those ideas and form
solutions of the group. Here, all the groups exchange their ideas, so the quality
of the ideas can be increased. But here computation charge is more as compared
to the ‘personal-based’.
Application of CDOA for optimal tuning of the PID parameters. For finding
the optimal working of SOFC stack and SC, this paper proposes the CDOA-based
optimization technique to find the optimal PID parameter (K p, K i and K d) values.
Here, integral time absolute error (ITAE)-based performance criterion index is used
for minimization of the objective functions. The objective functions for SOFC (F1)
and SC (F2) are formulated as follows
Novel Collecting Decision Optimization Algorithm for Enhanced … 155

Fig. 5 Flow chart of CDOA

t
F1 = (e1 (t))2 dt (12)
0

t
F2 = (e2 (t))2 dt (13)
0

where e1 (t) = P = PSOFC − PGRID

where e2 (t) = SOC = SOCinitial − SOCfinal

4 MATLAB/Simulation Model and Analysis of Results

In this paper, the analysis of coordinated use of solid oxide fuel cell (SOFC) and
supercapacitor with power electronic application of DC to DC converters like boost,
156 S. Choudhury et al.

bidirectional buck/boost converter and three-phase voltage source inverter is imple-

mented in a grid-connected system using MATLAB/Simulink software. The char-
acteristics of SOFC and SC like voltage, current, power, duty ratio and SOC have
been studied. Further to evaluate the coordinated performance of the SOFC and SC
with the proposed technique, the microgrid system under consideration has been sub-
jected to a rectifier load for time, t = 0.1 to 0.2 s. The result obtained clearly justifies
the novelty, functionality and robustness of the coordinated microgrid system with
the proposed CDOA-based optimization technique as compared to the classical PID
control technique (Fig. 6).
Figures 7, 8, 9 and 10 depict the output voltage, current, power of SOFC and duty
ratio of boost converter, respectively. From Figs. 7, 8, 9 and 10, it can be concluded
that the proposed CDOA-tuned PID maintains the SOFC output voltage, current and

Fig. 6 MATLAB/Simulink model of the SOFC and SC based HPS

Fig. 7 SOFC voltage

Novel Collecting Decision Optimization Algorithm for Enhanced … 157

Fig. 8 SOFC current

Fig. 9 SOFC power

power at a desirable level with improved system stability, better peak time and rise
time as well as enhanced duty ratio of the boost converter in comparison with the
conventional PID controller.
Figures 11, 12, 13 and 14 illustrate the supercapacitor voltage, current, power and
SOC, respectively. The characteristic obtained signifies the efficiency of the proposed
method in improving the system stability with better control parameters as compared
to classical PID controller.
Figures 15 and 16 show the total DC voltage and power of coordinated use of
SOFC and supercapacitor when led to a fault (a rectifier load is switched on 0.1 s). It
can be observed in the above figures that during fault time the FC-based PID alone
cannot meet the demand of the load effectively. However, when the PID-based FC
is coordinated with SC the system response in meeting the load demand is slightly
158 S. Choudhury et al.

Fig. 10 Duty ratio of boost

converter

Fig. 11 Supercapacitor
current

improved. Nevertheless, the proposed CDOA-tuned PID with the coordinated FC

and SC is proved to be more efficient as it meets maximum load demand, stabilizes
system faster and enhances system reliability by reducing harmonics.
Figures 17, 18 and 19 represent the grid voltage during the fault time with SOFC-
based PID, SOFC- and SC-based PID and SOFC- and SC-based CDOA-tuned PID,
respectively. To check the effectiveness of the proposed controller with the grid
during the fault, the grid voltage with SOFC-based PID, SOFC- and SC-based PID
and SOFC- and SC-based CDOA-tuned PID has been plotted. In the above figures,
the robustness of the proposed controller with SOFC and SC is visible as the proposed
controller maintains the grid voltage at constant levels with proper sinusoidal nature
of the wave. However, the SOFC and SC with PID controller fail in maintaining
the sinusoidal nature of the curve of the grid voltage but compensate the system to
Novel Collecting Decision Optimization Algorithm for Enhanced … 159

Fig. 12 Supercapacitor
voltage

Fig. 13 Supercapacitor
power

maintain the grid voltage level nearly constant which shows the effect of addition
of supercapacitor with SOFC on the system during the fault. But the SOFC-based
PID alone fails in keeping the voltage level and nature of curve and rather consists
of harmonics.
Figures 20, 21 and 22 signify the THD values of the grid voltage during fault for
SOFC-based PID, SOFC- and SC-based PID and SOFC- and SC-based CDOA-tuned
PID. The computed values of THDs are found out to be 9.06%, 4.40% and 1.57%,
respectively, which indicates that the proposed controller is superior in comparison
with other compared controller in bringing the system stability faster by reducing
the harmonics occurring in the system.
160 S. Choudhury et al.

Fig. 14 Supercapacitor SOC

Fig. 15 DC voltage during

fault

Fig. 16 DC power during

fault
Novel Collecting Decision Optimization Algorithm for Enhanced … 161

Fig. 17 Grid voltage during fault with SOFC-PID

Fig. 18 Grid voltage during fault with SOFC-SC-PID

Fig. 19 Grid voltage during fault with SOFC-SC-CDOA-PID

162 S. Choudhury et al.

Fig. 20 THD of grid voltage

with SOFC-PID

Fig. 21 THD of grid voltage

with SOFC-SC-ID

Fig. 22 THD of grid voltage

with SOFC-SC-CDOA-PID
Table 1 System stability parameters
Parameters Types of controller
SOFC-SC-CDOA-PID SOFC-SC-PID SOFC-PID
Settling time Peak time Percentage Settling time Peak time Percentage Settling time Peak time Percentage
overshoot overshoot overshoot
SOFC power 1.23 0.91 86.75 1.9 2.9 94.7 2.3 4.1 104.5
SC power 1.36 0.98 98.2 2.12 3.3 104.6 3.73 4.67 110.8
SOFC voltage 1.78 1.02 97.2 2.33 3.56 109.5 3.87 4.71 112.5
SC voltage 1.9799 1.0299 112.55 2.639 0.826 116.26 3.9299 1.9299 125.699
Grid voltage 1.26 0.869 101.11 2.566 1.826 126.26 2.19 1.24 115.67
Novel Collecting Decision Optimization Algorithm for Enhanced …
163
164 S. Choudhury et al.

Table 2 PID control parameters

Parameters Controllers
CDOA-optimized PID PID
Kp Ki Kd Kp Ki Kd
SOFC 1.43 1.233 0.45 3.24 3.9 2.90
SC 1.34 1.56 0.98 21.35 23.72 15.64

Table 3 THD values of grid voltage

Types of controllers System parameter
Grid voltage THD (in percentage)
FC-SC-CDOA-PID 1.57
FC-SC-PID 4.40
FC-PID 9.06

Table 1 displays the statistical comparison of system stability parameters like

settling time, peak overshoot and percentage overshoot for various system parameters
of microgrid system. Table 2 shows the comparative evaluation of proposed CDOA-
tuned PID control parameters and classical PID control parameters. Table 3 illustrates
the analytical comparison of THD values evaluated for grid voltage through FFT
analysis.

5 Conclusions

This chapter investigates the optimization of the energy management for battery- and
ultracapacitor-based HESS for an autonomous microgrid system. For enhancing the
battery’s life span and efficient operation of the HESS, a novel COA-based optimiza-
tion technique has been proposed which dynamically tunes the PID parameters, thus
bringing about enhancement of system dynamic response and stability. To examine
the effectiveness of the technique proposed, the system has been tested by inclusion
of an unsymmetrical fault. Lastly, the stability analysis has been carried out through
THD calculation to verify the robustness of the proposed technique.
Novel Collecting Decision Optimization Algorithm for Enhanced … 165

Appendix

Parameters Values
SOFC V O = 40, E = 45, R1 + R2 = 0.4 , R3 = 0.06 and C = 0.25 F
SC V O = 30 V, supercapacitor internal resistance = 0.015 , supercapacitor-rated
capacitance = 50 F, number of series capacitors—4 and number of parallel
capacitors—1
Boost L = 0.8 mH, C = 100 µF, Fs = 30 kHz, V O = 100 and V in = 40 V
BBC L = 1 mH, C1 = C2 = 14 µF, Fs = 25 kHz, V O = 50 and V in = 30 V
Rectifier load Nominal frequency—50 Hz, power—2500 W
Grid V = 100, X/R = 7, f = 50 Hz
SOCinitial 50
SOCfinal 100

References

1. İnci, M., Türksoy, Ö.: Review of fuel cells to grid interface: configurations, technical challenges
and trends. J. Clean. Prod. 213, 1353–1370 (2019)
2. Yalcinoz, T., Alam, M.S.: Improved dynamic performance of hybrid PEM fuel cells and
ultracapacitors for portable applications. Int. J. Hydrogen Energy 33(7), 1932–1940 (2008)
3. Nesaraj, A.S.: Recent developments in solid oxide fuel cell technology—a review (2010)
4. Irshad, M., et al.: A brief description of high temperature solid oxide fuel cell’s operation,
materials, design, fabrication technologies and performance. Appl. Sci. 6(3), 75 (2016)
5. Lazarov, V. D., et al.: Hybrid power systems with renewable energy sources types, structures,
trends for research and development. In: Proceedings of International Conference ELMA2005,
Sofia, Bulgaria (2005)
6. Bhikabhai, Y.: Hybrid power systems and their potential in the Pacific islands. SOPAC
Miscellaneous, Report 406 (2005)
7. Zhao, X., et al.: The role of nanomaterials in redox-based supercapacitors for next generation
energy storage devices. Nanoscale 3(3), 839–855 (2011)
8. Burke, A., Zhao, H.: Applications of supercapacitors in electric and hybrid vehicles. ITS (2015)
9. Rawale, S., Kamble, C.: Study and analysis of supercapacitor with its applications. Int. Res. J.
Eng. Technol. 2(3), 2194–2196 (2015)
10. Duarte-Galvan, C., et al.: Advantages and disadvantages of control theories applied in
greenhouse climate control systems. Span. J. Agric. Res. 10(4), 926–938 (2012)
11. Rao, K.S., Mishra, R.: Comparative study of P, PI and PID controller for speed control of
VSI-fed induction motor. Int. J. Eng. Dev. Res. 2(2), 2740–2744 (2014)
12. Hidalgo, A., Albors, J.: Innovation management techniques and tools: a review from theory
and practice. R&D Manage. 38(2), 113–127 (2008); Das, S., et al.: Bacterial foraging opti-
mization algorithm: theoretical foundations, analysis, and applications. In: Foundations of
Computational Intelligence, vol. 3, pp. 23–55. Springer, Berlin (2009)
13. Dorigo, M., Birattari, M.: Ant colony optimization. Springer, US (2010)
166 S. Choudhury et al.

14. Passino, K.M.: Biomimicry of bacterial foraging for distributed optimization and control. IEEE
Control Syst. Mag. 22(3), 52–67 (2002)
15. Koza, J.R.: Genetic programming (1997)
16. Kennedy, J.: Particle swarm optimization. In: Encyclopedia of Machine Learning, pp. 760–766
(2010)
17. Dabbaghjamanesh, M., Kavousi-Fard, A., Mehraeen, S.: Effective scheduling of reconfigurable
microgrids with dynamic thermal line rating. IEEE Trans. Ind. Electron. 66(2), 1552–1564
(2018)
18. Kupecki, J., Jewulski, J., Milewski, J.: Multi-level mathematical modeling of solid oxide fuel
cells. In: Clean Energy for Better Environment. IntechOpen (2012)
19. Chaudhary, P., Rizwan, M.: Hybrid control approach for PV/FC fed voltage source converter
tied to grid. Int. J. Hydrogen Energy 43(14), 6851–6866 (2018)
20. Hasaneen, B.M., Mohammed, A.A.E.: Design and simulation of DC/DC boost converter. In:
2008 12th International Middle-East Power System Conference. IEEE (2008)
21. Inoue, S., Akagi, H.: A bidirectional DC–DC converter for an energy storage system with
galvanic isolation. IEEE Trans. Power Electron. 22(6), 2299–2306 (2007)
22. Liu, D., Li, H.: A ZVS bi-directional DC–DC converter for multiple energy storage elements.
IEEE Trans. Power Electron. 21(5), 1513–1517 (2006)
23. Jun, R., Kai, W., Liwei, L.: Characteristics analysis of ultracapacitor-battery hybrid energy
storage system. In: 2017 Chinese Automation Congress (CAC). IEEE (2017)
24. Åström, K.J., Hägglund, T., Astrom, K.J.: Advanced PID control, vol. 461. ISA-The
Instrumentation, Systems, and Automation Society, Research Triangle Park, NC (2006)
25. Messner, W.C., Tilbury, D.M., Hill, R.: Control Tutorials for MATLAB® and Simulink® (1999)
26. Ramadan, R.A.: Fuzzy brain storming optimisation algorithm. Int. J. Intell. Eng. Inform. 5(1),
67–79 (2017)
27. Blum, C.: Ant colony optimization: Introduction and recent trends. Phys. Life Rev. 2(4), 353–
373 (2005)
28. Das, S., et al.: Bacterial foraging optimization algorithm: theoretical foundations, analysis, and
applications. In: Foundations of Computational Intelligence, vol. 3, pp. 23–55. Springer, Berlin
(2009)
29. Holland, J.H.: Genetic algorithms. Sci. Am. 267, 66–72 (1992)
30. Zhang, Q., et al.: Collective decision optimization algorithm: a new heuristic optimization
method. Neurocomputing 221, 123–137 (2017)
31. Zhang, Q., et al.: Modified collective decision optimization algorithm with application in
trajectory planning of UAV. Appl. Intell. 48(8), 2328–2354 (2018)
Robust Controller Application to PV Fed
CUK Converter for Constant Voltage
Regulation: “A Review”

Nibedita Swain, Sadhna Mallik, and Gyana Ranjan Biswal

Abstract In this paper, two control techniques, namely sliding mode control and
H-∞ control, are proposed for a non-isolated CUK converter. The mathematical
model of CUK converter is designed using state-space averaging method. The two
controllers serve two main purposes, regulation of output voltage and set point track-
ing. The regulation and tracking of the plant are measured against the output load
and input voltage fluctuations. The solar PV module is used as a source at the
input side instead of DC voltage. The outcomes are presented in this paper using
MATLAB/Simulink platform.

Keywords Non-isolated CUK converter · State-space averaging technique · Solar

PV module · Sliding mode controller · H-infinity controller

1 Introduction

“This paper is an extension of work originally presented in 2018 2nd International

Conference on Trends in Electronics and Informatics (ICOEI) [1]”. The advancement
of switching converters is applied in many electrical and electronic applications. It
gives a regulated DC output voltage, while the output load and input voltage vary.
Due to nonlinear and time-invariant characteristics in DC-DC converter, classical
control techniques are not suitable for controlling the output voltage as there are
large fluctuations in system parameters and output load [2]. So, nonlinear controllers
like sliding mode controller [3] and H-infinity controller are used for their robustness
and stability. The sliding mode controller along with the converter is robust against
changing load and uncertain system parameters and is identified by a nice dynamic

N. Swain (B) · S. Mallik · G. R. Biswal

Silicon Institute of Technology, Silicon Hills, Bhubaneswar, Odisha 751024, India
e-mail: [email protected]
S. Mallik
e-mail: [email protected]
G. R. Biswal
e-mail: [email protected]
© Springer Nature Singapore Pte Ltd. 2020 167
G. Pradhan et al. (eds.), Advances in Electrical Control and Signal Systems,
Lecture Notes in Electrical Engineering 665,
https://doi.org/10.1007/978-981-15-5262-5_11
168 N. Swain et al.

response as in [4]. A voltage controller based on H-infinity and a repetitive control

technique is also proposed in this paper. The robust H-infinity control is used for
systems that need robustness to parameter variations and have high performance
requirements [5]. The formulation of the H-infinity control problem is based on the
minimization of a quadratic cost function which includes both the disturbance and
control input effects [6]. One of the notable features of H-infinity technique is the
reduction of noises and external disturbances without making assumptions on them.

2 Configuration and Modeling of CUK Converter

A CUK converter steps up or steps down the source voltage with negative polarity.
The circuit is made up of a switch, a load and other nonlinear elements. It has dual
switching states—ON state and OFF state of the switch.
During switch ON state, current will flow from V i -L1 -S-V i . Energy gets stored
in the inductors, and capacitor discharges to load through C 1 -R-S-C 1 . The circuit
configuration for ON state is depicted in Fig. 1.
The switching equations for ON state are described in (1), (2), (3) and (4),
respectively.

di L1
Vi − L 1 =0 (1)
dt
di L2
VC1 − L 2 − VC2 = 0 (2)
dt
dVC1
C1 + i L2 = 0 (3)
dt

iL1

iC1
L1
iL2 io
C1

Vi S C R

Fig. 1 Circuit configuration for ON state

Robust Controller Application to PV Fed CUK Converter … 169

iL1
iL2
iC1
L1 L2
io
C1

Vi D C R

Fig. 2 Circuit configuration for OFF state

dVC2 VC2
C2 − i L2 − =0 (4)
dt R
During the switch OFF state, diode is ON and C 1 is energized via L 1 -D-V i . Current
in both the inductors increases. The switching equations for OFF state are stated in
(5)–(8), respectively. The circuit configuration for switch OFF state is depicted in
Fig. 2.

di L1
Vi − L 1 − VC1 = 0 (5)
dt
di L2
L2 + VC2 = 0 (6)
dt
dVC1
i L1 = C1 (7)
dt
dVC2 VC2
C2 = i L2 − (8)
dt R
The output voltage for both switching states is the same and is stated in Eq. (9).

V0 = VC (9)

The state-space equations for CUK converter by combining ON and OFF states
are stated in (10), (11), (12) and (13), respectively, as in [7]. u is the control input
which signifies the position of the switching state. It is either 1 or 0. u = 1 for switch
ON state and u = 0 when diode is ON. Here, u is controlling parameter and treated
as duty ratio “d”.

di L1 1
= (Vi − VC1 (1 − u)) (10)
dt L1
170 N. Swain et al.

di L2 1
= (VC1 u − VC2 ) (11)
dt L2
dVC1 1
= (−ui L2 + (1 − u)i L1 ) (12)
dt C1
 
dVC2 1 VC2
= i L2 − (13)
dt C2 R

3 Sliding Mode Controller

SMC is a nonlinear controller that finds a control input u such that the state vector
tracks a desired trajectory in the presence of model uncertainties and external dis-
turbances. The control input has two components: a normal component (un ) and an
equivalent component (ueq ) stated in Eq. (14) referred to [8].

u = u n + u eq (14)

The equivalent component guarantees the motion of the system on the sliding
surface while the system is on the surface. The normal component makes the deviation
of the sliding surface to zero.
.
S=0

The sliding surface for CUK converter can be selected as

de
S = ke +
dt
“e” specifies the difference between reference voltage and output voltage given
by

e = Voref − Vo

The sliding surface can be specified by considering the dynamic equation of the
CUK converter referred to paper [9, 10]. S and Ṡ are calculated and are stated in
Eqs. (15) and (16), respectively.
 
S = K (V0ref − V0 ) + V̇0ref − V0
= K V0ref − K V0 − V̇C2
 
1 VC2
= K V0ref − K V0 − i L2 −
C2 R
Robust Controller Application to PV Fed CUK Converter … 171
 
i L2 1
= K V0ref − + − K VC2 (15)
C2 RC2
. dVC2 1 di L2 1 dVC2
S = −K − +
dt C2 dt RC2 dt
 
1 dVC2 1 di L2
= −K + −
RC2 dt C2 dt
    
1 1 1 1 u VC2
= −K + i L2 − VC2 − VC1 − (16)
RC2 C2 RC2 C2 L 2 L2

As Eq. (15) contains the input term “d”, hence the control
.
remains on the sliding
surface for a long span of time. ueq is obtained by putting S = 0.
   
i L2 −K R L 2 C2 + L 2 VC2 K R L 2 C2 + R 2 C2 − L 2
u eq = + (17)
VC1 RC2 VC1 R 2 C2

Substituting the values of R, L 2 and C 2, the above equation can be rewritten as in

Eq. (18).
   
i L2 −3.3K + 150 VC2 3.3K + 2199850
u eq = + (18)
VC1 22 × 103 VC1 22 × 105

To formulate a control law, the Lyapunov function V is specified as:

V = 0.5S 2

When V̇ is negative, it makes the system stable and the sliding surface becomes
more effective [11, 12]. Such condition leads to the inequality:

S Ṡ < 0

For satisfying above condition, the normal component is the signum function of
the sliding surface and is defined as

u n = sign(S)

where
   
1 i L2
u n = sign K V0ref + − K V0 −
RC2 C2
= sign[38K + (45.45 − K )V0 − 4545.45i L2 ]
172 N. Swain et al.

4 Design of H-Infinity Controller

The robust H-∞ controller has been designed, and its robustness is verified with
the help of γ iteration method [13]. While designing the controller, the weighting
functions are selected such that the system could meet the performance requirements.
Two weight functions (W S and W T ) are chosen such that
 
 1 
|S( jw)| <  
W S ( jw) 
 
 1 
|T ( jw)| <  
WT ( jw) 

S(jw) and T (jw) represent the sensitivity and the complementary sensitivity function,
respectively. The design criterion of lag–lead compensator is used to synthesize the
weight function of H-infinity controller. The sensitivity and complementary sensi-
tivity function are defined in Eqs. (19) and (20), respectively, representing G as the
plant transfer function and C as the controller transfer function [14]. Here, boost
converter is considered as plant or system.

S = (I + GC)−1 (19)

T = GC(I + GC)−1 (20)

In the above two equations, G represents the plant transfer function and C
represents the controller transfer function.
The two weight functions for this converter are given below in Eqs. (21) and (22),
respectively.

0.5s + 400
Ws (s) = (21)
s + 40
0.01s + 1
Wt (s) = (22)
0.01s + 2

By using “hinfsyn” in the MATLAB window, the controller transfer function is

calculated in Eq. (23).

0.0255s + 5.085
K (s) = , γ = 0.5354 (23)
0.006s 2 + 1.331s + 44.38
Robust Controller Application to PV Fed CUK Converter … 173

Fig. 3 Internal structure of PV section

5 Design of PV System

The switching converter is fed from a PV module [15]. The PV module consists of
20 solar cells; each cell has a short-circuit current of 4.75 A and open-circuit voltage
of 0.6 V. All the series-connected solar cells jointly produce an output of 12 V under
standard irradiance and temperature conditions. The inner structure of PV section is
depicted in Fig. 3.

6 Simulink Diagram and Waveforms

The schematic diagram of CUK converter along with the SMC is illustrated in Fig. 4.

Fig. 4 Schematic diagram of CUK converter with SMC

174 N. Swain et al.

Figure 5 shows output waveform at normal temperature of 25 °C and insolation of

1000 Wp /m2 for time t = 1 s without any controller.
Figure 6 shows output voltage waveform by using SMC. The output of CUK
converter with SMC by varying insolation from 600 to 1200 Wp /m2 at time t = 0.5 s
with a desired output voltage of 38 V is illustrated in Fig. 7.
The load varies from 80 to 120  for a time of 0.5 s, and a constant output of
approximately 38 V is attained. The design parameters for the converter are given in
Table 1.
Table 2 shows the output voltage of the converter with varying insolation, and it

10

-10
Vout(V)

-20

-30

-40

-50

-60
0 0.2 0.4 0.6 0.8 1
time(s)

Fig. 5 Output voltage waveform at 1000 Wp/m2 without controller

-5

-10
Vout(V)

-15

-20

-25

-30

-35

-40
0 0.1 0.2 0.3 0.4 0.5
time(s)

Fig. 6 Output voltage waveform at 1000 Wp/m2 using SMC

Robust Controller Application to PV Fed CUK Converter … 175

0
1200Wp/m 2
-5 1000Wp/m 2
80Wp/m 2
-10
600Wp/m 2

-15
Vout(V)

-20

-25

-30

-35

-40
0 0.1 0.2 0.3 0.4 0.5
time(s)

Fig. 7 Converter output using SMC for different insolations

Table 1 Converter parameter

Circuit parameters Values
specifications
Input voltage (V in ) 12 V
Set point voltage (V s )desired 38 V
Inductance (L 1 ) 180 μH
Inductance (L 2 ) 150 μH
Capacitance (C 1 ) 200 μF
Capacitance (C 2 ) 220 μF
Load resistance (R) 100 
Switching frequency (f s ) 25 kHz

Table 2 Variation of output

Insolation (Wp /m2 ) PV output (V) Converter output (V)
voltage by varying the
insolation 1200 12 −38.3
1000 11.8 −38
800 11.57 −37.32
600 11.15 −36.55

is found that the variation is very less in the range from 600 to 1200 Wp /m2 .
The output voltage almost remains constant for varying irradiance. It is illustrated
in Fig. 7.
Figure 8 shows the output voltage waveform for various loads. It remains constant
approximately.
The block diagram of CUK converter connecting H-infinity controller is depicted
in Fig. 9.
176 N. Swain et al.

5
R=80 ohm
0 R=100 ohm
R=120 ohm
-5

-10
Vout(V)

-15

-20

-25

-30

-35

-40
0 0.1 0.2 0.3 0.4 0.5
time(s)

Fig. 8 Converter output varying R = 80–120 

Fig. 9 Closed-loop configuration of CUK converter with H-infinity controller

Figure 10 shows the singular value plot for sensitivity and complementary
sensitivity function of CUK converter.
From singular value plot analysis, it is found that the peak gain is 47.4, and it
occurs at a frequency of 3.95 × 104 rad/s. The system is highly unstable as the gain is
very large. Figure 11 shows the singular value plot of sensitivity and complementary
sensitivity function for closed-loop CUK converter.
From Fig. 11, it is noticed that the peak gain of sensitivity function is reduced to
0.0078, and peak gain of complementary sensitivity function is reduced to 3.34. The
system is highly robust in nature. To satisfy the condition for robustness as mentioned
Robust Controller Application to PV Fed CUK Converter … 177

Singular Values
50
System: T1 S1
Peak gain (dB): 47.4
At frequency (rad/s): 3.95e+04 T1

0
Singular Values (dB)

-50

-100 2 3 4 5 6
10 10 10 10 10
Frequency (rad/s)

Fig. 10 Singular value plots for open-loop plant

Singular Values
20
System: S
S
Peak gain (dB): 0.00781
At frequency (rad/s): 7.68e+03
T
0 System: T
Peak gain (dB): 3.34
At frequency (rad/s): 7.08e+03

-20
Singular Values (dB)

-40

-60

-80

-100

-120 0 1 2 3 4 5
10 10 10 10 10 10
Frequency (rad/s)

Fig. 11 Singular value plots for closed-loop plant

178 N. Swain et al.

Bode Diagram
10
S
5 (1/w1)

-5
Magnitude (dB)

-10

-15

-20

-25

-30

-35

-40 0 1 2 3 4 5
10 10 10 10 10 10
Frequency (rad/s)

Fig. 12 Bode diagram of S(jω) and 1/W s (jω)

earlier that the magnitude of sensitivity function must be less than the magnitude of
one by weight sensitivity function and complementary sensitivity magnitude must
be less than the magnitude of one by complementary sensitivity weight function for
all values of frequencies, Figs. 12 and 13 show the bode diagram showing
   
 1   1 

|S( jw)| <   
, |T ( jw)| <  
W S ( jw)  WT ( jw) 

for all values of ω.

The robustness of H-infinity control technique is investigated against load varia-
tion and insolation variation. Load varies from 80 to 120 . The fluctuation in load
affects the system performance. If the load is varied, the output voltage is varied;
hence, regulation is less using H-infinity controller. The output voltage for varying
load is shown in Fig. 14.
The voltage output is −46 V for R = 120  and −32 V for R = 80 . The output
voltage for varying irradiance is shown in Fig. 15.
The performance comparison between two controllers is compared in Table 3 for
Vin-12 V and R = 100 .
Robust Controller Application to PV Fed CUK Converter … 179

Bode Diagram
20
T
(1/w2)
0

-20
Magnitude (dB)

-40

-60

-80

-100

-120 1 2 3 4 5
10 10 10 10 10
Frequency (rad/s)

Fig. 13 Bode diagram of T (jω) and 1/W T (jω)

10
R=100 ohm
R=80 ohm
0 R=120 ohm

-10
Vout(V)

-20

-30

-40

-50
0 0.2 0.4 0.6 0.8 1
time(s)

Fig. 14 Output voltage for varying load

7 Conclusion

The sliding mode controller and the H-infinity controller are designed for the con-
verter, and the controlled system along with the controller is insensitive to parameter
180 N. Swain et al.

5
1000Wp/m 2
0 800Wp/m 2
600Wp/m 2
-5

-10
Vout(V)

-15

-20

-25

-30

-35

-40
0 0.2 0.4 0.6 0.8 1
time(s)

Fig. 15 Output voltage for varying irradiance

Table 3 Time domain

t s (s) Mp (%) ess (%)
specifications without and
with using controller Open-loop plant 0.55 44.73 13.15
With SMC 0.12 0 0
With H-∞ controller 0.5 0 2.5

variations. The robustness with respect to input irradiance and output load is described
with proper output waveforms. The sliding mode controller reduces the settling time
from 0.55 to 0.12 s with 0% overshoot and zero steady-state error, but H-infinity
controller reduces the overshoot to 0% with lower steady-state error and settling
time than open-loop converter. Among the two controllers, sliding mode controller
is the best by considering both stability and robustness.

References

1. Swain, N., Ali, S.M., Panigrahi, C.K., Praveen Kumar, D.: Sliding mode controller—a non-
linear approach to non-isolated Cuk converter for constant voltage application. In: 2018 2nd
International Conference on Trends in Electronics and Informatics (ICOEI-2018)
2. Forsyth, A.J., Mollov, S.V.: Modelling and control of DC-DC converters. IEEE Power Eng. J.
12, 229–236 (1998)
3. Bartoszewicz, A.: Sliding Mode Control. InTech
4. Gopal, M.: Digital Control and State Variables Method. Conventional and Intelligent Control
System, 4th edn. McGraw Hill Publication
5. Phurahong, N., Kaitwanidvilai, S., Ngaopitakkul, A.: Fixed structure robust 2DOF H-infinity
loop shaping control for ACMC buck converter using genetic algorithm. In: Proceedings of the
IMECS 2012, Hong Kong, vol. II, 14–16 Mar 2012, pp. 1030–1035
Robust Controller Application to PV Fed CUK Converter … 181

6. Duncan, M., Keith, G.: A loop shaping design procedure using H∞ synthesis. IEEE Trans.
Autom. Control 37(6), 759–769 (1992)
7. Modelling and control of DC-DC converters. Tutorial Power Eng. J. (1998)
8. Guldemir, H.: Modeling and sliding mode control of DC-DC buck-boost converter. In: 6th
International Advanced Technologies Symposium (IATS’11), Elazığ, Turkey, 16–18 May 2011
9. Lai, Y.-M., Tan, S.-C., Wu, C.-K.: Design of a PWM based sliding mode controlled buck-boost
converter in continuous-conduction-mode. ECTI Trans. Electr. Eng. Electron. Commun. 5(1)
(2007)
10. Malesani, L., Rossetto, L., Spiazzi, G., Tenti, P.: Performance optimization of Cuk converters
by sliding-mode control. IEEE Trans. Power Electron. 10(3), 302–309 (1995)
11. Muley, S.S., Nagarale, R.M.: Sliding mode control of boost converter. Int. J. Emerg. Technol.
Adv. Eng. 3(9) (2013)
12. Ahmed, M.: Sliding mode control for switched mode power supplies. PhD. thesis at
Lappeenranta University of Technology, Lappeenranta, Finland
13. Pati, N., Swain, N.: Application of H∞ controller to boost converter using model order
reduction. In: Proceedings of IEEE, INDICON-2015
14. Swain, N., Pati, N.: Comparative study of model reference adaptive control and H-infinity
control to non-isolated boost converter. In: 2018 5th IEEE Uttar Pradesh Section International
Conference on Electrical, Electronics and Computer Engineering (UPCON)
15. Villalva, M.G., de Siqueira, T.G., Ruppert, E.: Voltage regulation of photovoltaic arrays: small-
signal analysis and control design. IET Power Electron. 3(6), 869–880 (2010)
Crow Search Algorithm Optimized
3DOF-PID Controller for AGC
of Multi-area Power System Using RFB

Sk Alle Rasul Tarafdar, Digant Mishra, and D. P. Bagarty

Abstract In this paper, crow search algorithm (CS) is proposed for 3DOF-PID
for automatic generation control of multi-area multi-source power system. A two-
area reheat thermal system is considered and the gains of the 3DOF-PID controller
implemented individually in each area are tuned by a global search optimization
technique CS. At first, 2DOF/3DOF-PID is written in MATLAB/SIMULINK envi-
ronment. 3DOF-PID controller performs better in terms of transiency considering
three important aspects of undershoot, overshoot, and settling time at a disturbance
of 0.1 pu. Later on, redox flow battery (RFB) is introduced in both the areas addition-
ally to improve the system performance by minimizing the error in frequency and
achieving minimum objective function ITAE. Sensitivity analysis is carried out in
two ways by varying the SLP within +50% and −50% in steps of 25% whereas para-
metric variation is done by change in system parameters within ±50%. Percentage
improvement, bar plot, and tabular presentation are exhibited in order to clarify the
proposed 3DOF-PID controller tuned by CS algorithm for a two-area thermal power
system in addition to RFB units’ in each area outperforms better when transiency
condition is considered.

Keywords Automatic generation control (AGC) · Redox flow battery (RFB) ·

2DOF-two degree of freedom · 3DOF-three degree of freedom · Crow search
algorithm (CS)

S. A. R. Tarafdar (B) · D. Mishra · D. P. Bagarty

CET BBSR-03, Bhubaneswar, India
e-mail: [email protected]
D. Mishra
e-mail: [email protected]
D. P. Bagarty
e-mail: [email protected]

© Springer Nature Singapore Pte Ltd. 2020 183

G. Pradhan et al. (eds.), Advances in Electrical Control and Signal Systems,
Lecture Notes in Electrical Engineering 665,
https://doi.org/10.1007/978-981-15-5262-5_12
184 S. A. R. Tarafdar et al.

1 Introduction

Contemporary power plants are maneuvered with interconnection to give shape for
a large power system. Scheduled voltage and frequency are the imperative aim of
power system when transmitting and delivering electrical energy. Generation and
consumption of real power are depended upon frequency [1]. A bottleneck provision
crops up when instability of power system creates havocs during load disturbance.
Nature of frequency owing to above-discussed consent creates a hazardous situation
in an interconnected part. Hence, the automatic generation control concept creates
a bloom in the picture. Different control structures along with intelligent techniques
sauced in survey. In Ref. [2], Kothari et al. presented discrete controller for reheat
thermal power system. Nanda et al. [3] revealed superiority of Bacteria Foraging
Optimization technique (BFOA) upon genetic algorithm. In Ref. [4], effect of RFB is
conducted by system mode study to a hydro-thermal gas system with/without RFB. In
Ref. [5], type-2 (T2) fuzzy system is scrutinized upon a thermal system. Application
of ant lion optimizer controlled by PID plus second-order derivative controller is
suggested in Ref. [6]. Relevant to hybrid algorithm of differential evolution and
particle swarm optimization, Sahu et al. [7] devised fuzzy-PID controller on a reheat
thermal power system. Two thermal and one diesel units’ describe the effectual
control of fuzzy-PID controller using TLBO in Ref [8]. Rahman et al. [9] analyzed
the use of two degree of freedom and three degree of freedom in a solar-thermal
integrated power system. Demonstration of different structures for PID using DEPSO
techniques published on [10] by authors. For a thermal-hydro-gas power system
authors in [11] demonstrated a novel LUS-TLBO algorithm via parallel AC/DC
tie-line. In [12] Mohanty et al. applied DE algorithm to optimally tune the gain
parameters of conventional I, PI and PID controller for AGC of a two-area multi-
source power system. Uses of FACTS devices in ref. [13] upon non-linearity in power
system. Alopex based DE algorithm [14] was proposed for a hydro-thermal power
system controlled by cascaded PD-PI. Nayak et al. [15] proposed a maiden attempt
by application adaptive SOS algorithm based on fuzzy-PID controller with derivative
filter.

2 Power System Modeling

Representation of a under labeled figure name Fig. 2 is an interconnected power

system of a reheat thermal power system [5]. Secondary controller named 3DOF-
PID controller is endowed in each area with the input area control errors (ACEs) and
outputs are u1 and u2. Linear amalgamation of ACE is frequency and tie-line error.
System parameters are given in appendix. ACEs for the two-area system shown in
Fig. 3 are given by

ACE1 = P12 + B1  f 1 (1)

Crow Search Algorithm Optimized 3DOF-PID Controller … 185

ACE2 = P21 + B2  f 2 (2)

P12 /P21 are the alteration in tie-line power in area 1/2, B1 /B2 are the frequency bias
factor of area 1/2 and  f 1 / f 2 are the frequency deviations in area 1/2, respectively.
When the system is subjected to a small disturbance, ACEs are used as actuating
signal to reduce P and f to zero when steady state is reached.

3 Crow Search Algorithm

Memorization of faces, using of tools, communication in a sophisticated ways, and

retrievation of food from corner to corner [16, 17] are some surplus amount of
behavior that makes crows the widely distributed genus of birds. Competition always
persists in the crow flock to obtain better food source as it is a strident task. In view
of optimization, environment acts as search space, flocks are the searchers, a piece
pose of environment corresponds to a feasible solution and quality of food acts as
the global solution of problem. Intellectual behavior of crows endeavors CSA to
simulate the pros of optimization.
Development of CSA [18] is based on the above-discussed part of the crows. It is
a population-based met-heuristic algorithm.
The principles of CSA are listed as follows:
1. Crows live in the form of flock.
2. Crows memorize the position of their hiding places.
3. Crows follow each other to do thievery.
4. Crows protect their caches from being pilfered by a probability.
Steps involved in CSA:
Step 1: Delineation of optimization problem, decision variables, and constraints
are done. Then, the adjustable parameters of CSA [flock size (N), maximum
number of iterations (itermax ), flight length (f l ), and alertness probability (AP)]
are valued.
Step 2: Initialize position and memory of crows
N crows are randomly positioned in a d-dimensional search space. Each crow
symbolizes a feasible solution of the impenetrability and d is the number of
resolute variables.
The memory of each crow is initialized. Since, at the initial iteration, the crows
have no experiences and assumed to have hidden their foods at their initial
positions.
Step 3: Evaluate fitness (objective) function
By interleaving the decision variable values into the objective function, the
excellence of its position is computed for each crow.
186 S. A. R. Tarafdar et al.

Step 4: Generate new position

Suppose, crow i wants to engender a new position. For this aim, the crow randomly
selects one of the flock crows’ j and follows it to discover the position of the foods
hidden by this crow (mj ). The new position of crow i is obtained by Eq. (3). Process
is repeated.
  
x i,iter+1 = x i,iter + ri × f l i,iter × m i,iter − x i,iter r j ≥ APi,iter
a random position otherwise (3)

Step 5: The new position feasibility is checked of each crow, and if validated,
updating is completed or else stays in contemporary position.
Step 6: Computation of new fitness function and positions is done.
Step 7: Update memory
It is seen that if the fitness function value of the new position of a crow is recovered
than the fitness function value of the memorized position, the crow keep posted
its memory by the new position.
Memory updating is as follows:
 i,iter+1    
m i,iter+1 = mx i,iter f x i,iter+1 is better than f m i,iter (4)

Step 8: Check termination criterion

Steps 4–7 are reiterated until itermax is reached. The best position of the memory in
terms of the objective function value is reported as the solution of the optimization
problem.

4 3-DOF-PID Controller

Degree of freedom of a control system is defined as number of closed-loop transfer

functions that can be fiddled with autonomously. There are three constituents in a
three degree of freedom control structure, and they are (i) concerned with closed-
loop stability, (ii) concerned with shaping the closed-loop response, and (iii) used
to relinquish the disturbance [19]. Schematic diagram and structure of three degree
of freedom controller structure are shown in Figs. 1 and 2, respectively, where R(s)
symbolizes input reference signal, feedback from measured system output symbol-
ized as Y (s), P(s) represents process plant model, C(s) represents single degree of
freedom controller, D(s) implies load disturbance, R(s) is the reference controller,
and ffc (s) is the feed-forward controller. The combination of controller C(s) and
R(s) take care of the static and dynamic qualities of the output response Y (s)/R(s).
The location of feed-forward control in three degree of freedom controller structure
varies according to the need. The feed-forward control in the proposed controller is
so located that the controller FFc (s) available in this structure is used to eliminate
the load disturbance D(s). This can happen, if it satisfies the equation given by (3).
Crow Search Algorithm Optimized 3DOF-PID Controller … 187

D (s )

FFc ( s )

R (s )
+−
−
C (s )
++ P (s )
Y (s)

Fig. 1 Schematic diagram of 3DOF controller

Fig. 2 Structure of 3DOF

controller

P(s) − C(s)P(s)FFc (s) = 0 (5)

Thus, it makes the compulsion to update feed-forward controller FFc parameter

upon each change in controller C(s) parameter. The single-order controller C(s)
contains K p , K I as proportional gain and integral gain, respectively, and K D as the
derivative gain. RC(s) has ci as derivative set point weight for reference signal R(s).
G ff is the gain parameter of the feed-forward controller.

5 Results and Discussion

As shown in Fig. 3, a step-load disturbance of 10% is provided in area-1 to observe

the transiency of the system by using 2DOF-PID and 3DOF-PID tuned by CS indi-
vidually for a comparative study. Figures 4, 5 and 6 show the deviations of frequency
in each area and tie-line power deviation for small change in load controlled by dif-
ferent controllers tuned by crow search algorithm. In this paper, matter of concern
for AGC in a two-area interconnected reheat thermal power system along with radox
flow battery is tackled. Each area of the interconnected power system consists of a
governor, turbine, and a reheater in addition to RFB in both the areas. The proposed
power system depicted in Fig. 3 is build up in MATLAB/SIMULINK environment
crow search program is written in.m file to optimize the gains of proposed controllers
portrayed in Table 1. Exercising these controller gains’ dynamic performance of the
considered power system is studied by applying a sudden step-load change of 0.1 pu
in area 1. Pictorial presentations are depicted from Figs. 4, 5 and 6, i.e., frequency
188 S. A. R. Tarafdar et al.

Fig. 3 Transfer function model of the proposed power system [7]

0
Frequency Deviation in area1 (Hz)

-0.01

-0.02

-0.03
CS 3DOF-PID with RFB
-0.04 CS 3DOF-PID without RFB
CS 2DOF-PID

-0.05

-0.06

-0.07

-0.08
0 5 10 15 20 25 30
Time (sec)

Fig. 4 Frequency deviation in area-1

deviation in area-1  f 1 & area-2  f 2 and tie-line power deviation Ptie after endur-
ing a step-load perturbation of 0.1 pu in area-1 with different proposed controllers
and with/without RFB. Undershoot (Ush ), overshoot (Osh ), and settling time (Ts )
(with 0.5% band for  f 1 and  f 2 and 0.05% band for Ptie ) of  f 1 ,  f 2 , and
Ptie with various controllers, optimized through global crow search algorithm are
depicted in Table 2.
Crow Search Algorithm Optimized 3DOF-PID Controller … 189

x 10 -3
5
Frequency Deviation in area2 (Hz)

-5
CS 3DOF-PID with RFB
CS 3DOF-PID without RFB
CS 2DOF-PID
-10

-15

-20
0 5 10 15 20 25 30
Time (sec)

Fig. 5 Frequency deviation in area-2

x 10 -3
5

0
Power Deviation in ptie line (pu)

-5
CS 3DOF-PID with RFB
CS 3DOF-PID without RFB
CS 2DOF-PID
-10

-15

-20
0 5 10 15 20 25 30
Time (sec)

Fig. 6 Tie-line power deviation

Four performance indices such as Ush , Osh , Ts , and the value of ITAE fitness func-
tions are chosen to compare the performance of various controllers and with/without
RFB. The proposed 3DOF-PID controller with RFB ensures stupendous performance
indices of 0.5658 compared 3DOF-PID controller without RFB of 0.7835. Less is
190 S. A. R. Tarafdar et al.

Table 1 Optimum controller gains of two-area reheat thermal system

Area Parameters Controllers
3DOF-PID with RFB 3DOF-PID without RFB 2DOF-PID
Area-1 β 1.9999 1.9904 2.0000
γ 0.0101 0.0101 0.7145
Gff 0.0103 0.0172 –
KP 2.0000 1.8775 1.2281
KI 1.9898 1.9937 1.8291
KD 1.9989 1.9715 0.6621
N 499 302.7789 372.8549
Area-2 β 1.9997 1.1829 0.1672
γ 0.0108 0.0101 0.5230
Gff 1.9989 1.9940 –
KP 0.0101 1.9729 1.3619
KI 0.0103 1.9945 1.9590
KD 0.0105 0.3750 1.2281
N 302 301 383.1812

Table 2 Ush , Osh , and Ts (0.5% band for  f 1 and  f 2 and 0.05% band for Ptie ) of  f 1 ,  f 2 ,
and Ptie with different controllers
Controllers Parameters  f1  f2 Ptie
3DOF-PID with RFB Ush −37.9109 −12.3238 −10.7049
Osh 0.2341 0.1979 0.1897
Ts 1.18 2.10 5.93
3DOF-PID without RFB Ush −48.9957 −14.4933 −11.8680
Osh 4.9759 0.3085 0.7627
Ts 1.88 2.57 11.76
2DOF-PID Ush −75.6578 −15.9986 −17.1770
Osh 5.9356 0.7167 1.1087
Ts 2.40 2.17 12.11

the ITAE value better is the system performance and vice versa. It is observed from
Figs. 4, 5 and 6 and Table 2 that the proposed CS-based 3DOF-PID controller with
RFB performs noteworthy improvements in all aspects compared to 3DOF/2DOF-
PID controllers without RFB. So, it can be added as concept of implementing RFB
in both the areas of power system and improves the controller performance to a great
coverage. A bar plot view of Ush , Osh , and Ts are is shown in Fig. 6 for healthier
contrast of the system under various aspects (Fig. 7).
Percentage improvement in Ush , Osh , and Ts with CS-based 3DOF-PID controller
without RFB as compared to 2DOF-PID are given in Table 3.
Crow Search Algorithm Optimized 3DOF-PID Controller … 191

x 10-3
0 6

-0.01
5
-0.02
CS 3DOF-PID with RFB
4 CS 3DOF-PID without RFB
-0.03 CS 2DOF-PID
CS 3DOF-PID with RFB
CS 3DOF-PID without RFB
-0.04 CS 2DOF-PID 3

-0.05
2
-0.06
1
-0.07

-0.08 0
1 2 3 1 2 3

a b
0.014

0.012 CS 3DOF-PID with RFB

CS 3DOF-PID without RFB
0.01 CS 2DOF-PID

0.008

0.006

0.004

0.002

0
1 2 3
c

Fig. 7 Bar plot representation of a undershoot, b overshoot, c settling time

From Table 3, it is comprehensible that proposed CS-based 3DOF-PID controller

with RFB improves Ush of  f 1 ,  f 2 , and Ptie by 22.62%, 14.97%, and 09.8%,
respectively, Osh of  f 1 ,  f 2 , and Ptie by 95.30%, 35.85% and 75.13% and Ts of
 f 1 ,  f 2 , and Ptie by 37.36%, 18.42%, and 49.57%, respectively, in comparison
with CS-based 3DOF-PID controller without RFB. Similarly, with the 3DOF-PID
controller improvement in Ush of  f 1 ,  f 2 , and Ptie are 49.89%, 22.97%, and
37.68%, respectively, in Osh of  f 1 ,  f 2 , and Ptie are 96.06%, 72.39%, and 82.89%,
respectively, and in Ts of  f 1 ,  f 2 , and Ptie are 50.76%, 03.3%, and 51.04%,
respectively.

6 Sensitivity Analysis

Sensitive/robustness analysis of the proposed CS-based 3DOF-PID controller with

RFB is done in order to prove its efficacy under system parametric variation. With
optimal controller gains as depicted in Table 1, robustness analysis is done by (i)
varying the loading pattern in the range of −50% to +50% in steps of 25% of their
nominal values and (ii) varying one at a time, all the parameters of the proposed part
in range of [−50% +50%] of their nominal values of power system described below.
 192

Table 3 Percentage improvement in Ush , Osh , and Ts with crow search-based 3DOF-PID controller with RFB
 f1  f2 Ptie
Ush Osh Ts Ush Osh Ts Ush Osh Ts
In comparison with 3DOF-PID controller without RFB 22.62 95.30 37.36 14.97 35.85 18.42 9.8 75.13 49.57
In comparison with 2DOF-PID controller 49.89 96.06 50.76 22.97 72.39 3.3 37.68 82.89 51.04
S. A. R. Tarafdar et al.
Crow Search Algorithm Optimized 3DOF-PID Controller … 193

6.1 Sensitive Analysis by Randomly Varying the Load

Patterns

Variation in steps of 25% is mandated in order to attest the rigidity of the controller
besides the variation in electrical power demand, i.e., SLP. This study is done by
taking the nominal system parameters as given in appendix. Stable dynamic perfor-
mance is put on display from Figs. 8, 9 and 10 with crow search algorithm-optimized

0.01

+25%
0 -25%
Frequency Deviation in area 1 (Hz)

Normal
+50%
-0.01 -50%

-0.02

-0.03

-0.04

-0.05

-0.06
0 5 10 15 20 25 30
Time (sec)

Fig. 8 Frequency deviation in area-1 with varying step input 0.1 pu from −50% to +50% in steps
of 25%

-3
x 10
5
Frequency Deviation in area 2 (Hz)

0 +25%
-25%
Normal
+50%
-5 -50%

-10

-15

-20
0 5 10 15 20 25 30
Time (sec)

Fig. 9 Frequency deviation in area-2 with varying step input 0.1 pu from −50% to +50% in steps
of 25%
194 S. A. R. Tarafdar et al.

-3
x 10
2

0 +25%
-25%
-2 Normal
+50%
Power Deviation in ptie-line (pu)

-50%
-4

-6

-8

-10

-12

-14

-16

-18
0 5 10 15 20 25 30
Time (sec)

Fig. 10 Power deviation in ptie-line with varying step input 0.1 pu from −50% to +50% in steps of
25%

3DOF-PID controller amid RFB under varying system load of 0.1 pu within the range
of −50% to +50% in the steps of 25%. Therefore, finally, it can be concluded that
the proposed CS-based 3DOF-PID controller with RFB units’ AGC system is robust
against varying step-load perturbation.

6.2 Sensitivity Analysis by Varying All the Parameters

of the Proposed Power System

In this study, all the system parameters are varied for -50% and 50% to prove the
robustness of the proposed CS-based 3DOF-PID controller with RFB against para-
metric variation. A step-load perturbation of 10% is applied in area-1 in this study and
Ush , Osh and Ts of  f 1 ,  f 2 , and Ptie are depicted in Table 4. From tables, figures,
and discussion, it is concluded that undershoot, overshoot, and settling time of  f 1 ,
 f 2 , and Ptie of the proposed two-area reheat thermal power system with CS-based
3DOF-PID controller with RFB units’ varies within acceptable range. Hence, it can
be concluded that the proposed CS-based 3DOF-PID controller with RFB units’ is
robust against variation in SLP as well as parametric variation.
Table 4 Robustness analysis with parametric variation using the proposed CS-optimized 3DOF-PID controller with RFB
Parameters % age Ush  f 1 Osh  f 1 Ts  f 1 (s) Ush  f 2 Osh  f 2 Ts  f 2 (s) Ush Osh Ts
deviation (Hz) (Hz) (Hz) (Hz) Ptie (pu) Ptie Ptie (s)
(%) (pu)
TG +50 −42.78 0.97 1.75 −12.50 0.60 3.06 −10.75 0.19 5.88
−50 −31.31 0.91 1.80 −12.19 0.59 3.14 −10.67 0.18 5.97
TT +50 −44.49 1.21 1.65 −13.50 0.62 2.95 −11.31 0.20 5.71
−50 −28.66 0.87 1.89 −11.58 0.57 3.25 −10.39 0.18 6.12
KP +50 −43.03 0.96 1.78 −12.01 0.58 3.06 −10.37 0.18 6.00
−50 −31.21 0.90 1.76 −12.73 0.64 3.26 −11.76 0.21 5.72
TP +50 −33.71 0.91 1.77 −12.66 0.62 3.17 −11.24 0. 20 5.81
Crow Search Algorithm Optimized 3DOF-PID Controller …

−50 −47.07 1.01 1.78 −11.79 0.57 3.04 −10.23 0.17 6.04
TR +50 −37.93 0.75 1.84 −12.56 0.54 3.26 −10.89 0.19 6.46
−50 −37.84 1.85 1.63 −11.69 0.42 2.79 −10.20 0 5.32
KR +50 −31.54 0.15 1.63 −8.58 0 2.73 −8.15 0 8.02
−50 −51.31 4.43 1.909 −20.24 3.29 3.29 −16.03 1.80 11.81
195
196 S. A. R. Tarafdar et al.

7 Conclusion

In this present work, in order to illustrate the load versus frequency nature with an
increment in load, two-area reheat thermal system transfer function model is deemed
in addition with a secondary controller. So, initially, a two-area single-unit reheat
thermal system is considered and the gains of 2DOF/3DOF-PID controllers are tuned
by a population-based met-heuristic CS algorithm. 3DOF-PID controlled out forms
better in terms of transiency when subjected to a disturbance of 0.1 pu. Later on,
redox flow battery (RFB) is introduced in both the areas additionally to improve the
system performance by minimizing the error in frequency and achieving minimum
objective function ITAE. Robustness analysis is performed for ±50% by varying
system parameters and SLP of area-1.

Appendix

Tg = 0.08 s; Tt = 0.3 s; Tr = 10.0 s; K r = 0.5; TPS = 20 s; K PS = 120; R =

2.4 Hz/MW; B = 0.425 MW/Hz; T12 = 0.086.

References

1. Elgard, O.I.: Electric Energy Systems Theory. McGraw-Hill, New York, NY (1982)
2. Kothari, M.L., Nanda, J., Kothari, D.P., Das, D.: Discrete-mode automatic generation control
of a two-area reheat thermal system with new area control error. IEEE Trans. Power System
4(2), 730–738 (1989)
3. Nanda, J., Mishra, S., Saikia, L.C.: Maiden application of bacterial foraging based optimization
technique in multi-area automatic generation control. IEEE Trans. Power Syst. 24, 602–609
(2009)
4. Arya, Y., Kumar, N.: Optimal AGC with radox flow batteries in multi-area restructured power
systems. Int. J. Eng. Sci. Technol. 19(3), 1145–1159 (2016)
5. Nayak, J.R., Sahu, B.K., Pati, T.K.: Load Frequency control of a two-area non-reheat thermal
system using Type-2 Fuzzy system optimized DEPSO algorithm. In: International Conference
on Energy, Power and Environment: Towards Sustainable Growth (ICEPE), pp. 1–5 (2015)
6. Saikia, L.C., Sinha, N.: Automatic generation control of a multi-area system using ant-lion
optimizer algorithm based PID plus second order derivative controller. Int. J. Electr. Power
Energy Syst. 80, 52–63 (2016)
7. Sahu, B. K., Pati, S., Panda, S.: Hybrid differential evolution particle swarm optimization
optimized fuzzy proportional–integral derivative controller for automatic generation control of
interconnected power system. IET Gener. Transm. Distrib. 1–12 (2014)
8. Pati, T.K., Nayak, J.R., Sahu, B.K.: Application of TLBO algorithm to study the performance of
automatic generation control of a two-area multi-units. In: 2015 IEEE International Conference
on Signal Processing, Informatics, Communication and Energy Systems (SPICES), pp. 1–5
(2015)
9. Rahman, A., Saikia, L.C., Sinha, N.: AGC of dish-sterling solar thermal integrated system with
biogeography based optimized three-degree of freedom PID controller. IET Renew. Power
Gener. 10(8), 1161–1170 (2016)
Crow Search Algorithm Optimized 3DOF-PID Controller … 197

10. Pati, T.K., Sahu, B.K.: Application of Hybrid DEPSO algorithm to study the performance
of different PID controller structures for automatic generation control of a two-area multi-
units interconnected power system. In: 2014 International Conference on Circuit, Power and
Computing Technologies [ICCPCT], pp. 787–791 (2014)
11. Sahu, B.K., Pati, T.K., Nayak, J.R., Panda, S., Kar, S.K.: A novel hybrid LUS–TLBO optimized
fuzzy-PID controller for load frequency control of multi-source power system. Electr. Power
Energy Syst. 74, 58–69 (2016)
12. Mohanty, B., Panda, S., Hota, P.K.: Controller parameters tuning of differential evolution
algorithm and its application to load frequency control of multi-source power system. Int. J.
Electr. Power Energy Syst. 54, 77–85 (2014)
13. Pradhan, P.C., Sahu, R.K., Panda, S.: Firefly algorithm optimized fuzzy PID controller for AGC
of multi-area multi-source power systems with UPFC and SMES. Int. J. Eng. Sci. Technol. 19,
338–354 (2016)
14. Mishra, D., Pati, T.K., Nayak, J.R., Mohanty, S.: Performance evaluation of a two-area hydro
thermal power system using ALOPEX-BASED DE algorithm. In: International Conference on
Information Technology (ICIT) (2017)
15. Nayak, J.R., Shaw, B., Sahu, B.K.: Application of adaptive-SOS (ASOS) algorithm based
interval type-2 fuzzy-PID controller with derivative filter for automatic generation control of
an interconnected power system. Int. J. Eng. Sci. Technol. 21(3), 465–485 (2018)
16. Rincon, P.: Science/nature|crows and jays top bird IQ scale. BBC News
17. Prior, H., Schwarz, A., Güntürkün, O.: Mirror-induced behavior in the magpie (pica pica):
evidence of self-recognition. PLoS Biol. 6(8), e202 (2008)
18. Askarzadeh, A.: A novel meta heuristic method for solving constrained engineering optimiza-
tion problems: crow search algorithm. Comput. Struct. 169, 1–12 (2016)
19. Sahu, R.K., Goripottu, T.S., Panda, S.: A hybrid DE–PS algorithm for load frequency control
under deregulated power system with UPFC and RFB. Ain Shams Eng. J. 6, 893–911 (2015)
Improvement of Transient Stability
of AC-DC Power System Using RPSO
Based Sliding Mode Controller

Tanmoy Parida and Niranjan Nayak

Abstract The research is attractive in application of robust controller in VSC-

HVDC-based power system due to strong nonlinearity, coupling, and multi-input
multi-output (MIMO) system. The power system is highly nonlinear and complex
in nature. Thus, stability is a major issue in the interconnected power system. Many
robust controller techniques have been applied to solve the stability issues. In the
majority of the controller plan, the determination of gains likewise influences the
strength and productivity of the controller. The determination of legitimate gain is
profoundly entangled in a multi-machine power control system. Here, a four-machine
two-area power system interconnected with VSC-HVDC system is taken for study.
As a matter of first importance, proportional integral (PI) controller is applied toward
enhancement of the stability. The same system is put under sliding mode controller
with same working condition. Further, another particle swarm optimization method,
known as Regularized particle swarm optimization (RPSO), is applied to locate the
best estimations of gains of the sliding mode controller, and the model is simulated
in MATLAB/SIMULINK programming. It is seen that the RPSO-HVDC performs
superior to SMC and PI controller.

Keywords VSC-HVDC · PSO · RPSO · PI controller · Sliding mode controller

1 Introduction

VSC-based high voltage direct current (VSC-HVDC) modules are an important por-
tion of interconnected power system in present modern power system. It has high
complexity and nonlinearity due to its interconnected structure. Thus, stable power
flow is still a great challenge. The bulk power transportation over long distance has

T. Parida (B) · N. Nayak

Institute of Technical Education and Research, Siksha ‘O’ Anusandhan (Deemed to be
University), Bhubaneswar 751030, India
e-mail: [email protected]
N. Nayak
e-mail: [email protected]

© Springer Nature Singapore Pte Ltd. 2020 199

G. Pradhan et al. (eds.), Advances in Electrical Control and Signal Systems,
Lecture Notes in Electrical Engineering 665,
https://doi.org/10.1007/978-981-15-5262-5_13
200 T. Parida and N. Nayak

become a key issue regarding stability problems. VSC-HVDC technology ensures

huge power harvesting and transporting over long distance [1, 2].
The robust controllers like internal model control (IMC) and H-inf control are
applied to a weak AC system connected to VSC-HVDC and their capability is verified
successfully. Various control strategy of VSC-HVDC system has been analyzed in
controlling the parameters like voltage, current, power, and the control parameters
[3, 4]. In the article [5], PI controller is tested on VSC-HVDC-based power system
with LG fault in different states of the power system using PSCAD/EMTDC software
package. A back stepping controller is designed for voltage balance and active and
reactive power control via Lyapunov theorem of asymptotic stability [6]. In the study
of literature, many authors have exhibited an improved DC voltage control in order
to improve system stability. To improve the robustness and stability, the second-
order sliding mode control is exhibited in VSC-HVDC system [7–10]. A combined
approach with PSO and fuzzy logic is used in a VSC-HVDC system for reducing
the low frequency damping in various operating conditions [11, 12].
In this paper, a four-machine power system is considered for stability analysis. The
PI controller and sliding mode controller (SMC) are investigated for their robustness
under various fault conditions. Since, in the four-machine system, the stability is very
tough challenging which is affected by the gains of the sliding mode controller. Thus,
a new type of particle swarm optimization known as regularized PSO is applied to
find the optimized gains [13–15]. The simulation of the model with optimized gains
exhibits better stability than conventional sliding and PI controller.
The remaining part of the article is organized as follows. The mathematical mod-
eling of the multi-machine power system is described in the second section. The
third section of this article contains analysis of PI controller and sliding mode con-
troller. The fourth section discusses about the structure of PSO and regularized PSO.
The fifth section shows the simulation results of optimized sliding mode controller
performance and the sixth section concludes the work.

2 Mathematical Modeling

The system consists of two equivalent power systems, a DC link, and two VSC
converter, one acts as rectifier and other works as inverter, with PWM control strategy
with impedances Z r = rr + xr and Z i = ri + xi . The AC system in both sides consists
of two synchronous generators, two transformers, three phase transmission lines, and
passive loads. Two synchronous generators are connected through a VSC-HVDC
transmission system interface. The VSC-HVDC systems consist of transmission
lines of transmission impedances of Z r t1 = Rr t1 + X t1 , and Z r t2 = Rt2 + j X t2 . The
DC capacitors are utilized to enhance the DC voltage, relieve the effect of impulse
current when the bridge is shut and can likewise constrict the harmonics on DC side.
An AC filter is utilized in both sides of AC system to eliminate harmonics (Fig. 1).
The VSC-HVDC is transformed to a d-q reference frame,
Improvement of Transient Stability of AC-DC Power System … 201

Fig. 1 Schematic diagram of four machine VSC-HVDC system

VSC-I: Converter station

di d1
L1 = ω1 L 1 i q1 − R 1 i d1 + u r d1 − u sd1 (1)
dt
di q1
L1 = −ω1 L 1 i q1 − R 1 i q1 + u rq1 − u sq1 (2)
dt
VSC-II: Converter station
di d2
L2 = ω2 L 2 i d1 − R 2 i q2 + u r d2 − u sd2 (3)
dt
di q1
L2 = −ω2 L 2 i q2 − R 2 i d2 + u rq1 − u sq1 (4)
dt
Here, ω1 , ω2 represent the frequencies of the AC network. i dk , i qk are the d-axis
and q-axis grid side current. u dk , u sqk are d- & q-axis grid voltages. urqk are the
d-axis and q-axis voltages in the AC side of the VSC (where k = 1, 2).
Neglecting converter, the active power balance equation can be described as
follows.

Pdc = Pr 1 + Pr 2 (5)

where Pr1 , Pr2 are the active power supplied by VSC-1 and VSC-2, respectively, and
Pr1 , Pr2 can be given as

3 
Pr 1 = u r d1 i d1 + u rq1 i q1 (6)
2
3 
Pr 2 = u r d2 i d2 + u rq2 i q2 (7)
2
202 T. Parida and N. Nayak

The transformer losses are very small; thus, it can be neglected and the equations
are modified as
3 
Pr 1 ≈ P1 = u sd1 i d1 + u sq1 i q1 (8)
2
3 
Pr 2 ≈ P2 = u sd2 i d2 + u sq2 i q2 (9)
2
The d-axis and the ac voltages are on the same line by phase-locked loops (PLL).
Thus, u sq1 and u sq2 are zero. u sd1 and u sd2 are same as u sk .
Further the Eqs. (6–7) are modified as

3 3
P1 = (u sd1 i d1 ), P2 = (u sd2 i d2 ) (10)
2 2
Similarly, the reactive power expressions are as following

3  3 
Q1 = − u sq1 i q1 , Q2 = − u sq2 i q2 (11)
2 2
From Eq. (4)

Pdc = Pr 1 + Pr 2 or Pdc = P1 + P2 (12)

3
u dc i dc = (u sd1 i d1 + u sd2 i d2 ) (13)
2
du dc 3
u dc Cdc = (u sd1 i d1 + u sd2 i d2 ) (14)
dt 2

Remark 1 In steady state, usd1 and usd2 are constant, so i d1 and i q1 play key role
for exchanging active and reactive power between VSC-I and AC-I. Similarly, the
power exchange between VSC-II and AC-II is performed by i d2 and i q2

Remark 2 During the operation of the system, u dc should be maintained its rated
value. From Eq. (14), it is observed that the AC and DC power must be balanced.

3 Controller Design

PI controller is reasonable for this sort of framework to damp out the system dynam-
ics. In essential PI controller, d-axis and q-axis voltages u r d and u rq are the control
inputs of the rectifier and u id and u iq are the same of the inverter. Here, PI controller
is applied to stifle the system oscillations. Likewise, two vigorous controllers like
Improvement of Transient Stability of AC-DC Power System … 203

sliding mode controller are applied to this framework so as to balance out with a less
overshoot and settling time.

3.1 PI Controller

In general, there is wide use of PI control in a power system control sector because
it is simple and easy to implement and moderate robust. The greatest short fall of
such a controller is selection of PI gains because it is not constant for wide range of
operations. The trial and error method is adopted to suppress these difficulties.

t
y(t) = k p er(t) + ki er(t)dt (15)
0

Figures 2 and 3 show the block representation of PI controller of rectifier and

inverter, respectively.
For minimization of steady-state error, gradation of power system and elimination
of forced oscillation PI controller are extensively used in power system control. The
errors in PI control are given as following.

er1 = u dcref − u dc and er2 = Vr ref − Vr (16)

er3 = P1ref − P1 and er4 = Q 2ref − Q 2 (17)

u 1 , u 2 , u 3 and u 4 are the controller outputs.

udcref er 1 u1 Vrref er2 u2

Ki 1 + K i2
+ Kp 1 + s Kp 2 +
− − s
Vr
udc

Fig. 2 PI controller for converter

P1 ref er 3 Ki 3 u3 Q 2 REF er4

+
Kp3 +
K i4 u4
− s + Kp 4 +
Q2 − s
P1

Fig. 3 PI controller for inverter

204 T. Parida and N. Nayak

3.2 Sliding Mode Controller

It is a nonlinear control technique tested fruitfully in various engineering applications.

In this control conspire, an error in chosen state direction is brought close to the sliding
surface and settles at an equilibrium point.
The general condition anticipated before to choose the favored sliding surface
σ (x) is as follows
 qn−1
d
σ (x) = + ki er(x) (18)
dx

where ki = factor which decides the control band width.

er(x) = error of the output.
qn = degree of state variable.
The sliding surface is picked utilizing a positive scalar Lyapunov cost work v(x) >
0 as

v(x) = σ 2 (x)/2 (19)

The required thing for the function minimization is

σ1 (x)σ̇ (x) < 0. (20)

u(t) = u eq (t) + u st . (21)

Which must satisfy the convergence condition?

i.e., σ1 (x)σ̇ (x) < 0. (22)

Rectifier station using SMC

As per Eq. (19), the sliding surface can be represented for the output u dc1 .

d(er1 )
σ1 = + k1 (er1 ) (23)
dt

σ2 = er2 (24)

From the above equations, the fleeting derivative can be derived as

  
σ̇1 = d2 (er1 ) dt 2 + k1 d(er1 ) dt, σ̇2 = d(er2 ) dt (25)

The design of invariant sliding surface is as follows.

Improvement of Transient Stability of AC-DC Power System … 205

σ̇1 = −ρ1 sign(σ1 ) and σ̇2 = −ρ2 sign(σ2 ) (26)

σ̇1 = −ρ1 tanh(σ1 ) and σ̇2 = −ρ2 tanh(σ2 ) (27)

  
λ1 = − ρ1 tanh(σ1 ) + k1 d(er1 ) dt (28)

λ2 = ρ2 tanh(σ2 ) (29)

er1 , er2 are same as PI controller.

Inverter station using SMC

The SMC for inverter can be derived same as rectifier. The ideal surface that fulfills
the picked output states can be composed as

σ3 = er3 and σ4 = er4 (30)

where

er3 = Ii D − Ii Dref and er4 = Ii Q − Ii Qref (31)

The new sliding surface elements chosen are as per the following

σ̇3 = −ρ3 tanh(σ3 ) and σ̇4 = −ρ4 tanh(σ4 ) (32)

Further

λ3 = −ρ3 tanh(σ3 ), λ4 = −ρ4 tanh(σ4 ) (33)

3.3 Particle Swarm Optimization

It is swarm evolutionary computational algorithm followed by bird’s food searching

process. Each bird in a flocking group represents as an individual element in PSO.
They hold the data of the position and its velocity of each individual element. They
are made to learn from both of their own best positional data pbest about their ability
and the searching space.
PSO is an iterative process. After some iterative loops, PSO starts converging and
reaches at the optimum result. The process of calculation of velocity and position is
given by,
    
velik+1 = ωvik + c1r1 pbesti − xik + c2 r2 gbest − xik
(34)
xik+1 = xi + vik+1
206 T. Parida and N. Nayak

Here, xik and vik are the position and velocity of ith element for kth iterative step.
pbest is the ith element’s individual location, and gbest is the universal best location
in the whole swarm of particles in the search space.
Furthermore, ω is the inertia weight feature, which affects the performance search
techniques.c1 is the cognitive and c2 is the social factor whose value controls the
search progress.r1 and r2 are consistently scattered arbitrary numbers in [0, 1], which
resolve the algorithm randomness.

3.4 Regularized PSO

During working of PSO, if the velocity of each particle is constrained, then the
velocity may increase to an undesirable level within a few iterations. Thus, this
algorithm is modified by introducing constriction coefficients to regulate the element
velocities. The coefficient regulates the element’s motion and proceeds them toward
the convergence. The adapted velocities are shown as
   
velik+1 = w × C F × velik + k1r1 pik − xik + k2 r 2 pgk − xik (35)

Here, CF, C 1 , and C 2 , affects the computational speed and used to locate the ideal
point in the hunt space. Their mathematical form is given as below.

C1 = CF × ϕ1 (36)

C2 = CF × ϕ2 (37)

where
1
CF = (38)
ϕ−2− ϕ 2 − 4ϕ

ϕ = ϕ1 + ϕ2 ≥ 4 (39)

where CF and ϕ are represented as constriction factor and coefficient, respectively.

Now, the constriction elements lead the solution toward optimum value without
using velocity limits. However, it is a better approach is to use limits of velocities
and positions.
Improvement of Transient Stability of AC-DC Power System … 207

4 Simulation Results

Here, in this paper, a VSC-HVDC-based four machine power system is simulated in

different operating conditions. The PI controller, sliding mode controller, and RPSO-
based controller are investigated with their own strength and the results are shown
in the following figures.

Case-1: Voltage Dip at Inverter Bus

The HVDC system is operated alongside a disturbance created by reducing 20% of
inverter bus voltage and after 1.2 s again regained the original voltage. Further, when
the system is simulated with optimized SMC controller gains, then it settles down to
the steady-state condition with a quick settling time and a small overshoot as shown
in Fig. 4.

Case-2 Parallel AC line fault

Figure 5 shows the performance of RPSO-based SMC as compared to PI and con-
ventional sliding mode controller. The system stabilizes in 1.5 s approximately at
all the case of simulation. The result clearly indicates that the RPSO-based sliding
mode controller performs better than SMC and PI controller.

Case-3: Fault in DC line

The regularized particle swarm optimization-based sliding mode controller

0.5 1.3
PI PI
SMC 1.2
SMC
OSMC OSMC
1.1
dw1(rad/sec)

Pe1(pu)

0
1

0.9

-0.5 0.8
0 1 2 3 4 5 6 0 1 2 3 4 5 6
Time(second) Time(second)

1
PI PI
SMC 0.95 SMC
0.5
OSMC OSMC
0.9
Pe2(pu)

Qe3(pu)

0.85
0.4
0.8

0.75
0.3
0.7
0 1 2 3 4 5 6 0 1 2 3 4 5 6
Time(second) Time(second)

PI
1.0003 SMC
0.9
OSMC
0.85
Vdc1(pu)

1.0002
V2(pu)

0.8

0.75
1.0001
0.7 PI
SMC
0.65 OSMC 1
0 1 2 3 4 5 6
0 1 2 3 4 5 6
Time(second) Time(second)

Fig. 4 Performance of the proposed controller at 3-phase fault

208 T. Parida and N. Nayak

0.394
PI
SMC
0.84 OSMC 0.392

delta2(pu)
delta1(pu)

0.835
0.39

0.83
0.388

0.825 0 1 2 3 4 5 6
0 1 2 3 4 5 6 Time(second)
Time(second)

PI 0.02 PI
0.05 SMC SMC
OSMC 0.01 OSMC

dw2(rad/sec)
dw1(rad/sec)

0
0
-0.01

-0.02
-0.05
-0.03
0 1 2 3 4 5 6 0 1 2 3 4 5 6
Time(second) Time(second)

1.05
PI PI
0.82
SMC SMC
1.04 OSMC 0.815 OSMC

0.81
Qe1(pu)
Pe1(pu)

1.03 0.805
0.8
1.02 0.795
0.79
0 1 2 3 4 5 6 0 1 2 3 4 5 6
Time(second) Time(second)

0.44
PI PI
1.218
SMC SMC
0.435
OSMC OSMC
1.216
Pe2(pu)

0.43
V1(pu)

1.214
0.425
1.212
0.42
1.21
0 1 2 3 4 5 6 0 1 2 3 4 5 6
Time(second)
Time(second)

Fig. 5 Behavior of the proposed controller at 3-phase fault in parallel AC line

(RPSOSMC) performs fast in damping oscillation and able to reduce the steady-
state error. The PI controller performance is worst in this case to settle the change
in rotor angular frequency and SMC performs better and the proposed controller
settles the system into original condition nearly in 1.5 s as shown in Fig. 6. The
outcomes portrayed that the damping of oscillations accordingly is quick and the
related error to it is least in accomplishing the steady state. The effect on the reactive
power variations likewise was seen as negligible.

Case-4: Alteration in short-circuit Ratio

The short circuit ratio has been reduced from 10 to 3. The results are presented in
Fig. 7.
Improvement of Transient Stability of AC-DC Power System … 209

0.842 PI PI
SMC SMC
0.391
0.84 OSMC OSMC
0.838
delta1(rad)

delta2(radian)
0.39
0.836

0.834 0.389
0.832

0.83 0.388
0 1 2 3 4 5 6 0 1 2 3 4 5 6
Time(second) Time(second)

0.04
PI 0.01 PI
SMC SMC
0.02 OSMC OSMC
0.005

dw2(rad/sec)
dw1(rad/sec)

0 0

-0.005
-0.02
-0.01
-0.04 0 1 2 3 4 5 6
0 1 2 3 4 5 6
Time(second)
Time(second)

1.04 PI 0.82
SMC
OSMC 0.81
1.035
Qe2(pu)

0.8
Pe(pu)

1.03
0.79

1.025 0.78

1.02 0.77
0 1 2 3 4 5 6 0 1 2 3 4 5 6
Time(second) Time(second)

0.432 0.2
PI PI
0.431 SMC SMC
OSMC 0.195 OSMC
0.43
Pe2(Pu)

Qe2(Pu)

0.429 0.19

0.428
0.185
0.427

0.426 0.18
0 1 2 3 4 5 6 0 1 2 3 4 5 6
Time(second) Time(second)

Fig. 6 Behavior of the controller under 3-phase fault at Dc line

5 Conclusion

Here, in this paper, RPSO-based SMC for a VSC-HVDC system has been displayed
within the sight of dynamic demonstration and parametric vulnerabilities of differ-
ent magnitudes. The sliding mode control drops the nonlinearity and considers the
parametric vulnerabilities and exogenous contributions to the dc-interface converter
model. The said controller is seen as powerful, creating noteworthy damping and a
decrease of overshoots for an assortment of working conditions. The real testing of
SMC comes, when the determination of gains is done by the experimentation tech-
nique, which influences the controller yield. Thus, a transformative enhancement
known as regularized PSO (RPSO) is executed to discover advanced controller picks
up which improves the dependability. In correlation, the proposed control system
shows better execution contrast with ordinary PI and sliding mode controller.
210 T. Parida and N. Nayak

0.86 PI PI
SMC 0.4 SMC
OSMC OSMC
delta1(radian)

0.85

delta2(radian)
0.395

0.84
0.39

0.83 0.385

0.82 0.38
0 1 2 3 4 5 6 0 1 2 3 4 5 6
Time(second) Time(second)

0.15 PI 0.06 PI
0.1 SMC SMC
OSMC 0.04
dw1(rad/sec)

OSMC

dw2(rad/sec)
0.05
0.02
0
0
-0.05
-0.02
-0.1
-0.04
0 1 2 3 4 5 6 0 1 2 3 4 5 6
Time(second) Time(second)

1.1 0.84
PI
SMC 0.82
OSMC
1.05
Pe1(pu)

0.8
Qe1(pu)
0.78

1 0.76
PI
0.74 SMC
OSMC
0.95 0.72
0 1 2 3 4 5 6 0 1 2 3 4 5 6
Time(second)
Time(second)

Fig. 7 Behavior of the proposed controller beneath alternation in short-circuit ratio

References

1. Asplund, G.: Application of HVDC light to power system enhancement. In: IEEE Power
Engineering Society Winter Meeting, vol. 4, pp. 2498–2503, Jan 2000
2. Tzann-Shin, L.: Input-output linearization and zero-dynamics control of three-phase AC/DC
voltage-source converters. IEEE Trans. Power Electron. 18(1), 11–22 (2003)
3. Chen, H., Xu, Z., Zhang, F.: Non-linear control for VSC based HVDC system. In: 2006 IEEE
Power Engineering Society General Meeting
4. Zhang, Y.-A., Mi, Y.-L., Zhu, M., Lu, F.-L.: Adaptive sliding mode control for two-link flexible
manipulators with H/sub/spl infinite tracking performance. In: International Conference on
Machine Learning and Cybernetics, 2005, 18–21 Aug 2005, vol. 2, pp. 702–707
5. Hannan, M.A, Hussin, I., Ker, P.J., Hoque, M.M., Hossain Lipu, M.S., Hussain, A., Abd
Rahman, M.S., Faizal, C.W.M., F. Blaabjerg, F.: Advanced control strategies of VSC based
HVDC transmission system. IEEE Access 6, 1–18 (2018)
6. Karthi, K., Radhakrishnan, R., Baskaran, J.M., Titus, L.S.: Performance analysis on various
controllers of VSC—HVDC transmission systems. In: IEEE Xplorer, pp. 1–6 (2016)
7. Huang, J., Xu, D., Yan, W., Ge, L., Yuan, X.: Nonlinear control of back-to-back VSC-HVDC
system via command-filter back stepping. J. Control Sci. Eng. 2017, 1–11 (2017)
8. Sabba, S., Chikhi, S.: A discrete binary version of bat algorithm for multidimensional knapsack
problem. Int. J. Bio-Inspired Comput. 6(2), 140–152 (2014)
9. Sambariya, D., Prasad, R.: Robust tuning of power system stabilizer for small signal stability
enhancement using metaheuristic bat algorithm. Int. J. Electr. Power Energy Syst. 61, 229–238
(2014)
10. Taha, A.M., Mustapha, A., Chen, S.-D.: Naive bayes-guided bat algorithm for feature selection.
Sci. World J. 2013 (2013)
11. Auger, A., Hansen, N.: A restart CMA evolution strategy with increasing population size. In:
Proceedings of the 2005 IEEE Congress on Evolutionary Computation, pp. 1769–1776 (2005)
Improvement of Transient Stability of AC-DC Power System … 211

12. Liu, S.-H., Mernik, M., Hrnčič, D., Črepinšek, M.: A parameter control method of evolutionary
algorithms using exploration and exploitation measures with a practical application for fitting
Sovova’s mass transfer model. Appl. Soft. Comput. 13, 3792–3805 (2013)
13. Zhang, X., Zou, D., Shen, X.: A novel simple particle swarm optimization algorithm for global
optimization. Mathematics 6(287), 1–34 (2018)
14. Tighzert, L., Fonlupt, C., Mendil, B.: A set of new compact firefly algorithms. Swarm Evol.
Comput. 40, 92–115 (2018)
15. Hossain, M.A., Pota, H.R., Squartini, S., Abdou, A.F., Md Alamgir Hossain: Modified PSO
algorithm for real-time energy management in grid-connected microgrids. Renew. Energy 136,
746–757 (2019)
Adaptive Resource Allocation
for Cognitive Radio-Enabled Smart Grid
Network

Deepa Das and Niranjan Behera

Abstract Demand response management (DRM) in smart grid plays an important

role in balancing electricity demand and supply between the consumers and power
supplier. The paper mainly deals with the home area network (HAN) where smart
meter of each consumer is enabled with cognitive radio (CR) technology. The reliable
communication between the consumer and supplier is ensured by maximizing the
aggregated benefit by optimizing real-time demands at the consumer’s side under the
constraints of interference to the primary user, probability of detection and minimum
achievable data rate. The above objective problem is solved by our proposed adaptive
resource allocation approach based on genetic algorithm (GA). In this approach,
individual consumers maximize their own profit iteratively with optimal transmission
power resulting in maximized aggregated benefit. Simulation results demonstrate the
effectiveness of the proposed algorithm providing benefits to both power supplier and
consumer with optimal power and demands allocation, and with reasonable value of
electricity price.

Keywords Demand response management · Smart grid · Cognitive radio · Genetic

algorithm

1 Introduction

The concept of smart grid (SG) is evolved as the solutions to the issues arise in the
traditional power grid by integrating advanced bi-directional communication system
between the consumers and power provider, intelligent transmission and distribu-
tion system [1, 2]. Hence, reliability, security and efficiency of the power grid are
levitated significantly in smart grid technology. Further, SG balances energy genera-
tion and consumption with real-time electricity pricing, power scheduling and many

D. Das (B) · N. Behera

Department of Electrical Engineering, Government College of Engineering, Kalahandi,
Bhawanipatna 766002, India
e-mail: [email protected]

© Springer Nature Singapore Pte Ltd. 2020 213

G. Pradhan et al. (eds.), Advances in Electrical Control and Signal Systems,
Lecture Notes in Electrical Engineering 665,
https://doi.org/10.1007/978-981-15-5262-5_14
214 D. Das and N. Behera

more operations considering both power suppliers and energy demands of the con-
sumer. This requires incorporation of intelligent smart meter at the consumer side
which mainly supports all network functions and acts as relay between the home
appliances and power supplier. Usually, the data from the smart meter are huge,
and it requires the fastest transmission with higher security. Therefore, a number of
communication technologies have been adopted to meet this purpose. But cognitive
radio (CR) is introduced as a promising candidate which overcomes the static fre-
quency allocation policy by opportunistically accessing the licensed band without
interfering the primary user (PU). Consequently, SG is a multilayer topology pro-
viding real-time interaction between the home area network (HAN) to neighborhood
area network (NAN) and wide area network (WAN) [3]. Smart meters which are
facilitated with CR technology present at HAN, gather all the information from the
appliances regarding energy demands and aim on balancing the energy demands
by accounting the electricity price. The advantages of CR technology integration
to the SG network have been extensively discussed in [4, 5]. However, in spite of
advancement and effectiveness, CR paradigm in the SG leads to several challenges.
It inspired the authors to propose various techniques for obtaining optimal resource
parameters in CR-enabled SG networks. In [6], Fadel et al. proposed clustering, chan-
nel assignment and routing algorithm to reduce the packet loss ensuring high quality
link between the sensor nodes. The authors developed a hybrid energy-efficient spec-
tral honey bee mating optimization-based clustering (ESHC) algorithm to solve the
clustering problem in an energy-efficient manner. Further, quality-aware channel
assignment algorithm (QCA) and energy-efficient honey bee mating optimization
routing algorithm (EHRA) were proposed to efficiently select the spectrum band to
route the data packets from source to destination nodes. In [7], Zhang et al. pro-
posed resource allocation in energy harvesting heterogeneous cognitive radio sensor
networks (HCRSNs). To do this, two separate algorithms such as spectrum sensor
scheduling algorithm for allocating channels to the spectrum sensors and data sensor
resource allocation algorithm for jointly optimizing transmission time, power so as
to minimize the energy consumption of the data sensors were proposed. Abuadbba
et al. proposed a novel hybrid model which combined advanced steganographic algo-
rithms with error detection and correction technique (BCH syndrome code) in CR
smart meter to detect and recover any loss of data that raised due to CR shared spec-
trum noise ensuring security and privacy of the hidden confidential information [8].
In [9], Althunibat et al. proposed a channel selection mechanism in a CR-based SG
networks. Either reliable or high SNR-based channel was selected based on the type
of data to be transmitted. Further, in a CR-based SG network, sensing time plays
a crucial role. Increasing sensing time gives the sensing accuracy, but it leads to
degradation of transmission capacity. Hence, Cacciapuoti et al. proposed a sensing
time optimization technique which simultaneously maximized the channel capac-
ity and reduced the channel outage [10]. Then, the authors designed a switching
procedure between T.V. white space spectrum and industrial scientific and medical
(ISM) band considering a mobile SG such as electric vehicle [11]. The sensing time
was optimized to maximize the achievable throughput under the constraint of outage
rate. In [12], joint spatial and temporal spectrum sharing approach was proposed
Adaptive Resource Allocation for Cognitive … 215

for demand response management in CR-based SG network, which improved the

spectral efficiency while lowering the communication outage probability.
Though those papers discussed various techniques on spectrum allocation, sens-
ing time optimization and energy management schemes, the CR-based SG network
is still a challenging task at the consumer side. This is because the real-time price
increases with the increase in energy demand, and the consumers require some extra
power for spectrum sensing and information transmission over the spectrum band.
To solve this problem, this paper introduces a novel adaptive algorithm to optimize
power demand and transmission power that benefits both power provider and con-
sumer under the constraints of interference power to the PU network and minimum
achievable throughput.
The rest of the paper is organized as follows. System model is discussed in Sect. 2.
Section 3 describes the problem formulation and proposed solution approach. Simu-
lation results are analyzed to demonstrate our algorithm in Sect. 4. Finally, the paper
is concluded in Sect. 5.

2 System Model

This Section provides the system model describing the distribution of consumers in
the network and application of the CR in SG.

2.1 Demand Response Management

A CR-based SG network is consisting of one power supplier and M consumers as

shown in Fig. 1. M consumers are facilitated with intelligent sensors called smart
meter having the cognitive capability. Let S be the total power supply in a day and
D be the total power demands of the M consumers in a day. The entire day duration
is divided into four slots, depending on the characteristic behavior of the consumers
in an usual working day. Hence, the four time slots are T = {t1 , t2 , t3 , t4 } where t1 =
morning hour (6 AM–10 AM), t2 = office hour (10 AM–5 PM), t3 = evening hour
(5 PM–10 PM) and t4 = night hour (10 PM–6 AM). Therefore, D can be defined as


4 
M
D= dtm (1)
t=1 m=1

where dtm is the power demand of the mth consumer at tth instant. Further, the
supplier declares the electricity price ρ based on the power demands of the consumers.
The total expenses of supplying power S are represented by the cost function
F(S) which is increasing and convex. Therefore, the net profit of the power supplier
is given by
216 D. Das and N. Behera

Fig. 1 System model showing the distribution of consumers

R S = ρ S − F(S) (2)

where F(S) is given as F(S) = x2 S 2 + y S + z. x, y and z are constants [13].

Total expenses of the M consumers are ρ D. Further, the gain function of the mth
consumer at tth instant is G tm which is given by
 ψm 2
cm dtm − d
2 tm
0 ≤ dtm ≤ cm
ψm
G tm (dtm , cm ) = cm2 (3)
2
dtm ≥ cm
ψm

Always, the individual consumer aims to maximize its own profit, i.e.,

G tm (dtm , cm ) − ρdtm (4)

The total profit at the consumers side at t th slot is given by


M 
M
G tm (dtm , cm ) − ρ dtm (5)
m=1 m=1

where cm parameter characterizes the behavior of the mth consumer. ψm is a constant.

Higher value of ψm requires lower energy consumption of the mth consumer to reach
the saturation.
Adaptive Resource Allocation for Cognitive … 217

2.2 Cognitive Radio

The communication quality improvement and faster data transmission are aided by
integrating CR into demand response management. Generally, the gateway supports
two-way communication. The data transfer occurs either on the licensed band denoted
as chl through CR spectrum sensing or on the unlicensed band chu . The smart meter
is present in the consumer’s house and is enabled with CR which performs spec-
trum sensing before selecting the channel for data transmission. The smart meter
opportunistically selects chl on the absence of the PU. Let P1 and P0 represent the
probability that chl is occupied and not occupied by the PU, respectively.
Let Sw represents the channel switching probability to chl . Sw occurs in two cases;
(i) when the PU is absent and is correctly detected by the M CRs (ii) PU is in operation
on the licensed band but is missed by the CRs. Therefore, Sw is represented as
 
Sw = P1 (1 − Q d ) + P0 1 − Q f (6)

where Q d and Q f represent the cooperative detection probability and false alarm
probability, respectively, which are denoted as [14]


M
Qd = 1 − (1 − Pdm ) (7)
m=1

or


M
 
Qf = 1− 1 − Pf m (8)
m=1

where Pdm and P f m are detection probability and false alarm probability of the
mth consumer. The overall channel outage considers the individual licensed channel
outage Ol and unlicensed channel outage Ou . The probability of selecting chl for
data transmission is Sw Ol , and the probability of selecting chu on the unavailability
of chl is (1 − Sw )Ou . Therefore, the overall outage OV is

OV = (1 − Sw )Ou + Sw Ol
= (Ol − Ou )Sw + Ou (9)

It is obvious to assume that communication quality of licensed band is better than

the unlicensed. Therefore, Ol is less than Ou resulting in OV < Ou . From (9), it is
obvious that OV decreases with increase in Sw .
218 D. Das and N. Behera

3 Problem Formulation and Solution Approach

3.1 Problem Formulation

The system model is consisting of M = 4 consumers. Their demands vary according

to consumer types and their appliances’ usages which are different for four different
time slots t1 , t2 , t3 and t4 over the day duration T = 24 h. The four consumer types
are working person, shopkeeper, consumer staying at home and the corporate office.
If the consumer type is working, then in t2 slot, most of the appliances are in off
condition, and hence, demand is lesser during this slot than t1 and t3 . Electricity
demand from the shopkeeper type consumer is mostly for the duration t2 and t3 . In
general, electricity demands of the consumer staying at home is more in t2 slot than
t1 , t2 and t4 . The demands of the corporate office are almost constant for t1 , t2 , t3
and t4 duration. Further, the smart meter in each consumer is facilitated with CR
technology which requires certain power Ptm for data transmission over the licensed
band. The transmission power assigned to the mth consumer depends upon minimum
achievable throughput Tth and maximum power Ptmax . This is obvious that system
throughput of the mth consumer THm depends on the transmission power and the
wireless channel medium over which the data to be transferred is given by
 
P t |h cm |2
TH m = log2 1 + tm (10)
N0

where h cm is the channel coefficient between the mth consumer and the gateway. N0
t
is the noise power. Ptm is the transmission power of m th consumer at t th slot. Hence,
t
Ptm is time dependent. But Tth is assumed to be constant for all consumers irrespective
of time slots. Therefore, the minimum power required by the mth consumer at tth
slot is given by
 T 
2 th − 1 N0
t
Ptm = (11)
min
|h cm |2

Further,


M
Ptm = Ptmax (12)
m=1

where Ptmax is the maximum transmission power allowed on the licensed band.
Therefore, the transmission power assigned to the mth consumer must satisfy the
constraints (11) and (12). It is also necessary that the CR secondary users or the
CR-enabled smart meter are allowed to transmit on the licensed band until it does
not cause any interference to the legitimate user. Therefore,
Adaptive Resource Allocation for Cognitive … 219


M
2
t
Ptm h sp ≤ Ith (13)
m=1

where h sp is the channel between the smart meter and the primary receiver (PR). Ith
is the minimum interference threshold that the PR can tolerate.

3.2 Solution Approach

The objective of the individual consumer is to maximize

 
OV t .dtm
Ytm = G tm (dtm , cm ) − ρ (14)
+SW t .Ptm
t
+ Ps

where Ps is the sensing power. The consumers cooperatively try to achieve maximum
profit which is given by


M M 
 
OV t .dtm
Yt M = G tm (dtm , cm ) − ρ (15)
+SW t .Ptm
t
+ Ps
m=1 m=1


M
St ≥ (1 − OV t ) dtm (16)
m=1

where OV t and SW t are overall outage probability and switching probability at tth
slot. The power supplier declares price ρ and power supply St1 , St2 , St3 and St4 for
t1 , t2 , t3 and t4 , respectively. The objective function is dependent on Ptm
t
and dtm .
Hence, these both the terms must be updated for each time slots so as to maximize
the individual and overall profit (14) and (15), respectively.

Preposition The objective function (14) is strictly concave in dtm for the constant
Ptm .

Proof Keeping Ptm constant, the first derivative of (14) is obtained by

dYtm
= cm − ψm dm,tl − ρ OV t (17)
dtm | Ptm =constant

If the Hessian matrix H (Ytm ) is negative, then Ytm is strictly concave in dtm . The
second derivative of (14) with respect to dtm for a constant Ptm is obtained as

d 2 Ytm −ψm , t = current time

= (18)
2
dtm Ptm =constant
0, t = current time
220 D. Das and N. Behera

It is observed from (18) that diagonal elements of H (Ytm ) are negative and other
elements are zero. So, (14) is a concave function of dtm for constant Ptm but decreases
with increase in Ptm . Hence, the objective function (14) needs to be analyzed by using
metaheuristic algorithm. To solve this problem, a novel algorithm based on GA is
proposed which maximizes (15) by simultaneously optimizing Ptm and dtm . The
detailed steps of the proposed Algorithm 1 are given as follows.

Algorithm 1
ρ is declared.
for t = 1: T
Step 1: St is declared satisfying the constraint (16).
Step 2: Initialize the chromosomes randomly of size M and dimension I over the
search space. Each chromosome represents the power demand. Also, each consumer
generates transmission power Ptm satisfying the constraints (11) and (12).
Step 3: Evaluate R1 of each i chromosome.
⎧
⎪ 
M
⎪
⎨ 1 if Ith ≥ Ptm h sp
2

R1 = m=1
⎪
⎪ 
M
2
⎩ 0 if Ith < Ptm h sp
m=1

If R1 = 0, then discard that ith chromose.

Initialize Fm1 1 , Fm1 2 , Fm1 3 and Fm1 4 .
Initialize Fm2 1 , Fm2 2 , Fm2 3 and Fm2 4 .
 4: 1j ← 1 1
Step   1  
F ( j) > Fm 1 ( j − 1) Fm 2 ( j) > Fm1 2 ( j − 1)
if  m11   
Fm 3 ( j) > Fm1 3 ( j − 1) || Fm1 4 ( j) > Fm1 4 ( j − 1)
Assign rank 1 to the ith chromosome.
else
Replace the ith chromosome by ith chromosome of ( j − 1) iteration.
end if 2   2  
F ( j) > Fm2 1 ( j − 1) Fm 2 ( j) > Fm2 2 ( j − 1)
if  m21   
Fm 3 ( j) > Fm2 3 ( j − 1) || Fm2 4 ( j) > Fm2 4 ( j − 1)
Assign rank 2 to the ith chromosome
else
Replace the ith chromosome by ith chromosome of ( j − 1) iteration.
end if
Step 5: The chromosomes of same rank are assigned with one extra scaling factor
which decides the best chromosomes for offspring generation. The scaling factor is
C1 =  1j .
1+ j=1 OA

1 ifYt M ( j) ≥ Yt M ( j − 1)
OA =
0 ifYt M ( j) < Yt M ( j − 1)
Adaptive Resource Allocation for Cognitive … 221

Step 6: The chromosomes with higher rank are selected. If the rank is same, then chro-
mosomes with lower scaling factors are chosen. The selection is based on tournament
selection.
Step 7: The offspring is reproduced using simulated binary crossover and polynomial
mutation.
Step 8: The offspring is added with the parent chromosome to generate new
chromosomes.
Step 9: These chromosomes are arranged in descending order of their fitness function,
and I chromosomes are chosen for the next generation.
Step 10: j ← j + 1
end for.
Algorithm 1 describes the pseudocode for obtaining power demand and trans-
mission power assigned to the consumers that brings individual profit as well as
total profit of the consumers. Initially, the supplier declares the price in cents/kWh
uniform for all the consumers. Then, for each tth slot, GA algorithm is applied. The
supplier declares the total demand to the consumers. In Step 2, chromosomes of
size M = 4 and I = 50 are generated representing power demands of the consumer.
Further, demands of the consumers generated are ±20% of their targeted demand.
In Step 3, all initialization and iteration begin. Ytm and TH M are calculated in Step
4. If any of the consumer’s fitness function Ytm is greater than the last iteration of
Ytm for the same i th position of chromosome, then rank 1 is assigned. Here, Fm1 1
and Fm2 1 represent the system function (14) and system throughput (10), respectively.
Step 5 maximizes the overall system throughput. Suitable parent chromosomes are
selected for next offspring generation in Step 6–Step 9. From both parent and child
chromosomes, suitable I = 50 chromosomes are selected based on their fitness func-
tion and throughput. This process continues until the stopping criterion is achieved.
If Itr is the maximum iteration, then the complexity of the proposed algorithm is
approximately O(Itr I M).

4 Simulation Results

The system model is consisting of four consumers and a PU which are distributed in a
rectangular area of width 1 km. Here, the four consumers are consumer 1 is working
person, consumer 2 is shopkeeper, consumer 3 is a home, and the consumer 4 is a
corporate office. The PR is present at a circular radius of 100 m from the PU. The
location of power supplier is present at a distance approximately 2 km away from
the consumer’s area. In F(S), x, y and z are set as 0.02, 0.2 and 0, respectively. cm is
5, 4, 5 and 6 for consumer 1, consumer 2, consumer 3 and consumer 4, respectively.
ψm is 0.1 for all the consumers. The electricity price ρ = 2 cents/kWh is declared
by the supplier initially. The wireless channels between any users are assumed to

be distant dependent Rayleigh distributed channel and are given by h ∼ ℵ 1, d −α ,
where d is the distance between the users and α is the path-loss exponent. The target
222 D. Das and N. Behera

detection probability is kept 0.9. Table 1 lists the target demands of all the consumers
for different slots. Some important system parameters are listed in Table 2.
Figure 2 shows the variation of the total profit and throughput of the consumers
with time index. It is observed that profit and throughput are contradictory to each
other. Figure 3 shows the optimal consumers’ power demand with respect to time.
From Table 1 and Fig. 3, it is concluded that optimal demands differ from the targeted
demand. At some instances, it is greater or lesser than the targeted demand.

Table 1 Consumer’s demands for different time slots

Time slots Consumer 1 Consumer 2 Consumer 3 Consumer 4
demand (kWh) demand (kWh) demand (kWh) demand (kWh)
t1 20 2 6 30
t2 4 30 15 50
t3 10 30 10 50
t4 7 2 7 20

Table 2 System parameters

Simulation parameters Values
α 3
N0 −90 dBm
Tth 10 bits/s/Hz
Ptmax 35 dBm
Ps −20 dBm
Ith −50 dBm
Ou 0.3
Ol 0.2

Fig. 2 Variation of profit 300 55

and throughput of consumers

with respect to time
Total throughput (bits/sec/Hz)
Total Profit of Consumers

250 50

200 45

150 40
1 1.5 2 2.5 3 3.5 4
Time Index
Adaptive Resource Allocation for Cognitive … 223

Fig. 3 Variation of 60
consumer’s demand with Consumer 1
Consumer 2
respect to time
50 Consumer 3
Consumer 4

Consumer's Demand (kWh)

40

30

20

10

0
1 1.5 2 2.5 3 3.5 4
Time Index

Figure 4 illustrates the variation of aggregated profit of consumers and supplier

with respect to time for different SNR. As received SNR at the consumers increases,
SW increases, thereby the outage of the licensed band decreases. So, profit of the
supplier increases. But Yt M depends on both SW and OV t , so Yt M decreases slightly
with increased SNR. Figure 5 illustrates the impact of Tth on Yt M and the overall
throughput for different values of ρ. Though, Ptm t
min
increases with increase in Tth ,
Ptmax is fixed. Therefore, decrease in Yt M is not noticeable observed. For the above
reason, throughput increases slightly with increase in Tth .
Figure 6 shows the impact of Ou on the profit of the supplier and the consumers.
It is obvious from (9) that OV increases with increase in Ou . From (15) and (16), it
is clear that both these profits decrease with increase in Ou keeping Ol constant at
0.2.

Fig. 4 Effect of SNR on

Total Profit of Consumers

300
both profit of consumers and SNR=-25dB
supplier 250 SNR=-10dB
SNR=0dB

200

150
1 1.5 2 2.5 3 3.5 4
Time Index

100
Profit of Supplier

SNR=-25dB
80 SNR=-10dB
SNR=0dB

60

40
1 1.5 2 2.5 3 3.5 4
Time Index
224 D. Das and N. Behera

Fig. 5 Effect of T th on
throughput and profit for
different values of ρ

Fig. 6 Effect of Ou on profit 80 300

of supplier and consumers

Total Profit of Consumers

60 250
Profit of Supplier

40 200

20 150

0 100
0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Outage of Unlicensed band

5 Conclusions

This paper addressed the adaptive resource allocation for obtaining optimal power
demand and transmission power of the consumers in a CR-enabled smart grid net-
work. A novel algorithm based on GA was proposed to maximize overall profit of the
consumers emphasizing individual profit as well. Further, the corresponding trans-
mission power was assigned to achieve maximum data rate on the licensed band
without causing interference to the PU network. It was observed from the simulation
results that consumer’s power demand and assigned transmission power simulta-
neously brought profit to the consumers. Further, the impact of different system
parameters on the power demand and profit was studied.
Adaptive Resource Allocation for Cognitive … 225

References

1. Khan, A.A., Rehmani, M.H., Reisslein, M.: Cognitive radio for smart grids: survey of architec-
tures, spectrum sensing mechanisms, and networking protocols. IEEE Commun. Surv. Tutorials
18(1), 860–898 (2016)
2. Erol-Kantarci, M., Mouftah, H.T.: Energy-efficient information and communication infrastruc-
tures in the smart grid: a survey on interactions and open issues. IEEE Commun. Surv. Tutorials
17(1), 179–197 (2015)
3. Rehmani, M.H., Rachedi, A., Kantarci, M.E., Radenkovic, M., Reisslein, M.: Cognitive radio
based smart grid: the future of the traditional electrical grid. Ad Hoc Netw. (Elsevier) 100(41),
1–4 (2016)
4. Emmanuel, M., Rayudu, R.: Communication technologies for smart grid applications: a survey.
J. Netw. Comput. Appl. (Elsevier) 74, 133–148 (2016)
5. Le, T.N., Chin, W.-L., Chen, H.-H.: Standardization and security for smart grid communica-
tions based on cognitive radio technologies—a comprehensive survey. IEEE Commun. Surv.
Tutorials 19(1), 423–445 (2016)
6. Fadel, E., Gungor, M.V.C., Nassef, L., Akkari, N., Malik, M.G.A., Almasri, S., Akyildiz,
I.F.: Spectrum-aware bio-inspired routing in cognitive radio sensor networks for smart grid
applications. Comput. Commun. 101, 106–120 (2017). Elsevier
7. Zhang, D., Chen, Z., Ren, J., Zhang, N., Awad, M.K., Zhou, H., Shen, X.: Energy harvesting-
aided spectrum sensing and data transmission in heterogeneous cognitive radio sensor network.
IEEE Transactions on Vehicular Technology 66(1), 831–843 (2017)
8. Abuadbba, A., Khalil, I., Ibaida, A., Atiquzzaman, M.: Resilient to shared spectrum
noise scheme for protecting cognitive radio smartgrid readings−BCH based steganographic
approach. Ad Hoc Netw. 41, 30–46 (2016). Elsevier
9. Althunibat, S., Wang, Q., Granelli, F.: Flexible channel selection mechanism for cognitive radio
based last mile smart grid communications. Ad Hoc Netw 41, 47–56 (2016). Elsevier
10. Cacciapuoti, A., Caleffi, M., Marino, F., Paura, L.: Sensing-time optimization in cognitive
radio enabling smart grid. In: 2014 Euro Med Telco Conference (EMTC), IEEE, Naples, Italy,
pp. 12–15 (2014)
11. Cacciapuoti, A.S., Caleffi, M., Marino, F., Paura, L.: Mobile smart grids: exploiting the TV
white space in urban scenarios. IEEE Access 4, 7199–7211 (2016). IEEE
12. Li, Q., Feng, Z., Li, W., Gulliver, T.A., Zhang, P.: Joint spatial and temporal spectrum sharing
for demand response management in cognitive radio enabled smart grid. IEEE Trans Smart
Grid 5(4), 1993–2001 (2014). IEEE
13. Deng, R., Chen, J., Cao, X., Zhang, Y., Maharjan, S., Gjessing, S.: Sensing-performance tradeoff
in cognitive radio enabled smart grid. IEEE Trans. Smart Grid 4(1), 302–310 (2013). IEEE
14. Zhang, W., Mallik, R.K., Letaief, K.B.: Cooperative spectrum sensing optimization in cognitive
radio networks. In: 2008 IEEE International conference on communications, IEEE, Beijing,
China, 19–23 May (2008)
Power Quality Improvement of a Fuel
Cell-Based Distributed Generation
System Using Unified Power Quality
Conditioner

Sarita Samal, Tarakanta Jena, and Prasanta Kumar Barik

Abstract This paper deals with power quality profile improvement of fuel cell
(FC)-based distributed generation (DG) system using unified power quality con-
ditioner (UPQC). Despite of the several benefits of DG like excellent energy supply,
reducing expansion of power distribution system, environmental friendly, and so on,
there are several challenges existing due to the integration of DG with the grid or
operating it in stand-alone mode. Power quality (PQ) issue is one of the main tech-
nical challenges in DG power system. In order to provide improved PQ of energy
supply, it is necessary to analyze the harmonics distortion of the system as well as
the voltage sag and swell. The UPQC has been extensively useful and it is verified
to be the best solution to diminish this PQ issue. This paper explores the detail of
PQ impacts in FC-based DG system operating in stand-alone mode. The voltage sag
compensation with current and voltage harmonics are estimated at varying load con-
ditions by using control modified synchronous reference frame (MSRF) technique.
The proposed model is developed in MATLAB/SIMULINK® and the result obtained
validates the superiority of proposed technique over others in terms of harmonics
elimination and sag compensation.

Keywords Distributed generation · Power quality · Harmonics · Sag · MSRF

S. Samal
School of EE, KIIT Deemed to be University, Bhubaneswar, Odisha, India
e-mail: [email protected]
T. Jena (B)
Department of EE, SOA University, Bhubaneswar, India
e-mail: [email protected]
P. K. Barik
Department of MEE, CAET, OUAT, Bhubaneswar, India
e-mail: [email protected]

© Springer Nature Singapore Pte Ltd. 2020 227

G. Pradhan et al. (eds.), Advances in Electrical Control and Signal Systems,
Lecture Notes in Electrical Engineering 665,
https://doi.org/10.1007/978-981-15-5262-5_15
228 S. Samal et al.

1 Introduction

Distributed generation (DG) can be represented as a small-scale power system that

contains loads, energy sources, energy storage units, and control and protection
systems [1]. Using DG is more attractive as it improves the system quality, decreases
the carbon emission, and reduces the losses in transmission and distribution systems
[2]. The control of active and reactive power is easy, when DG connected to grid
rather that standalone system. However, under autonomous operation, the DG is
disconnected from the utility grid and operates in islanded condition. Usually, a stand-
alone DG system is used to supply power to isolated areas or places interconnected
to a weak grid. The application of above DG, on other hand, reduces the probability
of energy supply scarcity. The proposed DG consists of a fuel cell (FC) along with
controllable loads [3, 4]. The FC is the mostly developed energy source used now
[5]. However, electric power system is mostly affected by nonlinear loads, mostly
arc furnaces, power electronics converters, and house hold electronic equipment
plays a key role in polluting the supply voltages and currents. The increase of power
electronics-based equipment in household appliances and industries are the main
cause of pollution of power system [6]. The research in the area of power electronics
makes sure that unified power quality conditioner (UPQC) plays a vital role for
achieving superior power quality levels.
In the present scenario, the series active power filters (APFs) and shunt APF,
normally termed as SAPF, alone do not meet the requirement for compensating the
PQ distortions. A UPQC consists of two inverter integrated with the DC-link capac-
itor where the series APF is integrated though a series transformer and the shunt
is through interfacing inductor. The series inverter acts as a voltage source where
as the shunt one acts as a current source. Simultaneous compensation of voltage
and current-related PQ distortions using UPQC is achieved by proper controlling
of series APF and shunt APF. The shunt APF is employed for providing compen-
sating currents to PCC for generation/absorption of reactive power and harmonics
suppression. Moreover, the operation of SAPF depends on three main parts which
are momentous in its design; these consist of the control method used for gener-
ation of reference current, technique used for switching pulses generation for the
inverter, and the controller used for DC-link capacitor voltage regulation. Different
control strategy explained in literature as follows. The use of SAPFs for current har-
monic compensation typically in domestic, commercial, and industrial applications
has explained in Montero et al. [7]. The experimental study and simulation design
of a SAPF for harmonics and reactive power compensation are explained by Jain
et al. [8]. The power balance theory for active and reactive power compensation has
been developed by Singh et al. [9]. The instantaneous reactive power techniques of
three phase shunt active filter for compensation of source current harmonics have
been explained by Akagi et al. [13]. Sag is the most significant PQ problem faced
by lots of industrial consumers. The control for such a case can be analyzed by
protecting sensitive loads in order to preserve a load voltage without sudden phase
shift [10]. Different control strategies for series APF are analyzed by Benachaiba
Power Quality Improvement of a Fuel Cell-Based … 229

et al. [11] with importance on the reimbursement of voltage sags with phase jump.
Different control techniques to reimburse voltage sags with phase jump are also pro-
jected and compared by Jowder et al. [12]. To ensure stable operation and improve
the system performance of DG in island mode, a comparative study of two different
control techniques used in UPQC like reference current generation, i.e., synchronous
reference frame (SRF) method and modified synchronous reference frame (MSRF)
method in conjunction with pulse width modulation-based hysteresis band controller
is proposed in this paper by using MATLAB simulation software. The PQ issues like
voltage sag compensation, current, and voltage harmonics were analyzed both at
linear and nonlinear loads.

2 Proposed System

The projected DG system (comprising of solar and fuel cell-based energy sources)
is shown in Fig. 1 where DG system generates DC power to the DC bus and by using
a power inverter, this DC power is converted to AC. The AC bus delivers the power
to the load which may be a linear or nonlinear. The UPQC is located in between the
DG and nonlinear load manages the power quality of the system by using different
control techniques.

2.1 Modeling of Fuel Cell System

Proton exchange membrane (PEM) fuel cell is considered as another energy source
of the DG. The fuel cell consists of two electrodes, i.e., positive cathode, negative

Fig. 1 Basic block diagram of DG with UPQC

230 S. Samal et al.

Fig. 2 Fuel cell model

Fig. 3 Fuel cell with boost

converter

anode, and an electrolyte. The pressurized hydrogen gas enters the anode of the fuel
cell and oxygen enters the cathode [14, 15]. The basic model of PEM FC is shown
in Fig. 2 and its chemical reactions are given in Eqs. (1–3) [16].

(2H+ + 2e− ) → H2 (1)

(2H+ + 2e− + 1/2O2 ) → H2 O (2)

H2 + 1/2O2 → H2 O (3)

The simulation fuel cell with boost converter is shown in Fig. 3 and the output
voltage which is match with the output voltage of other DGs is shown in Fig. 4.
Table 1 represents different parameters of fuel cell.

3 Modeling of UPQC

This chapter begins with system configuration and detailed description on UPQC.
The basic structure of UPQC is shown in Fig. 5 which consists of two inverter
connected to a common dc-link capacitor. The series inverter is connected though a
Power Quality Improvement of a Fuel Cell-Based … 231

Fig. 4 Output voltage of boost converter

Table 1 Parameters of the

Different parameters Values
fuel cell
Load resistance (Rl ) 5
Oxygen percentage in air (O2 ) 59.3%
Each cell voltage (vs ) 1.128 V
Cell resistance (Rt ) 0.70833 
Number of cell (k) 65
Hydrogen percentage in fuel (H2 ) 99.56%
Fuel cell voltage (V f c ) 230 V

Fig. 5 Basic UPQC system Source

block diagram VC NL

ic

UPQC

series transformer and the shunt inverter is connected in parallel with the point of
common coupling. The series inverter acts as a voltage source where as the shunt
one acts as a current source. The main function of UPQC is to control the power flow
and reduce the harmonics distortion both in voltage and current waveforms.
The series APF topology is shown in Fig. 6. The series APF protects load from
the utility side disturbances. In case of series APF, Park’s transformation method is
used for generation of unit vector signal. A PWM generator, generating synchronized
switching pulses, is given to the six switches of the series converter.
Figure 7 shows the basic structure of shunt active filter. The shunt active power
filter injects compensating current to the PCC such that the load current becomes
harmonics free. The SAPF generates compensating current which is in opposition to
232 S. Samal et al.

Tr
Rs NL

Vs

Series APF

Fig. 6 Block diagram of series APF

Non Linear Load

IL2

I L1
Is
LS
Linear Load
PCC

LC

VSI

Fig. 7 Block diagram of shunt active filter

the harmonic current generated by nonlinear load. This compensating current cancel
out the current harmonics caused and makes the load current sinusoidal. So, the
SAPF is used to eradicate current harmonics and reimburse reactive power at the
source side so as to make load current harmonics free.
Equations 4 and 5 show instantaneous current and the source voltage.

Is (t) = I L (t) − IC (t) (4)

Vs (t) = Vm sin ωt (5)

Fourier series method is used for expressing the nonlinear load current as shown
in Eq. 6.
ε

Is (t) = I1 Sin(ωt + Φ1 ) + In Sin(nωt + Φn ) (6)
n=2
Power Quality Improvement of a Fuel Cell-Based … 233

The ic is expressed by

Ic (t) = I L (t) − Is (t) (7)

Hence, for the exact compensation of reactive power and harmonics, it is essen-
tial to determine Is (t). The instantaneous value of source, load, and compensation
current can be expressed by, Is (t), I L (t), and IC (t) where Vs (t) and Vm correspond
to instantaneous value and peak value of source voltage.

3.1 UPQC Design

The MSRF controller scheme works in steady state as well as in dynamic condi-
tion exquisitely to manage the active, reactive power, and reduce the harmonics in
load current. The literature in review reveals that MSRF technique has much more
advantages as compare to SRF scheme, so the authors have selected this control
scheme for UPQC operation. The control scheme not uses the PLL circuit as used by
SRF scheme, which makes the system more compatible and may be operated in load
changing condition. The MSRF scheme with its control algorithm is given below.

3.1.1 MSRF Scheme

Figure 8 shows the block diagram of modified SRF method for unit vector generation.
The unit vector is generated by vector orientation method and not by PLL. Figure 9
shows the block diagram to generate unit vector by sensing the supply voltage.

Vref
PI
Vdc Controller

abc α-β LPF d-q α-β Isa*

Ila
Ilb I sb *
α-β d-q LPF α-β
Ilc abc I sc*

Vsa Unit
Vsb Vector
Vs\c

Fig. 8 Block diagram of modified SRF method

234 S. Samal et al.

Cos
1/S ÷

Vsa abc Vα

Vsb w Est.
α-β Mag.
Vsc
Sin
Vβ ÷
1/S

Fig. 9 Unit vector generation block diagram

Vα
cos θ = √   (8)
(Vsα 2) + Vsβ 2
Vβ
sinθ = √   (9)
(Vsα 2) + Vsβ 2

3.1.2 HCC Technique

Figure 10 shows the block diagram of hysteresis current regulator which generates
the required pulses for inverter. In the current regulator, the error signal is generated
by comparing the reference current I *sa and actual current I sa .
The switching pulses required for the inverter is designed in such a way that when
the error signal go beyond the upper band of hysteresis loop, the lower switches of
inverter are ON and upper switches are OFF and similarly, the upper switches are

Upper Band

Hystersis Band Reference

Current
isa*
isa
S1
S2
isb* Switching S3
isb Analogy S4
S5
S6 Actual
VDC Current
isc* Lower Band
isc
-VDC

Fig. 10 Hysteresis current controller scheme

Power Quality Improvement of a Fuel Cell-Based … 235

ON and lower switches OFF when the error signal exceeds the lower band [17, 18].
So, the actual current is always tracked with respect to reference current inside the
hysteresis band.

4 Results and Discussion

Performance analysis of DG connected to nonlinear load with MSRF-based

UPQC
In this case, the system performance is analyzed by connecting nonlinear load with the
DG system first without UPQC and then with MSRF-based UPQC. The performance
of series APF can be evaluated by introducing voltage sag into the system. The profile
of load voltage shown in Fig. 11a conforms that voltage sag is introduce from 0.1
to 0.3 s of the load voltage waveform. For sag condition, the series APF detects the
voltage drop and injects the required voltage through the series coupling transformer.
It maintains the rated voltage across the load terminal. In order to compensate the
load voltage sag, UPQC (employing MSRF scheme) is turned on, which injects
compensating voltage at the PCC as displayed in Fig. 11b as a result, the load voltage
is same as that of source voltage. The load voltage after compensation is shown in
Fig. 11c. In general, the operation of the series part of the UPQC can be described as
rapid detection of voltage variations at source and it injects the compensation voltage
which maintains rated voltage across the load terminal.
The shunt VSI in the UPQC is realized as shunt APF and is applied to solve the
current-related PQ distortions, current harmonic distortion, reactive power demand,
etc. In order to investigate the performance of shunt APF, a rectifier-based nonlinear
load is introduced into the system and the level of harmonics is checked. It is observed
from Fig. 12a that the source current waveform has a total harmonic distortion (THD)
of 16.60% as per the FFT analysis of the source current shown in Fig. 12b. In order to
make source current to be sinusoidal, the shunt APF of the UPQC with conventional
MSRF technique is turned on at t = 0.1 s which injects compensating current as
displayed in Fig. 12c. Hence, the THD level comes down to 2.54% as shown in
Fig. 12d.
236 S. Samal et al.

Fig. 11 Profile obtained under (sag compensation). a Load voltage before compensation.
b Compensating voltage injected by UPQC. c Load voltage after compensation

5 Conclusion

The research reveals that MSRF technique of UPQC makes possible for improving
the power quality of a DG system connected with nonlinear load. The advantage of
MSRF technique is that the production of sine and cosine angles for synchronization
purpose instead of using PLL circuit, it uses a basic unit vector generation scheme.
The suggested method delivers superior output than the existing method in terms of
harmonic mitigation and compensation of active and reactive powers.
Power Quality Improvement of a Fuel Cell-Based … 237

Fig. 12 Profile obtained under (harmonics mitigation). a Source current before compensation.
b Harmonics content before compensation. c Compensating current injected by UPQC. d Harmonics
content after compensation
238 S. Samal et al.

References

1. Badoni, M., Singh, B., Singh, A.: Implementation of echo-state network-based control for
power quality improvement. IEEE Trans. Industr. Electron. 64(7), 5576–5584 (2017)
2. Mahmoud, M.S., Rahman, M.S.U., Fouad, M.S.: Review of microgrid architectures–a system
of systems perspective. IET Renew. Power Gener. 9(8), 1064–1078 (2015)
3. Samal, S., Hota, P.K.: Design and analysis of solar PV-fuel cell and wind energy based microgrid
system for power quality improvement. Cogent Eng. 4(1), 1402453 (2017)
4. Suresh, M., Patnaik, S.S., Suresh, Y., Panda, A.K.: Comparison of two compensation control
strategies for shunt active power filter in three-phase four-wire system. In: Innovative Smart
Grid Technologies (ISGT), IEEE PES, pp. 1–6 (2011)
5. Tang, Y., Loh, P.C., Wang, P., Choo, F.H., Gao, F., Blaabjerg, F.: Generalized design of high
performance shunt active power filter with output LCL filter. IEEE Trans. Industr. Electron.
59(3), 1443–1452 (2012)
6. Hosseinpour, M., Yazdian, A., Mohamadian, M., Kazempour, J.: Desing and simulation of
UPQC to improve power quality and transfer wind energy to grid. J. Appl. Sci. 8(21), 3770–3782
(2008)
7. Montero, M.I.M., Cadaval, E.R., Gonzalez, F.B.: Comparison of control strategies for shunt
active power filters in three-phase four-wire systems. IEEE Trans. Power Electron. 22(1),
229–236 (2007)
8. Jain, S.K., Agarwal, P., Gupta, H.O.: Simulation and experimental investigations on a shunt
active power filter for harmonics and reactive power compensation. IETE Techn. Rev. 20(6),
481–492 (2003)
9. Singh, B.N., Singh, B., Chandra, A., Al-Haddad, K.: Design and digital implementation of
active filter with power balance theory. IEE Proc. Electr. Power Appl. 152(5), 1149–1160
(2005)
10. Dixon, J.W., Venegas, G., Moran, L.A.: A series active power filter based on a sinusoidal current-
controlled voltage-source inverter. IEEE Trans. Industr. Electron. 44(5), 612–620 (1997)
11. Ferdi, B., Benachaiba, C., Dib, S., Dehini, R.: Adaptive PI control of dynamic voltage restorer
using fuzzy logic. J. Electr. Eng. Theor. Appl. 1(3) (2010)
12. Jowder, F.A.L.: Design and analysis of dynamic voltage restorer for deep voltage sag and
harmonic compensation. IET Gener. Transm. Distrib. 3(6), 547–560 (2009)
13. Akagi, H., Kanazawa, Y., Nabae, A.: Instantaneous reactive power compensators comprising
switching devices without energy storage components. IEEE Trans. Industr. Electron. Appl.
20(3), 625–630 (1984)
14. Noroozian, R., Abedi, M., Gharehpetian, G.B., Bayat, A.: On-grid and off-grid operation of
multi-input single-output DC/DC converter based fuel cell generation system. In: Electrical
Engineering (ICEE), 18th Iranian conference, pp. 753–758 (2010)
15. Nergaard, T.A., Ferrell, J.F., Leslie, L.G., Lai, J.S.: Design considerations for a 48 V fuel cell
to split single phase inverter system with ultra-capacitor energy storage. In: Power Electronics
Specialists Conference, Pesc 02, vol. 4, pp. 2007–2012 (2002)
16. Bucci, G., Ciancetta, F., Fiorucci, E, Vegliò, F.: An experimental approach to the modeling of
PEM fuel cells in dynamic conditions. In: Power Tech, pp. 1094–1099 (2007)
17. Bose, B.K.: An adaptive hysteresis-band current control technique of a voltage-fed PWM
inverter for machine drive system. IEEE Trans. Industr. Electron. 37(5), 402–408 (1990)
18. Karuppanan, P., Mahapatra, K.K.: A novel control strategy based shunt APLC for power quality
improvements. In: IEEE International Conference on Power, Control and Embedded Systems
(ICPCES), pp. 1–6 (2010)
Numerical Approach to the Fractional
Optimal Control Problem
of Continuous-Time Singular System

Tirumalasetty Chiranjeevi and Raj Kumar Biswas

Abstract In this work, numerical approach to the optimal control problem (OCP)
of the fractional order singular system (FOSS) is considered. Dynamics of the sys-
tem is expressed by the fractional order differential equations (FDEs) in the sense
of Riemann–Liouville (RL) fractional-order derivative (FD). General form of per-
formance index (PI) is considered. First, we convert FOSS into the fractional order
non-singular system (FONSS) using coordinate transformation and then obtain the
necessary conditions. The solution of optimal conditions is carried out by using
the Grunwald–Letnikov approximation (GLA)-based numerical approach. In order
to check the applicability of the formulation and efficacy of numerical method, an
example is illustrated.

Keywords Fractional-order singular system · Lagrange multiplier approach ·

Riemann–Liouville derivative · Grunwald–Letnikov approximation

1 Introduction

From the last few decades, the interest in fractional-order systems increasing rapidly
because these systems give accurate behavior compared with the integer-order sys-
tems [1–5]. We can find several definitions of FDs in literature [6]. Most widely used
FDs are Caputo and Riemann–Liouville (RL) derivatives. In this work, FDEs are
described in the sense of RL derivative.
OCP is defined as the problem of finding the optimal control, for minimizing the
specified PI subjected to system dynamics [7–9]. If system dynamics is described by
FDEs, then OCP is considered as fractional optimal control problem (FOCP).

T. Chiranjeevi (B) · R. K. Biswas

Electrical Engineering Department, NIT Silchar, Silchar, Assam, India
e-mail: [email protected]
T. Chiranjeevi
Electrical Engineering Department, REC Sonbhadra, Churk, Uttar Pradesh, India

© Springer Nature Singapore Pte Ltd. 2020 239

G. Pradhan et al. (eds.), Advances in Electrical Control and Signal Systems,
Lecture Notes in Electrical Engineering 665,
https://doi.org/10.1007/978-981-15-5262-5_16
240 T. Chiranjeevi and R. K. Biswas

Considerable work has been reported in the literature regarding FOCPs of non-
singular systems. In this context, Agrawal [10] proposed general formulation and vir-
tual work-based solution scheme for FOCPs of both time-invariant and time-variant
fractional-order systems. Different numerical schemes based on hat basis functions
[11], central difference numerical scheme [12], combination of epsilon penalty and
variational methods [13], GLA of FDs [14], the operational matrix and Bernoulli
polynomials [15], gradient projection technique [16], neural network approach [17],
the Legendre orthonormal polynomials [18], quadratic numerical scheme [19], a
semidefinite programming approach [20], hybrid functions [21], nonstandard finite
difference method [22], reflection operator [23], modified Adomian decomposition
method [24], shifted Legendre orthonormal polynomials [25], variational iteration
method [26], fixed point approach [27], Ritz method [28–32], etc., are presented in
the literature for the solution of FOCPs. Authors in [33–37] discussed formulation
and solution schemes of FOCPs at different end-point conditions. Chiranjeevi and
Biswas [38] proposed closed-form solution scheme for solving FOCPs in the sense of
conformable FD. Authors in [39] presented formulation of FOCPs with constraints
on control.
Singular systems having some special features which are not observed in classical
systems are consistent initial conditions, nonproperness of transfer matrix, impulse
terms and input derivatives in the state response, noncausality between input and state
or input and output, etc., because of these features singular systems have attracted
many researchers from last few decades [40]. We can use singular systems in system
modeling because of their form. We can find singular systems in many applications
such as network analysis, economical systems, engineering systems, biological sys-
tems, and social systems [40]. Few works have been reported in the literature regard-
ing OCP of singular systems [41–46]. Very limited work has been done on optimal
control of FOSS. In this respect, Biswas and Sen [47, 48] proposed pseudo-state
space approach for optimal control of FOSS. Ashayeri et al. [49] presented solution
method based on block pulse functions for solving OCP of FOSS. Authors in [50]
proposed GLA-based solution method for solving linear quadratic OCP of FOSS.
Chiranjeevi et al. [51] presented formulation and numerical approach for OCPs of
FOSS with fixed final time-free final state end-point condition.
In this work, a formulation and GLA-based numerical method for OCP of FOSS in
terms of RL derivative has been presented. Because of the complex nature, analytical
and numerical solutions are difficult for SSs. Therefore, first we convert the SS into
non-singular system by using coordinate transformation mentioned in [40], and then
apply the OCP formulation for obtaining necessary conditions. Thereafter, solution
of necessary conditions can be carried out by using GLA based solution technique.
An example is illustrated to check the applicability of the formulation and solution
scheme.
The remaining part of the paper as follows. Various definitions of FDs are given
in Sect. 2. In Sect. 3, transformation from FOSS to fractional-order non-singular
system and OCP formulation is presented. Numerical example is illustrated in Sect. 4.
Conclusions are given in Sect. 5.
Numerical Approach to the Fractional Optimal Control … 241

2 Preliminaries

Several definitions of FDs exit in the literature [6] are given in this section.
The “left and right RL derivative” of order α is given as [6]

t
RL α 1 d
0 Dt g(t) = (t − ξ )−α g(ξ )dξ
(1 − α) dt
0

  1
RL α 1 −d
t D1 g(t) = (ξ − t)−α g(ξ ) dξ
(1 − α) dt
t

The “left and right Caputo derivative” of order α is given as [6]

t  
C α 1 −α d
0 Dt g(t) = (t − ξ ) g(ξ )dξ
(1 − α) dξ
0

1  
C α 1 −α −d
t D1 g(t) = (ξ − t) g(ξ ) dξ
(1 − α) dξ
t

The “left and right Grunwald–Letnikov derivative” of order α is given as [6]

1  (α)
k
GL α
0 Dt g(t) = Lt w g(t − i h)
h→0 h α i=0 i
kh=t
N −k
GL α 1  (α)
t D1 g(t) = Lt w g(t + i h)
h→0 h α i=0 i
kh=t

 
α
where wi(α) = (−1)i
.
i

3 Formulation and Numerical Scheme

Consider a FOSS described in terms of FDE as

E 0R L Dtα x(t) = Ax(t) + Bu(t) (1)

242 T. Chiranjeevi and R. K. Biswas

Consider a PI

1
1
J (u) = [x T (t) M x(t) + u T (t) Su (t)]dt (2)
2
0

where x ∈ m×1 , u ∈ n×1 , A ∈ m×m , B ∈ m×n , E ∈ m×m is the singular

matrix, 0R L Dtα x(t) is the RL derivative, M ∈ m×m ≥ 0, and S ∈ n×n > 0.
OCP is defined as the problem of finding the optimal control, for minimizing the
specified PI subjected to system dynamics using the initial and terminal conditions
x(0) = x0 and x(1) = x f .
We can adopt F and G for satisfying the following relations based on the Lemma
given in [40]

FEG = diag(Im 1 , k), FAG = diag( Â, Im 2 ) (3)

F and G must be nonsingular.

Where  is a new state matrix of order m 1 × m 1 , k is a nilpotent matrix of order
m 2 × m 2 , m 1 = rank(E) and m 1 + m 2 = m.
In order to convert the FOSS to FONSS, the coordinate transformation has been
chosen as follows [40]
 
x1 (t)
x(t) = G , x1 (t) ∈ m 1 , x2 (t) ∈ m−m 1 (4)
x2 (t)

From Eqs. (3) and (4), Eq. (1) is modified as

RL α
0 Dt x 1 (t) = Â x1 (t) + 1 u(t) (5)

0 = x2 (t) + 2 u(t) (6)

 
1
where FB = .
2
By combining Eq. (4) with Eq. (6), we obtain
⎡ ⎤
    I 0  
x(t) G0 ⎣ ⎦ x1 (t)
= 0 −2 (7)
u(t) 0 I u(t)
0 I

By using Eq. (7), the PI given in Eq. (2) can be modified as

1  
1 
J (u) = x1T (t) M x1 (t) + w (t) Ŝw (t) dt
T
(8)
2
0
Numerical Approach to the Fractional Optimal Control … 243

where
⎡ ⎤T ⎡ ⎤
  I 0     I 0
M̂ Y G0 M 0 G0 ⎣
= ⎣ 0 −2 ⎦ 0 −2 ⎦ ,
Ŷ Ŝ 0 I 0 S 0 I
0 I 0 I

M = M̂ − Y Ŝ −1 Ŷ and w(t) = u(t) + Ŝ −1 Ŷ x1 (t).

Substitute u(t) = w(t) − Ŝ −1 Ŷ x1 (t) in Eq. (5), and we get

RL α
0 Dt x 1 (t) = ( Â − 1 Ŝ −1 Ŷ ) x1 (t) + 1 w(t) =  x1 (t) + 1 w(t) (9)

Therefore, by using coordinate transformation, we have converted FOSS (1) into

FONSS (9). Now, find w(t) in order to minimize the transformed PI (8) subject to
transformed system dynamics (9).
The augmented PI using Lagrange multiplier λ(t) is
⎧   ⎫
1 ⎪
⎨ 1 x T (t) M
 ⎪
⎬
1 x1 (t) + w (t) Ŝ w(t)
T
Ja (u) = 2 dt (10)
⎩ + λT (t) x (t) +  w(t) − R L D α x (t)⎪
⎪ ⎭
0 1 1 0 t 1

Take δ Ja (u) and further simplification, we get

⎧ T  T ⎫

1 ⎪
⎪
⎨ M x1 (t) + T λ(t) − C α T ⎪
⎪
t D1 λ(t) δ x 1 (t) + Ŝ w(t) + 1 λ(t) δ w(t)⎬
δ Ja (u) =
⎪   ⎪
dt − λ(1) δ x(1) (11)
⎪ ⎪
0 ⎩ + x (t) +  w(t) − R L D α x (t) δ λ(t) ⎭
T
1 1 0 t 1

The δ Ja (u) = 0, for necessary condition of optimality [7, 8]. In order to satisfy
δ Ja (u) = 0, the coefficients of δ x1 (t), δw (t), and δ λ(t) in Eq. (11) become zero
[50]. Then, the necessary conditions are
RL α
0 Dt x 1 (t) = x1 (t) + 1 w(t) (12)


C α
t D1 λ(t) = M x1 (t) + T λ(t) (13)

w (t) = − Ŝ −1 T1 λ(t) (14)

Thus, the first variation finally reduces to

λ(1) δx(1) = 0 (15)

To solve the optimal conditions (12)–(14), the entire time domain is divided into
N equal subdomains and size of each subdomain is h = 1/N. The time at node k is
244 T. Chiranjeevi and R. K. Biswas

tk = kh. The necessary conditions (12) and (13) are approximated by using GLA of
FDs as [50]

1  (α)
k
w x1 (kh − i h) =  x1 (kh) − 1 Ŝ −1 T1 λ(kh), k = 1, 2, . . . , N (16)
h α i=0 i
N −k
1  (α) 

α
wi λ(kh + i h) = M x1 (kh)
h
i=0
(1 − kh)−α
+ T λ(kh) + λ(1), k = N − 1, N − 2, . . . , 0 (17)
(1 − α)

The above set of Eqs. (16) and (17) can be solved by using any related software
like MATLAB. Ones x1 (t) and λ(t) are known, we can obtain x2 (t) and u(t).

4 Example

An example is illustrated in this section by using GLA-based numerical scheme for

checking efficacy of formulation and solution scheme.
Consider a FOSS (1) with
     
10 −1 0 1
E= , A= , B=
00 0 −1 1

Consider a PI (2) with

 
10
M= , S=2
01

Initial and terminal conditions are

x1 (0) = 1 and x(1) = 0

F and G are chosen as [50]

   
10 1 0
F= and G =
01 0 −1

Using the foregoing considerations, we obtain the following results.

Figures 1 and 2 show the optimal states and Fig. 3 shows the optimal control for
distinct α and N = 5. From these figures, we can observe that amplitude of optimal
states and optimal control decreases as α is decreased like in [10, 33, 51]. These
results also show that control effort increases as α approaches to one.
Numerical Approach to the Fractional Optimal Control … 245

Fig. 1 x1 (t) for distinct α,

N =5

Fig. 2 x2 (t) for distinct α,

N =5

5 Conclusions

In this work, a formulation and GLA-based numerical approach for OCP of FOSS is
presented. Dynamics of the system is expressed by FDEs in terms of RL derivative.
General form of PI is considered. Solution of the singular system is difficult because
of its complex nature. In this respect, we have converted FOSS into FONSS using
coordinate transformation similar to integer order systems and then applied GLA-
based solution method for solving FOCP. An example is illustrated to check the
efficacy of the numerical approach. For distinct values of α, results of optimal states
246 T. Chiranjeevi and R. K. Biswas

Fig. 3 u(t) for distinct α,

N =5

and optimal control are obtained. From the numerical simulation, we conclude that
control effort decreases as α is decreased.

References

1. Dzielinski, A., Sierociuk, D., Sarwas, G.: Some applications of fractional order calculus. Bull.
Pol. Ac. Tech. 58(4), 583–592 (2010)
2. Ding, Y., Wang, Z., Ye, H.: Optimal control of a fractional-order HIV-Immune system with
memory. IEEE Trans. Control Syst. Technol. 20(3), 763–769 (2012)
3. Abro, K.A., Memon, A.A., Abro, S.H., Khan, I., Tlili, L.: Enhancement of heat transfer rate
of solar energy via rotating Jeffrey nanofluids using Caputo-Fabrizio fractional operator: an
application to solar energy. Energy Rep. 5, 41–49 (2019)
4. Rangaig, N.A., Convicto, V.C.: On fractional modelling of dye removal using fractional deriva-
tive with non-singular kernel. J. King Saud Univ. Sci. https://doi.org/10.1016/j.jksus.2018.
01.006 (2018)
5. Suribabu, G., Chiranjeevi, T.: Implementation of fractional order PID controller for an AVR
system using GA and ACO optimization techniques. In: IFAC-ACODS Conference, vol. 49,
no. 1, pp. 456–461. NIT Trichy, India, 1–5 Feb 2016
6. Podlubny, I.: Fractional Differential Equations. Academic Press, San Diego, California, USA
(1999)
7. Lewis, F.L., Syrmos, V.L.: Optimal Control. Wiley-Interscience Publication, Canada (1995)
8. Kirk, D.E.: Optimal Control Theory: An Introduction. Dover Publications, Mineola, New York,
USA (2004)
9. Naidu, D.S.: Optimal Control Systems. CRC Press, New York, USA (2003)
10. Agrawal, O.P.: A general formulation and solution scheme for fractional optimal control
problems. Nonlinear Dyn. 38, 323–337 (2004)
11. Mohammed, O.H.: A direct method for solving fractional order variational problems by hat
basis functions. Ain Shams Eng. J. 9(4), 1513–1518 (2018)
12. Baleanu, D., Agrawal, O.P.: A central difference numerical scheme for fractional optimal
control problems. J. Vib. Control 15(4), 583–597 (2009)
Numerical Approach to the Fractional Optimal Control … 247

13. Lotfi, A.: A combination of variational and penalty methods for solving a class of fractional
optimal control problems. J. Optim. Theory Appl. 174(1), 65–82 (2017)
14. Agrawal, O.P., Baleanu, D.: A Hamiltonian formulation and a direct numerical scheme for
fractional optimal control problems. J. Vib. Control 13(9–10), 1269–1281 (2007)
15. Behroozifar, M., Habibi, N.: A numerical approach for solving a class of fractional optimal
control problems via operational matrix Bernoulli polynomials. J. Vib. Control 24(12), 2494–
2511 (2018)
16. Wu, S.L., Huang, T.Z.: A fast second-order parareal solver for fractional optimal control
problems. J. Vib. Control 24(15), 3418–3433 (2018)
17. Sabouri, K.J., Effati, S., Pakdaman, M.: A neural network approach for solving a class of
fractional optimal control problems. Neural Process. Lett. 45, 59–74 (2017)
18. Ezz-Eldien, S.S., Doha, E.H., Baleanu, D., Bhrawy, A.H.: A numerical approach based on Leg-
endre orthonormal polynomials for numerical solutions of fractional optimal control problems.
J. Vib. Control 23(1), 16–30 (2017)
19. Agrawal, O.P.: A quadratic numerical scheme for fractional optimal control problems. ASME
J. Dyn. Sys. Meas. Control 130, 0110101–0110106 (2008)
20. Dehghan, R., Keyanpour, M.: A semidefinite programming approach for solving fractional
optimal control problems. Optimization 66(7), 1157–1176 (2017)
21. Mashayekhi, S., Razzaghi, M.: An approximate method for solving fractional optimal control
problems by hybrid functions. J. Vib. Control 24(9), 1621–1631 (2018)
22. Zahra, W.K., Hikal, M.M.: Non standard finite difference method for solving variable order
fractional optimal control problems. J. Vib. Control 23(6), 948–958 (2017)
23. Biswas, R.K., Sen, S.: Numerical method for solving fractional optimal control problems. In:
ASME IDETC/CIE Conference, pp. 1–4. San Diego, California, USA, 30 Aug–3 Sept 2009
24. Alizadeh, A., Effati, S.: Modified Adomian decomposition method for solving fractional
optimal control problems. T. I. Meas. Control 40(6), 2054–2061 (2018)
25. Ezz-Eldien, S.S., Hafez, R.M., Bhrawy, A.H., Baleanu, D., El-Kalaawy, A.A.: New Numer-
ical Approach for Fractional Variational Problems Using Shifted Legendre Orthonormal
Polynomials. J. Optim. Theory Appl. 174(1), 295–320 (2017)
26. Alizadeh, A., Effati, S.: An iterative approach for solving fractional optimal control problems.
J. Vib. Control 24(1), 18–36 (2018)
27. Zeid, S.S., Kamyad, A.V., Effati, S., et al.: Numerical solutions for solving a class of fractional
optimal control problems via fixed-point approach. SeMA 74(4), 585–603 (2017)
28. Nemati, A., Yousefi, S.A.: A numerical method for solving fractional optimal control problems
using Ritz method. ASME J. Comput. Nonlinear Dyn. 11, 0510151–0510157 (2016)
29. Nemati, A., Yousefi, S.A.: A numerical scheme for solving two-dimensional fractional optimal
control problems by the Ritz method combined with fractional operational matrix. IMA J.
Math. Control Inf. 34(4), 1079–1097 (2017)
30. Jahanshahi, S., Torres, D.F.M.: A simple accurate method for solving fractional variational and
optimal control problems. J. Optim. Theor. Appl. 174(1), 156–175 (2017)
31. Nemati, A., Yousefi, S., Soltanian, F., SaffarArdabili, J.: An efficient numerical solution of
fractional optimal control problems by using the Ritz method and Bernstein operational matrix.
Asian J. Control 18(6), 2272–2282 (2016)
32. Lotfi, A., Yousefi, S.A.: A generalization of Ritz-variational method for solving a class of
fractional optimization problems. J. Optim. Theor. Appl. 174(1), 238–255 (2017)
33. Biswas, R.K., Sen, S.: Fractional optimal control problems with specified final time. ASME J.
Comput. Nonlinear Dyn. 6, 0210091–0210096 (2011)
34. Biswas, R.K., Sen, S.: Free final time fractional optimal control problems. J. Franklin Inst. 351,
941–951 (2014)
35. Dzielinski, A., Czyronis, P.M.: Fixed final time and free final state optimal control problem
for fractional dynamic systems-linear quadratic discrete-time case. Bull. Pol. Ac. Tech. 61(3),
681–690 (2013)
248 T. Chiranjeevi and R. K. Biswas

36. Dzielinski, A., Czyronis, P.M.: Fixed final time optimal control problem for fractional dynamic
systems-linear quadratic discrete-time case. In: Mikolaj, B., Krzysztof (eds.), Advances in Con-
trol Theory and Automation, pp. 71–80. Printing House of Bialystok University of Technology,
Poland (2012)
37. Chiranjeevi, T., Biswas, R.K.: Discrete-time fractional optimal control. Mathematics 5(2), 1–12
(2017)
38. Chiranjeevi, T., Biswas, R.K.: Closed-form solution of optimal control problem of a fractional
order system. J. King Saud Univ. Sci. https://doi.org/10.1016/j.jksus.2019.02.010 (2019)
39. Chiranjeevi, T., Biswas, R.K.: Formulation of optimal control problems of fractional dynamic
systems with control constraints. J. Adv. Res. Dyn. Control Syst. 10, 201–212 (2018)
40. Dai, L.: Singular control systems. In: Thoma, M., Wyner, A. (eds.) Lecture Notes in Control
and Information Sciences, pp. 1–340. Springer, Berlin (1989)
41. Cobb, D.: Descriptor variable systems and optimal state-regulation. IEEE Trans. Autom.
Control 28(5), 601–611 (1983)
42. Arora, S., Chauhan, S.: Optimal control of singular system via Block pulse function. Int. J. Ind.
Electr. Electron. Eng. 2, 64–67 (2014)
43. Bender, D., Laub, A.: The linear-quadratic optimal regulator for descriptor systems. IEEE
Trans. Autom. Control 32(8), 672–688 (1987)
44. Zhaolin, C., Huimin, H., Jifeng, Z.: The optimal regulation of generalized state-space systems
with quadratic cost. Automatica 24(5), 707–710 (1988)
45. Mohan, B.M., Kar, S.K.: Optimal control of singular systems via orthogonal functions. Int. J.
Control Autom. Syst. 9(1), 145–150 (2011)
46. Mohan, B.M., Kar, S.K.: Continuous Time Dynamical Systems: State ESTIMATION and
optimal Control with Orthogonal Functions. CRC Press, Taylor & Francis Group, Boca Raton,
London (2012)
47. Biswas, R.K., Sen, S.: Fractional optimal control problems: a pseudo-state-space approach. J.
Vib. Control 17(7), 1034–1041 (2010)
48. Biswas, R.K., Sen, S.: Fractional optimal control within Caputo’s derivative. In: ASME
IDETC/CIE Conference, pp. 1–8, Washington, DC, USA, 28–31 Aug 2011
49. Ashayeri, L., Shafiee, M., Menhaj, M.B.: Optimal control of fractional-order singular system
via Block pulse functions. In: 20th Iranian Conference of electrical engineering, Tehran, Iran,
15–17 May 2012
50. Moubarak, M.R.A., Ahmed, H.F., Khorshi, O.: Numerical solution of the optimal control for
fractional order singular systems. Differ. Equ. Dyn. Syst. 26, 279–291 (2018)
51. Chiranjeevi, T., Biswas, R.K., Chudamani, S.: Optimal control of fractional order singular
system. Int. J. Electr. Eng. Educ. (2019). https://doi.org/10.1177/0020720919833031
Direct Torque Controlled Induction
Motor Drive Using Super-Lift Converter
for Performance Improvement

P. Elangovan, V. Maheswari, G. Nithiyanandham, and S. Prabhu

Abstract The maintenance of ripple-free DC link voltage is obligatory for

satisfactory performance of induction motor (IM) drives. Because the IM drives
exhibit nowadays uses voltage source inverter (VSI) in the obverse end of IM which
converts DC link voltage into variable AC voltage. This paper proposes an active
front end (AFE) controlled elementary super-lift Luo converter (ESC) in the DC link
of IM drive. Unlike a usual DC–DC converter, the suggested ESC enhances the per-
formance of IM drive by offering undulation-less DC link voltage and also improves
supply end power quality. Among various control schemes, the direct torque control
(DTC) is broadly used scheme for IM drives. Hence, the proposed IM drive uses
basic DTC model and the control parameters of DTC are correlated with DC link
parameters mathematically. Finally, the simulation is conceded for the developed IM
drive using MATLAB/Simulink software. The simulation observations such as DC
link voltage, stator flux trajectory, electromagnetic torque and speed response of IM
validates the superiority of the proposed IM drive using ESC.

Keywords Active front end · DC–DC converter · Direct torque control · Induction
motor

1 Introduction

The speed regulation is mandatory for the motors used in the industrial applications.
On considering the cost, robustness and the performance characteristics, the induction
motor (IM) is the most preferred one for adjustable speed drives. The speed regulation
approaches for IM drives are classified into two types and they are scalar and vector
control. IM’s scalar control is easy to implement and trouble free [1]. However, due

P. Elangovan (B) · V. Maheswari · G. Nithiyanandham

Sreenivasa Institute of Technology and Management Studies, Chittoor, Andhra Pradesh 517127,
India
e-mail: [email protected]
S. Prabhu
Sree Vidyanikethan Engineering College, Tirupati, Andhra Pradesh 517501, India

© Springer Nature Singapore Pte Ltd. 2020 249

G. Pradhan et al. (eds.), Advances in Electrical Control and Signal Systems,
Lecture Notes in Electrical Engineering 665,
https://doi.org/10.1007/978-981-15-5262-5_17
250 P. Elangovan et al.

to constraints on torque vacillations and non-uniform flux dispersal during sudden

changes in load, IM scalar control is not suitable for more number of applications.
The vector control approaches such as field-oriented control (FOC) and direct torque
control (DTC) [2, 3] overcome the abovementioned drawback. However, the low
speed regulation with constant switching frequency can be offered by implementing
space vector modulation (SVM) for voltage source inverter (VSI).
Conventionally, the DTC-IM drives are cascaded with diode bridge rectifier
(DBR) at the façade end for transformation of input AC voltage to DC voltage and
a VSI for the transformation of that DC to adjustable AC. Due to the presence of
uncontrolled DBR, the supply end current is exceedingly distorted [4]. The various
problems associated with non-sinusoidal supply end current are poor power factor
(PF); overheating of the power electronics equipment, DC link ripple voltage and
the insulation failure which occurs in the stator winding of IM.
Firstly, the poor PF which leads to the greenhouse gas emission and huge electric-
ity bill on consumers can be reduced by using the PF correction converter [5]. For the
small and medium power variable frequency drives, the PF correction converters are
widely preferred in order to mitigate the power quality issues. But, the control on DC
link voltage for such systems tends to be an unsolved technical barrier. The regulation
of DC link voltage without ripple is enhanced by using a large capacitor [6] in the
DC link part. But, under or overcorrection of the DC link voltage makes it unsuit-
able for the adjustable speed drives whose load varies continuously. In reference [7],
the IM drive is incorporated with the suggested ESC but, zero speed regulation and
ripple-free torque production are not possible with that motor end control.
This paper proposes an active façade end (AFE) control using an elementary ESC
for DTC-IM drive. The ESC is an integrated DC–DC converter that is part of the Luo
converter series [8]. The different features of the proposed drive are the achievement
of unit PF at the utility end, the ripple content of the formed drive’s DC bus voltage is
less than 1% and the proposed IM operation is achieved at low speed and zero speed.
The paper is structured such that Sect. 2 explains the mathematical correlation of
key parameters in the projected IM drive along with the control strategy of suggested
ESC, Sect. 3 explains the uniqueness of anticipated AFE control, Sect. 4 explains the
simulation work and the corresponding results. Finally, the conclusion is presented
in Sect. 5.

2 Mathematical Correlation and Control Strategy of ESC

The schematic diagram of the proposed IM drive is presented in Fig. 1. Three junc-
tures alienate the proposed system. The first converts the utility end of single-phase
AC power by means of a DBR into unchanged DC power. The second part employs
an ESC to transform the varying DC from the DBR to an increased DC. The final
juncture incorporates a VSI to convert the DC power from the ESC to a variable
voltage and variable frequency AC power. The two main elements of the developed
Direct Torque Controlled Induction Motor Drive … 251

DBR ESC VSI

L11 D5 D6

L21 C21
S1 S3 S5
D1 D2
Three
Single C0
S Phase
Phase
IM
AC
Supply S4 S6 S2
D3 D4

Gate Pulse to ESC Gate Pulse

Speed
to
Inductor (L21) Current
Comparator Direct VSI
Torque
Control
Predictive Current Controller
Multiplier
Zero Cross Detection PI Regulator Stator Stator
Voltage Current
Per Unit Conversion
DC Link Voltage
Sample & Hold Circuit
Comparator

Supply AC Voltage
AFE Scheme

Fig. 1 Schematic of the suggested IM drive

model are façade end ESC control and rear end VSI control. This section explains
the control approach of the projected ESC and the control scheme of the VSI.
The super-lift converter with a positive output is categorized into five different
subseries [9]. For the proposed IM drive, the elementary circuit from main series,
which is a direct evolution of the voltage lift converter, is instigated.
The DBR’s input voltage to ESC makes the voltage across the condenser C 21 to
V i(dc) , in the ON period and the current seeping over the inductor L 21 increases with
particular regard to V i(dc) . The condenser C 21 discharges and the current seeping
through the inductor L 21 diminishes with—(V o(dc) − 2V i(dc) ) in the course of the
ESC switch OFF period.
The voltage across the output terminals of ESC is

2−α
Vo(dc) = Vi(dc) (1)
1−α

where V o(dc) is the typical ESC output voltage and V i(dc) is the DBR rectified ESC
input voltage. The voltage transferred from ESC is three times V i (dc) when replacing
the value of α = 0.5 in Eq. (1). It means that the ESC produces one and a half times
252 P. Elangovan et al.

elevated voltage transfer for the similar conductive duty compared to the traditional
boost converter.
It is required to correlate the parameters of ESC, VSI and IM in the proposed
model to justify the impact of voltage in DC link on the performance of IM. The
expression for the torque developed [10] by IM is given by,
 
P Rr Vs2
Torque,T = 3 .  (2)
2 Sωe Rs + Rr 2
S
+ ωe2 (L ls + L lr )2

where P is the pole count of IM, Rr is the resistance offered by rotor, Rs is the
resistance offered by stator, L ls is the inductance offered by stator, L lr is the inductance
offered by rotor, V s is stator terminal voltage, S is slip and ωe is stator supply
frequency.
Considering that the inverter is operating in 180° mode, the output voltage (stator
terminal voltage) of VSI is given by,

2−α
Output voltage (stator terminal voltage) Vs = 0.8165Vo(dc) = Vi(dc) (3)
1−α

Substituting Eqs. (3) in (2), we get

   2
P Rr 1 2−α
T =3 .  Vi(dc) (4)
2 Sωe Rs + Rr 2
+ ωe2 (L ls + L lr )2 1−α
S

From Eq. (4), it is proven that the torque produced by IM is highly dependent on
the DC connection voltage and ESC’s conduction duty. Hence, it is mathematically
validated that the performance of IM in the drive arrangement depends on DC link
voltage.
The control algorithm (AFE scheme) for the proposed ESC is exposed in Fig. 1.
The triggering signal to the suggested ESC is extracted by means of extrapolative
algorithm. In the offered system, the base current (I ref ) for the inductor (L 21 ) is
generated from the facade end AC voltage, which points toward the utility end current
sticks to the similar phase of the facade end voltage signal.
As seen from Fig. 1, there are two parallel paths such as the path that carries utility
end AC voltage signal and the path with the DC link voltage signal are processed
simultaneously to engender the base inductor current I ref .
In the primary path, the utility end AC voltage is tasted by the sample and hold
circuit, then the sampled signal is converted in terms of per unit for simplification,
and finally, for the attainment of the utility end AC voltage in terms of fully rectified
waveform, the per-unit samples are processed through zero-cross detection.
In the secondary path, the real DC bus voltage is matched with the DC bus voltage
reference and the error is treated by the proportional plus integral (PI) controller.
The regulating mechanism of PI regulator’s depends on the values of proportionality
constant (K P ) and integral time (T I ). K P and T I standards are calculated using the
Direct Torque Controlled Induction Motor Drive … 253

tuning process of Ziegler-Nichol [11]. The constants K P and T I are determined from
the delay time L and the time constant T, which are perceived from the phase reaction
S-shaped curve of ESC, according to the tuning method of the Ziegler-Nichol. The
ESC’s state-space model [12] is to be built to achieve the S-shaped curve.
ESC’s state model is framed by considering the state variables A1 (current through
L 21 ), A2 (C 21 voltage) and A3 (C O voltage). By ignoring the resistance values at input
and the output terminals of the recommended ESC, the state model formula of the
ESC during turn ON time is specified by,
⎡ ⎤ ⎡ ⎤⎡ ⎤ ⎡ ⎤
Ȧ1 0 0 0 A1 1/L 21
⎣ Ȧ2 ⎦ = ⎣ −1/C21 0 0 ⎦⎣ A2 ⎦ + ⎣ 1/C21 ⎦ B (5)
Ȧ3 0 0 −1/C O A3 0

where B is the input variable (i.e.,) voltage across ESC’s input terminals. ESC’s state
equation during OFF time is given by,
⎡ ⎤ ⎡ ⎤⎡ ⎤ ⎡ ⎤
Ȧ1 0 −1/L 21 −1/L 21 A1 1/L 21
⎣ Ȧ2 ⎦ = ⎣ −1/C21 0 0 ⎦⎣ A2 ⎦ + ⎣ 0 ⎦ B (6)
Ȧ3 −1/C O 0 −1/C O A3 0

The averaging state model of the anticipated ESC is articulated as,

⎡ ⎤ ⎡ ⎤⎡ ⎤ ⎡ ⎤
Ȧ1 1/L 21 (α − 1)/L 21 (α − 1)/L 21 A1 1/L 21
⎣ Ȧ2 ⎦ = ⎣ (1 − 2α)/C21 −α/C21 0 ⎦⎣ A2 ⎦ + ⎣ α/C21 ⎦ B
Ȧ3 (1 − α)/C O 0 1/C O A3 1/C O
(7)

Applying a unit DC signal for the complete ESC state model represented in Eq. (7),
the constants K P (9.36) and T I (0.016) are directly picked from the tuning chart of
Zeigler-Nichols for the correlating T and L values. With the attained values of K P
and T I , the PI regulator is designed and it will remove the deviations in the voltage
signal of DC bus.
At last, the product output of parallel path signals will generate the reference
current for the L 21 inductor. The actual and reference current fleeting through the
L 21 inductor are compared, and the hysteresis current regulator processes the error
to produce the ESC triggering signal.

3 Uniqueness of the Recommended AFE Mechanism

In this section, the novelty exists in the projected AFE mechanism is explained.
The base of the anticipated AFE topology is the inductor current tracking technique.
Another technique, which is widely used for PF correction purpose at the utility
254 P. Elangovan et al.

end, is the diode current sensing technique. In the diode current sensing technique,
the average current information on the diode is attained, and with respect to a digital
integrated circuit, the sampling instant of the diode current is identified. Even though
the peripheral circuit structure for diode current sensing technique is simple, it is not
feasible for the operation of such circuit in regenerative mode. The proposed inductor
current sensing technique overcomes such problem when power flows bidirectional
in the front end.
The expression for the base inductor (L 21 ) current I ref , which is used to produce
the triggering signals for ESC is derived as follows,
At turn ON period (for t(m) ≤ t < t (m + αT ) of ESC, the mesh equation based
on Kirchhoff’s voltage law can be written as,

di L 21 1
Vidc = L 21 = i C21 dt (8)
dt C21
di L 21
L 21 = Vidc (9)
dt
At turn OFF period [for t(m) + αT ≤ t < t(m + 1)] of ESC, the mesh equation
based on Kirchhoff’s voltage law can be written as,

di L 21 1
Vidc − Vodc = L 21 + i C21 dt (10)
dt C21

Here, t(m) and t(m + 1) are the beginning instant of mth and (m + 1)th turning
cycle; and T is the overall turning period of ESC. Substituting (8) in (10), we get,

di L 21
Vidc − Vodc = 2L 21 (11)
dt
It is possible to rewrite Eq. (11) as,

di L 21
2L 21 = Vidc − Vodc (12)
dt
The differential equations (9) and (12) can be expressed by the following Eqs. (13)
and (14) because the switching frequency of the ESC is much higher than the line
frequency.

i L 21 [t (m) + αT ] − i L 21 [t (m)]
L 21 = Vidc (13)
αT
i L 21 [t (m + 1)] − i L 21 [t (m) + αT ]
2L 21 = Vidc − Vodc (14)
(1 − α)T

where i L 21 [t (m)] and i L 21 [t (m + 1)] are the current through inductor at the origination
time of mth and (m + 1)th conduction cycle, respectively.
Direct Torque Controlled Induction Motor Drive … 255

The current in L 21 during the turn OFF time is dogged from (13) and articulated
as,

Vidc
i L 21 [t (m) + αT ] = αT + i L 21 [t (m)] (15)
L 21

The current in L 21 during the opening of (m + 1)th conduction cycle is obtained

using (14) and as follows,

1
i L 21 [t (m + 1)] = i L 21 [t (m) + αT ] + [Vidc − Vodc ](1 − α)T (16)
2L 21

The current in L 21 during the commencement of (m + 1)th cycle in terms of current

in L 21 during the commencement of mth cycle can be obtained by substituting (15)
in (16) and it is expressed as,

1 1
i L 21 [t (m + 1)] = i L 21 [t (m)] + [Vidc − Vodc (1 − α)]T − Vidc .α.T (17)
2L 21 L 21

The discrete form of Eq. (17) is denoted as,

Vidc (1 − 2α)T − Vodc (1 − α)T

i L 21 (m + 1) = i L 21 (m) + (18)
2L 21

From Eq. (18), it is noticeable that the inductor current during the beginning of the
succeeding switching cycle is determined from the inductor current at the beginning
of the existing switching cycle, firing angle, input voltage and output voltage. Further,
the firing angle (α) is derived from (19) and is represented by,

2L 21 [i L 21 (m + 1) − i L 21 (m)] Vodc − Vidc

α = + (19)
(Vodc − 2Vidc )T Vodc − 2Vidc

The above Eq. (19) defines the firing angle or conduction duty cycle of ESC, in
which the i L 21 (m + 1) and V odc trails I ref and V ref . Therefore, the conduction duty
cycle is redrafted as,

2L 21 [i ref (m + 1) − i L 21 (m)] Vref − Vidc

α= + (20)
(Vref − 2Vidc )T Vref − 2Vidc
256 P. Elangovan et al.

4 Results and Discussions

The ESC-adapted IM drive is tested using MATLAB 2012a software for authenti-
cating the efficacy of the proposed scheme. Table 1 presents the specifications and
their ranges assigned to the developed drive.
In the non-existence of the offered AFE mechanism, the tested result on facade end
voltage and current is exposed in Fig. 2. It is very clear that, in the non-appearance
of the proposed AFE mechanism, the façade side PF is very poor and the line current
is also extremely distorted.
The simulated result on supply side current and voltage is obtained in the presence
of the projected AFE mechanism and is presented in Fig. 3. The voltage and current

Table 1 Specifications and

Specification Range
their ranges of developed IM
drive AC Input supply Single phase, 100 V, 50 Hz
IM Three phase, 0.5 hp, 50 Hz, 400 V,
squirrel cage type
Rs 11.1 
Ls 18.8 mH
Rr 12.3 
Lr 26.7 mH
Mutual inductance of IM 467 mH
Inductors L 11 = 3 mH, L 21 = 2.56 mH
Capacitors C 21 = C o = 2000 µF
VSI Insulated-gate bipolar transistor

Fig. 2 Supply side current and voltage without AFE mechanism. a actual, b zoomed
Direct Torque Controlled Induction Motor Drive … 257

Fig. 3 Supply side current and voltage with AFE mechanism. a actual, b zoomed

signals have zero phase difference, indicating the function at the utility end of the
projected drive with unit PF.
The novelty in the presented AFE control by using inductor current sensing tech-
nique is authenticated by representing the consistent result, and therefore, the actual
and reference current through L 21 are perceived and are shown in Fig. 4.
The result shown in Fig. 4 indicates that the actual current tracks the reference
current phase, the phase of which is similar to the AC supply voltage phase.
In the presence and absence of the developed AFE mechanism, the DC link voltage
of the IM drive is shown in Fig. 5. At the time of proposed AFE mechanism absent,
the output voltage of the implemented ESC shown in Fig. 5a is uncontrolled due
to the fixed gate signal and the fixed value is not set. This unregulated ESC DC

Fig. 4 Actual and reference

current through L 21
258 P. Elangovan et al.

Fig. 5 DC link voltage. a without proposed AFE mechanism, b with proposed AFE mechanism

bus voltage will affect VSI’s performance, which is cascaded with it. At the time
of proposed AFE mechanism present, the DC bus voltage is flawlessly tracking the
reference 300 V DC voltage when observing Fig. 5b. The actual DC link voltage
signal has also enhanced raise time together with the removal of transient and static
state error. The applied ESC also raises the voltage three times higher than the 100 V
AC supply.
The findings addressed up to this point justify the effectiveness of the developed
ESC AFE control in the IM drive, and to authenticate its influence on the implemented
IM drive in the back end control (DTC), results such as stator d-q axis flux trajectory,
speed response and electromagnetic torque are provided and are shown in Figs. 6, 7
and 8.

Fig. 6 Flux route. a without anticipated AFE control, b with anticipated AFE control
Direct Torque Controlled Induction Motor Drive … 259

Fig. 7 Speed of IM. a without integration of AFE mechanism, b with integration of AFE mechanism

Fig. 8 Electromagnetic torque. a without integration of AFE control, b with integration of AFE
control

Figure 6 shows the IM’s stator flux trajectory with the absence and presence of
developed AFE control. As shown in Fig. 6a, at the time of proposed AFE mechanism
absent, the locus of the stator flux is not accurately circular, whereas the perfect
circular trajectory of stator flux is achieved for the IM drive incorporated with the
developed AFE control and shown in Fig. 6b. The developed DTC-based back end
control provides low speed (100 rpm) and zero speed regions for IM operation.
Figure 7 shows the speed response of IM with and without suggested AFE control.
When observing Fig. 7a, the actual speed of IM follows the reference speed without
integrating the established AFE mechanism, but it is not possible to eliminate the
260 P. Elangovan et al.

Table 2 Comparative
Parameter IM drive in [7] Proposed IM drive
analysis
IM control technique V/f control Direct torque control
Low speed Not reported Reported and
regulation excellent
Speed regulation Poor Excellent
during load
DC link voltage Less than 1% Less than 1%
ripple
Flux trajectory Not reported Reported and
excellent

occurrence of oscillations throughout the operation. The integration of suggested

AFE mechanism facilitates the actual speed signal without oscillation to follow the
reference speed signal and is shown in Fig. 7b.
The projected IM drive operates in a state of no load. The electromagnetic torque
developed by IM with and without the suggested AFE control is depicted in Fig. 8.
On comparing Fig. 8a, b, it is justified that the presence of suggested AFE control in
the developed IM drive influences over elimination of torque ripple of IM.
To authenticate the developed IM drive, a comparative study between the proposed
work and the work reported in [7] has been made and it is presented in Table 2.

5 Conclusion

This manuscript presents an AFE control using ESC embedded with a three-phase
VSI-fed DTC-based IM drive. In this system, the AFE mechanism works in prog-
nostic algorithm and the influence of the anticipated AFE mechanism on the supply
end, DC linking part, and IM end parameters are discussed. A comparison for the
utility end PF, DC link voltage and motor end parameters (IM speed, flux trajectory
and electromagnetic torque) between the premeditated IM drives with and without
AFE are shown via simulation. The results indicate that the supply end PF is in unit
for IM drive integrated with the indicated AFE, whereas in the non-appearance of
the designed AFE mechanism, the supply end PF is less than 0.5. The ESC’s ability
to reduce the DC bus voltage ripple by less than 1% is a significant advantage that
brings out it as an appropriate replacement for the conventional DC–DC boost con-
verters. For the anticipated unit, the IM response for the speed variance from 100 to
0 rpm validates the suitability of the control based on DTC at the motor end.
Direct Torque Controlled Induction Motor Drive … 261

References

1. Naik, N.V., Singh, S.P.: Improved dynamic performance of direct torque control at low speed
over a scalar control. In: IEEE Proceeding INDICON-13, pp. 1–6, IIT Bombay (2014)
2. Guo, Y., Wang, X., Guo, Y., Deng, W.: Speed-sensorless direct torque control scheme for matrix
converter driven induction motor. J. Eng. 13, 432–437 (2018)
3. Lascu, C., Jafarzadeh, S., Fadali, M.S., Blaabjerg, F.: Direct torque control with feedback
linearization for induction motor drives. IEEE Trans Power Electron. 32(3), 2072–2080 (2016)
4. Buja, G.S., Kazmierkowski, M.P.: Direct torque control of PWM inverter-fed AC motors—A
survey. IEEE Trans. Industr. Electron. 51(51), 744–757 (2004)
5. Xie, X., Wang, J., Zhao, C., Lu, Q., Liu, S.: A novel output current estimation and regulation
circuit for primary side controlled high power factor single-stage fly-back LED driver. IEEE
Trans. Power Electron. 27, 4602–4612 (2012)
6. Maheshwari, R.L., Munk-Nielsen, S., Lu, K.: An active damping technique for small DC-link
capacitor based drive system. IEEE Trans. Industr. Inf. 9(2), 848–858 (2013)
7. Elangovan, P., Mohanty, N.K.: PI controlled active front end super-lift converter with ripple
free DC link for three phase induction motor drives. J. Power Electron. 16(1), 190–204 (2016)
8. Luo, F.L., Ye, H.: Advanced Dc/DC Converters. CRC Press, London (2003)
9. Luo, F.L., Ye, H.: Positive output super-lift converters. IEEE Trans. Power Electron. 18(1),
105–113 (2003)
10. Bose, B.K.: Modern Power Electronics and AC Drives, 4th edn. Pearson Education, South Asia
(2007)
11. Comines, P., Munro, N.: PID controllers: recent tuning methods and design to specification.
Proc. IEE Control Theor. Appl. 149(1), 46–53 (2002)
12. Panov, Y., Rajagopalan, J., Lee, F.C.: Analysis and control design of N paralleled DC-DC con-
verters with master-slave current sharing control. In: Proceeding of Applied Power Electronics
Conference, pp. 436–442 (1997)
Design of Synchronous Reluctance Motor
for Ceiling Fan Application

Pratik Kochgabay, P. Ramesh, and N. C. Lenin

Abstract Ceiling fans are one of the widely used household appliances. In this paper,
a 22 W, ferrite-assisted synchronous reluctance motor (SRL) is proposed for ceiling
fan application. The performance of the designed SRL is analyzed and compared to
PMBLDC motor. Due to the usage of non-rare earth material, the proposed motor
decreases the dependency of rare earth magnets, which is being used in PMBLDC
motors for ceiling fans. The proposed design also offers low cost, low power con-
sumption and high power density. These advantages make the proposed SRL as an
optimal choice for commercial usage.

Keywords Synchronous reluctance motor · BLDC · Ceiling fan

1 Introduction

Ceiling fans are the most generic application widely used around the world. It con-
sumes about 20% of the electricity in households, and their numbers are growing
rapidly. The production of all fans in India is about 30 million units per year [1].
Reducing 20% to reflect sales of table and pedestal fans and 10% for exports, annual
sales of ceiling fans in India are about 29 million. These fans require power gener-
ation of about 2000 MW each year [2]. Past many years, majority of fan industries
rely on single phase induction motor (IM), which consumes 60–80 W power with
the efficiency less than 40% [3, 4]. Poor efficiency and high power consumption lead
to energy wastage and increase heat in the system. After the invention of PMBLDC
fan, power consumption was brought down by 40–50% [3–6]. In spite of better per-
formance, cost and availability of rare earth magnets have always been a foremost
problem that increases the manufacturing and selling prices.
In this paper, a low cost, low weight SRL is proposed with ferrite-assisted magnets
to reduce the power consumption with increased efficiency [7]. This machine offers
high torque density and reduced copper loss when compared to pure synchronous

P. Kochgabay · P. Ramesh · N. C. Lenin (B)

Vellore Institute of Technology, Chennai Campus, Chennai 600127, India
e-mail: [email protected]

© Springer Nature Singapore Pte Ltd. 2020 263

G. Pradhan et al. (eds.), Advances in Electrical Control and Signal Systems,
Lecture Notes in Electrical Engineering 665,
https://doi.org/10.1007/978-981-15-5262-5_18
264 P. Kochgabay et al.

reluctance and induction machine [8–11]. In Sect. 2, the torque production of SRL
is discussed. The design, electromagnetic and thermal analysis of SRL for ceiling
fan application is carried in Sects. 2, 3 and 4. Finally, the comparative analysis of the
proposed motor with existing BLDC motor is carried in Sect. 5.

2 Modeling of SRL

In SRL, three phase symmetrical winding in stator creates sinusoidal rotating mag-
netic field in the air gap. The flux lines flows through rotor d-axis and q-axis and
produces the reluctance torque.
Figure 1 shows the cross section of the proposed SRL rotor. This rotor is designed
in such a way to have maximum inductance in the d-axis and minimum inductance
in the q-axis. The ferrite magnets in the rotor help to improve the power density. The
phasor diagram of SRL motor is shown in Fig. 2 from which the torque equation can
be derived as given in Eq. (1).

3 P   
T = L d i d i q − L q i q i d + ϕa i d (1)
22
where,
id —Direct axis current
iq —Quadrature axis current
ϕ a —Magnetic flux
ϕ o —Optimal magnetic flux

Fig. 1 Cross section of the

rotor of proposed SRL

Rotor core

d-axis

Ferrite magnets q-axis

Flux barrier
Design of Synchronous Reluctance Motor for Ceiling … 265

Fig. 2 Phasor diagram of

SRL Motor [12] q-axis
Ldid
Lqiq
a
Ia

iq o

id d-axis

L d —Direct axis inductance

L q —Quadrature axis inductance.
From Eq. (1), the torque developed in this motor is mainly due to the saliency
(inductance difference between d-axis and q-axis), which is termed as reluctance
torque. Additionally, the permanent magnet flux linkage produces the magnet torque,
which aids the reluctance torque.

2.1 Design Specifications of Fan Motor

The design specifications and dimensions of conventional BLDC motor are given
in Table 1 [3, 13]. The SRL motor is also designed for the same specifications and
dimensions, in order to adopt the same enclosure.
In order to avoid the manufacturing and assembling difficulties, the air gap
between the stator and rotor is constrained as 0.5 mm. The other parameters like
number of poles, bore diameter, slot depth, slot opening width and tooth width are
optimized to achieve better performance. The slot fill factor is limited less than 35%
with the conductor diameter of 0.25 mm.

Table 1 Design
S. No. Parameter Value Unit
specifications and dimensions
[3, 13] 1 Voltage 230 Volts (AC)
2 Air delivery 230 CMM
3 Rated torque 0.57 Nm
4 Rotor speed 370 rpm
5 Rotor outer diameter 152 mm
6 Stack length 13 mm
266 P. Kochgabay et al.

3 Design Verification Through Numerical Study

The finite element analysis (FEA) helps in improving the torque–speed character-
istics, reducing the loss, analyzing the thermal behavior and operating the machine
in optimal region. In order to visualize the real-time product, the CAD model of the
designed SRL is shown in Fig. 3. The materials used for various components of the
motor are listed in Table 2. The study has been carried out with the help of MagNet
v7.7 software in Dell Workstation T1700.
The meshing is an important step in FEA, during which the designed model is
subdivided into smaller elements, over which the set of equations are solved. The
designed SRL has been meshed using two-dimensional triangular elements. Figure 4
shows the generated mesh, which has 16,562 nodes and 33,086 elements.
The flux density in the stator and rotor core is limited less than the saturation level
of the material, 2.3 T. The magnetic flux distribution at the rated load condition is
shown in Fig. 5. The unused core in the stator back iron can be reduced by creating
ducts.
In the air gap, the maximum flux density is about 0.37 T. The air gap flux density
mainly depends on the air gap thickness, magnet position, magnetization direction,
stator’s slot and tooth structure. Figure 6 shows the flux distribution in the air gap
for one pole pair.

Rotor Core
Ferrite
Magnet

Stator Shaft
teeth

Copper Stator back

Windings iron

Fig. 3 SRL-CAD model

Design of Synchronous Reluctance Motor for Ceiling … 267

Table 2 Materials used

S. No. Component Material
1 Rotor core M-43 26 Ga
2 Magnet Ceramic 11
3 Stator core M-43 26 Ga
4 Windings Copper
5 Shaft EN8

Fig. 4 Mesh model of SRL

Figure 7 represents the input current waveform of the designed SRL at the rated
condition. The peak current value is 0.214 A, with the current density of 3 A/mm2 .
Generally, the machines with current density less than 5 A/mm2 do not require
external cooling.
The torque generated, while operating the SRL at the rated speed of 370 rpm, is
plotted in Fig. 8. The average value of the torque developed is 0.58 Nm.

4 Thermal Study

For the applications like ceiling fan, the lifetime plays a vital role. The thermal
characteristics of the motor mainly decide the lifetime of the fan. The losses such
as ohmic loss and core loss are the heat sources of the motor. High temperature rise
268 P. Kochgabay et al.

Fig. 5 Magnetic flux distribution–partial geometry of SRL (at rated load)

0.6

0.4
Air gap flux (T)

0.2

-0.2

-0.4

-0.6
0 60 120 180 240 300 360
Position (electrical degrees)

Fig. 6 Air gap flux density

may also lead to insulation failure, reduces the efficiency and subsequently leads to
malfunctioning of the machine.
A coupled thermal analysis has been carried out by ThermNet v7.7, in order
to predict the temperature rise in the designed SRL. In which, the instantaneous
losses are considered directly from the electromagnetic analysis. The motor has
been operated at the rated load continuously for 10 h, with the ambient temperature
of 30 °C. The maximum temperature 42 °C is reached at the end windings whereas
39 °C in magnets. The temperature rise obtained in various components of the motor
is plotted in Fig. 9.
Design of Synchronous Reluctance Motor for Ceiling … 269

0.3
Current reference (upper)

0.2 Current reference (lower)

Current
0.1
Current (A)

-0.1

-0.2

-0.3
0 60 120 180 240 300 360
Source phase angle (in degree)

Fig. 7 Current waveform

0.8
Torque(Nm)

0.6

0.4

0.2

0
0 60 120 180 240 300 360
Source phase angle(degrees)

Fig. 8 Torque versus rotor angle

5 Performance Comparison of Proposed Motor with BLDC

Motor

In order to validate the advantages of designed SRL motor, the performance of SRL
motor is compared with BLDC motor from Crompton Greaves Ltd [3]. Figure 10
shows the input power consumption of SRL and BLDC with NdFeB and ferrite
magnets. SRL consumes 27.9 W of input power, which is 5% less than the BLDC
(Ferrite).
The output power and efficiency of the discussed motors are plotted in Figs. 11
and 12. The proposed SRL motor is delivering the output power of 22.3 W with an
270 P. Kochgabay et al.

Temperature (degree celcius) 45

40

Rotor Core (ºC)

Rotor Magnet (ºC)
35
Stator Back Iron (ºC)
Stator Coil Side (ºC)
Stator End Winding (ºC)
30
0 100 200 300 400 500 600
Time (in minutes)

Fig. 9 Temperature rise

Fig. 10 Input power 29.5 29.34

consumption 29.13
29
Input power (W)

28.5
27.9
28

27.5

27
BLDC BLDC SRL
(NdFeB) (Ferrite)

Fig. 11 Output power 22.4 22.3

22.2
Output power (W)

22.08

22
21.89

21.8

21.6
Design of Synchronous Reluctance Motor for Ceiling … 271

Fig. 12 Efficiency 80 79.7

comparison
78

Efficiency (%)
75.8
76 74.6
74

72

70
BLDC BLDC SRL
(NdFeB) (Ferrite)

Table 3 Material weight and cost

Material Cost per BLDC BLDC BLDC BLDC SRL SRL
kg (Rs) (NdFeB) (NdFeB) (Ferrite) (Ferrite) weight cost (Rs)
[3] [3] [3] [3] (kg)
weight cost (Rs) weight cost (Rs)
(kg) (kg)
Copper 600 0.213 128 0.367 220 0.288 172.8
Stator 100 0.457 46 0.513 51 0.6 60
steel
Rotor 90 0.114 10 0.285 25 0.181 16.29
steel
NdFeB 2500 0.100 250 – – – –
Ferrite 300 – – 0.135 42 0.095 28.5
Total – 0.884 434 1.302 338 1.164 277

efficiency of 79.7%. At the rated load condition, the BLDC (NdFeB) and BLDC
(Ferrite) motor are operating with the efficiency of 75.8% and 74.6%, respectively,
which are 3.9 and 5.1% lesser than the proposed SRL motor.
The material weight and cost of various components of the discussed motors are
shown in Table 3. The designed SRL motor has the weight of 1.16 kg, which is 11%
less than the BLDC (Ferrite) motor. This helps to reduce 18% of the material cost.
Due to reduced weight, the power to mass ratio of the SRL motor is improved by
14%.

6 Conclusion

A 22 W, ferrite-assisted SRL motor is proposed for ceiling fan. The proposed SRL
motor is analyzed and compared with existing BLDC motor. Compared to the BLDC
motor with ferrite magnets, the proposed motor has the following advantages,
272 P. Kochgabay et al.

• Reduces the power consumption by 5%, which helps to improve the efficiency.
• 11% reduction in weight, which enhances the power to mass ratio by 14%.
• Reduces the material cost by 18%, which helps to reduce the overall cost of the
fan.

References

1. Mahesh, R.: Energy efficient ceiling fans using BLDC motors-a practical implementation.
In: Singh, R.K. (eds.) International Conference on Advances in Computer Electronics and
Electrical Engineering, pp. 59–63. Switzerland (2018)
2. Prayas Homepage: http://www.prayaspune.org/peg/publications/item/81-ceiling-fans-the-
overlooked-appliance.html. Last accessed 30 Aug 2019
3. Anchal, S.: Performance and cost comparison of PM BLDC motors for ceiling can. In: IEEE
International Conference on Power Electronics Drives and Energy Systems, pp. 1–5. IEEE,
Mumbai (2014)
4. Izhar, T., Mohammad, A., Adil, I.: Energy efficient five phase BLDC ceiling fan. In: 2nd
International Conference on Energy Conservation and Efficiency. IEEE, Lahore (2018)
5. Anil, D.M.: Energy conservation in ceiling can using BLDC motor. Int. J. Adv. Eng. Res. Dev.
4(7), 197–199 (2017)
6. Tejas, H.P., Amit, N.P., Rajesh, M.P.: Reduction of cogging torque of radial flux permanent
magnet brushless dc motors with application of dual permanent magnet materials. Int. J. Eng.
Adv. Technol. 8(4), 1714–1716 (2019)
7. Lenin, N.C., Ramesh, P., Pratik, K.: Low cost stator, rotor and hub structures for energy efficient
fans. Patent application number—201941007806, Ipr India (2019)
8. Lonna, H., luha, P.: Permanent magnet assisted synchronous reluctance motor an alternative
motor in variable speed drives. In: Parasiliti, F., et al. (eds.) Energy Efficiency in Motor Driven
Systems 2003. Springer, Heidelberg (2003)
9. Xiao, C., Jiabin, W., Panagiotis, L., Liang, C.: Permanent magnet assisted synchronous reluc-
tance machine with fractional-slot winding configurations. In: International Electric Machines
and Drives Conference 2013, pp. 374–381. IEEE, USA (2013)
10. Emir, P., Cristian, B., Nicola, B., Luca, F.: The study of permanent magnet assisted reluctance
machine with the adoption of NdFeB bonded magnets. In: International Symposium on Power
Electronics Electrical Drives Automation and Motion 2018, pp. 274–279. IEEE, Italy (2018)
11. Enrico, C., Micheie, D., Nicola, B.: Permanent magnet volume minimization in permanent mag-
net assisted synchronous reluctance motors. In: 8th International Conference and Exhibition
on Ecological Vehicles and Renewable Energies 2013. IEEE, Monaco (2013)
12. Dehghani, A.J., Hassan, K., Mohsen, N., Jafar, M.M.: Optimum design and operation analysis
of permanent magnet-assisted synchronous reluctance motor. Turk. J. Electr. Eng. Comput.
Sci. 25(2017), 1894–1907 (2016)
13. Superfan Homepage: https://www.superfan.in/product-x1.html. Last accesses 30 Aug 2019
Modified Empirical Mode Decomposition
and Teager–Kaiser Energy
Operator-Based Phasor Estimation
in Presence of DC Offset for Digital
Relaying Application

Debadatta Amaresh Gadanayak and Ranjan Kumar Mallick

Abstract Conventional discrete Fourier transform algorithm which is commonly

used for phasor estimation in digital protective relays exhibits large estimation error
and long convergence time in presence of exponentially decreasing DC components.
This paper presents an efficient algorithm for phasor estimation using a modified
empirical mode decomposition and Teager–Kaiser energy operator. The knot-based
empirical mode decomposition efficiently separates the decreasing DC component
from the signal and the Teager–Kaiser energy operator estimates the amplitude with
minimum delay. The performance is evaluated using an ideal signal with double
decreasing dc component generated in MATLAB and fault signals from a 66 kV
transmission line model created in Simulink. Simulation results show promising
results in terms of estimation accuracy and convergence time as compared to the
Fourier transform-based method. Because of low-computational complexity, higher
accuracy and satisfactory convergence time, this method is practicable and proficient
for fast digital relaying applications.

Keywords Phasor estimation · Empirical mode decomposition · Knot-based

empirical mode decomposition · Teager–Kaiser energy operator · Hilbert transform

1 Introduction

In the field of modern digital protective relaying accurate measurement of phasor

parameters of current and voltage signals is extremely important. For this purpose,
digital filtering technologies and estimation theories have a major role in the develop-
ment of protection algorithms. Discrete Fourier transform (DFT) is commonly used
for estimation of the phasor parameters of the fundamental frequency component.
DFT can accurately evaluate the magnitude, phase and frequency when the signal
contains only the fundamental and the harmonic components whose frequencies are
integral multiples of the fundamental frequency with only a signal-window length

D. A. Gadanayak (B) · R. K. Mallick

Siksha O Anusandhan deemed to be University, Bhubaneswar, India
e-mail: [email protected]

© Springer Nature Singapore Pte Ltd. 2020 273

G. Pradhan et al. (eds.), Advances in Electrical Control and Signal Systems,
Lecture Notes in Electrical Engineering 665,
https://doi.org/10.1007/978-981-15-5262-5_19
274 D. A. Gadanayak and R. K. Mallick

of one fundamental cycle. However, fault current signals normally contain exponen-
tially decaying DC (DDC) offsets which have non-integer harmonic components.
Hence, DFT algorithm often exhibits large estimation error and long convergence
time. Under the worst case, the amplitude estimation of such signals can differ from
the actual value by more than 20% [1]. Therefore, it is necessary to remove the decay-
ing DC component from the signal for high-performance protection algorithms and
other protection-related applications such as fault location.
Several techniques have been proposed in the literature to deal with the adverse
effects of DDC offset on phasor measurement error. An FIR-type digital mimic filter
is proposed in [2]. However, the filtering performance depends upon the choice of
time constant which is generally not offered in a real power system [3]. An even-
sample-set DFT and odd-sample-set DFT-based estimation to eliminate the adverse
effects of exponentially DDC offset is developed in [4]. But the method suffers
from high-computational burden due to the complexity of the algorithm. A partial
summation based algorithm is proposed in [1]. It uses three simplifying algorithms to
compromise between speed of computation and accuracy. But its performance is not
consistent in case of double decaying DC offsets. A quick convergence invariant filter
for estimation of the fundamental is proposed in [5]. The characteristic harmonic
digital filter uses an approximation for the exponential term in DDC. However, it
cannot be used for signals with multiple DDC. Also, the estimation error is sometimes
significant due to approximation error. A mathematical morphology-based filter is
used to eliminate DDC in [3]. However, it also assumes the fault current to have
only single DDC, hence is not suitable for signals with multiple DDC. In [6], a
modified empirical mode decomposition and half-cycle Hilbert transformation based
methodology are proposed. However, the algorithm can be modified using the Teager
energy operator-based amplitude estimation method for faster response.
In this work, a recently developed, fully data-driven modified empirical
mode decomposition (EMD) called the knot-based empirical mode decomposition
(KEMD) with Teager–Kaiser energy-based estimation technique is used for phasor
estimation. The proposed method is used for signals with single as well as double
DDC and has been compared with the classical DFT.

2 Theoretical Background

In this section, few signal processing concepts that will be used for the estimation
of fundamental phasor in presence of DDC is briefly presented. Firstly, the concept
of classical empirical mode decomposition and the modified knot-based empirical
mode decomposition is reviewed. Next, the Teager energy operator is presented and
use of it to estimate instantaneous amplitude and frequency is discussed.
Modified Empirical Mode Decomposition … 275

2.1 Empirical Mode Decomposition

Empirical mode decomposition [7] is a signal processing tool used for decomposition
of a complex signal into simpler components called intrinsic mode functions (IMFs),
having either mono-frequency or narrow-band frequency content. The IMFs must
satisfy the following conditions:
(a) The difference between the number of extrema and the number of zero crossings
must be either zero or one.
(b) At every point, the mean of the upper and lower envelops characterized by the
local maxima and minima respectively must be zero.
EMD uses a specific procedure, often termed as the EMD sifting process to extract
the IMFs [8], as given below:
(a) Find the positions of maxima (P j ) and minima (Q j ) in the signal
 y(t).
(b) Create the upper envelope H (t) by connecting points P j , y P j using cubic
spline interpolation.
 Similarly
  lower envelope L(t) is found out by cubic spline
connection of Q j , y Q j .
(c) Now the mean envelope operator S can be defined as,

1
S(y) = y − (H (t) + L(t))
2
(d) The first IMF can be found as,

IMF1 = lim S n
n→∞

where S n is the output after application of S operator n times. Equation (2) can be
stated as, IMF1 is the residue after operating S operator on the signal y again and
again until the further processing will not be able to change the signal.
(e) The stopping criteria for the above process as proposed in [8], can be written as,

 n 
 S (t) − S n−1 (t)2
SD =   ≤ α,
 S n−1 (t)2

where 0.1 ≤ α ≤ 0.4.

(f) Once the first IMF is obtained, the residue y 1 (t) can be extracted as y 1 (t) =
y − IMF1 . Now y 1 (t) is treated as a fresh signal and the rest of the IMFs are
produced by steps (i) to (v).

IMFk = lim S n (y − IMF1 − IMF2 − · · · − IMFk−1 )

n→∞
276 D. A. Gadanayak and R. K. Mallick

(g) The above procedure is terminated when total number of extrema in y k (t)
becomes less than or equals to 3. Now y(t) is represented as,


k
y(t) = IMF j + y k (t)
j=1

2.2 Knot-Based Empirical Mode Decomposition

An alternative algorithm for EMD is proposed in [9] to establish flat envelope signals.
It is based on the use of particular knot points depending upon the positions of
maxima and minima for the calculation of the mean envelopes. The sifting procedure
is the same as the classical EMD with two loops in which the inner loop calculates
individual IMFs and the outer loop decides the stopping criteria. The KEMD of a
signal y(t) can be given as below:
(a) Find the positions of maxima (P j ) and minima (Q j ) in the
 signal y(t).

(b) Create the upper envelope H (t) by connecting points P j , y(P j ) using lin-
ear
 interpolation.
 Similarly lower envelope L(t) is found out by connecting of
Q j , y(Q j ) .
(c) Find the vector of extrema positions(U j ) by sorting the position vectors
Pj , Q j .
(d) Find the knot points at each maxima (P j ) as

  1    
N Pj = y Pj + L Pj
2
(e) Find the knot points at each minimum(Q j ) as,

  1    
N Qj = y Qj + H Qj
2
(f) Now
 the
 mean
 envelope M is extracted by cubic spline interpolation of
Uj, N Uj .
(g) Define the operator S as,

S(y) = y − M.

(h) The first IMF can be found as,IMF1 = lim S n where S n is the output after
n→∞
application of S operator n times.
(i) The stopping criteria is the same as classical EMD,
Modified Empirical Mode Decomposition … 277

Fig. 1 Derivation of knot points in knot-based empirical mode decomposition

  2
S n (t) − S n−1 (t)
SD =   ≤ α.
 S n−1 (t)2

And all IMFs are obtained in the same way as the classical EMD.
(j) The process stops when the residue has less than 3 extrema.
KEMD sifting process differs from the classical EMD only in the extraction of
the mean envelope which is obtained by the cubic spline interpolation of the knot
points. The knot point (K ) extraction process from upper (H (t)) and lower (L(t))
envelopes is shown in Fig. 1.

2.3 Teager–Kaiser Energy Operator

Teager–Kaiser energy operator (TKEO) was formulated in [10] for instantaneous

energy tracking of AM-FM signals. It was later used for the estimation of instan-
taneous amplitude and frequency of AM-FM signals with excellent accuracy [11].
The amplitude and frequency estimations require only three samples of the signal,
and hence, it has a very high time resolution to even minor deviations in the signal
and is faster.
The discrete Teager–Kaiser operator for a signal y(t) is defined as,

Ψ [y(n)] = y 2 (n) − y(n − 1)y(n + 1)

The instantaneous amplitude and frequency estimation algorithm using TKEO

developed in [11] can be given as below:
When D(n) = y(n) − y(n − 1) and G(n) = 1 − Ψ [D(n)]−Ψ [D(n+1)]
4Ψ [y(n)]
, the
instantaneous frequency is given as,

1
f (n) = arccos[G(n)],
2π Ts
278 D. A. Gadanayak and R. K. Mallick

where Ts is the sampling period of the signal.

The instantaneous amplitude is given as,

Ψ [y(n)]
|A(n)| =
1 − G 2 (n)

3 Methodology

Let us take the Thevenin equivalent circuit of the point at which the fault occurs
as shown in Fig. 2 with an open-circuit voltage of E and an internal impedance of
Z s = Rs + j X s . Practically, the values of Rs and X s values are not constant and
change with the operating conditions of the power system. The fault impedance is
also another variable quantity. If the fault occurs at t = 0 s, the instantaneous fault
current i f can be written as,

i f (t) = Imax sin(ωt + θv − θz ) − sin(θv − θz )e−t / τ ,

ωL s
where τ = Ls
Rs
, Imax =√ Vmax
, and θz = arctan Rs
.
Rs2 +(ωL s )2
So the maximum value of DDC will occur when sin(θv −θz ) = 0, that is θz −θv =
(2n + 1)π/2 n = 0, 1, 2, 3, . . ..
At this condition, the fault current equation will become

i f (t) = Imax − cos(ωt) + e−t / τ .

However, it has been established in [12] that when current is measured by the
current transformer (CT), the CT introduces an additional exponentially decreasing
DC component in the signal.
A test signal with double DDC is given by,

Fig. 2 Thevenin equivalent

circuit at fault point
Modified Empirical Mode Decomposition … 279

Fig. 3 The test signal with double exponentially decreasing DC components


1.0 cos(2π f t) (t ≤ 0)
y(t) =
−5 cos(2π f t) + d1 e−t / T1 + d2 e−t / T2 (elsewhere)

where f = 50 Hz, d1 = 4, d2 = 1, T1 = 0.15, T2 = 0.05, sampling frequency

( f s ) = 1200 Hz. Figure 3 shows the test signal for 0.7 s.
Despite, several advantages of classical empirical mode decomposition and meth-
ods based on EMD like ensemble empirical mode decomposition (EEMD), complete
ensemble empirical mode decomposition with adaptive noise (CEEMDAN), down-
sampling empirical mode decomposition (DEMD), and knot-based empirical mode
decomposition (KEMD), all of them share a common disadvantage, that is, for useful
decomposition, they require long data windows. Therefore, for faster phasor estima-
tion, it is essential to find the minimum data window required for error-free decom-
position. In [6], based on several trials, the optimized window length for KEMD is
found out to be 2 14 cycles.
Figures 4 and 5 show the decomposition of the test signal by KEMD with the
window at two different positions. Figure 4 shows the current waveform decom-
position when the window composes of post-fault currents only that is the current
waveform consists of a sinusoidal signal of constant magnitude and the exponentially
decreasing DC components. The decomposition gives directly the fundamental as
the first IMF and the DDC as the second IMF and the residue. Figure 5 shows the
decomposition of the current waveform when the window consists of both pre-fault
and post-fault quantities. It gives 3 IMFs and the residue. However, the first IMF is
having the maximum magnitude. Hence, although the complete KEMD can be used
in other applications for retrieving more hidden information about the waveform
under analysis, for the case of amplitude estimation of the fundamental, only the first
IMF is sufficient. Hence, the KEMD algorithm is required only until the first IMF
is obtained. That further decreases the computational complexity of the proposed
scheme to a great extent.
280 D. A. Gadanayak and R. K. Mallick

Fig. 4 Decomposition of steady-state signal

Fig. 5 Decomposition of transient state signal

The proposed amplitude and frequency measurement scheme composes of two

steps.
(a) Step 1: Removal of DDC from the signal by KEMD.
(b) Step 2: Estimation of instantaneous amplitude and frequency by Teager–Kaiser
energy operator.
Once, DDC components are removed; various methods can be used for instan-
taneous amplitude and frequency estimation, such as the classical DFT, the Hilbert
transform (HT), etc. However, practically, the DFT requires at least one complete
cycle and the Hilbert transform requires at least half-cycle of data for estimation of the
fundamental amplitude. But the Teager–Kaiser operator requires only three samples
for correct estimation of these instantaneous quantities. In other words, calculation
of instantaneous quantities by DFT and HT is delayed by one cycle and half-cycle
respectively, whereas, estimation by TKEO is delayed by only three samples.
Modified Empirical Mode Decomposition … 281

4 Performance Evaluation

In this section, the proposed algorithm is verified and compared with the classical
DFT for the signal with double DDC discussed in Sect. 3. The advantage of Teager
energy-based estimation is compared with that of Hilbert transform which is normally
used for estimation of instantaneous quantities in connection with IMFs of empirical
mode decomposition. The algorithm is also verified with a second set signals obtained
from the simulation of fault in a 66 kV transmission line modeled in SIMULINK.
All signals are sampled at 1200 Hz.

4.1 Basic Signal with Double DDC

For the signal with double DDC discussed in Sect. 3 (Fig. 3), amplitude estimation
by the proposed method is compared with the estimation by classical DFT in Fig. 6.
It can be observed that the estimation by the proposed method becomes almost
constant at about 0.033 s. The error becomes below 1% of the actual amplitude. But
in the case of estimation by DFT to achieve error below 1%, it requires 0.2 s.
Normally, the instantaneous amplitude estimations of the IMFs obtained via EMD
are done using the Hilbert transform (HT). Practically, a window length of half-cycle
of the fundamental is required to be correctly estimated by HT. Figure 7 shows
a comparison between the Hilbert transform-based estimation and Teager–Kaiser
energy-based estimation on the DDC-free first IMF obtained by KEMD. It can be
observed that the HT-based estimations are delayed by 0.005 s that is six samples
as compared to TKEO-based estimations. It is because HT requires a window of 10
samples as compared to three samples by TKEO.

Fig. 6 Comparison between DFT and the proposed scheme

282 D. A. Gadanayak and R. K. Mallick

Fig. 7 Comparison between Teager–Kaiser energy operator and Hilbert transform

4.2 Simulink Generated Signal

Using Simulink, a simulation model of a 66 kV, 42 km overhead transmission line

shown in Fig. 8 is designed to evaluate the performance of the scheme. The transmis-
sion lines are designed using the PI-model. Line parameters are given in Table 1. The
faults are created at the mid-point of transmission line with different fault resistances
(R f ) at 0° fault inception angle. So, DDC of different fault currents has different val-
ues of time constants. Figure 9 shows three different fault currents at R f = 0.01Ω,
10 and 40 .

Fig. 8 Transmission line model

Table 1 Transmission line

Sequence components Parameters Value/km
parameters
Positive sequence R1 0.1153
L1 1.048 mH
C1 11.33 nF
Zero sequence R0 0.3963
L0 2.730 mH
C0 5.338 nF
Modified Empirical Mode Decomposition … 283

Fig. 9 Fault currents at different values of fault resistance

Fig. 10 Comparison between DFT and the proposed scheme for Simulink generated signal

It is observed that in each case, the proposed technique is able to estimate mag-
nitude within 0.035 s with an error boundary of 1%. Figure 10 shows a comparison
of DFT and the proposed scheme for the fault current at Rf = 0.01 .

5 Conclusion

It is a well-established fact that the conventional full-cycle Fourier-based phasor esti-

mation techniques show undesirable oscillations in the estimated amplitude results
due to presence of exponentially decreasing DC components. It often exhibits large
estimation error and long convergence time. This work presents an improved phasor
estimation algorithm using newly developed knot-based empirical mode decompo-
sition and Teager–Kaiser operator. The performance evaluation using a signal with
284 D. A. Gadanayak and R. K. Mallick

double DDC shows improved accuracy and shorter convergence time as compared to
conventional DFT. The Teager–Kaiser energy operator-based instantaneous ampli-
tude estimation requires only a 3 sample window as compared to the minimum
half-cycle window length for popular Hilbert transform-based scheme. The superi-
ority of TKEO as compared to HT in terms of response time is also established. In
addition, the proposed algorithm is tested for a single line to ground fault current on
a 66 kV transmission line modeled in Simulink. It should be noted that the KEMD
is applied only partially up to the first IMF and TKEO-based calculations are com-
putationally efficient due to smaller window size. Hence, the overall algorithm can
be assumed to be less complex and suitable for fast digital relaying applications.

References

1. Guo, Y., Kezunovic, M., Chen, D.: Simplified algorithms for removal of the effect of exponen-
tially decaying DC-offset on the Fourier algorithm. IEEE Trans. Power Deliv. 18(3), 711–717
(2003)
2. Benmouyal, G.: Removal of DC-offset in current waveforms using digital mimic filtering. IEEE
Trans. Power Deliv. 10(2), 621–630 (1995)
3. Godse, R., Bhat, S.: Real-time digital filtering algorithm for elimination of the decaying DC
component using mathematical morphology. IET Gener. Transm. Distrib. 13(15), 3230–3239
(2018)
4. Kang, S., Lee, D., Nam, S., Crossley, P.A., Kang, Y.: Fourier transform-based modified phasor
estimation method immune to the effect of the DC offsets. IEEE Trans. Power Deliv. 24(3),
1104–1111 (2009)
5. Domínguez, J.L., Argüelles, J.F., Arrieta, M.A., Jaurrieta, B.L., Benito, M.S., Zugazaga, I.A.:
New quick-convergence invariant digital filter for phasor estimation. Electric Power Syst. Res.
79(5), 705–713 (2009)
6. Jiang, Z., Miao, S., Liu, P.: A modified empirical mode decomposition filtering-based adaptive
phasor estimation algorithm for removal of exponentially decaying DC offset. IEEE Trans.
Power Deliv. 29(3), 1326–1334 (2014)
7. Huang, N.E., Shen, Z., Long, S.R., Wu, M.C., Shih, H.H., Zheng, Q., Yen, N.-C., Tung, C.C.,
Liu, H.H.: The empirical mode decomposition and thehilbert spectrum for nonlinear and non-
stationary time series analysis. Proc. R. Soc. Lond. A: Math. Phys. Eng. Sci. 454(1971), 903–
995 (1998)
8. Lin, L., Wang, Y., Zhou, H.: Iterative filtering as an alternative algorithm for empirical mode
decomposition. Adv. Adapt. Data Anal. 01(04), 543–560 (2009)
9. Xu, Z., Huang, B., Zhang, F.: Envelope approach based on special knots for empirical mode
decomposition. Electron. Lett. 45(9), 480–481 (2009)
10. Kaiser, J.F.: On a simple algorithm to calculate the ‘energy’ of a signal. In: International
Conference on Acoustics, Speech, and Signal Processing, vol. 1, pp. 381–338 (1990)
11. Maragos, P., Kaiser, J.F., Quatieri, T.F.: On amplitude and frequency demodulation using energy
operators. IEEE Trans. Signal Process. 41(4), 1532–1550 (1993)
12. Horowitz, S.H., Phadke, A.G.: Power System Relaying, p. 56. Research Studies Press, Taunton,
UK (1992)
Improvement of Power System Stability
Using Optimized Modified Robust
Controller Based STATCOM

Niranjan Nayak

Abstract Power system electromechanical oscillation is a significant challenge in

the present power system scenario. The quick-acting of a static synchronous com-
pensator (STATCOM) which is an efficient reactive power control device capable to
increase the system stability. PI controller, in general, plays a key role in stability
analysis. But the PI controller is not capable enough to settle the system quickly and
reducing overshoots in different operating conditions. Thus, in this study, a modified
sliding mode controller, known as a terminal sliding mode controller(TMC), using
harmony search algorithm, is designed to eliminate the oscillations during various
operating conditions of a power system quickly with less overshoot. The performance
of the proposed controller is compared with PI controller and justified its superiority.

Keywords STATCOM · Power system stability · Harmony search. · Sliding mode

control · Terminal sliding mode controller (TMC)

1 Introduction

In the present situation, Flexible AC Transmission System (FACTs) family is a key

factor in supplying stable power. Static compensator is an important shunt com-
pensating device which belongs to FACTs family and regulates the voltage at the
point of common coupling (PCC) by injecting or absorbing reactive power. This
device supplies reactive power (capacitive) at low line voltage and it absorbs the
VAR (inductive) at high line voltage [1]. The STATCOM becomes more efficient
by changing magnitude and phase of bus voltage and line reactance. In addition to
normal control of the power system, the FACTS also improves stability itself dur-
ing power swings [2]. The active power in FACTS can be controlled to improve

N. Nayak (B)
ITER, Siksha ‘O’ Anusandhan (Deemed to be University), Bhubaneswar, Odisha 751030, India
e-mail: [email protected]

© Springer Nature Singapore Pte Ltd. 2020 285

G. Pradhan et al. (eds.), Advances in Electrical Control and Signal Systems,
Lecture Notes in Electrical Engineering 665,
https://doi.org/10.1007/978-981-15-5262-5_20
286 N. Nayak

the stability caused by low-frequency oscillation. Newly, some FACTS are installed
and investigated in practice around the globe. STATCOM is a primary member of
shunt connected FACTS family. SVC, instant exchange of VAR with the system, it
can pick up fluctuation stability better than other shunt connected FACTS like SVR,
STATCOM, etc. [3]. The STATCOM works on the principle of ac controllable volt-
age produced by the transformer-leakage reactance of the voltage source inverter.
The differential voltage in the reactance is responsible for power exchange. In the
extensive literature study, it is seen that the required controllers have been designed
to regulate power flow and voltage fluctuations [4, 5]. In [6] a STATCOM is installed
and analyzed its damping effect by a robust controller technique. Additionally seems
that no attempt is taken to recognize the most appropriate STATCOM control struc-
ture, to get a robust controller. Fuzzy controllers used for STATCOM parameters
to improve system stability [7]. The design of control scheme for a large system
is really challenging. Various optimization methods like PSO, GA, and HBMO are
applied to design optimal control for stability improvement. Genetic Algorithm is
employed effectively to solve complex optimization problems, but its deficiencies in
its performance are highlighted in the literature [8]. To surmount these defects, the
new control schemes are proposed to pick up the stability of the power system [9,
10], which has a strong switching transient effect.
In this study, a new robust controller known as the terminal sliding mode controller
(TMC) is designed to eliminate the damping of a power system whose parameters
are optimized by the harmony search algorithm. The efficiency and robustness of
new controller is investigated through MATLAB/SIMULINK. The result analysis
depicts that the suggested controller performs better than PI and traditional sliding
mode controller.
The rest of the paper is organized as follows. The second section represents the
mathematical model of the power system with STATCOM. Design of various con-
trollers is discussed in Section 3. The harmony search algorithm is explained in
fragment Sect. 4. The results are presented in Sect. 5. The concluding remarks are
described in Sect. 6.

2 Mathematical Modeling of Statcom

In this study, a single machine infinite bus power system, integrated with a STATCOM
is investigated for improving the dynamic stability with the help of an optimized
robust controller technique. The model is a three bus power system, in which the
STATCOM and a linear load are connected to bus-2.
In the model shown,
E s = Generator voltage.
Vb = Infinite bus voltage.
i d , i q = direct and quadrature axis generator current.
i bd , i bq = Infinite bus current.
Improvement of Power System Stability … 287

Fig. 1 STATCOM model under study

Pl , Q l = active and reactive power of the load.

i L = load current (Fig. 1).

2.1 Generator

The dynamic equations of synchronous generators are expressed by the following

equations.

dδ
= ω (1)
dt
dω 1 
 
 
= Pm − E q i q − xq − xd i d i q (2)
dt J

1    

dE q  
=  E f d − E q − x d − x d id (3)
dt Td0

where ω = δ − δ0 , ω = ω − ω0 , δ, are the machine parameters (Fig. 2).

2.2 Statcom

The essential structure of six-pulse STATCOM, interconnected with a single machine

infinite bus power system with passive loads is shown in Fig. 3.
where R p and L p = Converter transformer parameters. Rdc = Dc line resistance.
V p = Inverter output voltage which is regulated by inverter angle.
  transmits due to fixed Vdc .
No real power
If |Vs | > V p  the VSI receives VAR.
288 N. Nayak

Fig. 2 The generator with AVR and PSS

Fig. 3 Structure of STATCOM

 
Whereas, if |Vs | < V p , it injects VAR.

2.3 Statcom Modeling

The dynamic equation of STATCOM is represented by.

 
d
Rp + L p i p = Vs − V p (4)
dt

where
T
i p = ia ib ic (5)

Equation (4) can be remodeled as

 
d T
Rp + L p ia ib ic = Vs − V p (6)
dt
Improvement of Power System Stability … 289

where
T
Vs = Vsa Vsb Vsc

T
V p = V pa V pb V pc

Substituting in Eq. (6)

 
d T T T
Rp + L p ia ib ic = Vsa Vsb Vsc − V pa V pb V pc (7)
dt
⎡ ⎤ ⎡ ⎤
  Vsa V pa
d
= ⎣ Vsb ⎦ − ⎣ V pb ⎦
T
Rp + L p ia ib ic (8)
dt
Vsc V pc

where
⎡ ⎤ ⎡ ⎤
Rp 0 0 Lp 0 0
Rp = ⎣ 0 Rp 0 ⎦ L p = ⎣ 0 L p 0 ⎦ (9)
0 0 Rp 0 0 Lp

Substituting
⎛⎡ ⎤ ⎡ ⎤ ⎞ ⎡ ⎤ ⎡ ⎤
Rp 0 0 Lp 0 0 Vsa V pa
d
⎝⎣ 0 R p 0 ⎦ + ⎣ 0 L p 0 ⎦ ⎠ i a i b i c = ⎣ Vsb ⎦ − ⎣ V pb ⎦
T
(10)
dt
0 0 Rp 0 0 Lp Vsc V pc
⎛⎡ ⎤ ⎡ ⎤⎞⎡ dia ⎤ ⎡ ⎤
Rp 0 0 Lp 0 0 dt Vsa − V pa
⎝⎣ 0 R p 0 ⎦ + ⎣ 0 L p 0 ⎦⎠⎣ dib ⎦ = ⎣ Vsb − V pb ⎦ (11)
dt
0 0 Rp 0 0 Lp di c
dt
Vsc − V pc

No zero sequence components are taken by the power system. The Park’s
transformation is used for the conversion of currents and voltages into d-q frame.
⎡ ⎤
cos θ cos θ − 2π cos θ + 2π
2⎢ 3 3
⎥
T = ⎣ − sin θ − sin θ − 2π3
− sin θ + 2π3 ⎦ (12)
3 √1 √1 √1
2 2 2

Thus, the transformed dynamic equations are

di pd Rp 1
= − i pd + ωi pq + Vsd − V pd (13)
dt Lp Lp
di pq Rp 1
= − i pq − ωi pd + Vsq − V pq (14)
dt Lp Lp
290 N. Nayak

ω is the generator speed and VSI voltage components are neglected.

The modulation index (m) and phase angle (φ) are expressed as

2
V pd + V pq
2  
−1 V pq
m= φ = tan (15)
kVdc V pd

where k is a constant.
For effective dc voltage control, the power balance equation is expressed as

3
p= Vsd i pd + Vsq i pq − i 2pd + i 2pq R p
2
dVdc V2
= C Vdc + dc (16)
dt Rdc

Hence,
 
dVdc V i
3 sd pd + V i
sq pq − i 2
pd + i 2
pq R p Vdc
= − (17)
dt 2 C Vdc C Rdc

Equation (17) gives dc voltage dynamics.

The mathematical model of STACOM can be derived from Eq. (14), (15), and
(17) and summarized in Eq. (7).
⎡ ⎤ ⎡ ⎤
⎡ ⎤ R
− L pp i pd + ωi pq 1
i  0 
d ⎣ pd ⎦ ⎢ R
− L pp i pq − ωi pd
⎥ ⎢
⎥+⎣ 0 1 ⎥
Lp
Vsd − V pd
i pq = ⎢
⎣   ⎦ Lp ⎦ V − V
dt V i +Vsq i pq − i 2pd +i 2pq R p sq pq
Vdc 3 sd pd 0 0
2 C Vdc
− Vdc
C Rdc
(18)

3 Controller Design

Under various disturbances, the power system undergoes unstable conditions. Differ-
ent types of controllers are normally designed to return back the system into original
condition. In this work PI and a new type of sliding mode controller are used to
regain the system after the fault is cleared. There are some parameters of the con-
trollers which are assigned randomly on which the performance of the controllers
depends. Hence harmony search algorithm is used here to select the proper value of
the controller parameters.
Improvement of Power System Stability … 291

Fig. 4 Block diagram of PI

controller

3.1 PI Controller

It is a common practice that application of PI controller is broadly used power

system issues in different operating conditions. This is due to its simple structure,
easy operation, and robust nature. Proper selection of PI controller is a tough task,
and the same is selected in hit and trial method. The PI controllers are presented by

t
y(t) = k p er(t) + ki er(t)dt (19)
0

The block diagram of the PI controller is shown in Fig. 4.

For minimization of steady-state error, gradation of power system, and elimination
of forced oscillation PI controller is extensively used in power system control. The
errors in PI control are given as following.

er = Ipref − I p (20)

u, is the controller input.

3.2 Sliding Mode Controller for Statcom

Here the STATCOM output voltage error is considered as controller output.

er = Ipref − I p

Let us assume σ = er.

For a balanced three-phase ac network a vector control strategy could be adopted
by assuming the d and q-axis voltage components with vsd and vsq . vsd = vs , vsq = 0
and. With these simplifications, the STATCOM model can be expressed a nonlinear
form as
·
x1 = f 1 + g1 u 1 (21)
292 N. Nayak

where
Rp (vsd − vsq )
f1 = − I pd + ωs L p + , (22)
Lp Lp
1
g1 = (23)
Lp

Using an integral form of terminal sliding mode controller (TSMC), the sliding
surface cannot provide finite-time control. On the other hand, the TSMC can provide
convergence in finite time to the steady-state condition.
To develop the TSMC for taking care of STATCOM dynamics, the system
equations can be written as an affine nonlinear system as given below.

t
p
σ1 = er + β1 er q (τ )dτ and σ1 = 0 (24)
0

and its derivative is

p
σ̇1 = eṙ + β1 er q (25)

and β 1 > 0 is a design constant and “p” and “q” are positive and
p
1< <2
q

Solving the error dynamics of the equation (), the time required to converge to
the equilibrium condition is found as
q
|e1 (0)|1− p
τ=   (26)
β1 1 − qp
1 p

u= − f 2 − β1 er q − ks 1 sign(σ1 ) (27)
g1

Further, the dynamic function can be expressed in form of uncertainty

f 2 = f 2 +  f 2 , where | f 2 | ≤ F2 (28)

g1 = ĝ1 + g1 where |g1 | ≤ G 1 (29)

where F2 and G 1 represents the upper bound of the dynamic function f 2 and g1
respectively.
Improvement of Power System Stability … 293

The derivative of the sliding surface can be derived by substituting Eq. (25) in
Eq. (24) as follows:
 
g1  ˆ p
q  ks1 g1
σ̇1 = | f 2 | − f 2 + β1  − sign(σ1 ) (30)
g1  ĝ1

where F2 ∈ R n
The above control scheme converges in a fixed time so it is called the terminal
sliding mode controller.

4 Harmony Search Algorithm

In the fundamental Harmonic Search optimization, each result is named as har-

mony and is presented by real vector of “n” dimension. The harmony populations
are arbitrarily selected is stored in harmony memory (HM). By using a selection
reinitialization rule called pitch rule, from the HM a new resolution is produced.
Then
the new candidate is compared with the worst candidate, the HM is updated.
Consecutively, these phases are discussed below.

4.1 Initialization

The global optimization problem can be formulated as follows.

min h( p) so that p j ∈ [paramin
k , parak ] Where j = 1, 2, 3, . . . , n and h(x) is
max

the objective function p = [ p1 , p2 , . . . , pn ] is the set of “n” design parameter. Here

k , parak
paramin max
are the upper and lower bounds of the variables p k respectively.
The HS parameters and harmony memory are the same.. the number of solutions
contained in harmony memory. The parameters like harmony memory consideration
rate (HMCR), peach adjusting rate (PAR), Bandwidth (BW), and the improvisation
number or the fitness function evaluation may be adjusted to provide a stopping
criterion.

4.1.1 Initialization of HM

The HM is consists of HMS solution vectors. Let the randomly generated kth
harmonic solution is given by

pk = [ p1k , p2k , . . . , pnk ] (31)

Within the boundary paramin

k and paramax
k . Then the HM matrix is formulated as
294 N. Nayak

⎡ ⎤
p11 p21 . . . pn1
⎢ p1 p2 . . . pn2 ⎥
HM = ⎢
⎣ .
2 2 ⎥ (32)
. . . ⎦
p1Hms p2Hms . . . pnHms

4.1.2 Improvisation of New Harmony

A fresh harmony may be improvised by selecting new values of each harmony vector
such as

p new = [ p1new , p2new , . . . , pnnew ] (33)

The improvisation includes methods like memory selection, pitch adjustment, and
random selection.
Step-1: a random number r1 set is selected from [0,1].
Step-2: if r1 < HMCR, then the decision variable p new j is selected from the
memory consideration. Or else p new
j is selected randomly is called reinitialization.
Step-3: every decision variable p new
j will undergo a pitch change with a probability
of PAR.
Step-4: The pitch adjustment is achieved by the equation

x new
j = x new
j ± r3 × BW. (34)

where r3 is a random between [0,1]

4.2 Updating of HM

After the selection of updated solution vector p new

j is selected, the HM will be updated
by a vector between the new value p and most evil solution p worst . Depending on
new

the fitness value p new may be replaced by p worst .

4.2.1 HS Algorithm

Step-1: The parameters like HMS,HMCR,PAR,BW,NI and n are assigned to initial

values.
Step-2: The HM is initialized and objective function of the harmony vector is
calculated.
Step-3: Improve the HM with p new
Improvement of Power System Stability … 295

for (k = 0; k < 1, k + +)
if r < HMCR,then
pknew = pka a ∈ (1, 2, 3 . . . , hms)
if r2 < P A R, then pknew ± r3 × bw, (r1 , r2 , r3 ∈ [0, 1])
end if
else
pknew = paramin
k + r × (paramax
k − paramin
k ), r ∈ [0, 1]
end if
end for

Step-4: update the HM as

p worst = p new if h( p new ) < h( p worst ).
Step.5: If NI is computed, return the best solution
Vector p best in HM, or else go back to step.

4.2.2 Optimized Sliding Mode Controller

PI gains k p , ki and sliding mode controller constants ks1 , β1 , p and q are selected
randomly which affects the performance of the controllers. Thus harmony search
algorithm is applied to tune these parameters, which improves system stability.

4.2.3 Objective Function

The design of the optimized SMC based SSSC damping controller is formulated as a
single objective constrained optimization problem. The objective function considered
in this paper is:

tsim
J = (er)dt
0

where t sim represents computational time. For a set of parameters, the objective here
is to reduce the J value subject to the following conditions:

γmin < γ < γmax

K min < K < K max

Optimized gains of PI and Sliding Mode Controller are given in the Table 1.
296 N. Nayak

Table 1 Optimized controller gains

S. No. PI gains Optimized PI gains SMC Parameters Optimized SMC gains
1 k p = 0.2 0.28 ks1 131
2 ki = 0.01 0.016 β1 121
p 3.241
q 2.111

5 Simulation Results

A STATCOM based power system, is simulated in different operating conditions like

three-phase fault, increase in load, etc. with PI, modified sliding mode, and optimized
modified sliding mode controller and the results are depicted in the following figures.

5.1 STATCOM Parameters

R p = 0.04 pu, X p = 0.1 pu, Rdc = 150, C = 5000 µ F,

L s = 0.0318 H, capacitive suscitive susceptance, gc = 0.0067 pu
capacitive susceptance = bc = 0.0157 pu

5.2 AVR and PSS Parameters

Te = 0.1H = 12, K e = 50, K p = 5, K i = 12, K s = 0.01,

Ts = 0.2, Ts1 = 0.05, Ts2 = 0.0076, Ts3 = 0.276, Ts4 = 0.01;
E f dmax = 6, E f dmin = −6, Umax = 0.15, Umin = −0.15,

5.3 Synchronous Machine

X d = 1.2, X q = 0.6, X dd = 0.1, Td0 = 2.314, D = 0.0075.

Improvement of Power System Stability … 297

5.4 Controller Gains

k p = 0.2, ki = 0.01. K = 130, γ = 1130

Here the system is connected to a load of P = 1.25 pu and Q = 0.8 pu which

heavily affects the system under various disturbances. At this operating conditions,
PI controller gains are given as in Table 1, given.

Case-1: LLLG Fault at bus-2

A three-phase fault is initiated at the infinite bus of the system, which creates dynamic
instability in the power system. The fault continued for 0.2 s and then the circuit
breaker is reclosed. The action of PI controller, SMC optimized sliding mode con-
troller are shown in Fig. 5. The performance of the harmony search optimization based
modified sliding mode controller is superior than classical SMC and PI controller in
regulating the overshoot and settling time.

Case-2: Increase in load

The load connected to the system is increased by 30% for 0.2 s and the system is
simulated. It observed the PI controller needs more than 4 s to achieve steady state.
The SMC settles the system at approximately 3 s whereas optimized sliding mode

4
PI
2
dw(radian/sec)

2 SMC
delta(radian)

OSMC
1.5 0
PI
SMC -2
1
OSMC
-4
0 1 2 3 4 0 1 2 3 4
Time(second) Time(seconds)
2
PI PI
2
SMC SMC
1.5 OSMC 1.5 OSMC
Qe(pu)
Pe(pu)

1
1
0.5

0 0.5
0 1 2 3 4 0 1 2 3 4
Time (seconds) Time (seconds)

PI PI
SMC SMC
V1(pu)

OSMC 0.8 OSMC

V2(pu)

0.6
0.8
0.4
0 1 2 3 4 5 0 1 2 3 4 5
Time (seconds) Time (seconds)

Fig. 5 Response of the proposed controller at three phase fault at bus-2

298 N. Nayak

1
PI PI
SMC 0.5 SMC
1.6 OSMC

dw1(radian)
OSMC
delta(radian)

0
1.5
-0.5
1.4
-1
0 1 2 3 4 0 1 2 3 4
Time (seconds) Time (seconds)
1
PI PI
1.6 SMC SMC
OSMC 0.9 OSMC
Pe(pu)

1.5

Qe(pu)
1.4 0.8

1.3 0.7
0 1 2 3 4 0 1 2 3 4
Time(seconds) Time (seconds)
PI PI
0.88 SMC 1.02 SMC
OSMC OSMC
V2(pu)
1
V1(pu)

0.86

0.84 0.98

0 1 2 3 4 0 1 2 3 4
Time (seconds) Time (seconds)

Fig. 6 Response of controller under active power change

controller settles the system before 1 s. The HS-based modified SMC controller
performs better than only SMC and PI controller (Fig. 6).

Case-3: Three-phase Fault in Transmission line near generator bus

The generator power is increased by 20% and the model is simulated for 0.2 s.
From Fig. 7, it is seen that the recital of the STATCOM based power system is very
vigorous. The PI controller is not capable to settle the system in more than 4 s.
The SMC brings the system into stable zone before 3 s approximately. In the same
operating condition, the optimized SMC is more efficient than other controllers.

6 Conclusion

This paper presents an optimized and modified sliding mode controller for a single
machine infinite bus power systems integrated with STATCOM. The chosen control
inputs are the STATCOM current error at PCC. The system is simulated under heavy
load with various disturbances. The performance of optimized sliding mode con-
troller is tested under different operating conditions like three-phase fault, increase
of generator output, and system load increase. The proposed controller produces less
overshoot and settling time than the other conventional controllers.
Improvement of Power System Stability … 299

1.7 PI 1
PI
SMC SMC
delta(radian)

1.6 0.5
OSMC

dw(radian/sec)
OSMC
0
1.5
-0.5
1.4
-1
0 1 2 3 4 0 1 2 3 4
Time(seconds) Time ( seconds)

PI 1.1
PI
1.6 SMC SMC
1
1.5 OSMC OSMC
Pe1(pu)

Qe1(pu)
0.9
1.4
1.3 0.8
1.2 0.7
0 1 2 3 4 0 1 2 3 4
Time (seconds) Time (seconds)

0.92 PI PI
SMC 1.05 SMC
0.9
OSMC OSMC
V1(pu)

0.88 V2(pu)
0.86 1

0.84
0.82 0.95
0 1 2 3 4 0 1 2 3 4
Time (seconds) Time (seconds)

Fig. 7 Response of controller under transmission line fault

References

1. Lehn, P.W., Iravani, M.R.: Experimental evaluation of STATCOM closed loop dynamics. IEEE
Trans. Power Deliv. 13(4), 1378–1384 (1998)
2. Sahoo, N.C., et al.: Multivariable nonlinear control of STATCOM for synchronous generator
stabilization. Int. J. Electr. Power Energy Syst. 26(1), 37–48 (2004)
3. Hingorani, N.G., Gyugyi, L.: Understanding FACTS: Concepts and Technology of Flexible AC
Transmission Systems. In: Mohamed El-Hawary (Ed.), vol. 1. IEEE press, New York (2000
4. Pereira, R.M.M., Ferreira, C.M.M., Barbosa, F.P.M.: Comparative Study of STATCOM and
SVC Performance on Dynamic Voltage Collapse of an Electric Power System with Wind
Generation
5. Sruthi, G., Naveen Kumar, G.: Comparison of series and shunt facts controllers for voltage
stability in wind power network. Int. J. Comput. Appl. (0975 – 8887) 125(11) (2015)
6. Hedayati, M., Mariun, N., Hizam, H., Bahari, S.: Design of robust controller for STATCOM
applied to large induction motor using normalized coprime factorization approach. Arab. J.
Sci. Eng. 38(10), 1–12 (2013)
7. Nayak, N., Routray, S.K., Rout, P.K.: Design of a novel fuzzy PI controller based on artificial
bee colony algorithm for VSC-HVDC transmission link. Int. Rev. Autom. Control (IREACO)
6(4), 450–463 (2013)
8. Safari, A., Ahmadian, A., Golkar, M.A.: Controller design of STATCOM for power system
stability improvement using honey bee mating optimization. J. Appl. Res. Technol. 11(1),
144–155 (2013)
9. Melin, P.E., Guzman, J.I., Hernandez, F.A., Baier, C.R., Muñoz, J.A., Espinoza, J.R., Espinosa,
E.E.: Analysis and control strategy for a current-source based D-STATCOM towards minimum
losses. Int. J. Electr. Power Energy Syst. 1(116), 105532 (2020)
10. Jia, K., Xuan, Z., Chen, J., Feng, T., Zhao, Q., Bi, T.: Transient switching performance of VSC
and the DC fault partitions. Int. J. Electr. Power Energy Syst. 1(116), 1–10 (2020)
Comparative Analysis of Different
Problem Formulation for Optimal
Capacitor Allocation in Distribution
Systems

Shivangi Upadhyay and Sachin Singh

Abstract In Radial Distribution Networks (RDNs), reactive power compensation is

provided commonly through the installation of capacitor banks. Proper allocation of
capacitors in the network can bring several advantages which include improvement
of feeder voltage and significant power loss reduction which will then provide a good
amount of savings in energy as well as cost reduction. The explained formulation is
known as the Optimal Capacitor Placement (OCP) problem. In this paper, an attempt
has been made to study and analyze different OCP problem formulations. The power
membership function and voltage membership function of the fuzzy-based method
has been used for the optimal location identification and with help of available sizes
of the capacitor to minimize the annual cost of 9-bus RDN and voltage, improvement
has been done. MATLAB coding has been used to simulate the test cases and the
results obtained for different OCP problems have been compared.

Keywords Radial distribution networks (RDNs) · Optimal capacitor placement

(OCP) · Node voltage · Power loss

1 Introduction

Rapid rural electrification and ever-growing consumer demand are increasingly

putting a burden on distribution feeders which causes a reduction in the voltage
at load ends. Moreover, since most of the electrical loads are inductive, the overall
system power factor also decreases. This reduction in power factor leads to incre-
ment in power losses, voltage regulation problems, and violation of system capacity
constraints. Therefore, corrective measures are required to be adopted on time before
the system stability jeopardizes.
Through several methods, the easier one is the application of shunt capacitors to
improve the system power factor. The benefits of using capacitors are voltage boost
uniformly along the feeder and the heavy reduction of the system losses, which will

S. Upadhyay (B) · S. Singh

Department of Electrical Engineering, Institute of Engineering and Technology, Lucknow, India
e-mail: [email protected]

© Springer Nature Singapore Pte Ltd. 2020 301

G. Pradhan et al. (eds.), Advances in Electrical Control and Signal Systems,
Lecture Notes in Electrical Engineering 665,
https://doi.org/10.1007/978-981-15-5262-5_21
302 S. Upadhyay and S. Singh

lead to increased revenue and there will be no investment in new system facilities,
etc. The placement of capacitors can be done of several possible places, i.e., customer
load points, at distribution substation secondary/primary points, and along with the
feeder mains.
The capacitor placement at the customer load points provides maximum advan-
tages but also sometimes it may lead to the condition of resonance and the problems
of self-excitation in the induction motor, which will cause overheat and damage the
insulation of the winding. The practical utility point of view says that the installed
capacitors at customer ends, 50% of them were found disconnected from the service
mains and results in the loss of investment, because of this the placement of capacitor
along with the feeder mains is much effective instead of capacitors at the substations.
The feeder main is designed for certain load growth and it follows the growth
in the load factor always, which will cause the increment in the power and energy
cost. As the load growth surely influences the optimal choice of the capacitor, so the
growth factor consideration is necessary to achieve an overall economy.
In distribution systems, the problem of capacitor placement includes the determi-
nation of location, number and size of to be installed capacitors so that the operational
and total installation cost is kept a minimum of the system with the load variation
at the different hours of the day and each node voltage will be within its specified
limits.
OCP is a combinatorial optimization and the structure and size of RDNs tell its
complexity. If a distribution network consists of n buses, and q is the number of
available sizes of capacitor then (q + 1)n possible solutions to implement and solve
the OCP problem. Schmill’s [1] has proposed a procedure to search the capacitor
size and location along the feeder in a proposed model and his work has been the
beginning in this direction. An optimized formula to estimate the exact energy loss
reduction achieved by the capacitors is given by Cook [2]. Based on Cook’s formula, a
dynamic programming approach for the optimization of location, sizes, and numbers
of capacitors is suggested by Duran [3] with a fine number of theorems regarding
the allocation of capacitors. Grainger and Lee in [4, 5] defined the problems for
three aspects, i.e., location, sizing and switching time by treating one aspect out of
three as a variable and other two remaining aspects as fixed and for optimal sizing
and location of the capacitor on radial feeders they also developed an equal area
criterion. Many nonlinear optimizations [5–8] and heuristic techniques [9, 10] are
used to solve the combinatorial problem and techniques such as Genetic Algorithm
[12] and Simulated Annealing [11].
Goswami et al. has given an approximate method for capacitor placement based
on heuristic and greedy search technique is given in [13], the determination of the
number of probable locations of capacitor is done by the heuristic rules and for the
optimum size at the selected location determination a basic search technique is used.
A greedy search technique is designed using this search technique to determine the
optimal capacitor sizes and locations in a distribution network. A fuzzy-reasoning
optimization method, to minimize the power loss and installation cost within the
voltage limit constraints of the formulated problem is given by Su and Tsai in [14].
Comparative Analysis of Different Problem … 303

The fuzzy method makes the operation less computational. Under the practical load’s
condition, the solution is optimized. Some assumptions are as follows:
1. The feeders are balanced.
2. The reactive and real power on the nodes of the feeder is known.
3. Harmonics’ currents are ignored. But none of the above-mentioned approaches
considered the effect of the growth factors like load growth except [8].
Load growth has a noticeable influence on the optimal capacitor allocation. In
this paper, a study has been carried out to compare the two different OCP problem
formulations from [13, 14], with the consideration of percentage loading and fixed
installation cost of capacitors and the other one is without them. MATLAB coding
has been done by using a fuzzy-reasoning method [14] to improve the voltage profile
and power loss reduction for a given load structure.
The first step of the procedure is to perform a power flow to estimate the bus volt-
ages. Later then the membership power loss and voltage functions are specified. The
optimum locations and sizes of the capacitor with the lowest cost within the voltage
limits are identified and determined. Finally, after the reactive power compensation,
power flow is performed again to calculate the improved bus voltages. To show the
effectiveness of the mentioned method, an example of a 9-bus RDN is considered
from [13, 14]. The computed results suggest the proposed method is effective in
solving the OCP problem with accuracy and better robustness.

2 Mathematical Modeling for Optimal Capacitor

Placement Problem

In RDNs, the OCP problem is a combinative problem in which the finding of the
number, location, and size of the capacitors to be allocated. To solve the OCP for-
mulation many various objective functions have been acknowledged. The primary
objective is to compensate for the reactive power in the installed feeder while min-
imizing the total system cost. This is done by keeping the voltage profile within its
permissible limits and reducing the energy and power loss. In this work, two different
objective function has been compared: the first function is introduced in [14], which
is stated as:


n
S = K p · Ploss + k cj . (1)
j=1

where S denotes the total system cost ($) for an n—buses RDN. K p is energy cost
per unit ($/kW), Q cj is the size of the installed capacitors at bus ‘j’, whereas k cj is
its corresponding cost per kVAR ($/kVAR) and Ploss is the network’s total power
loss (kW). Equation 1 neglects the consideration of the operating, installation, and
maintenance costs as well as the load level changes at the different hours of the day.
304 S. Upadhyay and S. Singh

The second objective function in our study is derived from [13], which is stated
as follows:


m 
n
S = K p Ploss [PL(i)] T (i) +
2
k cj Q cj + kc (2)
i=1 j=1

When compared to Eq. 1, the installation cost of the capacitor, as well as the
change in load levels at the different hours of the day for the calculation of network
total cost, has also been considered. K p is energy cost per unit ($/kW), Q cj is the rating
of the allocated capacitors at bus ‘j’, while k cj is its cost per kVAR ($/kVAR) and
Ploss is the complete power loss (kW) of the test network. T (k) is the time duration
of different load levels (where k = 1, 2, …, NL) with percentage loading PL(k) in
time duration T (i).
A definite number of standard ratings of capacitors which are integer multiple
of the smallest size Q c0 exist is assumed and cost per kVAR varies from one size to
another. The capacitor size will usually limit to:

Q maxc = L Q c0 (3)

where L is an integer.

2.1 Power Loss Calculation

For the calculation of the complete power loss as stated in the above objective func-
tions, the calculation method developed in [14] is applied. A set of simplified line
flow equations with a significant amount of iterations is proposed to calculate the
exact line flow. A balanced three-phase distribution system’s single line diagram is
given in Fig. 1.
The power loss of the line connecting kth and (k + 1)th buses while considering
the above equations is calculated as follows:

Fig. 1 Schematic diagram of a typical radial distribution network conspired by n nodes

Comparative Analysis of Different Problem … 305

Pk2 + Q 2k
Ploss(k,k+1) = Rk,k+1 · (4)
|Vk |2

where Pk and Qk are the real and reactive line power respectively flowing from bus
‘k’, while PLk and QLk indicate the real and reactive load power respectively, and
the line reactance and resistance between kth and (k + 1)th buses is X k,k+1 and Rk,k+1
respectively. |V k | is the kth bus voltage magnitude.
By adding the power loss of each line, the total power loss of all the n buses can
be calculated as


n
k,k+1
Ploss = Ploss (5)
k=0

2.2 Voltage Constraints

For the safety measures, the voltage at each bus should reside within its permissible
limits. Therefore, the voltage constraint for each bus in a radial distribution system
is given as:

Vmin ≤ |Vk | ≤ Vmax i = 1, 2, 3, . . . , n. (6)

where V min = minimum bus voltage limit and V max = maximum bus voltage limit.

3 Fuzzy Set Operators

3.1 The AND Operator

Let two fuzzy sets A and B with uA (x) and uB (x), respectively, as their membership
function and the universal set is X. The membership function of intersection (AND)
is defined by:

u A∩B (x) = min(u A (x), u B (x)) x ∈ X (7)

306 S. Upadhyay and S. Singh

Fig. 2 Bus voltage

membership function

3.2 The OR Operator

Let two fuzzy sets A and B with uA (x) and uB (x), respectively, as their membership
function and the universal set is X. The membership function of union (OR) is defined
by:

u A∪B (x) = max(u A (x), u B (x)) x ∈ X (8)

3.3 Membership Function of the Bus Voltage

A membership function (U v ) of voltage magnitude V as shown in Fig. 2.

It is defined as a high value is given to the bus with low voltage deviation. Similarly,
a low value is given to the bus voltage with a high deviation. The given equation shows
the membership function:

1
Uv (i) =   ; i = 1, 2, 3, . . . , n. (9)
vr −10
1+ Vmax −Vmin

3.4 Membership Function of the Power Loss

A membership function (U P ) of power loss Ploss is shown in Fig. 3.

It is defined as a low value is given to the high loss branch. Similarly, a high value
is given to the low loss branch. The given equation shows the membership function:
Comparative Analysis of Different Problem … 307

Fig. 3 Power loss

membership function

1 i,i+1
U p (i) = − · Ploss + 1; i = 1, 2, . . . , n; Ploss ≤ C p (10)
cp

U p (i) = 0; i = 1, 2, . . . ., n; C p < Ploss (11)

4 Computational Procedure

For the computational process, the first step is to perform the load flow to estimate
the bus voltages and their corresponding membership function values as well as the
section losses. Later, then AND operator is applied on the U v and U p ’s membership
function to estimate the optimum bus with the lowest membership function for the
compensating capacitor installation.
Determining the optimum ith bus to install capacitor is done by performing the
following operations:
 
Ul (i) = min Uv (i), U p (i) i = 1, 2, 3, . . . , n. (12)

and

Ul (i∗) = min{Ul (1), Ul (2), . . . , Ul (n)} (13)

The compensating reactive power which provides the lowest system cost within
the permissible voltage constraints is considered to be the optimum size. The above
process is continued until no power loss can be reduced and no bus voltage is violated
further. The procedure is described in the flowchart as shown in Fig. 4.
308 S. Upadhyay and S. Singh

Read System Data

1. Perform a Load flow study to calcu-

late bus voltages and power losses.
2. Find the membership functions.

Determining the optimum bus ith for

installing capacitor by performing the
following operations:
Ul (i)=min{Uv (i) ,Up(i)} i=1,2,3…..,n.
And, Ul (i*)=min{Ul (1),Ul (2),…..Ul (n)}

Determine and install the optimum size

of reactive compensation.

Perform load flow again and find the

new bus voltage and power losses.

Are Voltage Remove the bus

losses violation? capacitor installed.
con-

Find
optimal
End
capacitor
location.

Fig. 4 Flow chart

5 Numerical Example and Computational Results

In this demonstrative example, a nine-node radial distribution feeder network [13]

shown in Fig. 5, is used as a test case and the line and load data of the network
is mentioned in Table 3 [13] and the load duration data for test system is listed in
Tables 1 [13] and 2 [13].
Total no. of buses = 9.
Base voltage = 23 kV.
Comparative Analysis of Different Problem … 309

Fig. 5 Nine bus radial network

Table 1 Parameters required for both formulation

Parameters Case 1 Case 2
Kp 168$/kW 0.06$/kW
Kc – 1000$

Table 2 Load duration data for the test system

Load level 1 2 3
Per unit load 0.3 0.6 1.1
Duration(h) 1000 6760 1000

Table 3 Line and load data for the test system

From bus To bus Impedance Load connected at ‘to bus’
R () X () KW KVAR
0 1 0.1233 0.4127 1840 460
1 2 0.0140 0.6051 980 340
2 3 0.7463 12,050 1790 446
3 4 0.6984 0.6084 1598 1840
4 5 1.9831 1.7276 1610 600
5 6 0.9053 0.7886 780 110
6 7 2.0552 1.1640 1150 60
7 8 4.7953 2.7160 980 130
8 9 5.3434 3.0264 1640 200

Total reactive load = 4186 kVAR.

V min = 0.9 p.u.
V max = 1.1 p.u.
The suitable sizes of available capacitors and the yearly cost of fixed capacitors
[14] are listed in Tables 4 and 5. The required simulation parameters are given in
Table 1.
310 S. Upadhyay and S. Singh

Table 4 Available capacitor sizes and $/kVAR

Size (kVAR) 150 300 450 600 900 1200
Cost ($/year) 750 975 1140 1320 1650 2040

Table 5 Commercial
Capacitor index (j) Capacitor sizes Cost per unit of
capacitor sizes (kVAR) and
(Qc ) kVAR reactive power
their corresponding cost per
($/kVAR)
unit ($/kVAR)
0 0 0
1 150 0.500
2 300 0.350
3 450 0.253
4 600 0.220
5 750 0.276
6 900 0.183
7 1050 0.228
8 1200 0.170
9 1350 0.207
10 1500 0.201
11 1650 0.193
12 1800 0.187
13 1950 0.211
14 2100 0.176
15 2250 0.197
16 2400 0.170
17 2550 0.189
18 2700 0.187
19 2850 0.183
20 3000 0.180
21 3150 0.195
22 3300 0.174
23 3450 0.188
24 3600 0.170
25 3750 0.183
26 3900 0.182
27 4050 0.179
Comparative Analysis of Different Problem … 311

Table 6 Annual cost ($) and

Total power losses (KW) Annual cost ($)
total losses (KW) after
reactive power compensation Case 1 Case 2
696.7716 119,420 171,943

Table 7 Voltage profile and capacitor allocation

Node number Uncompensated voltage Reactive power Compensated voltage
(p.u.) allocated (kVAR) (p.u.)
1 0.992901 0 0.997162
2 0.987378 0 0.997890
3 0.963409 1050 0.986671
4 0.948017 1050 0.976671
5 0.917171 1950 0.957769
6 0.907169 0 0.950215
7 0.888957 0 0.935827
8 0.858695 0 0.914346
9 0.837505 900 0.902869

The annual cost and total loss after compensation through the proposed method
has been mentioned in Table 6 for both the cases and the voltage profile and optimal
capacitor allocation is given in Table 7.
Figure 6 given below shows the voltage profile before compensation, i.e., data
1 and voltage profile after compensation, i.e., data 2. The X-axis is the number of
nodes/bus and the Y-axis is the node voltage before and after the capacitor placement
method. From the plot we can see the improvement in voltage profile, as well as the
reduction in power loss, can also be seen as before compensation it is 783.77 kW and
after compensation it is 696.77 kW. Therefore from the three sets of locations for
capacitor placement set 3, 4, 5, 9 with a capacitor of sizes 1050 kVAR, 1050 kVAR,
1950 kVAR, and 900 kVAR, respectively, will provide a V min = 0.902 p.u.

6 Conclusion

This paper proposed a study to compare the two different OCP problem formulations
using a fuzzy-reasoning technique to solve the capacitor placement problem in the
radial distribution system. The used method search for the most effective buses
to install the capacitors of optimum reactive power, so that maximum annual cost
saving can be achieved. Further research to incorporate a fuzzy load model into the
system model using switched capacitors for compensation has been exploring and
the implication of the second equation considering the load growth effect can be
done for different OCP problem-solving techniques.
312 S. Upadhyay and S. Singh

Fig. 6 Voltage profile comparison

References

1. Schmill, J.V.: Optimum size and location of shunt capacitors on distribution feeders. IEEE
Trans. Power Appar. Syst. 84(9), 825–832 (1965)
2. Cook, R.F.: Calculating loss reduction afforded by shunt capacitor application. IEEE Trans.
Power Appar. Syst. PAS-83, 1227–1230 (1964)
3. Dura, H.: Optimum number, location, and size of shunt capacitors in radial distribution feeders
a dynamic programming approach. IEEE Trans. Power Appar. Syst. PAS-87(9), 1769–1774
(1968)
4. Grainger, J.J., Lee, S.H.: Capacity release by shunt capacitor placement on distribution feeders:
a new voltage-dependent model. IEEE Trans. Power Appar. Syst. PAS-101(5), 1236–1244
(1982)
5. Civanlar, S., Grainger, J.J.: Volt/var control on distribution systems with lateral branches using
shunt capacitors and voltage regulators: part I, part II, part III. IEEE Trans. Power Appar. Syst.,
pp. 3278–3297 (1985)
6. Baran, M.E., Wu, F.F.: Optimal capacitor placement on radial distribution systems. IEEE Trans.
Power Deliv., pp. 725–734 (1990)
7. Baran, M.E., Wu, F.F.: Optimal sizing of capacitors placed on a radial distribution system.
IEEE Trans. Power Deliv., pp. 735–743 (1990)
8. Ponnavaiko, M., Prakasha Rao, K.S.: Optimal choice of fixed and switched capacitors on
radial distribution feeders by the method of local variations. IEEE Trans. Power Appar. Syst.,
pp. 1607–1614 (1983)
9. Abdel-Salam, T.S., Chikhani, A.Y., Hackam, R.: A new technique for loss reduction using
compensating capacitors applied to distribution systems with varying load condition. IEEE
Trans. Power Deliv. 9(2), 819–827 (1994)
10. Salama, M.M., Chikhani, A.Y.: A simplified network approach to the VAR control problem
for radial distribution systems. IEEE Trans. Power Deliv. 8(3), 1529–1535 (1993)
Comparative Analysis of Different Problem … 313

11. Chiang, H.D., Wang, J.C., Cockings, O., Shin, H.D.: Optimal capacitor placements in distri-
bution systems. I. A new formulation and the overall problem. IEEE Trans. Power Deliv. 5(2),
634–642 (1990)
12. Sundhararajan, S., Pahwa, A.: Optimal selection of capacitors for radial distribution systems
using a genetic algorithm. IEEE Trans. Power Syst. 9(3), 1499–1507 (1994)
13. Goswami, S.K., Ghose, T., Basu, S.K.: An approximate method for capacitor placement in
distribution system using heuristics and greedy search technique. Elsevier Science (1999)
14. Su, C.-T., Tsai, C.-C.: A new fuzzy-reasoning approach to optimum capacitor allocation for pri-
mary distribution systems. In: Proceedings of the IEEE International Conference on Industrial
Technology, pp. 237–241 (1996)
Self-tuned PI Controller Based Hybrid
Shunt Active Power Filter for Power
Quality Enhancement

Alok Kumar Mishra, Prakash Kumar Ray, Akshaya Kumar Patra,

Ranjan Kumar Mallick, and Soumya Ranjan Das

Abstract The aim of this paper is to design a Self-Tuned Proportional and Integral
Controller (STPIC) for Hybrid Shunt Active Power Filter (HSAPF) used for reactive
power and harmonic compensation. In this control strategy, the Conventional Propor-
tional and Integral Controller (CPIC) is re-formulated with an automated self-tuned
approach to improve the control performance. The STPIC is a novel approach in
which, according to the error signal the gains vary vigorously. Comparing Passive
Filter with Active Filter the former is being bulky and design is complex and the
later is not cost-effective hence a hybrid structure of Shunt Active and Shunt Passive
filter is considered. MATLAB/SIMULINK models of HSAPF are developed to get
a low distorted and in phase source current waveform at the input side. The compen-
sation process for Shunt Active Power Filter (SAPF) is based only on source current
extraction an approach different from conventional methods such as (p-q) theory or
(id-iq) theory. The proposed system is tested at both steady-state, transient condition
and its performance is then estimated and compared in terms of various parameters
like Total Harmonic Distortion (THD), input power factor, percentage overshoot for
CPIC and STPIC used for reference current estimation.

A. K. Mishra (B) · A. K. Patra · R. K. Mallick

EEE Department, Siksha ‘O’ Anusandhan (Deemed to be University), Bhubaneswar, Odisha
751030, India
e-mail: [email protected]
A. K. Patra
e-mail: [email protected]
R. K. Mallick
e-mail: [email protected]
P. K. Ray
EE Department, College of Engineering and Technology, Bhubaneswar 751003, India
e-mail: [email protected]
S. R. Das
EE Department, IIIT, International Institute of Information Technology, Bhubaneswar 751003,
India
e-mail: [email protected]

© Springer Nature Singapore Pte Ltd. 2020 315

G. Pradhan et al. (eds.), Advances in Electrical Control and Signal Systems,
Lecture Notes in Electrical Engineering 665,
https://doi.org/10.1007/978-981-15-5262-5_22
316 A. K. Mishra et al.

Keywords CPIC · STPIC · Reactive power · Total harmonic distortion · Power

quality

1 Introduction

Nowadays, in almost all application power electronics based loads are being used
such as UPS, SMPS, ASD, etc. All of these power electronics based equipment draws
the current from the supply which is of nonlinear in nature and consists of a lot of
harmonic [1]. Due to this nonlinear current, the utility system has to supply a quite
a huge quantity of reactive volt-ampere from the supply as the current drawn from
the supply is not at all in phase with the supply voltage, again the Point of Common
Coupling (PCC) gets affected due to the same [2, 3]. To mitigate these harmonic
issues traditionally Passive Filters are being used [4]. Fixed compensation with res-
onance problem are the demerits of passive filters, which are being overcome by a
power electronic based device, called active filter [5]. Active Filters are known to be
costly for higher rating, hence a low-rated SAPF and a low-cost Shunt Passive Filter
(SPF) are formed called HSAPF for a three-phase three wire system [6]. To estimate
the reference current for the SAPF, many control techniques such as instantaneous
power theory (p-q) or synchronous reference frame theory (id-iq) are available in the
literature [7]. The present paper focused on a novel control method for estimation
of the reference current which does not require the load reactive volt-ampere [8].
A CPIC or STPIC is adopted to control the inverter dc-side capacitor voltage and
to extract the reference current. Various simulations are performed and results are
analyzed both for CPIC and STPIC of the proposed HSAPF system [9, 10].

2 Hybrid Shunt Active Power Filter System

The SAPF provides a better solution for harmonic mitigation. However, shunt active
filtering is costly. So, in many applications even these days Passive Filters are being
used for providing a cost-effective solution for harmonic mitigation. To construct the
passive filter resistors, inductors and capacitors are used which makes them massive.
Hence HSAPF is created by low-rated and low-cost SAPF and SPF respectively,
which in turn removes the demerits of pure passive and pure active filters while the
advantages of both remain intact. In this paper, an HSAPF is formed by the use of
a low-rated three-phase voltage source PWM converter and a shunt connected LC
passive filter, for designing the passive filters the reactive power requirement is taken
as 2 KVAR which is divided as 1000 VAR for single tuned 5th harmonic filter and
500 VAR each for single tuned 7th harmonic filter and 11th and 13th double-tuned
filter as shown in Fig. 1.
Self-tuned PI Controller Based Hybrid Shunt Active … 317

distortion.
is iL

is
Rs, Ls iL

Three Phase iaf ipf

Source
iaf Non Linear Load

5th single tuned, 7th single ipf

tuned, 11th & 13th double
Rc, Lc tuned shunt passive filter

S1 S3 S5

+
Vdc,act
Cdc

S4 S6 S2

Current controlled
Voltage source converter
Hybrid Shunt Active Power Filter

ig1 ig4 ig3 ig6 ig5 ig2

Is,abc

Hysteresis Controller

Vs,abc

+
Vdc,ref - CPIC or STPIC

Vdc,act

Fig. 1 Hybrid shunt active power filter system

3 Control Algorithms

At PCC equal magnitude with opposite polarity harmonic current has to be injected
in order to accomplish harmonic compensation, which in turn improves the power
quality of the distribution system by canceling the original distortion.
318 A. K. Mishra et al.

From Fig. 1, the instantaneous source current can be represented as

i s (t) = i L (t) − i a f (t) (1)

The source voltage is represented as:

Vs (t) = Vm sin ωt (2)

In case of nonlinear load, load current is having harmonic component as well as

fundamental component that can be represented as
∞

i L (t) = In sin(nωt + φn )
n=1
∞ 

= I1 sin(ωt + φ1 ) + In sin(nωt + φn ) (3)
n=2

The instantaneous load power can be calculated as:

p L (t) = vs (t) ∗ i L (t)

= Vm I1 sin2 ωt ∗ cos φ1 + Vm I1 sin ωt ∗ cos ωt ∗ sin φ1
∞ 

+ Vm sin ωt ∗ In sin(nωt + φn ) (4)
n=2

p L (t) = p f (t) + pr (t) + ph (t). (5)

This load power is having active, reactive and harmonic power. From (4), the real
power drawn from the load is represented as

p f (t) = Vm I1 sin2 ωt ∗ cos φ1 = vs (t) ∗ i s (t). (6)

If the active filter gives the harmonic and reactive power, then the source current
i s (t) will be pure sinusoidal and in phase with the source voltage. The three-phase
source currents after compensation can be expressed as:

∗ p f (t)
i sa (t) = = I1 cos φ1 sin ωt = Imax sin ωt. (7)
vs (t)

where Imax = I1 cos φ1 .

Similarly,
∗
i sb (t) = Imax sin(ωt − 120◦ ). (8)
Self-tuned PI Controller Based Hybrid Shunt Active … 319

Kp

E(s) + U(s)
R(s) + C(s)
Ki/s G(s)
+
-

Fig. 2 Basic structure of a CPIC

∗
i sc (t) = Imax sin(ωt + 120◦ ). (9)

CPIC or STPIC is used to regulate the inverter DC-link capacitor voltage and to
estimate this peak value of the reference current Imax .

4 CPIC for Reference Current Estimation

A Proportional Integral (PI) control system is represented in Fig. 2. In a PI controller,

the error signal e(t) generates the proportional and integral actions which are applied
to the plant model. Mathematically, PI controller is represented in (10).

t
u(t) = K p e(t) + K i e(t) (10)
0

5 STPIC for Reference Current Estimation

In self-tuning PI control, depending on error signal e(t) the required amount of

control signal u(t) is generated. The control signal u(t) can be represented as:

u(t) = K p (t)e(t) + K i (t) e(t)dt (11)

where error signal e(t) is calculated as the difference between reference and actual
DC-link capacitor voltage. The proportional gain K p (t) and the integral gain K i (t)
can be expressed in terms of error signal e(t) as follows:

K p (t) = K p(Max) − (K p(Max) − K p(Min) )e−[K e(t)] (12)

320 A. K. Mishra et al.

The pre-specified minimum and maximum values of proportional gains are

denoted as K p(Min) and K p(Max) . Where K is a constant that regulates the com-
putation of K p (t) within the specified range according to the existing error. In case
of large error signal e(t), the corresponding computed K p (t) value also large to
speed up transient response damping. Similarly, in case of small error signal e(t) the
corresponding computed K p (t) is the minimum value and responsible to eliminate
overshoots and undesirable oscillations. The integral gain K i (t) for the self-tuned PI
controller is expressed in terms of error signal e(t) and K i(Max) as follows:

K i (t) = K i(Max) e−[K e(t)] (13)

This is because under the approaching stage of steady-state conditions when the
value of the error signal e(t) is small, the controller uses large integral gain and
K i(Max) that overcome the existing associated steady-state error quickly. Similarly,
in case of large error signal e(t) the computed value of integral gain is small and that
eliminates the associated overshoots and undesirable oscillations. The two controller
gains K p (t) and K i (t) are optimally varied according to the error signal e(t) as men-
tioned in (12) and (13), respectively. It is observed from the (12) that the exponential
term approaches zero (e−[K e(t)] → 0) only when the error signal e(t) is large and
hence K p (t) = K p(Max) . In other way when the error e(t) is small the exponential
term approaches one (e−[K e(t)] → 1) and hence K p (t) = K p(Min) . From the (13), it is
found that the integral gain is allowed to vary within the range 0 ≤ K i (t) ≤ K i(Max) .
The controller parameter values are represented in Table 1. The Simulink structure
of the STPIC as mathematically described in (11) to (13) is illustrated in Fig. 3.

Table 1 STPIC parameter values

K p(Max) K p(Min) K i(Max) K i(Min) K
14 8 0.005 0 0.001

Fig. 3 Simulink diagram of STPIC

Self-tuned PI Controller Based Hybrid Shunt Active … 321

6 Results and Discussion

To investigate the performance of the proposed system, a Simulink model of an

HSAPF System for reactive power and harmonic compensation is developed as
shown in Fig. 4. For nonlinear load selection diode bridge rectifier with R load
on its dc-side is taken for which filtering is required. The designed parameter used
in the simulation is given in Table 2. The system performance is investigated in four
steps. Step: 1 When Hybrid Filter is not connected to the system. Step: 2 When SPF
is connected alone to the system. Step: 3 When HSAPF is connected to the system
with CPIC is used for reference to current estimation. Step: 4 When HSAPF is con-
nected to the system and to estimate the reference current STPIC is used. The load or
source current waveform and its harmonic spectrum along with various waveforms
are shown in Fig. 5a–f. The load or source current is having a THD of 25.14%. From
Fig. 6a–f, it can be seen that the THD is reduced from 25.14% to 17.5% when SPF
is connected to the system. To further improve the performance of the system active
filter is inserted into the existing system makes the system hybrid. Figures 7a–h and
8a–h gives various waveforms of HSAPF system with CPIC and STPIC used for ref-
erence current estimation, respectively. To test the validity of the CPIC and STPIC
a load change is applied at t = 0.3 s and t = 0.5 s. From Figs. 7a–h and 8a–h, it
is clear that Self-Tuned Proportional and Integral Controlled HSAPF system gives
better performances compared to Conventional Proportional and Integral Controlled
HSAPF system. Various performance parameter obtained in simulation are given in
Table 3 for comparison.

Fig. 4 Simulink model of hybrid shunt active power filter system

322 A. K. Mishra et al.

Table 2 Designed parameter

Parameters Values
of hybrid shunt active power
filter system Source voltage (V rms ) 230 V
System frequency (f ) 50 Hz
Source impedances
Source resistance (Rs ) 0.1 
Source inductance (L s ) 0.5 mH
Nonlinear load
Diode rectifier 6-diode
Load resistance (RL ) 40 
Shunt active power filter
Power converter 6-MOSFETs/6-diodes
Filter resistance (Rc ) 0.4 
Filter inductance (L c ) 1.35 mH
DC-side capacitance (C dc ) 2000 µF
Reference voltage (V dc,ref ) 680 V
Shunt passive power filter
5th harmonic filter K var = 1, Quality factor Q =
50
7th harmonic filter K var = 0.5, Quality factor Q
= 50
11th and 13th harmonic filter K var = 0.5, Quality factor Q
= 50
PI controller K P = 12, K I = 0.005

7 Conclusion

The design, modelling and simulation of Hybrid Shunt Active Power Filter sys-
tem for reactive power and harmonic compensation purposes considered in MAT-
LAB/Simulink environment. The outcomes of the simulation work provided low
THD of the source current with upgraded AC mains power factor and reduced sup-
plied mains reactive power. Comparing with CPIC and STPIC used for estimating
the reference current, STPIC gives better results in terms of THD of supply current,
input power factor, reactive power compensation and reduced percentage overshoot
of capacitor voltage. From the obtained result it is clear that STPIC gives a well-
regulated capacitor voltage with reduced overshoot when load disturbance occurs.
The prototype of the proposed HSAPF can be developed, which can be aimed as
future work.
Self-tuned PI Controller Based Hybrid Shunt Active … 323

Fig. 5 Simulation results of the proposed system without any filter

324 A. K. Mishra et al.

Fig. 6 Simulation results of the proposed system when passive filter connected alone
Self-tuned PI Controller Based Hybrid Shunt Active … 325

Fig. 7 Simulation results of conventional PI controlled HSAPF system

326 A. K. Mishra et al.

Fig. 8 Simulation results of self-tuned PI controlled HSAPF system

Self-tuned PI Controller Based Hybrid Shunt Active … 327

Table 3 Performance parameters of the system under different condition

Different Performance parameter
condition THD % Active power Reactive Power factor % Overshoot
Kw power VAR of V dc
Without any 25.14 8.609 581.30 0.9673 –
filter
Passive filter 17.50 8.372 69.30 0.9679 –
connected
alone
CPIC HSAPF 02.32 8.730 14.22 0.9925 4.41
STPIC 01.98 8.72 7.569 0.9971 3.67
HSAPF

References

1. Haugan, P.T.S., Tedeschi, E.: Reactive and harmonic compensation using the conservative
power theory. In: International Conference on Ecological Vehicles and Renewable Energies,
Elsevier (2015)
2. Thirumoorthi, P., Yadaiah, N.: Design of current source hybrid power filter for harmonic current
compensation. Simul. Model. Pract. Theor. 52, 78–91 (2015)
3. Mishra, A.K., Pathak, M.K., Das, S.: Isolated converter topologies for power factor correction—
a comparison. In: IEEE International Conference on Energy, Automation and Signal, pp. 1–6,
28 Dec 2011
4. Lam, S., Wong, M.C., Choi, W.H., Cui, X.X., Mei, H.M., Liu, Z.J.: Design and performance of
an adaptive low-DC-voltage-controlled LC-hybrid active power filter with a neutral inductor
in three-phase four-wire power systems. IEEE Trans. Ind. Electr. 61(6), 2635–2647 (2014)
5. Rahmani, S., Hamadi, A., Al-Haddad, K., Dessaint, L.A.: A combination of shunt hybrid power
filter and thyristor-controlled reactor for power quality. IEEE Trans. Industr. Electron. 61(5),
2152–2164 (2013)
6. Das, S.R., Ray, P.K., Mohanty, A.: Comparative assessment for power quality improvement
using hybrid power filters. In: Recent Findings in Intelligent Computing Techniques, pp. 257–
265. Springer, Singapore (2018)
7. Das, S.R., Ray, P.K., Mohanty, A.: Fuzzy sliding mode based series hybrid active power filter
for power quality enhancement. In: Advances in Fuzzy Systems (2018)
8. Das, S.R., Ray, P.K., Mohanty, A., Panigrahi, T.K.: Application of artificial intelligence tech-
niques for improvement of power quality using hybrid filters. In: Computational Intelligence
in Data Mining, pp. 719–729. Springer, Singapore (2020)
9. Ray, P.K., Das, S.R., Mohanty, A.: Fuzzy-controller-designed-PV-based custom power device
for power quality enhancement. IEEE Trans. Energy Convers. 34(1), 405–414 (2018)
10. Mohamed, S.Z.: A self-tuning PI controller for the speed control of electrical motor. Electr.
Power Syst. Res. 110, 293–303 (2015)
Performance Analysis of Si-Channel
Nanosheet FETs with Strained SiGe
Source/Drain Stressors

E. Mohapatra , T. P. Dash, J. Jena, S. Das, J. Nanda, and C. K. Maiti

Abstract Silicon channel Nanosheet Field Effect Transistors (NSFETs) are inte-
grated with diamond-shaped embedded-Si1−x Gex source/drain (S/D) stressors to
boost the electrical performance. The effects of stressor geometry (shape and size)
on device performance are studied in detail. The carrier mobility enhancement due
to process-induced strain in NSFET is found to increase the drain current. Further,
enhancement in other electrical parameters is also obtained by increasing the Ge
mole fraction in Si1−x Gex . The impact of higher Ge mole fraction in the S/D region
resulted in an optimum V th , SS, and DIBL values of 141.78 mV, 69.49 mV/dec,
and 24.11 mV/V, respectively. The Efficiency of S/D stress transfer is found to be
dependent on the Ge mole fraction in the source/drain stressors p-channel NSFETs.

Keywords Nanosheet FETs · Strained SiGe · Source/drain stressors

E. Mohapatra (B) · T. P. Dash · J. Jena · C. K. Maiti

Department of Electronics and Communication Engineering, Siksha ‘O’ Anusandhan (Deemed to
be University), Bhubaneswar, Odisha 751030, India
e-mail: [email protected]
T. P. Dash
e-mail: [email protected]
J. Jena
e-mail: [email protected]
C. K. Maiti
e-mail: [email protected]
S. Das
Department of Electronics and Communication Engineering, Silicon Institute of Technology,
Bhubaneswar, Odisha 751024, India
e-mail: [email protected]
J. Nanda
Department of Physics, Siksha ‘O’ Anusandhan (Deemed to be University), Bhubaneswar, Odisha
751030, India
e-mail: [email protected]

© Springer Nature Singapore Pte Ltd. 2020 329

G. Pradhan et al. (eds.), Advances in Electrical Control and Signal Systems,
Lecture Notes in Electrical Engineering 665,
https://doi.org/10.1007/978-981-15-5262-5_23
330 E. Mohapatra et al.

1 Introduction

The scaling of FinFETs become more complex as well as to achieve higher perfor-
mance becomes harder. So, it is necessary to search for the next generation tran-
sistor architecture. Nanosheet Field Effect Transistors (NSFETs) have emerged as a
promising candidate to substitute the FinFETs or nanowires for ultimate scaling due
to superior electrostatic integrity, better gate control, high I ON as well as RC-delay
and an optimized formation process [1–3]. NSFET can be a strong contender and
better replacement for Nanowire Field Effect Transistors (NWFETs) or even stacked
NWFET structures regarding drive current [4].
Over the past decade, the semiconductor industry has relied on stress engineering
which is very critical for optimization of device performance in every successive
technology node. To enhance the device performance, the first production of the
strained-silicon technology was reported in the 90 nm technology node [5]. The
introduction of embedded SiGe stressors in the source/drain (S/D) region provided
the necessary uniaxial compressive stress in the channel to improve hole mobility.
Since then, various innovations on strained-Si technology have been made to boost the
transistor scaling performance according to Moore’s Law. P-FET performances were
highly enhanced by uniaxial stress-induced by e-SiGe in the S/D region [6]. However,
the transition to nonplanar devices like FinFETs, Nanowire FETs, or Nanosheet FETs
has raised the effectiveness of embedded SiGe for the latest technology nodes [7].
In this paper, the integration of Si1−x Gex (10% ≤ x ≤ 50%) on S/D regions in
14 nm gate length NSFET has been made. The induced stress by SiGe stressor was
simulated by technology computer-aided (TCAD) tool and the impacts of Ge mole
fraction on various electrical parameters such as threshold voltage (V th ), on current
(I ON ), off current (I OFF ), subthreshold swing (SS) and drain induced barrier lowering
(DIBL) are discussed.

2 Simulation Environment

2.1 Device Details

The proposed 3D device structure of NSFET is shown in Fig. 1. The channel length,
sheet height, and sheet width of NSFET are considered as 14 nm, 6 nm, and 12 nm,
respectively. The High-K and oxide thicknesses are 1.5 and 0.5 nm, respectively.
The various key parameters and the dimensions of the proposed device are listed in
Table 1.
Performance Analysis of Si-Channel … 331

Fig. 1 A 3D view of NSFET

with diamond-shaped SiGe
S/D stressor

Table 1 Geometrical
Parameters Value
parameters of NSFETs
considered in this work Nanosheet height (NS_H) 6 nm
Nanosheet width (NS_W) 12 nm
Channel length 14 nm
High-k (HfO2 ) thickness 1.5 nm
Oxide (SiO2 ) thickness 0.5 nm
Spacer length 7 nm
Channel doping 1 × 1016 cm−3
Source/drain doping concentration 1 × 1020 cm−3

2.2 Stress Engineering

To extend the life of Moore’s Law, multi-gate structures that provide the best possible
electro statistics and superior carrier transport properties are becoming the trend for
future technology nodes. The relation between strain and no-strain is widely evaluated
by the piezoresistive approach. Then the impact of stress on mobility enhancement
has been investigated by the following approach. The lattice parameters related to
transport properties has been provided. When this approach is applied to carrier
mobility, different terms of piezo coefficients are evaluated based on the doping
and electric field. The device characteristics are investigated by using the sub-band
Boltzmann transport equation, including phonon and surface roughness scattering.
The elastic tensor constants can be calculated from the material property specifically
332 E. Mohapatra et al.

stiffness matrix [S] [7]. The inverse matrix of the [S] can be expressed in terms of
the compliance matrix [S] = [C]−1 .
The compliance coefficient (C ij ) for both Si and Ge can be expressed as follows:
For Si,
⎧  
⎨ c11 = 163.8 GPa − TL · 0.0128 K ,
GPa

c = 59.2 GPa − TL · 0.0048 GPaK 

, (1)
⎩ 12
c44 = 81.7 GPa − TL · 0.0059 K ,
GPa

where T L denotes the lattice temperature, and for Ge, the values of c11 , c12 and c44
are 126.0 GPa, 44.0 GPa, and 67.7 GPa, respectively.
⎧
⎨ c11 = 126.0 GPa,
c = 44.0 GPa, (2)
⎩ 12
c44 = 67.7 GPa

The isotropic strain in the device can be calculated from the lattice constants using
Victory Stress [8] when there will be no stress for SiGe in three 3D systems:
  
1 + νSiGe aSiGe − aSi
εx x = ε yy = εzz = (3)
1 − νSiGe aSiGe

Where ν SiGe and ε represent Poisson’s ratio and strain tensor components, respec-
tively. Similarly, aSiGe and aSi denote the lattice constant for SiGe and silicon,
respectively.
For SiGe, the lattice constant (a) and Poisson’s ratio (ν) can be related by linear
interpolation between the following values of Si and Ge:

aSi = 5.43102 Å + (1.41 × 10−5 Å/K) . (TL − 300 K), and

aGe = 5.6579 Å + (3.34 × 10−5 Å/K) . (TL − 300 K),
νSi = 0.28, and νGe = 0.273. (4)

The stress engineering has been implemented in the simulation environment using
the VSP tool [9] for analysis of the stress profile in the device. A 3D stress profile of
the sheet (14 nm long) has been shown in Fig. 2.
The longitudinal stress distribution in the channel regions of the device with 30%
Ge mole fraction in S/D SiGe stressor is shown in Fig. 2a. The line in Fig. 2a shows
the exact locations where the exact effective stress values are extracted and plotted in
Fig. 2b. Figure 2b shows the stress distribution in the channel of NSFET for different
Ge contents in the stressor. With the increase in the Ge mole fraction, the carrier
effective mass decreases, and the hole mobility increases due to the S/D stressor.
This results in strain on the channel which enhances the hole mobility of the Si
channel. The longitudinal stress ZZ for different Ge mole fraction content of 10%,
30%, and 50% are 0.512, 1.49 and 2.41 GPa, respectively.
Performance Analysis of Si-Channel … 333

Fig. 2 a Stress distribution in sheet with 30% Ge mole fraction, b 1-D longitudinal stress profiles
in the channel, due to diamond-shaped SiGe S/D stressor for different Ge mole fraction

3 Results and Discussion

In the TCAD simulations, quantum confinement effects are considered with the clas-
sical drift-diffusion-model [10]. Fermi-Dirac distribution, Lombardi model, and SRH
recombination have been used [11, 12]. Bandgap narrowing model was included for
all the Si and SiGe regions [13]. MINIMOS-NT [14] has been used in the simulation
of the proposed device to extract various electrical parameters.
The impact of the Ge mole fraction on the I D –V G characteristics for the proposed
NSFET design is shown in Fig. 3. For the proposed device, drain current increases
with raising the Ge mole fraction in the S/D region at V D = 0.05 V (linear) and
V D = 0.7 V (saturation). This is due to the fact of higher stress in SiGe S/D region
indicates a higher hole mobility enhancement due to the larger induced stress with an

Fig. 3 I D –V G characteristics of NSFET devices with SiGe S/D stressors having a gate length of
14 nm
334 E. Mohapatra et al.

Fig. 4 I D –V D characteristics of NSFET devices with SiGe S/D stressors having a gate length of
14 nm

increase in the Ge mole fraction. Superior I D –V G characteristics are demonstrated by

the NSFET with 50% Ge mole fraction for both linear and saturation regimes. And
the enhancement in higher saturation driving current due to 50% Ge mole fraction
is 75% than the Si channel device with no-stress effect in the S/D region.
The device output characteristics I D –V D is shown in Fig. 4. A clear enhancement
in drain current is observed as strain in the sheet increases as Ge content increases
from 10 to 50%. The enhancement of IDSAT is around 9%, 55%, and 89% compared
to control Si NSFET without any stress by increasing the Ge content to 10%, 30%,
and 50%, respectively.
Figure 5 shows the variation of V th , SS, and DIBL for NSFETs with different
Ge mole fractions. The threshold voltage decreases significantly with an increase in
the Ge mole fraction from 0 to 50% as shown in Fig. 4. In no-stress condition, the
value of V th is maximum which is according to the requirements of ITRS [15]. This
decrease in V th with the increase in the Ge mole fraction is because of the reduction
in the energy bandgap which results in a decrease in flat-band voltage, and hence an
early onset of inversion [16].
The DIBL is extracted at V DS = 0.05 V and V DS = 0.7 V by constant current
method. The DIBL decreases with an increase in the Ge mole fraction as shown in
Fig. 5. As mole fraction increases, the reduction of the bandgap at source and drain
regime happens and carrier mobility increases which results in a decrease in DIBL.
The SS value decreases with an increase in the Ge mole fraction in Si1−x Gex due to
higher carrier mobility as shown in Fig. 2.
The effect of variation of Ge mole fraction on I ON and I OFF is shown in Fig. 6.
It is clearly observed that the values of I ON and I OFF increase with raising in the Ge
Performance Analysis of Si-Channel … 335

Fig. 5 Impact of Ge mole fraction on V th , SS, and DIBL of Si1−x Gex S/D stressors in NSFET. The
lines with stripped marks represent saturation region (V D = 0.7 V) and the plain lines represent the
linear region of operation (V D = 0.05 V)

Fig. 6 Impact of Ge mole fraction on I ON and I OFF of Si1−x Gex S/D stressors in NSFET

mole fraction. With the increase in Ge mole fraction, the bandgap decreases which
results in enhancement of drain current both in linear and saturation regions. The
short channel effect (SCE) and subthreshold performance are serious concerns for
devices at this small dimension. The critical comparison of all electrical parameters
(I ON , I OFF , V th , SS) is also studied and plotted in Fig. 7.
336 E. Mohapatra et al.

Fig. 7 Relative different Ge mole fraction for the normalized five different electrical parameters in
the saturation regime. Red, blue, green and purple color correspond to 0, 10, 30 and 50% Ge mole
fraction values

As shown in Fig. 7, the extracted normalized electrical parameters are in satura-

tion. The variations of the five electrical parameters with different Ge mole fractions
are normalized with respect to the control Si channel NSFET without S/D stressors.

4 Conclusions

A novel 14 nm long Si channel NSFET with diamond shape embedded SiGe source
and drain has been designed and simulated. A compressive stress profile has been
mapped in the channel (sheet) which enhances the device performance. The advan-
tages of SiGe S/D junctions result in a significant compressive strain in the channels
of NSFETs. 76% drive current enhancement was observed in comparison with an
NSFET with 0% Ge content in the saturation regime. The subthreshold regional
performance of NSFET is presented.

References

1. Loubet, N., Hook, T., Montanini, P., Yeung, C.W., Kanakasabapathy, S., Guillom„ M.,
Yamashita, T., Zhang, J., Miao, X., Wang, J., Young, A., Chao, R., Kang, M., Liu, Z., Fan, S.,
Hamieh, B., Sieg, S., Mignot, Y., Xu, W., Seo„ S.-C., Yoo, J., Mochizuki, S., Sankarapandian,
M., Kwon, O., Carr, A., Greene, A., Park, Y., Frougier, J., Galatage, R., Bao, R., Shearer, J.,
Conti, R., Song, H., Lee, D., Kong, D., Xu, Y., Arceo, A., Bi, Z., Xu, P., Muthinti, R., Li, J.,
Performance Analysis of Si-Channel … 337

Wong, R., Brown, D., Oldiges, P., Robison, R., Arnold, J., Felix, N., Skordas, S., Gaudiello,
J., Standaert, T., Jagannathan, H., Corliss, D., Na, M.-H., Knorr, A., Wu, T., Gupta, D., Lian,
S., Divakaruni, R., Gow, T., Labelle, C., Lee, S., Paruchuri, V., Bu, H., Khare, M.: Stacked
nanosheet gate-all-around transistor to enable scaling beyond FinFET. In: Symposium on VLSI
Technology, Kyoto, pp. T230–T231 (2017)
2. Zhang, J., Ando, T., Yeung, C. W., Wang, M., Kwon, O., Galatage, R., Chao, R., Loubet, N.,
Moon, B. K., Bao, R., Vega, R. A., Li, J., Zhang, C., Liu, Z., Kang, M., Miao, X., Wang,
J., Kanakasabapathy, S., Basker, V. S., Jagannathan, H., Yamashita, T.: High-k metal gate
fundamental learning and multi-VT options for stacked nanosheet gate-all-around transistor.
In: IEEE International Electron Devices Meeting, San Francisco, CA, pp. 22.1.1–22.1.4 (2017)
3. Yakimets, D., Bardon, M.G., Jang D., Schuddinck, P., Sherazi, Y., Weckx, P., Miyaguchi, K.,
Parvais, B., Raghavan, P., Spessot, A., Verkest, D., Mocuta, A.: Power aware FinFET and lateral
nanosheet FET targeting for 3 nm CMOS technology. In: IEEE International Electron Devices
Meeting, San Francisco, CA, pp. 20.4.1–20.4.4 (2017)
4. Jang, D., Yakimets, D., Eneman, G., Schuddinck, P., Bardon, M.G., Raghavan, P., Spessot, A.,
Verkest, D., Mocuta, A.: Device exploration of nanosheet transistors for sub- 7-nm technology
node. IEEE Trans. Electron Dev. 64(6), 2707–2713 (2017)
5. Ghani, T., Armstrong, M., Auth, C., Bost, M., Charvat, P., Glass, G., Hoffmann, T., Johnson,
K., Kenyon, C., Klaus, J., McIntyre, B., Mistry, K., Murthy, A., Sandford, J., Silberstein,
M., Sivakumar, S.; Smith, P., Zawadzki, K., Thompson, S., Bohr, M.: A 90 nm high volume
manufacturing logic technology featuring novel 45 nm gate length strained silicon CMOS
transistors. In: IEEE International Electron Devices Meeting, Washington, DC, pp. 11.6.1–
11.6.3 (2003)
6. Thompson, S.E., Armstrong, M., Auth, C., Alayi, M., Chau, R., Cea, S., Ghani, T., Glass, G.,
Hoffman, T., Jan, C.H., Kenyon, C., Klaus, J., Kuhn, K., Ma, Z., Mcintyre, B., Mistry, K.,
Murthy, A., Obradovic, B., Nagisetty, R., Nguyen, P., Sivakumar, S., Shaheed, R., Shifren, L.,
Tufts, B., Tyagi, S., Bohr, M., El-Mansy, Y.: A 90-nm logic technology featuring strained-
silicon. IEEE Tans. Electron Dev. 51(11), 1790–1797 (2004)
7. Dash, T.P., Jena, J., Mohapatra, E., Dey, S., Das, S., Maiti, C.K.: Stress-induced variability
studies in tri-gate FinFETs with source/drain stressor at 7 nm technology nodes. J. Electron.
Mater. 48(8), 5348–5362 (2019)
8. VictoryStress User’s manual (2017)
9. VSP User’s manual (2018)
10. Ancona, M.G., Iafrate, G.J.: Quantum correction to the equation of state of an electron gas in
a semiconductor. Phys. Rev. B. 39(13), 9536–9540 (1989)
11. Lombardi, C., Manzini, S., Saporito, A., Vanzi, M.: A physically based mobility model for
numerical simulation of nonplanar devices. IEEE Trans. Comput. Aided Design 7(11), 1164–
1171 (1988)
12. Shockley, W., Read Jr., W.T.: Statistics of the recombinations of holes and electrons. Phys. Rev.
87(5), 835–842 (1952)
13. Slotboom, J.W., Graaff, H.C.: Measurements of bandgap narrowing in Si bipolar transistors.
Solid-State Electron. 19(10), 857–862 (1976)
14. MINIMOS-NT, User’s manual (2018)
15. Semiconductor Industry Association (SIA), International Technology Roadmap for Semicon-
ductors (ITRS) (2005)
16. Venkataraman, V., Nawal, S., Kumar, M.J.: Compact analytical threshold-voltage model of
nanoscale fully depleted strained-Si on silicon–germanium-on-insulator (SGOI) MOSFETs.
IEEE Trans. Electron Devices 54(3), 554–562 (2007)
Digital Implementation of OS-ELM
for Data Classification in Real-Time

Susanta Kumar Rout and Pradyut Kumar Biswal

Abstract Field-programmable gate array (FPGA) has been used as a very effec-
tive hardware platform in different research area as it enhances the efficiency of the
embedded module. The accessibility of minimized, fast circuitry for a artificial neu-
ral networks (ANNs) is the most important and utmost necessity for many critical
applications. In this paper, a single layer feed-forward neural network (SLFN) named
as online sequential extreme learning machine (OS-ELM) is conferred and realized
in digital platform for real-world data classification. The digital employment of OS-
ELM supports to form an efficient hardware unit for data classification in real-time
as the classifier has high learning speed and promising accuracy. Finally, the digital
architecture of OS-ELM is implemented on a Virtex-5 FPGA hardware platform to
validate the feasibility, efficacy, and vitality of the proposed classifier in real-time.

Keywords Field-programmable gate array (FPGA) · Artificial neural networks

(ANNs) · Single layer feed-forward neural network (SLFN) · Online sequential
extreme learning machine (OS-ELM) · Digital architecture · Hardware
implementation

1 Introduction

Digital architecture of artificial neural networks (ANNs) acknowledges the design

of hardware circuitry that validates the results as performance, power consumption
and hardware resources [1]. Several domain such as sensor networks [2], image
processing [3], robotics [4], and EEG signal classification requires the support of

S. K. Rout (B) · P. K. Biswal

International Institute of Information Technology, Bhubaneswar, India
e-mail: [email protected]
P. K. Biswal
e-mail: [email protected]
S. K. Rout
Siksha “O” Anusandhan deemed to be University, Bhubaneswar, India

© Springer Nature Singapore Pte Ltd. 2020 339

G. Pradhan et al. (eds.), Advances in Electrical Control and Signal Systems,
Lecture Notes in Electrical Engineering 665,
https://doi.org/10.1007/978-981-15-5262-5_24
340 S. K. Rout and P. K. Biswal

neural networks (NNs) with compact fast circuitry [5]. The advantages of field-
programmable gate array (FPGA) as compared to ASIC are very flexible, reusable,
and quicker to acquire. The hardware implementation of ANNs design on reconfig-
urable digital environment to provide high flexibility and high processing speed as
it implemented in parallel structure [1].
Machine learning is used to extract meaningful information from the dataset. Arti-
ficial neural network is information processing system stimulated neurons. It is used
in different domain to extract the information from real-time dataset. SLFN has capa-
bility to transfer the information from input node to output node through the hidden
node sequentially, so it is used frequently. Unlike its counterpart, i.e., the recurrent
network, it is easy to design and train the network. It is widely used in supervised
learning like classification regression [6, 7]. An extreme learning machine (ELM) is
the most advanced tool to train a single layer feed-forward neural network (SLFN).
Reducing the computational requirement ELM enables the SLFN for real-time data
processing and manages timing constraints. It outperforms the conventional learning
algorithm on the basis of learning speed, use of wide range of activation functions
and simplicity in hardware realization. It also has good performance, capability to
avoid the local minima and comply ubiquitous alikeness axioms [8, 9]. The numbers
of hidden nodes are more in ELM than conventional algorithms; randomly, ELM
creates the hidden nodes in order to match the performance. The testing phase of
ELM has longer running time if the network size is large. Hence, several efforts have
been made to increase the compactness of ELM [10, 11]. It is not required that the
training data are ready before the training process in real-life applications. There are
certain cases where the data may come sequentially or step by step. Liang et al. [12]
proposed a different variant of ELM named as online sequential ELM (OS-ELM) is
widely used when data arrive one by one or group by group with different or fixed
step size. It provides efficient result for classification and prediction. In [13, 14],
introduced forgetting online sequential ELM (FOS-ELM), where the training data
have a certain period of validity.
Recently Basu et al. [15] presented the ELM implementation on silicon. The
silicon neuron is the major block fixated on converting input current to a frequency.
In [16, 17], the digital implementation of ELM has proposed on reconfigurable FPGA
and complex programmable logic devices (CPLDs). The ELM has been implemented
using L1 regularization with less complexity.
In this paper, the testing phase of OS-ELM classifier digital architecture using
cosine activation function is presented. A high-speed FPGA is used to implement
the proposed architecture considering the dataset as inputs. The novel hardware
architecture of OS-ELM classifier consumes less power and minimum hardware
resources with high efficiency. Figure 1 shows the real-time data classification block
diagram using SLFN.
Remaining part of the manuscript is prepared as follows. The OS-ELM classifier
is presented in Sect. 2. Section 3 explains the simulation result, hardware implemen-
tation, and validation of the proposed classifier. Finally, the conclusion of the paper
is drawn in Sect. 4.
Digital Implementation of OS-ELM … 341

Fig. 1 Real-time data

classification block diagram
using SLFN

2 OS-ELM Classifier

In case of ELM, all the training data are usable during training. In real-time data
classification, the training data may be arrived one by one or group by group, so
ELM is modified and named as OS-ELM, which is used to process the real-time data
[18]. The OS-ELM classifier has no earlier knowledge regarding the training dataset.
The layer representation of ELM is shown in Fig. 2.
To deal with N number of input data of the ith feature vector for output class
is [Ii1 , Ii2 , . . . , Ii N ]T ∈ R n . Consider L number of hidden nodes with weight wi
[wi1 , wi2 , . . . , wi L ]T , and the bias bi is [bi1 , bi2 , . . . , bi L ]T . The m number of output
classes has predefined target Ti = [Ti1 , Ti2 , . . . , Tim ]T ∈ R m , and the output weight
vector βi = [βi1 , βi2 , . . . , βim ]T is calculated from the ith hidden neuron and output
nodes. The network output Oi = [Oi1 , Oi2 , . . . , Oim ]T of ELM using kernel function
is given as

Fig. 2 Layer representation of ELM

342 S. K. Rout and P. K. Biswal


L 
N
βi K i (I j ) = βi K i (wi · I j + bi ) = O j , j = 1, 2, . . . , N (1)
i=1 i=1

To achieve zero mean error, consider the exact value of wi , bi and βi ,


L
 
i.e., O j − Tj  = 0 (2)
j=1

Such that


L
βi K i (wi · I j + bi ) = T j , for j = 1, 2, . . . , N
i=1

Above, N equations are represented as hidden layer output matrix H of ELM [6,
7] as,

H ·β = T (3)

N0
Let us consider the initial chunk size training set N0 = {(Ii , ti )}i=1 ∈ Rn ×
R and N0 > L . Using hidden layer output matrix, concept in batch ELM leads to
m

minimizing the factor H0 β − T0 , where

⎡ ⎤
K (w1 · I1 + b1 ) · · · K (w L · I1 + b L )
⎢ .. .. .. ⎥
H0 = ⎣ . . . ⎦ ;
K (w1 · I N0 + b1 ) · · · K (w L · I N0 + b L ) N ×L
0
⎡ T⎤ ⎡ T ⎤
β1 t1
⎢ .. ⎥ ⎢ .. ⎥
β=⎣ . ⎦ and T0 = ⎣ . ⎦
β NT t NT0 N0 ×m
0 L×m

The solution of minimization factor H0 β − T0  is β (0) = K 0−1 H0T T0 , where

N0 +N1
K 0 = H0T H0 . Now, consider the next chunk of data N1 = {(Ii , ti )}i=N 0 +1
.
Here, the problem leads to minimize the factor
 
 H 
 0 T0 
 β−  (4)
 H1 T1 

where
Digital Implementation of OS-ELM … 343
⎡ ⎤
K (w1 · I N0 +1 + b1 ) · · · K (w L · I N0 +L + b L )
⎢ .. .. .. ⎥
H1 = ⎣ . . . ⎦ ;
K (w1 · I N0 +N1 + b1 ) · · · K (w L · I N0 +N1 + b L ) (N0 +N1 )×L
⎡ T ⎤ ⎡ T ⎤
β1 t1
⎢ .. ⎥ ⎢ .. ⎥
β=⎣ . ⎦ and T0 = ⎣ . ⎦
β NT +N t NT0 (N0 +N1 )×m
0 1 L×m

For both training datasets N0 and N1 , the output weight β becomes

T
H0 T0
β =1
K 1−1 (5)
H1 T1

where
T
H0 H0
K1 =
H1 H1

To make the online learning, sequential β 1 must be represented in terms of

β , K 1 , H1 , and T1 .
0

If the parameter N0 = N1 , then the OS-ELM can be treated as batch ELM. It

signifies batch ELM is a unique type of OS-ELM, where all the training datasets are
used in one iteration.

3 Result and Discussion

To test the performance of the proposed classifier, authors use diabetes dataset avail-
able in UCI machine learning repository [19]. The aforementioned dataset contains
768 data with eight features and two classes from which 576 data are used for training
purpose and 192 data are used for testing purpose. We have obtained promising test-
ing classification accuracy of 80.72% using MATLAB given in Table 1. Moreover,

Table 1 Testing accuracy of

Researchers Database Function Accuracy (%)
OS-ELM using cosine
activation function Huang et al. [8] Diabetes Sigmoid 76.54
Huang et al. [10] Diabetes Sigmoid 77.95
Gaussian 77.52
Multiquad 78.09
Proposed Diabetes Cosine 80.72
344 S. K. Rout and P. K. Biswal

OS-ELM digital architecture is enforced on a Viretex-5 FPGA hardware platform to

classify the real-time data.

3.1 Hardware Implementation of OS-ELM on FPGA

Environment

The hardware implementation of OS-ELM classifier on FPGA environment is utmost

important to classify the signals and images in real-time. Figure 3 shows the OS-ELM
hardware implementation architecture. The hardware architecture of the OS-ELM
consists of number of stages. Using parallel and pipeline concept, all the stages are
executed at a time to achieve promising output. In this manuscript, authors use 16
bits to represent a numerical value, where integer value is represented by three bits,
fractional value is represented by 12 bits, with one sign bit.
First stage of Fig. 3 is consisting of RAM, which stores input testing data and
fetches the data as per select line value. In this paper, diabetes real-world dataset
of 192 testing data with eight features is used. So, eight RAMs are used, and each
RAM has 192 data, chooses eight bit input-select line to fetch 192 data one by one,
as 28 = 256. In input stage, parallel operation is performed to provide eight outputs,
and each output contained 192 data from which one value is selected as per the
select line value. In this article, authors use 18 hidden nodes, which represent barrel
shifters and stored the hidden layer weight is the power of two, i.e., 0, ±1/2, ±1/4,
±1/8, ±1/16, etc. All the eight outputs from the input stage are fed into each of
the shifter, each shifter has eight inputs and eight outputs, and each multiplexer has
18 inputs with one output, so five bit selection line is required for multiplexer. The
term wi · Ii is computed and fed into the adder unit sequentially to extract the output
L
i=1 wi Ii without considering the bias (b) to reduce the design intricacy. Adder
output is passed through the activation unit and set threshold value to classify the

Fig. 3 OS-ELM hardware implementation architecture

Digital Implementation of OS-ELM … 345

output into a particular class. MATLAB is used to calculate output weight and stored
in output weight memory. Conclusively, H (I ) and output weight β are fed to MAC
block to produce covet output H (I )·β = O(I ). The control signals used in OS-ELM
hardware architecture are generated using controller.
Conclusively, the OS-ELM architecture in digital domain is tested using Chip-
Scope on FPGA hardware platform. Figure 4 shows Virtex-5 FPGA with host-PC
hardware interfacing. Figures 5 and 6 show the simulation output of OS-ELM using
Xilinx tool and hardware output using ChipScope, respectively. We found 1017 H
for both the outputs, i.e., 1.0056 D, which classify the test data with 0.0056 errors.
The novel hardware architecture of OS-ELM produced the output after place and
route on Virtex-5 FPGA. To execute one test case 59 IOBs, 2 BlockRAMs of 72 KB,

Fig. 4 Virtex-5 FPGA with host-PC hardware interfacing

Fig. 5 Simulation output of OS-ELM using Xilinx tool

346 S. K. Rout and P. K. Biswal

Fig. 6 Hardware output of OS-ELM using ChipScope

Table 2 Hardware utilization

Parameter Used Available Utilization in %
summary of OS-ELM
Clock speed (MHz) 39.27 – –
Precision (bit) 16 – –
Bounded IOBs 59 480 12.29
Number of slice 1793 32,640 5.49
LUTs
Power consumption 911 – –
(mW)

1793 LUTs and 911 mW power are required. 0.0958 ms with 1.044 µs initial latency
is required to produce estimated output. Table 2 presents the hardware utilization
summary of OS-ELM classifier.

4 Conclusion

This paper presents the OS-ELM architecture in digital domain for real-time data
classification efficiently. To check the simulation output of the OS-ELM classifier,
Virtex-5 FPGA platform is used with minimum number of hardware resources. High
classification accuracy, effectiveness, and flexibility are the major utility to build the
proposed OS-ELM digital architecture is an efficient technique for real-time data
classification.

References

1. Misra, J., Saha, I.: Artificial neural networks in hardware: a survey of two decades of progress.
Neurocomputing 74(1–3), 239–255 (2010)
2. Dlugosz, R., Talaska, T., Pedrycz, W.: Current-mode analog adaptive mechanism for ultra-low-
power neural networks. IEEE Trans. Circuits Syst. II Express Briefs 58(1), 31–35 (2011)
3. Chen, J., Shibata, T.: A neuron-MOS-based VLSI implementation of pulse-coupled neural
networks for image feature generation. IEEE Trans. Circuits Syst. I Regul. Pap. 57(6), 1143–
1153 (2010)
4. Arena, P., et al.: A CNN-based chip for robot locomotion control. IEEE Trans. Circuits Syst. I
Regul. Pap. 52(9), 1862–1871 (2005)
Digital Implementation of OS-ELM … 347

5. Wang, Y., Li, Z., Feng, L., Bai, H., Wang, C.: Hardware design of multiclass SVM classification
for epilepsy and epileptic seizure detection. IET Circ. Dev. Syst. 12(1), 108–115 (2018)
6. Huang, G.B., Babri, H.A.: Upper bounds on the number of hidden neurons in feedforward
networks with arbitrary bounded nonlinear activation functions. IEEE Trans. Neural Networks
9(1), 224–229 (1998)
7. Huang, G.B.: Learning capability and storage capacity of two-hidden-layer feedforward
networks. IEEE Trans. Neural Networks 14(2), 274–281 (2003)
8. Huang, G.B., Zhu, Q.Y., Siew, C.K.: Extreme learning machine: a new learning scheme of feed-
forward neural networks. In: 2004 IEEE International Joint Conference on Neural Networks,
vol. 2, pp. 985–990 (2004)
9. Huang, G.B., Zhu, Q.Y., Siew, C.K.: Extreme learning machine: theory and applications.
Neurocomputing 70(1–3), 489–501 (2006)
10. Huang, G.B., Zhou, H., Ding, X., Zhang, R.: Extreme learning machine for regression and
multiclass classification. IEEE Trans. Syst. Man Cybern. Part B Cybern. 42(2), 513–529 (2012)
11. Huang, G.B., Wang, D.H., Lan, Y.: Extreme learning machines: a survey. Int. J. Mach. Learn.
Cybernet. 2(2), 107–122 (2011)
12. Liang, N.Y., Huang, G.B., Saratchandran, P., Sundararajan, N.: A fast and accurate online
sequential learning algorithm for feedforward networks. IEEE Trans. Neural Networks 17(6),
1411–1423 (2006)
13. Jianwei, Z., Wang, Z., Park, D.S.: Online sequential extreme learning machine with forgetting
mechanism. Neurocomputing 87, 79–89 (2012)
14. Ye, Y., Squartini, S., Piazza, F.: Online sequential extreme learning machine in nonstationary
environments. Neurocomputing 116, 94–101 (2013)
15. Basu, A., et al.: Silicon spiking neurons for hardware implementation of extreme learning
machines. Neurocomputing 102, 125–134 (2013)
16. Decherchi, S., Gastaldo, P., Leoncini, A., Zunino, R.: Efficient Digital Implementation of
Extreme Learning Machines for Classification. IEEE Trans. Circuits Syst. II Express Briefs
59(8), 496–500 (2012)
17. Frances-Villora, J.V., Rosado-Muñoz, A., Martínez-Villena, J.M., Bataller-Mompean, M.,
Guerrero, J.F., Wegrzyn, M.: Hardware implementation of real-time Extreme Learning Machine
in FPGA: analysis of precision, resource occupation and performance. Comput. Electr. Eng.
1(51), 139–156 (2016)
18. Sahani, M., Dash, P.K.: Variational mode decomposition and weighted online sequential
extreme learning machine for power quality event patterns recognition. Neurocomputing 310,
10–27 (2018)
19. Blake, C., Merz, C.J.: fUCIg Repository of Machine Learning Databases (1998)
Comparative Study of System
Performances Using Integral Type LQR
with DE and Z-N Optimized PID
Controller in AVR System

Ashribad Pattnaik, Bidyadhar Rout, and Akshaya Kumar Patra

Abstract The low-frequency system oscillation is inherently present in a power

system network. This oscillation arises due to slow and gradual changes of load
and power demand. The persistence of these oscillations may seize power flow and
decreases the terminal voltage in the synchronous generator of the power systems.
The required terminal voltage is improved by strengthening the field voltage in AVR
system of the exciter in the synchronous generator by minimizing the oscillations.
This paper presents the control design of AVR in SMIB power system using an
integral type LQR control technique with adjusting the state weighting matrix and
control the weighting matrix simultaneously. The terminal voltage of the generator
using integral type LQR technique has been compared with Z-N tuned PID and DE
optimized PID controller in AVR system. This paperwork is carried out in MATLAB
platform.

Keywords Small-signal stability · AVR · PID controller · LQR · DE algorithm

optimization

1 Introduction

The modern power system desires to supply power demanded by the consumer with-
out interruptions maintaining reliability, quality of power. However, the electrome-
chanical low-frequency oscillations are a common problem in power systems due
to mechanical inertia and small scale disturbances like gradual change of load and
power. On the other hand, large disturbances may also present in the power system,

A. Pattnaik (B)
Department of EEE, SUIIT, Jyoti Vihar, Burla, Odisha 768019, India
e-mail: [email protected]
B. Rout
Department of EE, VSSUT, Burla, Odisha 768018, India
A. K. Patra
Department of EEE, Institute of Technical Education and Research, Siksha ‘O’ Anusandhan
(Deemed to be University), Bhubaneswar, Odisha 751030, India
© Springer Nature Singapore Pte Ltd. 2020 349
G. Pradhan et al. (eds.), Advances in Electrical Control and Signal Systems,
Lecture Notes in Electrical Engineering 665,
https://doi.org/10.1007/978-981-15-5262-5_25
350 A. Pattnaik et al.

which is closely related to dynamic behavior of the system. Persistence of these

oscillations and occurrence of disturbances in power system leads to the power sys-
tem instability. A small disturbance instability and the dynamics of power systems
are easily analyzed considering all the sub-systems of power system in first-order
transfer function. The cause of low-frequency oscillations instability and the remedy
has been carried out in [1]. Moreover, the stability of a power system is divided into
two broad classes such as small scale and transient system stability.
The capability of power systems to maintain synchronism and voltage levels
within their limits at all buses under small disturbances is called small-signal stability.
The capacity of system to reach an acceptable steady-state working condition for
a large disturbance or sequence of disorders is called transient stability.
The lack of insufficient damping torque caused by changing operating condi-
tions, the complex power systems exhibit low-frequency electromechanical oscilla-
tions [2]. The small disturbance due to low-frequency oscillations causes decrement
of generated terminal voltage and objection to power flow. The automatic voltage
regulator (AVR) maintains the generator voltage by regulating the field voltage of
the excitation system in synchronous machines [3]. The dynamic performance of
system voltage can be improved using classical controllers with AVR in power sys-
tem [4]. The controller parameters have been tuned using various methods such as
pole placement [5], Z-N tuning method [6], LQR [7, 8], Due to large computational
time and difficulties the optimum value of controller parameters are tuned with var-
ious heuristic optimization techniques such as genetic algorithms (GA) [9], particle
swarm optimization (PSO) [10], Differential Evolution (DE) [11–13], and many
recent optimization techniques.
In this paper, the comparison of system performance has been carried out with
Z-N tuned PID-type AVR and designed integral type LQR based AVR. Further,
the improved effectiveness of the above system performances is studied with DE
optimized PID controller. The paperwork has been done in MATLAB environment
and studied with time-domain analysis for achiveing better performances of the
system.

2 System Description and Modeling

2.1 Automatic Voltage Regulator

A simple linearized model of first-order AVR system of a synchronous generator in

the power system is shown in Fig. 1. The AVR keeps the generator terminal voltage
constant at various load levels under normal operating conditions. Therefore, the
objective is the designing of an AVR system for voltage regulation [4].
An AVR system consists of an amplifier, an exciter system, and a sensor associated
with generator model. All the system components are first order and the parameters
such as the AVR gain and the time constants of AVR system are referred in [10].
Comparative Study of System Performances … 351

VF
Vs KA VR KE KG
TA.s+1 TE.s+1 TG.s+1
Ref. Voltage AMPLIFIER EXCITER GENERATOR

[Vt]

Generator
Terminal
voltage
KR
TR.s+1
SENSOR

Fig. 1 Block diagram of AVR in power system

2.1.1 Amplifier Model

The first-order dynamic of the amplifier is represented by a gain K A and correspond-

ing time constant T A which is given by

VR (s) KA
= (1)
Ve (s) 1 + TA s

where the range of K A is the range of 10–400 and T A varies from 0.02 to 0.1 s.

2.1.2 Exciter Model

The excitation system controls the field voltage VF with the input as regulator voltage
VR . The first-order exciter is shown by

VF (s) KE
= (2)
VR (s) 1 + TE s

where the exciter gain K E varies from 1 to 400 and time constant TE from 0.5 to
1.0 s.

2.1.3 Generator Model

The synchronous generator is the important component of power system from which
the terminal voltage is available. However, the generator model is represented by the
first-order dynamic considering the terminal voltage and its field voltage is

Vt (s) KG
= (3)
VF (s) 1 + TG s
352 A. Pattnaik et al.

where the constant K G which depends on load may vary in the range of 0.7–1.0, and
1.0 and 2.0 s varies from full load to its no load.

2.1.4 Sensor Model

The sensor is provided in the feedback path to measure the generator terminal voltage
and is compared with a given reference voltage. The error in voltage is given to
the amplifier for amplification. The sensor model is shown by the transfer function
consisting of gain and time constant is

Vs (s) KR
= (4)
Vt (s) 1 + TR s

where K R is the sensor gain and time constant TR ranging from 0.001 to 0.06 s.

3 Control Structure

In this paper, a classical PID controller is considered with the AVR to decrease the
steady-state error and improvement of transient performances of system response.
Since the parameters of the system is known, the PID controller is useful in system
for analysis [5].
PID controller output is given below:

Ki
C(s) = K p + + Kd s (5)
s
In this paper, the PID-type AVR controller of power system is considered and as
shown in Fig. 2 to improve the damping and hence the stability. The control objective
is to study the dynamic responses. The controller gain K p , K i , K d for PID controller
are optimized with Ziegler–Nichols tuning method and DE algorithm separately.

4 Control Techniques

4.1 Ziegler–Nichols (Z-N) Tuning Method

The controller parameters of PID structure may be found from the Z-N empirical
first method of non-first order with delay time and second method of rigorous Routh–
Hurwitz of manual computations [6].
In this study, the second method of Z-N tuning analysis is used as the system
dynamics is known. Using the proportional control action only, the close loop transfer
Comparative Study of System Performances … 353

KP

Proportional Gain
10 1 1
[Vt]
KI 1/s 0.1s+1 0.4s+1 s+1
Vref Integral Integrator Amplifier Saturation Exciter Generator Generator Terminal
Gain Voltage

KD du/dt

derivative Derivative
Gain 1
0.01s+1
Voltage Sensor

Fig. 2 Control structure of PID-AVR

Table 1 Rule for Z-N tuning

Type of controller Kp Ti Td
method
P 0.5K cr ∞ 0
PI 0.45K cr 0.833Pcr 0
PID 0.6K cr 0.5Pcr 0.125Pcr

function of AVR system was calculated. Then using the Routh - Hurwitz criterion
K p is calculated. Then the critical gain K cr is calculated. The frequency of oscillation
is obtained from the auxiliary equation.
The period of sustained oscillation is obtained as Pcr = 2π ω
where ω in rad/s.
Here K cr and Pcr were found to be 1.7016 and 1.089 respectively.
By referring to Table 1 of second method of Z-N, K p , Ti , Td are determined.

4.2 Linear Quadratic Regulator (LQR)

The Linear Quadratic Regulator theory provides the optimal gain using pole place-
ment method [7]. The LQR method determines the feedback gain of infinite margin
which leads to guarantee stable system. The optimal gains were calculated by mini-
mizing the cost function through solving the Algebraic Riccati Equation (ARE). The
LQR method is implemented in this system since it is controllable and observable.
A controllable LTI system is represented in state-space as shown in Eq. (6)

ẋ(t) = Ax(t) + Bu(t)

y(t) = C x(t) (6)

where A, B, C are system matrix, input, and output matrix respectively and x(t), y(t),
and u(t) are the state, control, and output matrices, respectively. The design problem
354 A. Pattnaik et al.

is to develop a LQR control law (7) as:

u(t) = −K x(t) (7)

In this paper, the amplifier output, the field voltage, the generator voltage, and
the feedback sensor output are considered as system states. The control output is the
generator voltage which needs to be controlled through the proposed integral type
LQR controller. The control law minimizes the cost function of LQR as shown in
Eq. (8) which comprises a sum of mean square state control weight and mean square
control signal weight [8]. The control law is shown in Eq. (8) minimizes the LQR
cost function.
∞
 
J= x T Qx + u T Ru dt (8)
0

where Q qualifies the semi-definite matrix and R qualifies the symmetric, definite
control matrix.
The state feedback gain K can be calculated as

K = R −1 × B T × S (9)

S is the solution for the algebraic Riccati equation which can be represented as:

AT × S + S × A − S × B × R −1 × B T × S + Q = 0 (10)

The system matrix, input matrix, and output matrix are obtained through
MATLAB code without a controller as
⎡ ⎤ ⎡ ⎤
−113.5 −43.3594 −7.373 −2.4414 4
⎢ 32 0 0 0 ⎥ ⎢0⎥
A=⎢
⎣ 0
⎥ ; B=⎢ ⎥ ;
⎦ ⎣0⎦
16 0 0
0 0 2 0 4×4
0 4×1
C = 0 0 0 6.1035 1×4

After augmenting the integrator with the plant, the augmented plant becomes
 
A 0 B
Ā = ; B̄ = ; C̄ = C 0
−C 0 5×5
0 5×1
1×5

The LQR weights for the augmented system

 
Q = diag 1 1 1 1 100 ; R = 0.01
Comparative Study of System Performances … 355

1 x' = Ax+Bu [Vt]

-K- s y = Cx+Du K*u
Vref IntegralIntegrator Generator
AVR System states Output
Gain Terminal Voltage
Matrix

K*u

Feedback Gain

Fig. 3 Control structure of integral type LQR with plant

As the LQR has its own limitations, i.e., it cannot produce zero offset for which
an integrator-based LQR technique is implemented in AVR system to enhance its
performance. In Fig. 3 the feedback gain block consists of a row matrix K =
K 1 K 2 K 3 K 4 , whose values are given in Table 4.

5 Differential Evolution Algorithm

The DE is found as an effective and understandable optimization technique for opti-

mization problems [5]. DE involves mainly three operations, i.e., mutation, crossover,
and selection. The operations are described below.
Initialization:
Each controller parameters are initialized with a lower and upper bound in the form
of, X JL , X UJ where, where J is the parameter and X JL is the lower limit and X UJ is
the upper limit of controlling parameter.
Mutation:
Three random vectors are selected and a resultant donor vector is created from the
weighted difference for any two vectors and added with the third vector as:
 
Vi,G+1 = Vr 1,G + F ∗ X r 2,G − X r 3,G

where (r1 , r2 , r3 ) are distinct values and F varies from 0.2 to 0.8.
Crossover:
A trial vector δ is created from the mutant vector and target vector X i,J with
probability ratio CR.

V j,i,G+1 if rand j,i ≤ CR
U j,i,G+1 =
X j,i,G+1 if rand j,i ≤ CR

Selection:
If trial vector has better fitness than target, this is chosen as desired vector for next
generation.
356 A. Pattnaik et al.

The detailed flowchart and vector diagram of DE are given in Figs. 4 and 5,
respectively.
Simulation and Results Analysis:
Scenario 1: Figure 6 in scenario 1 shows the four various states such as amplifier
voltage, field voltage, sensor voltage and generator voltage in the proposed AVR
system with the proposed integral based LQR controller. It is observed from the
figure that three states are converging to its initial condition except for generator
voltage that is converging at 0.16 p.u. However, the settling time of all states is at
1.1 s.
Scenario 2: The AVR model was disturbed by a step input as a reference voltage.
The terminal voltage is considered as output and is investigated in Fig. 7. It can be

Start

Parameters of DE were specified

Population was initialized

Gen.=1
Population was evaluated

Offsprings were created and their fitness was

evaluated

No
Is offspring’s fitness better than
parent’s fitness
Y es O ffsprings in
new popula tion
Parents were replaced by offsprings w ere disc arded
in the new popula tion

Yes
N ew population size <
O ld popula tion siz e
G en. = G en+1
No

No
G en. > M ax. G en ?

Yes
Stop

Fig. 4 Flow diagram of DE algorithm

Comparative Study of System Performances … 357

Fig. 5 Generation of new candidate solution from DE vector

Sensor voltage
0.3
Field voltage
Amplifier voltage
0.2 Generator voltage
States (p.u.)

0.1

-0.1
0 0.5 1 1.5 2
Time (sec)

Fig. 6 Scenario 1: System states of AVR system

observed that the response of AVR system is oscillatory in nature, which implies as
unstable system. Whereas the use of PID controller and LQR improves the system
significantly. The DE-based PID controller gives better results as compare to the Z-N
based tuning method. As it brings the dominant poles to −3.38 ± 7.32i. It is observed
from Fig. 7 that LQR technique makes the system faster as compare to system with
no controller, but it gives rise to a large steady-state error and is sluggish in nature.
Due to this reason, an integrator-based LQR is used in this system. It can be observed
that the integrator-based LQR produces zero offset and has better convergence char-
acteristics than the rest of other controllers following optimal control law. The LQR
techniques performance depends upon the Q and R weighting coefficients. In this
study, it has been observed that if weight coefficients are increased or decreased
from the value mentioned above in matrix then it hampers the system performance
by making the system sluggish or sometimes making the transient response worse.
358 A. Pattnaik et al.

2
LQR
No Controller
Z-N based PID Controller
1.5
DE optimized PID Controller
Integral based LQR
V t (p.u.)

0.5

0
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
Time in (sec)

Fig. 7 Scenario 2: Generator terminal voltage of AVR system with LQG control techniques and
PID controllers

So it’s very important to choose the Q and R matrix carefully. The quadratic function
of performance index J following Riccati equation is minimized by selecting the
proper values of Q and R (Tables 2, 3 and 4).

Table 2 Time-domain specifications of AVR system with different techniques

Control techniques Dominant eigen values Damping ratio (Zeta) Settling time in (s)
Z-N −1.23 ± 4.35i 0.272 3.71
DE −3.38 ± 7.32i 0.420 1.1
LQR with integrator −5.97 ± 8.45i 0.578 0.816
Note The bold letters indicates the best performnace among all controllers

Table 3 Controller parameters of AVR PID System with Z-N and DE tuning methods
Tuning methods Kp Ki Kd ITAE
X10e−3
Z-N 1.02096 1.875 0.1389 6.4
DE 1.5041 0.3946 0.2950 2.3
Note The bold letters indicates the best performnace among all controllers

Table 4 Gain of Proposed Control Technique

Proposed controller gain K1 K2 K3 K4 Integral gain
Integral type LQR 8.1936 27.0074 34.8523 149.3816 −100.00
Comparative Study of System Performances … 359

6 Conclusion

In this simulation work, the time-domain analysis is carried out for AVR system
under the MATLAB platform. The controller parameters of PID-type AVR controller
are optimized using Z-N tuning methods and differential evolution algorithms. The
system was also implemented with integral type LQR control technique and all the
results are compared with each other. The proposed integral type LQR controller
produces large negative eigenvalues and results in quick damping for the system
responses than optimized PID controller. In summary, LQR with integrator technique
is suitable for the system and is more effective for stabilizing than other controllers.
This proposed integral based LQR control technique may be further implemented in
multi-machine infinite bus (MMIB) system.

References

1. Demello, F.P., Concordia, C.: Concepts of synchronous machine stability as affected by

excitation control. IEEE Trans. Power Appar. Syst. 88(4), 316–329 (1969)
2. Padiyar, K.R.: Power System Dynamics Stability. BS Publications
3. Ghfarokhi, G.S., Arezoom, M., Mahmoodian, H.: Analysis and simulation of the single-
machine infinite-bus with power system stabilizer and parameters variation effects. In:
International Conference on Intelligent and Advanced System (2007)
4. Khalid, A., et al.: Comparative assessment of classical and adaptive controllers for automatic
voltage regulator. In: International Conference on Advanced Mechatronic Systems, Melbourne,
Australia, pp. 538–543, 30 Nov 30 2016–3 Dec 2016
5. Boyd, S., Barratt, C.: Linear Controller Design: Limits of Performance. Prentice-Hall
publication
6. Ogata, K.: Modern Control Engineering, 4th edn. PEARSON Prentice Hall
7. Pradhan, R., Rout, B.: Enhancement of stability of SMIB using ZN-PID and LQR-PSS
as affected by excitation control. In: 2015 IEEE Power, Communication and Information
Technology Conference (PCITC) Siksha ‘O’ Anusandhan University, Bhubaneswar, India
8. Bandal, V., Bandyopadhyay, B.: Robust decentralised output feedback sliding mode control
technique-based power system stabilizer (PSS) for multimachine power system. IET Control
Theory Appl. 1(5), 1512–1522 (2007)
9. N. Nimpitiwan, and C. Chaiyabut, Centralized Control of System Voltage/Reactive Power
Using Genetic Algorithm
10. Gaing, Z.L.: A particle swarm optimization approach for optimum design of PID controller in
AVR system. IEEE Trans. Energy Convers. 19(2) (2004)
11. Storn, R., Price, K.: Differential evolution—a simple and efficient heuristic for global
optimization over continuous spaces. J. Glob. Optim. 341–359 (1997)
12. Panda, S.: Differential evolution algorithm for SSSC-based damping controller design
considering time delay. J. Franklin Inst. 348, 1903–1926 (2011)
13. Sun, Z., Wang, N., Srinivasan, D., Bi, Y.: Optimal tuning of type-2 fuzzy logic power system
stabilizer based on differential evolution algorithm. Electri. Power Energy Syst. 62, 19–28
(2014)
Voltage Control Method of Isolated Wind
Power System

Gyana Ranjan Biswal and Banaja Mohanty

Abstract This paper deals with the voltage controller of an isolated wind power
system consisting of asynchronous generator driven by wind turbine. The proposed
voltage controller consists of IGBT switch-based voltage source converter along
with battery energy storage system at its DC link. The proposed controller controls
the system voltage with variation of consumer lagging loads with the bidirectional
active and reactive power flow capability. The proposed electro-mechanical system
along with its controller is modeled and simulated in MATLAB using Simulink and
SimPowerSytem block-set. Performance of the proposed controller is demonstrated
by taking different electrical load for the voltage control of a wind generator system
driven by wind turbine.

Keywords Asynchronous induction generator · Wind turbine · AC voltage

controller · Voltage source converter · Battery ESS

1 Introduction

Wind energy is one of the most important and promising sources of renewable energy
all over the world, mainly because it is treated to be non-polluting and economically
viable source of energy. Wind is a free, clean and inexhaustible energy source. Induc-
tion generators are widely preferable in wind energy conversion system because of its
robust construction, low maintenance requirements, low cost, self protection against
short-circuits.
In this paper, self-excited asynchronous generator (squirrel-cage type) has been
considered with the wind turbine for the electricity generation [1–3], to supply power
where grid supply is not available. We know that the voltage at the generator terminal

G. R. Biswal (B) · B. Mohanty

Department of Electrical Engineering, Veer Surendra Sai University of Technology (VSSUT),
Burla, Odisha 768018, India
e-mail: [email protected]
B. Mohanty
e-mail: [email protected]

© Springer Nature Singapore Pte Ltd. 2020 361

G. Pradhan et al. (eds.), Advances in Electrical Control and Signal Systems,
Lecture Notes in Electrical Engineering 665,
https://doi.org/10.1007/978-981-15-5262-5_26
362 G. R. Biswal and B. Mohanty

is a function of reactive power. In some applications, where prime mover provides

constant speed, the speed of isolated asynchronous generator (IAG) remains constant,
while the terminal voltage varies as per the variation in load because of the change
in reactive power requirement. In this case, the recommended solution is reactive
power compensators.
In some other applications, where input power is constant, frequency and voltage
vary according to the load change, the recommended solution for this type of appli-
cations is electronic load controllers, which will maintain the frequency by main-
taining the constant power at load terminal and maintain the voltage by maintaining
the excitation of the generator.
Now, in wind power applications [4], both speed (wind speed) and input power
(wind power) vary, so there will be variation of both voltage and frequency at the
terminal of the IAG. This paper shows the application of voltage controller for an
isolated wind power system, where a self-excited induction generator, preferably
squirrel-cage type, is connected. This overcomes the limitation of unavailability of
grid in rural/remote areas by maintaining the terminal voltage with respect to the
balanced/unbalanced load change.

2 Complete System Configuration

There are different configurations of voltage controller (VC) [5–7] are available
based on three-phase three-wire IAG system and three-phase four-wire IAG system.
Again, these classification can be subclassified based on which type of voltage source
converter (VSC) is being used in the proposed system.
The diagram shown in Fig. 1 depicts the voltage and frequency control of an
isolated wind energy conversion system [8] with excitation capacitor and connected to
three-phase three-wire balanced/unbalanced and linear loads. The voltage controller
(VC) mainly consists of a three-leg VSC with battery energy storage system (BESS)
[9]. Each part of the above schematic diagram is explained further as follows:

Asynchronous Generator
An induction machine can be used as motor and generator. In this case, an asyn-
chronous induction generator with squirrel-cage type rotor is taken which is self-
excited also. It is more appropriate to use this in wind power generation in remote
areas as no external excitation method is needed to generate the magnetic field. The
stator is three-phase star connected winding, across which the electricity is generated.

Excitation Capacitor Bank

As we know, capacitor provides reactive power, so in this case, the capacitor bank
will provide the additional reactive power requirement for the IAG to generate rated
voltage at the terminal .
Voltage Control Method of Isolated Wind Power System 363

Fig. 1 Schematic diagram of isolated wind system connected with VC and three-phase three-wire
load

Wind Turbine
To provide the mechanical power/to rotate the shaft of the induction generator [8],
the wind turbine is used, whose output power can be expressed as follows:

ρA 3
Pm = C p (λ, β) v
2 wind
where Pm = Mechanical output power of the turbine (W),
C p = Performance coefficient of the turbine,
ρ = Air density (kg/m3 ), A = Turbine swept area (m2 ),
vwind = Wind speed (m/s),
λ = Tip speed ratio of the rotor blade tip speed to wind speed,
β = Blade pitch angle (deg).

Three-leg VSC
It consists of a voltage source converter with three legs; each leg consists of two
switches as IGBT. This switch has the characteristic of both turn-on and turn-off can
be controlled. The gate signal is given to these switched through a two-level PWM
pulse generator. A DC link capacitor is connected at the input side of the converter.

Battery Energy Storage System (BESS)

It acts as an energy storage unit. It consists of a battery with open-circuit voltage of
V oc and a series resistance Rin . The parallel circuit of RB and C B is used to describe
the stored energy and voltage during charging or discharging. The parallel connection
of RB and C B shows the self-discharging of the battery.

Load
This load can be of different type like linear, nonlinear, balanced, unbalanced and
364 G. R. Biswal and B. Mohanty

dynamic load. A three-phase three-wire resistive and inductive load is taken in this
paper.

3 Control Strategy

In the proposed voltage control strategy, as shown in Fig. 2, the main objective is to
generate reference source currents (i sa r
, i sb
r
, i sc
r
). This reference current is compared
with the actual generation currents (i sa , i sb , i sc ), and the error is used for the gener-
ation of switching signals for the VSC. Now, this reference source current has two
components: one is active components (i da r
, i db
r
, i dc
r
), which will regulate the genera-
tion frequency; another is reactive components (i qa , i qb r r
, i qc
r
), which will regulate the
magnitude of terminal voltage.
To generate active component of source current
The line voltages (vla , vlb , vlc ) from the generator output terminal are used to calculate
the terminal voltage V tm . Then, unit amplitude templates (da , db , dc ) are calculated
by dividing line voltages with the V tm . Simultaneously, from the line voltage and
currents, the instantaneous power (PL ) is calculated. Now, the amplitude of active
power component of source current (I dm ) is calculated by dividing the difference of
filtered load power (PLfilter ), from the output of the second order filter, and Pc , which

From AC generation
Voltages

Vla Vlb Vlc

Computation of da Estimation of active

In-phase Component of
Unit Vectors db Reference source
dc current Idm Vtm
da db dc

Vtm Computation of quadrature

Unit Vectors r
- irda i db ir
+ qa qb qc dc
2nd order Instantaneous Vlabc
Vtmref PLfilter Active load power
irqa + Low
ilabc
Estimation of reactive + - Pass Filter estimation
Component of irqb
AC Voltage
Reference source +
PI Controller
Iqm current irqc PC

irsa irsb irsc Frequency

f
f to Pc measurement
Using PLL
PWM Signal
Generation
(SPWM)
Vlabc

ib
Rf Lf
RB CB
To generator
Terminals
Vb
Rs
Cdc

Voc

Voltage Source Battery Source

Converter

Fig. 2 Voltage control scheme of isolated wind power system

Voltage Control Method of Isolated Wind Power System 365

is from the measured frequency, to the terminal voltage. Now, the active components
of source current are found by multiplying the I dm with the unit templates.
To generate reactive component of source current
The quadrature unit template (qa , qb , qc ) is found by 90° phase shift of the in-phase
unit templates. Now, the reactive component of source current is found by multiplying
the quadrature templates with the output of PI AC voltage controller (I qm ).

4 Modeling of Components

4.1 AC Voltage PI Controller

It takes the input as voltage error, found from the difference of actual terminal voltage
and reference terminal voltage, and gives the output I qm , which is further used to
generate the reactive component of source current. It also uses a proportional and
integral (PI) controller with their corresponding gains. The block diagram is shown
in Fig. 3.

4.2 Generation Active Component of Source Current

The instantaneous active power is calculated as PL = vα i α + vβ i β , where vα , vβ , iα

and iβ can be found using following matrix:

  √  ⎡ v ⎤
−1 −1 la
vα 2 1 ⎣ vlb ⎦
= √2 √
2
3 − 3 (1)
vβ 3 0 2 2 vlc

Fig. 3 Block diagram of AC voltage PI Controller

366 G. R. Biswal and B. Mohanty

  √  ⎡ i ⎤
−1 −1 la
iα 2 1 ⎣ ilb ⎦
= √2 √
2
3 − 3 (2)
iβ 3 0 2 2 ilc

The terminal voltage is found as follows:

 
2 2
Vtm = v + vlb
2
+ vlc
2
(3)
3 la

The amplitude of active component of source current is found as follows:

2 (PLfilter − PC )
Idm = × (4)
3 Vtm

After this, the active (or in-phase) component is calculated by multiplying the
in-phase unit template with the I dm .

4.3 Generation Reactive Component of Source Current

The quadrature unit template can be found by following matrix expression:

⎡ ⎤
⎡ ⎤ 0 −1 √1
√ ⎡ ⎤
qa ⎢ √3 3 3 ⎥ da
⎣ qb ⎦ = ⎢ √1 −1
√ ⎥⎣ db ⎦ (5)
⎣ √2 2 3 2 3 ⎦
qc − 3 √1 −1
√ dc
2 2 3 2 3

Now, the reactive (or quadrature) component of reference source current can be
found by multiplying I qm (output of AC voltage PI controller) with the quadrature
unit templates.
Finally, the addition of in-phase and quadrature components gives rise to the
r
reference source current (i sa , i sb
r
, i sc
r
).

4.4 Switching Signal Generation

Based upon the comparisons between the actual generation current (i sa , i sb , i sc ) with
r
the reference source current (i sa , i sb
r
, i sc
r
), obtained from the control loops, error
source current signal is generated. These current errors are again compared with
a carrier signal having frequency 20 kHz, to generate pulse signals. These pulse
signals will be given to the gate terminals of the IGBT switches connected in the
Voltage Control Method of Isolated Wind Power System 367

voltage source converter (VSC), in the controller. After this, the output of the VSC is
filtered and connected with the generation terminals through the limiting inductors
and resistors (Rf , L f ).

5 Brief Description of Proposed System

The modeling of proposed wind energy system including IAG, wind turbine and
voltage controller is carried out in the MATLAB/Simulink platform with version
R2016a. All the components/blocks from the Simscape/Simpower system are used
in the system. The brief idea about the modeling is given for some of the components.

5.1 Modeling of Asynchronous Generator

An asynchronous generator of rating 15 kW, 400 V, 50 Hz is chosen with squirrel-

cage type rotor. In the output of this, one delta connected capacitor bank is connected
which will act as excitation capacitor [10–12] and maintain the no-load terminal
voltage across the generation terminals with the proper adjustment of its VAR value.

5.2 Modeling of Wind Turbine

A wind turbine model is taken which will act as a driving section for the asynchronous
machine. For the time-being, the wind speed (vw ) is taken constant, and blade pith
angle (β) is taken as zero. According to the speed of the generator, with the help of
wind speed, it will generate the electromagnetic torque which will be fed to the IAG
[4].

5.3 Modeling of Voltage Controller

Basically, the proposed voltage controller consists of IGBT-based VSC and the bat-
tery storage system [7, 9]. The switching technique of the IGBTs of VSC uses
simple PWM signal generation technique [13, 14], which is nothing but comparison
of source current error signal with the carrier signal. As battery is a storage unit,
the Thevenin equivalent [an open-circuit voltage (V oc ) with a series resistor (Rs )]
of the battery is used. Also, a parallel circuit of RB and C B is used to describe the
charging/discharging process and also the stored energy [15].
368 G. R. Biswal and B. Mohanty

5.4 Modeling of Load

In this case, a three-phase linear resistive load and inductive is used, which can be
extended to different other type load like nonlinear, dynamic, inductive loads later
part. Here, the load is added to the circuit only after a specific time with the help of
circuit breaker in the circuit.

800
Vabc

0
-800
1.4 1.45 1.5 1.55 1.6 1.65 1.7 1.75 1.8 1.85 1.9

20
Iabc

0
-20

1.4 1.45 1.5 1.55 1.6 1.65 1.7 1.75 1.8 1.85 1.9
Vtm

800
740
1.4 1.45 1.5 1.55 1.6 1.65 1.7 1.75 1.8 1.85 1.9
Vw

1.4 1.45 1.5 1.55 1.6 1.65 1.7 1.75 1.8 1.85 1.9

50
Ilabc

0
-50

1.4 1.45 1.5 1.55 1.6 1.65 1.7 1.75 1.8 1.85 1.9
Ib

950

1.4 1.45 1.5 1.55 1.6 1.65 1.7 1.75 1.8 1.85 1.9

742
Vb

1.4 1.45 1.5 1.55 1.6 1.65 1.7 1.75 1.8 1.85 1.9
4
10
PL & PLfilter

1.4 1.45 1.5 1.55 1.6 1.65 1.7 1.75 1.8 1.85 1.9

Fig. 4 Simulation transient waveforms after the application of resistive load

Voltage Control Method of Isolated Wind Power System 369

6 Result and Discussion

The proposed voltage controller is studied and simulated using MATLAB, and the
performance is studied for different load conditions. Figures 4 and 5 show the con-
troller performance of wind energy system supplying power to linear, balanced resis-
tive load and inductive load, respectively. Both the figures show the transient wave-
forms of generator voltage (V abc ), generator current (I abc ), load current (I labc ), termi-
nal voltage (V tm ), speed of the wind in m/s (V w ), battery current (ib ), battery voltage
(V b ) and instantaneous active load power and filtered power (PL and PL filter).

6.1 Performance of Controller for Static Resistive Load

At 1.5 s, a 15 kW resistive load is applied to the system, and it is observed from

Fig. 4 simulated waveform that, after sudden disturbance, the generation voltage,

800
Vabc

0
-800
1.4 1.45 1.5 1.55 1.6 1.65 1.7 1.75 1.8 1.85 1.9

20
Iabc

-20

1.4 1.45 1.5 1.55 1.6 1.65 1.7 1.75 1.8 1.85 1.9
Vtm

800
740
1.4 1.45 1.5 1.55 1.6 1.65 1.7 1.75 1.8 1.85 1.9
Vw

1.4 1.45 1.5 1.55 1.6 1.65 1.7 1.75 1.8 1.85 1.9

60
Ilabc

0
-60

1.4 1.45 1.5 1.55 1.6 1.65 1.7 1.75 1.8 1.85 1.9
Ib

900

1.4 1.45 1.5 1.55 1.6 1.65 1.7 1.75 1.8 1.85 1.9

742
Vb

1.4 4 1.45 1.5 1.55 1.6 1.65 1.7 1.75 1.8 1.85 1.9
10

1.4 1.45 1.5 1.55 1.6 1.65 1.7 1.75 1.8 1.85 1.9

Fig. 5 Simulation transient waveforms after the application of inductive (R-L) load
370 G. R. Biswal and B. Mohanty

generation current and the terminal voltage remain constant. This shows the load
balancing nature of the controller.

6.2 Performance of Controller for Static Inductive Load

To the same system, at 1.5 s, a 15 kW, 0.8 pf lagging load of three-phase is connected
with the system, and the simulation is as shown in Fig. 5.
The simulated transient waveform shows that after the application of inductive
load at 1.5 s for a fixed wind speed, after a very little disturbance, the generation
voltage, generation current and the terminal voltage remain constant thereafter.

7 Conclusion

The voltage controller is designed, and its performance is studied for isolated wind
power system containing an asynchronous slip-ring generator. Also, the simulation
transient waveform is examined, from which it is clear that the proposed voltage
controller maintains the generation and terminal voltage as well as the generator
current for different types of load like purely resistive load and inductive type load
discussed here. This work can be further extended to variation of wind speed, and
also, the battery storage can be replaced by new storage technologies.

Appendix

A 20 hp, 400 V, 50 Hz, 1460 rpm, 4-pole Asynchronous generator with squirrel-cage
type rotor: Rs = Rr = 0.02 pu, L s = L r = 0.03 pu, L m = 1.89, H = 0.08 s.
Wind Turbine: 15 kW, 8 m/s, pitch angle = 0.
Load: Resistive load—15 kW, unity pf; Inductive load—15 kW, 0.8 pf lagging.
Battery Specification: V oc = 750 V, Rin = 0.01 , C dc = 4 mF.
Voltage PI Controller: K p = 0.9, K i = 0.8, Rf = 0.45 , L f = 1.2 mH.
PWM Signal Generation: f carrier = 20 kHz, m = 1, amplitude = 1 V.

References

1. Wang, D., Ma, W., Xiao, F., Zhan, B., Liu, D., Hu, A.: A novel standalone dual stator winding
induction generator with static excitation regulation. IEEE Trans. Energy Convers. 20(4), 826–
835 (2006)
Voltage Control Method of Isolated Wind Power System 371

2. Schinas, N.A., Vovos, N.A., Poulos, G.B.G.: An autonomous system supplied only by a pitch-
controlled variable speed wind turbine. IEEE Trans. Energy Convers. 22(2), 325–331 (2007)
3. Bansal, R.C.: Three phase self excited induction generators: an overview. IEEE Trans. Energy
Convers. 20(2), 292–299 (2005)
4. Szabo, L., Biro, K.A., Nicula, C., Jurca, F.: Simulation of wind turbine driven autonomous squir-
rel cage induction generators. In: 2007 11th International Conference on Intelligent Engineering
Systems, pp. 213–218 (2007)
5. Aderson, M.D., Carr, D.S.: Battery energy storage technology. Proc. IEEE 81(3), 475–479
(1993)
6. Ceraola, M.: New dynamic models of lead-acid batteries. IEEE Trans. Power Syst. 15(4),
1184–1190 (2000)
7. Kansal, G.K., Singh, B.: Voltage and frequency controllers for an asynchronous generator-
based isolated wind energy conversion system. IEEE Trans. Energy Convers. 26(2), 402–416
(2011)
8. Bansal, R.C., Bhatti, T.S., Kothari, D.P.: On some of the design aspects of wind energy
conversion systems. Energy Convers. Manag. 43, 2175–2187 (2002)
9. Singh, B., Kansal, G.K.: Solid State Voltage and Frequency controller for a stand-alone wind
power generating system. IEEE Trans. Power Electron. 23(3), 1170–1177 (2008)
10. Das, D.C., Sriramoju, H., Ranjan, S., Sinha, N.: Voltage control of fuel cell-wind-diesel hybrid
power system using FA based SVC and AVR controller. In: IEEE Region 10 Humanitarian
Technology Conference, Dhaka (2017)
11. Singh, B., Kansal, G.K., Chandra, A.: Capacitor excited asynchronous generator for an iso-
lated wind energy conversion system. In: Joint International Conference on Power System
Technology and IEEE Power India (2008)
12. Amin, A.A., Tawfiq, K.B., Youssef, H., El-kholy, E.E.: Performance analysis of inverter fed
from wind energy system. In: Eighteenth International Middle East Power Systems Conference
(MEPCON), IEEE (2016)
13. Rashid, M.H.: Power Electronics, Circuits, Devices and Applications, 3rd edn. Pearson Prentice
Hall (2004)
14. Mohan, N., Undeland, T.M., Robbins, W.P.: Power Electronics: Converters, Applications and
Design, 3rd edn. Wiley, Singapore (2004)
15. Lopes, L.A.C., Almeida, R.G.: Wind-driven self excited induction generator with voltage and
frequency regulated by a reduced rating voltage source inverter. IEEE Trans. Energy Converters
21(2), 297–304 (2006)
Hybrid Sliding Mode Based Simplified
NFC for Fuel Cell-Powered Linearized
IM Drive
Rabi Narayan Mishra, Kanungo Barada Mohanty, Abhimanyu Sahu,
and Partha Sarathi Behera

Abstract This paper discusses the design of a fuel cell system operated hybrid
sliding mode based simplified neuro-fuzzy control (NFC) for feedback linearized
induction motor (IM) drive. The proposed sliding mode simplified NFC (SMSNFC)
with intuitive feedback linearization (FBL) extensively reduces torque ripple and
gives optimal performance. This proposed technique has also the high computational
efficiency over conventional SMNFC and thus can easily be applied for industrial
applications. A fuel cell followed by a boost regulator is treated as an external source
during power failure and maintain the supply to IM drive to improve the efficiency
of the system. Extensive simulation results with its analysis are investigated and it is
observed that the system is robust and gives an enhanced performance.

Keywords Feedback linearization (FBL) · Fuel cell · Induction motor (IM) ·

Sliding mode simplified neuro-fuzzy control (SMSNFC)

1 Introduction

Over the years, the most commonly used photovoltaic (PV) systems have been used
for applications as an external source in the area of sustainable energy [1]. In spite
of many advantages, the solar system has uniform irradiation and temperature which
are the major disadvantages. Therefore, an energy storage system (ESS) based fuel
cell (FC) integrated with feedback linearization (FBL) induction motor (IM) drive
incorporated with different control techniques have been designed here as a stand-
alone application. This hybrid system improves the efficiency of the motor drive as it
has the properties of injecting extra power during motor acceleration, energy regen-
erated during braking [2]. The clean and continuous FC as a stand-alone application
delivers an environmentally-friendly source.

R. N. Mishra (B) · P. S. Behera

Silicon Institute of Technology, Bhubaneswar 751024, India
e-mail: [email protected]
K. B. Mohanty · A. Sahu
National Institute of Technology, Rourkela 769008, India
© Springer Nature Singapore Pte Ltd. 2020 373
G. Pradhan et al. (eds.), Advances in Electrical Control and Signal Systems,
Lecture Notes in Electrical Engineering 665,
https://doi.org/10.1007/978-981-15-5262-5_27
374 R. N. Mishra et al.

One of the important stand-alone applications of IM drive is its inverter control.

But, industries require certain norms for implementation of adjustable drive with
high-performance. These are extensive torque chattering reduction, quick dynamic
response, better response to load, versatile with different operating modes, and insen-
sitivity to and system uncertainties and parameter variations [3]. Throughout the
years, IMs control has been a vital area of research in academia and in many indus-
tries. However, the implementation of controllers is bit difficult due to its nonlinear
dynamics [4]. Many robust control techniques like FBL control, variable structure
control were also applied in the drives field [5, 6]. The FBL with a classical PI has the
drawbacks of ripple, load disturbance, and parameter uncertainties resulting in weak
decoupling behavior and poor dynamic response [7, 8]. To overcome this, sliding
mode control (SMC) technique has been implemented in FBL IM drive leading to
robust performance even with system uncertainties [9]. But, with this, the chattering
phenomena is introduced which is the major drawback of SMC. These issues of
controllers are overcome by introducing soft computing approaches as it is based on
human knowledge of operating IM and thus making the plant independent of plant
model and robust.
In recent years, many research works have been carried out on Fuzzy SMC
(FSMC) with IM for performance enhancement. The adaptive tuned fuzzy logic
control (FLC) with the robust classical SMC has been proposed as an efficient adap-
tive hybrid FSMC and thus, this has been applied successfully in many industries
like adjustable drives, robotics, etc. [3, 10].
Further, hybrid neuro-fuzzy control (NFC) technique is applied in control system
to overcome the drawback of fuzzy system and neural network while handling the
IM as an adjustable speed drive [11–15]. However, some industries are still unwilling
to use this controller as large computational burden is imposed by more membership
functions (MFs), rules, particularly for self-tuning conditions [16, 17]. The large
sampling time because of high computational burden is not preferred for real-time
industrial applications as it produces greater torque ripple.
Usually, power is supplied to IM from a standard grid, but during the disruption
of power and voltage disturbance, the power is supplied to IM directly from DC form
as a stand-alone application to make the system more efficient. This paper proposes
a new hybrid FC-ESS powered FBL based IM drive using an adaptive sliding mode
simplified neuro-fuzzy control (SMSNFC) technique. Here, the optimized parame-
ters of simplified NFC (SNFC) and the adaptive technique are based on the sliding
mode surface. The simulated responses prove that the system is effective, robust, and
free from chattering with this SMC incorporated with the modified NFC.

2 Modeling of Fuel Cell-ESS Integrated with FBL IM Drive

Energy storage system (ESS) based fuel cell (FC) system incorporated with linearized
IM drive is shown in Fig. 1a. A boost converter is connected to primary power source
FC to increase the voltage level to a fixed DC voltage level. A battery as an ESS is
Hybrid Sliding Mode Based Simplified NFC … 375

(a)
Fuel Cell Boost Converter

E=Eoc-NAln(ifc/i0)
Eoc, i0, A Rint ifc

L D
ifc
E
Vfc S1 C1 Vc1

(b)
Fig. 1 a SMSNFC based IM drive with hybrid FC system, b FC system with a boost regulator

connected across DC link which supports FC during different working modes of

drive. Lastly, whatever energy is not absorbed by battery is dissipated through the
braking chopper during braking.
376 R. N. Mishra et al.

2.1 Hybrid FC System with Boost Converter

The fuel cell (FC) is based on an electrochemical process that generates DC-voltage
[18]. Proton exchange membrane (PEM) is one of the frequently used FC which has
some special characteristics of like high power at reduced temperature. The efficiency
of FC is observed to be 50% during normal operation. The FC produces reduced DC
voltage and regenerative braking are not possible as its output is positive DC.
So, an FC-ESS system is provided as mentioned to offer regenerative braking and
fast transient. The FC output is increased by a conventional dc/dc boost converter
[19] to bring the voltage level to inverter fed IM to a standard DC link voltage as
shown in Fig. 1b.
The equivalent circuit of the simplified FC model is shown in Fig. 1b. Under
normal pressure and temperature, this FC stack model is given by
 
V f c = E oc − N A ln i f c /i 0 − i f c Rint (1)

where V fc , ifc , E oc, i0 , Rint , N, and A are the FC output voltage and current, open-
circuit voltage, exchange current, internal resistance, number of cells, and Tafel slope,
respectively.
The V-I curve and P-I curve of the FC are shown in Fig. 2 under normal conditions.

Fig. 2 V-I and P-I curve for FC system

Hybrid Sliding Mode Based Simplified NFC … 377

3 Modeling of Feedback Linearized IM Drive

Feedback linearization (FBL) control is a linear control approach where the system
with nonlinear dynamics can be implemented efficiently. This is absolutely different
from classical linearization technique [20]. The following equations describe the IM
dynamics in the stationary d-q reference [4].

dψdr Rr L m Rr
= − ψdr − Pωr ψqr + i ds (2)
dt Lr Lr
dψqr Rr L m Rr
= − ψqr + Pωr ψdr + i qs (3)
dt Lr Lr
dωr B 1
= − ωr + (Te − Tl ) (4)
dt J J
3 Lm  
Te = P ψdr i qs − ψqr i ds (5)
2 Lr

where (ids , iqs ), (ψ ds , ψ qs ), (idr , iqr ), (ψ dr , ψ qr ) are the stator d-q current, stator d-
q flux, rotor d-q current, rotor d-q flux, respectively. (Rs , Rr ), (L s , L r , L m ) are the
stator and rotor resistances, stator and rotor inductances, and magnetizing  inductance,
L 2m
respectively. σ is the leakage coefficient having value σ = 1 − L s L r . The rotor
speed, torque, inertia, and friction coefficient, number of pole pairs and load torque
are represented by ωr , T e , J, B, P, T l , respectively.
Here, the selected output controlled parameter is given so that the IM behaves
like a DC motor with the decoupling speed and flux.

Y T = [ωr ψr ] (6)

Now the total rotor flux is expressed as

ψr2 = ψdr
2
+ ψqr
2
(7)
 
dψr 1 dψdr dψqr
= ψdr + ψqr (8)
dt ψr dt dt

The Eqs. (2), (3) and (4) are given by following equations when the stator currents
are directly controlled by a current controller,

dψdr Rr L m Rr ∧
= − ψdr − ωr ψqr + u1 (9)
dt Lr Lr
dψqr Rr L m Rr ∧
= − ψqr + ωr ψdr + u2 (10)
dt Lr Lr
378 R. N. Mishra et al.

dωr B 1 3 Lm  ∧ ∧
 1
= − ωr + P u 2 ψdr − u 1 ψqr − Tl (11)
dt J J 2 Lr J

Now from (8), (9) and (10), the state space linearized equations of rotor flux and
speed are obtained as

dψr Rr L m Rr
= − ψr + u1 (12)
dt Lr Lr
dωr B 1 3 Lm 1
= − ωr + Pu 2 − Tl (13)
dt J J 2 Lr J

Here (9), (10) and (11) can be written as linearized equations by the new control
inputs u1 and u2.
 ψdr ψqr ∧
u1 ψr ψr u1
= ∧ (14)
u2 −ψqr ψdr u2

From (14), the decoupled system derived from the nonlinear state controller is
given by
 ∗
 
i ds 1 ψdr −ψqr 1 0 u1
∗ = (15)
i qs ψr ψqr ψdr 0 ψ1r u2

The new defined inputs u1 and u2 as given below are the PI-controller outputs for
accurate rotor flux and speed tracking and stable error dynamics.

u 1 = K p1 (ψr ∗ −ψr ) + K i1 (ψr ∗ −ψr )dt (16)

0
t

u 2 = K p2 (ωr ∗ −ωr ) + K i2 (ωr ∗ −ωr )dt (17)

0

4 Design of Adaptive Sliding Mode with the Modified

Neuro-Fuzzy Controller

4.1 Design of SNFC

The design of the proposed simplified NFC (SNFC) as shown in Fig. 3a combines
Hybrid Sliding Mode Based Simplified NFC … 379

Fig. 3 NFC: a Modified simple structure b classical structure and c input MFs of SNFC

FLC with a four-layer ANN structure, unlike classical NFC as in Fig. 3b. Here, the
normalized error in speed is given by

ωr∗ − ωr
%es (t) = × 100 (18)
ωr∗

where ω*r and ωr are the command and measured speed, respectively.
Layer 1: This is a fuzzification layer with negative (N), zero (Z), and positive (P)
speed errors as the membership functions (MFs) and is represented by the adaptive
nodes.
⎧
⎪
⎨ 1,1 xi1 ≤ b1
−a
O11 = μ M1 (es (t)) = bi −a1 , b1 < xi1 < a1
x
(19)
⎪
⎩ 1 1 1
0, x i ≥ a1
  1
0, x  ≥ b2
|xi1 −a2 |  1 
i
O2 = μ M2 (es (t)) =
1
(20)
1− b2
, x − a2  ≤ b2
i
380 R. N. Mishra et al.
⎧
⎪
⎨ 0,1 xi1 ≤ a3
xi −a3
O31 = μ M3 (es (t)) = , a3 < xi1 < b3 (21)
⎪
⎩
b3 −a3
1, xi1 ≥ b3

Here to reduce the computational burden, symmetrical (a2 = 0) linear MFs

are chosen as in Fig. 3c instead of exponential function and are represented by
μ M1 (es (t)), μ M2 (es (t)), μ M3 (es (t)). Here, the layer and node of input (x) and output
(O) are denoted by its superscript and subscript, respectively.
Layer 2: Unlike two input classical NFC, the “AND” operator is not required here
for the weight calculation of the rules wi and can be written as
wi
Oi2 = wi =  , i = 1, 2, 3 (22)
i wi

Layer 3: The calculation of consequent value vi is done here, whose output is

given by

Oi3 = wi vi , i = 1, 2, 3 (23)

Layer 4: This is the final defuzzification layer, where the output is given by

wi vi 
v = Oi4 = i = wi vi , i = 1, 2, 3 (24)
i wi
i

4.2 Design of Sliding Mode Based SNFC

The structure of the proposed sliding mode based simplified NFC (SMSNFC) inter-
facing with FBL IM drive is shown in Fig. 4. Here, unlike classical NFC, single
input speed error is input to the fuzzifier block of the fuzzy sliding mode control.
The sliding surface S as defined below [9], is so controlled that the speed error (e)
and change in speed error (e) plot drives the system to slide through the surface.

Fig. 4 Proposed SMSNFC

structure
Hybrid Sliding Mode Based Simplified NFC … 381
 
d
S= + λωr e (25)
dt

where the design parameter λωr > 0. To design SMC for the FBL system of (13),
the sliding surface as the speed error

S = ωr∗ − ωr (26)

where ωr∗ is the command speed and ωr is the recorded speed. For selecting of sliding
line, the SMC used is given by

∧
ũ 2 = −k ωr − βωr sgn(S), βωr > 0 (27)

∧
where ũ 2 = 1J 23 LLmr Pu 2 ≈ Te∗ , k ωr = − BJ ωr corresponds to dynamics of uncertain-
ties which is derived from (13), and β ωr sgn (S) signifies the corrective switching
control. Here, to make the system simple, the zero load torque T l is considered. The
condition for the system trajectory to be slided through the sliding surface is given
by

1 d 2
S ≤ −ηωr |S| (28)
2 dt

where the Lyapunov function ‘V’ is considered to be S 2 /2 and the convergence of

speed depends upon the constant ηωr . Now
 
1 d 2 ∧
S = S kωr − k ωr − βωr |S| (29)
2 dt

For better convergence, the (29) must be less than zero during uncertainties. Here,
β ωr as in (29) is considered to satisfy the condition in (27).

βωr = K ωr + ηωr (30)

 
∧ 
where k ωr − kωr  ≤ K ωr is considered to be restricted error which is estimated one.
The chattering-free response is produced by proposing thickness of a boundary layer
of φωr > 0.
For parameters updation of the proposed NFC, the moving surface S is used here
for the adaptive mechanism based on gradient descent algorithm [17]. Here, the cost
function is defined as
1 •
E= ( S +λωr S)2 (31)
2
382 R. N. Mishra et al.

Fig. 5 Error of the proposed

SNFSMC

The parameters updating of the NFC is done to get the required control
performance are as follows:

∂
ai (k + 1) = ai (k) − ηai E(k) (32)
∂ai
∂
bi (k + 1) = bi (k) − ηbi E(k) (33)
∂bi
∂
wi (k + 1) = wi (k) − ηwi E(k) (34)
∂wi

where k is the time instant at which the signal is sampled. The node values and
their tuning rate are denoted as (ai , bi ) and (ηai , ηbi , ηwi ), respectively. The values
of tuning rate ηai , ηbi and ηwi are 0.004, 0.004, and 004, respectively.
The error of the conventional and simplified NFSMC as shown in Fig. 5 are
obtained as 0.00137 and 0.00125, respectively.

5 Simulation Results and Performance Assessment

Case 1: Under different controllers, no load speed of 800 rpm is observed for lin-
earized IM drive and the startup responses are given in Fig. 6. The torque response is
improved significantly in terms of ripple due to reduced stator current ripple (18 A)
compared to that of the classical PI-controller based drive. However, during starting,
the capacitor voltage V dc suddenly rises to (742 V) with the proposed SMSNFC
is observed as in Fig. 6c (iv). It is also observed that the low computationally bur-
dened SMNFC based IM drive does not deteriorate the system behavior compared to
the classical SMNFC. Rather, it outperforms compared to the classical PI-controller
based drive which is demonstrated in Fig. 6.
Case 2: The 50% step load responses from 1.5 to 2 s is shown in Fig. 7. It is
evident that the proposed controller has fast speed recovery time during external
disturbance. As compared to the PI-controller, the proposed controller reduces the
torque chattering remarkably and thus reducing the magnitude and stator current
Hybrid Sliding Mode Based Simplified NFC … 383

Fig. 6 Starting up responses of feedback linearized IM drive during with a PI-controller,

b conventional SMNFC and c proposed SMSNFC

Fig. 7 50% step load responses with a PI-controller, b conventional SMNFC and c proposed
SMSNFC
384 R. N. Mishra et al.

Fig. 8 Speed reversal responses of feedback linearized controlled IM with a PI-controller,

b conventional SMNFC and c proposed SMSNFC

ripple. Also, the speed distortion is almost disappeared with the proposed controller
as shown in Fig. 7c (i) compared to the conventional SMNFC, which still can have
tiny oscillations.
Case 3: The reversal of speed responses from 800 to −400 rpm at 2.5 s shown
in Fig. 8 and a large negative torque and stator current is observed. During braking,
the frequency decreases first and the responses are the mirror image of the responses
during startup due to the reversal of phase.

6 Conclusion

The proposed SMSNFC based linearized IM drive does not compromise the sys-
tem performance in terms of speed responses and torque ripple compared to that
of classical SMNFC based drive. Rather, it increases the computational efficiency
by reducing the computational burden due to less number of MFs. Moreover, the
responses prove that the proposed SMSNFC shows better responses compared to
that of PI-controller based drive. Thus, this control algorithm base SMSNFC can
be useful for industrial applications as the control is more precise under parameter
variations and external disturbance by the load.
Hybrid Sliding Mode Based Simplified NFC … 385

Appendix

Parameters of induction machine

Rated nominal power Pn = 3.7 kW, rated nominal voltage V n = 415 V, rated nomi-
nal speed nr = 1445 rpm, rated nominal frequency f n = 50 Hz, no. of pole pairs P
= 2, stator resistance Rs = 7.34 , leakage inductances of stator and rotor L ls = L lr
= 0.021 H, rotor resistance Rr = 5.64 , mutual inductance L m = 0.5 H, frictional
coefficient B = 0.035 kg m2 /s, moment of inertia J = 0.16 kg m2 .

References

1. Jia-Min, S., Hurng-Liahng, J., Jinn-Chang, W., Kuen-Der, W.: Single-phase three-wire grid-
connected power converter with energy storage for positive grounding photovoltaic generation
system. Electr Power Energy Syst. 54, 134–143 (2014)
2. Li, Q., Chen, W., Liu, S., You, Z., Tao, S.: Power management strategy based on adaptive
neuro-fuzzy inference system for fuel cell-battery hybrid vehicle. J. Renew Sustain Energy 4,
013106 (2012)
3. Lekhchine, S., Bahi, T., Abadlia, I., Bouzeria, H.: PV-battery energy storage system operating
of asynchronous motor driven by using fuzzy sliding mode control. Int. J. Hydrogen Energy
(2016). https://doi.org/10.1016/j.ijhydene.2016.05.298
4. Bose, B.K.: Modern Power Electronics and AC Drives. New Delhi, Prentice Hall Of India
(2008)
5. Chiasson, J.: A new approach to dynamic feedback linearization control of an induction motor.
IEEE Trans. Automatic Control. 43(3), 391–397 (1998)
6. Alonge, F., Cirrincione, M., Pucci, M., Sferlazza, A.: Input–output feedback linearization
control with on-line MRAS-based inductor resistance estimation of linear induction motors
including the dynamic end effects. IEEE Trans. Ind. Appl. 52, 254–266 (2016)
7. Zhang, Z., Tang, R., Bai, B., Xie, D.: Novel direct torque control based on space vector modu-
lation with adaptive stator flux observer for induction motors. IEEE Trans. on Magnetics 46(8),
3133–3136 (2010)
8. Mishra, R.N., Mohanty, K.B.: Implementation of feedback linearization modelled induction
motor drive through an adaptive simplified neuro-fuzzy approach. Sadhana (Springer) 42(12),
2113–2135 (2017)
9. Lascu, C., Jafarzadeh, S., Fadali, S., Blaabjerg, F.: Direct torque control with feedback
linearization for induction motor drives. IEEE Trans. Power Electron. 33(3), 2072–2080 (2017)
10. Gadoue, S.M., Giaouris, D., Finch, J.W.: MRAS sensorless vector control of an induction motor
using new sliding-mode and fuzzy logic adaption mechanisms. IEEE Trans Energy Conversion.
25(2), 394–402 (2010)
11. Lee, C.C.: Fuzzy Logic in Control Systems: Fuzzy Logic Controller—Part 2, IEEE (1990)
12. Uddin, M.N., Abido, M.A., Rahman, M.A.: Development and implementation of a hybrid
intelligent controller for interior permanent magnet synchronous motor drive. IEEE Trans. Ind.
Appl. 40(1), 68–76 (2004)
13. Lin, C.T., Lee, C.S.G.: Neural Fuzzy Systems: A Neuro-Fuzzy Synergism to Intelligent
Systems. Englewood Cliffs, NJ, USA, Prentice Hall (1996)
14. Uddin, M.N., Chy, M.I.: A novel fuzzy logic controller based torque and flux controls of IPM
synchronous motor. IEEE Trans. Ind. Appl. 46(3), 1220–1229 (2010)
15. Mishra, R.N., Mohanty, K.B.: Design and implementation of a feedback linearization controlled
IM drive via simplified neuro-fuzzy approach. IETE J. Res. 64(2), 209–230 (2017)
386 R. N. Mishra et al.

16. Uddin, M.N., Wen, H.: Development of a self-tuned neuro–fuzzy controller for induction motor
drives. IEEE Trans. Ind. Appl. 43(4), 1108–1116 (2007)
17. Uddin, M.N., Huang, Z.R., Hossain, A.B.M.S.: Development and implementation of a
simplified self-tuned neuro–fuzzy-based IM drive. IEEE Trans. Ind. Appl. 50(1), 51–59 (2014)
18. Sher, H.A., Addoweesh, K.E.: Power storage options for hybrid electric vehicles-A survey. J.
Renew Sustain Energ 4, 052701 (2012)
19. Kazimierczuk, M.K.: Pulse-width Modulated DC–DC Power Converters, Chichester. Wiley,
UK (2008)
20. Slotine, J.J.E., Li, W.: Applied Nonlinear Control. Prentice Hall (1991)
Kalman Filtering Sliding Mode
Controller Design for Stabilizing
and Trajectory Tracking of Inverted
Pendulum

Akshaya Kumar Patra, Alok Kumar Mishra, Anuja Nanda,

Lalit Mohan Satapathy, Amaresh Gantayet, Ramachandra Agrawal,
and Abhishek Patra

Abstract The aim of this paper is to design a Kalman filtering sliding mode
controller (KFSMC) for control of cart position (CP), and angular position (AP)
of the pendulum under uncertainties and disturbances. For designing of the KFSMC,
the fourth-order state-space model of the inverted pendulum (IP) is considered. In
this control strategy, the conventional sliding mode controller (CSMC) is reformu-
lated with a state estimator based on the Kalman filtering approach to improve
the control performance. The validation of the improved control performance of
KFSMC is established by comparative result investigation with other published con-
trol algorithms. The comparative results clearly reveal the better response of the
proposed approach to control the system dynamics within a stable range with respect
to accuracy, robustness, and ability to handle uncertainties.

Keywords Inverted pendulum · Angular displacement · Kalman filter · Sliding

mode control

A. K. Patra (B) · A. K. Mishra · A. Nanda · L. M. Satapathy · A. Gantayet · R. Agrawal · A. Patra

Department of EEE, Institute of Technical Education and Research, Siksha ‘O’ Anusandhan
(Deemed to be University), Bhubaneswar 751030, India
e-mail: [email protected]
A. K. Mishra
e-mail: [email protected]
A. Nanda
e-mail: [email protected]
L. M. Satapathy
e-mail: [email protected]
A. Gantayet
e-mail: [email protected]
R. Agrawal
e-mail: [email protected]
A. Patra
e-mail: [email protected]

© Springer Nature Singapore Pte Ltd. 2020 387

G. Pradhan et al. (eds.), Advances in Electrical Control and Signal Systems,
Lecture Notes in Electrical Engineering 665,
https://doi.org/10.1007/978-981-15-5262-5_28
388 A. K. Patra et al.

1 Introduction

The control of IP is a standard problem among all other problems in the control system
engineering field owing to non-minimum phase, nonlinear, and under-actuated char-
acteristics as reported in the references [1–3]. Additionally, the IP system exhibits a
significant number of industrial applications like self-balancing two-wheeled vehi-
cles or a kind of Sag-way, guided missiles, rockets, intelligent robots, and other
crane models [4]. In this present study, an adaptive law of control approach is being
tested and analyzed to choose the IP system as it possesses a significant relevance
with the control dynamics. During the past three decades, so many control strategy
techniques are suggested and tested for control of AP of pendulum within the sta-
ble range. Time-discrete and switching PID control strategy is implemented in IP
problems with variable control gains based on the measured AP of the pendulum
[5, 6]. However, the optimal gain parameter setting, a lesser range of robust control,
and need of change of gain setting with varying conditions are the major limitations
to limit the real-time application of these controllers. Among other projected robust
control algorithms implemented for limiting the AP of the pendulum are fuzzy [7],
linear quadratic regulator (LQR) [8–10], neurocontrol [11], backstepping control
[12], passivity control [13], state feedback control [14], H-infinity (H ∞) control
[15], sliding mode (SM) control [16], fuzzy sliding mode (FSM) control [17, 18],
and BLQG control [19]. However, even if these control techniques are implemented
effectively to control the AP of the pendulum with better accuracy, still they fail to
handle various constraints and random change found in a trajectory of motion in
the pendulum. These control techniques are not completely insensitive to the dis-
turbances and the uncertainties of the model in spite of the improved performance.
Hence, optimal control parameters setting for better performance and for avoiding
slow response following process disturbance, the current work suggests an alterna-
tive novel hybrid technique based on the Kalman filtering and sliding mode control
concept.
In this study, a hybrid concept based on Kalman filtering approach and principles
of sliding mode control is projected with the title Kalman filtering sliding mode
controller (KFSMC). The hybrid approach concept leads to an enhanced control
performance with respect to robustness, accuracy, stability, and better ability to handle
uncertainty. The novel idea followed in this formulation is to modify CSMC with a
state estimator according to the approach of Kalman filtering. Secondly, a stabilizing
control law is framed by using the Kalman filtering approach [20–22]. Application
of the proposed approach is to control the AP of pendulum results to ensure a better
robust controller in comparison with other contemporary well-accepted methods
under both harmonized and incompatible uncertainties.
The highlights of this manuscript are as follows:
• Development of a SIMULINK model of an IP.
• Design of a KFSMC to control the AP of the pendulum within a stable range.
Kalman Filtering Sliding Mode Controller Design … 389

• Estimation of the control actions of the KFSMC under huge deviation of process
disturbance.
• Comparative investigation to certify the better response of the KFSMC.
This manuscript is structured as follows: Sect. 2 concisely illustrates the IP system
with mathematical details reflecting to its dynamic characteristics. Also, it clearly
demonstrates the simulation execution of the system on MATLAB environment.
A detailed presentation on how the control technique is formulated and how it is
implemented for this problem is presented in Sect. 3. Comparative results of the
proposed approach with other published control techniques and the related analysis
are provided in Sect. 4. The concluding comments are summarized in Sect. 5.

2 Problem Formulation and Modeling

2.1 System Overview

The closed-loop model of the IP is depicted in Fig. 1a. The applied horizontal force
F(t) and v(t) are reflected as the process disturbance and the sensor noise, respectively,
in this study. The controller receives information about the AP of the pendulum as
an input to provide the optimal control force u(t), and it balances the pendulum.

2.2 System Modeling

Figure 1b reflects the cart-pendulum model connected to a flexible cart rail with a free
swinging pole. The CP is being controlled by a DC motor. The nonlinear IP modeling
through the Newton’s law-based mathematical equations has been carried out. It is
presumed that the vertical force does not affect the CP and the CP is disturbed by
the horizontal force F(t) based on the operation of the DC motor [23, 24]. All the
physical activities of the IP system are expressed mathematically and specified by
Eqs. (1) and (2). All nomenclature and specifications for IP are shown in Tables 1
and 2, respectively. The SIMULINK model of the IP is established with respect to
the Eqs. (1) and (2) as displayed in Fig. 1c.

d2 x(t) d2 θ (t)
(M + m) − ml cos θ (t)
dt 2 dt 2
 
dθ (t) 2 dx(t)
+ ml sin θ (t) + b = F(t) (1)
dt dt
  d2 θ (t) d2 x(t)
i + Ml 2 − mgl sin θ (t) = ml cos θ (t) (2)
dt 2 dt 2
390 A. K. Patra et al.

(a)

(b)

(c)

Fig. 1 a IP model with KFSMC; b schematic model of the IP system; c simulation model of the
nonlinear IP system; d the horizontal force acting on the cart at the time of 0.1 s
Kalman Filtering Sliding Mode Controller Design … 391

Fig. 1 (continued)

(d)

Table 1 IP model states and parameters

Symbol Description
F(t) Horizontal force acting on the cart
b dx(t)
dt Frictional force acting on the cart
dx(t) d2 x(t)
x(t), dt , dt 2 Cart position, cart velocity, and cart acceleration, respectively
dθ (t) d2 θ (t)
θ(t), dt , dt 2 AP, angular velocity, and angular acceleration of pendulum, respectively

Table 2 IP model
Symbol Physical meaning Value
specification values
M Cart mass 0.5 kg
m Pendulum mass 0.2 kg
i Inertia 0.3 kg m2
g Gravitational acceleration 9.8 m/s2
b Frictional coefficient 0.1 N.s/m
l Pendulum length 0.3 m

2.3 Linearization of IP Model

The nonlinear IP system is linearized surrounding the operating point for the design
of KFSMC to control the system dynamics within the stable range. The linearization
of the IP system based on Eqs. (1) and (2) is done by neglecting the higher-order
 2
terms such as dθ(t) dt
. For linearization of the nonlinear IP system, the dynamic
Eqs. (1) and (2) are reduced to Eqs. (3) and (4) based on the stable conditions such
as θ (t) = 0, d dtθ(t)
2
2
∼
= 0, and cos(0) = 1.
392 A. K. Patra et al.

d2 x(t) dx(t) d2 θ (t)

(M + m) + b − ml = F(t) (3)
dt 2 dt dt 2
  d2 θ (t) d2 x(t)
i + ml 2 − mglθ (t) = ml (4)
dt 2 dt 2
The transfer function (TF) of the CP and AP of the pendulum is derived as follows
[23]:

(i+ml 2 )s 2 −mgl
X (s) ψ
= (5)
U (s) b(i+ml 2 ) 3
s +
4
ψ
s − (M+m)mgl ψ
s2 − bmgl
ψ
s
ml
θ (s) ψ
s
= (6)
U (s) s 3 + ( ψ ) s 2 − (M+m)mgl
b i+ml 2

ψ
s− bmgl
ψ

  
where ψ = (M + m) i + Ml 2 − ml 2 . The linearized equation of the IP system
with F(t) and v(t) based on the Eqs. (5) and (6) can be expressed as follows [23]:

dxm (t)
dt
= ẋm (t) = Am xm (t) + Bm u(t) + Bd F(t)
(7)
y(t) = Cm xm (t) + Dm u(t) + v(t)

where xm (t), u(t), and y(t) are represented as the state variable, control input, and reg-
ulated output, respectively. The state-space matrices of the IP system are represented
as Am , Bm , C m , Dm and Bd . The detailed description of linearization is described
in the literature [23]. In MATLAB, the command ‘linmod’ is used to evaluate the
state-space matrices on the SIMULINK model of IP as depicted in Fig. 1c.

2.4 Response of IP Model

There are four roots in IP system. One of them lies in right-hand side of the complex
plane. As a result, the system becomes unstable. This needs the design of an adaptive
controller for improving the stability of the system by means of shifting the roots
into the left-hand side of the complex plane. The IP system SIMULINK model in
the open-loop form is depicted in Fig. 1c. The IP system consists of two inputs and
four outputs. The control force u(t) and applied horizontal force F(t) are the two
inputs of the IP system. The CP and AP of the pendulum are the four outputs of the
IP system. An uncontrolled system dynamics such as AP of the pendulum and CP
are being observed owing to the application of 1 N impulsive horizontal force F(t)
on the cart at the time t = 1.0 s.
The uncontrolled system dynamics are illustrated in Fig. 2a, b. Figure 2a, b illus-
trates the unstable dynamics under various model uncertainties and disturbances.
The unstable dynamics can be reduced by applying the suitable control techniques.
Kalman Filtering Sliding Mode Controller Design … 393

Fig. 2 a AP of pendulum with application of F(t); b CP with application of F(t)

In this case, the AP of the pendulum is the most essential outcome that needs to
be controlled within a stable range through suitable control techniques, and CP is
analyzed in order to view the motion trajectory.

3 Control Algorithm

The KFSMC control algorithm is demonstrated in this section. The closed-loop

system response with respect to robustness, accuracy, and stability are analyzed.
The control specifications such as settling time t s , steady-state error ess , maximum
overshoot Omax , and maximum undershoot U max are also evaluated and examined
with proper validation of the controller action.

3.1 Design of KFSMC

The linearized model of the IP as discussed in Sect. 2.3 has been taken for the formu-
lation of the suggested control algorithm to regulate the CP and AP of the pendulum.
For accomplishing an upgraded performance and the adjustment of controller param-
eters of the suggested KFSMC, it is integrated into the linearized model of the IP as
illustrated in Fig. 3. The KFSMC is designed by integrating a state estimator with
the CSMC as shown in Fig. 1a. The state estimator is implemented to estimate all the
states of the system in a recursive manner to enhance the control performances such
as robustness, accuracy, and stability. The linearized model of the IP with F(t) and
v(t) is formulated as represented in Eq. (7), where both v(t) and F(t) are represented
as the Gaussian noise. The relationship between v(t) and F(t) is as follows:
394 A. K. Patra et al.

Fig. 3 Closed-loop IP model with KFSMC in the state-space representation

E{F(t)} = 0
(8)
E{v(t)} = 0

E F(t)F T (τ ) = Q 2 δ(t − τ ) (9)

E v(t)v T (τ ) = R2 δ(t − τ ) (10)

F(t) is uncorrelated to the v(t); therefore, it can be defined as follows:

E F(t)v T (τ ) = 0 (11)

where Q2 and R2 are symbolized as the positive semi-definite intensity matrices of

the F(t) and v(t), respectively. Figure 3 illustrates the linearized model of the IP with
the feedback gain k c and Kalman filtering gain k f . The mathematical expressions of
k c and k f are presented in Sect. 3.1.2. The calculation of the TF of the KFSMC is
described in Sect. 3.1.3.
The control signal u(t) of proposed controller has two major components such
as switching function ust (t) and equivalent control input ueq (t). This is defined as
represented in Eq. (12).

u(t) = u st (t) + u eq (t) (12)

For the formulation of the proposed control algorithm, two basic steps are required.
In the first step, the switching function ust (t) is to be ensured reaching condition by
pulling the all states of a nominal model inside the boundary layer near the sliding
hyperplanes; as a result, chattering is reduced and also robustness of a system is
improved. In the second step, the desired ueq (t) is to be formulated in such a manner
that accuracy, and stability of a system is enhanced. The design procedures of first
and second steps of the KFSMC are analyzed in Sects. 3.1.1 and 3.1.2, respectively.
Kalman Filtering Sliding Mode Controller Design … 395

3.1.1 Reaching Condition (RC)

To achieve RC, all the states of the nominal model are to be inside the boundary layer
near the sliding hyperplanes. Secondly, these states should remain inside the bound-
ary layer after that [20]. The robustness of a model is enhanced by achieving the RC.
Under this situation, the system dynamics are independent to the model uncertainty
and disturbances. It can be possible through the sliding hyperplane parameter Gs .
(A) Sliding hyperplane parameter:
The sliding hyperplane parameter Gs for a system with α number of inputs is related
to the equal number of α hyperplane vectors as follows [20]:

si (t) = giT xm (t) (13)

s(t) = G s xm (t) (14)

where s(t) stands for si (t), Gs stands for giT , si (t) = [s1 (t), s2 (t), s3 (t), . . . , sα (t)],
si (t) is the ith hyperplane vector, giT = [g1 , g2 , g3 , . . . , gα ]T , giT is the transpose of
ith hyperplane parameter, α is the number of inputs of a system, and i varies from 1 to
α. When si (t) = 0, α number of hyperplanes move through the origin in state-space
as shown in Fig. 4. These hyperplanes are called as sliding hyperplanes.
Under the above condition, s(t) and Gs are known as the sliding hyperplane vector
and sliding hyperplane parameter of the IP model, respectively. Equation (14) can
be rewritten as follows:

ṡ(t) = G s ẋm (t) (15)

The Gs is derived by using the similarity transformation controllable canonical

form (STCCF) of an IP model. The STCCF of an IP model is defined as follows [20]:

q̇(t) = Am q(t) + Am u(t) + Ad w(t) (16)

Fig. 4 Location of boundary

layer near the sliding
hyperplane
396 A. K. Patra et al.

where q(t) = H xm (t), Am = H Am H −1 , B m = H Bm , B d = H Bd . The H matrix is

computed by using Eq. (17).
 T
H = F Bm (17)
 
where F = null BmT . The q(t) can be decomposed as follows:

q1 (t)
q(t) = (18)
q2 (t)

where q1 (t) and q2 (t) are vectors with size of (n −α)×α and α ×α, respectively. The
n represents the order of the Am . When the rank of a system matrix Am is not matched
with the rank of the controllable matrix c(Am , Bm ) of the system, the appropriate
system is uncontrollable in nature. The system can be decomposed into controllable
and uncontrollable parts. The decomposed controllable and uncontrollable parts of
the system with respect to Eq. (16) are defined as follows [20]:
    
q̇1 (t) Am11 Am12 q1 (t) 0 0
= + u(t) + w(t) (19)
q̇2 (t) Am21 Am22 q2 (t) B mr B dr

where q1 (t) and q2 (t) are the controllable and uncontrollable parts of the system,
respectively. The B mr consists of the last α-rows of the matrix B m , and B dr consists
of the last α-rows of the matrix B d . The sliding hyperplane vector can be expressed
in the form of q1 (t) and q2 (t) as follows [20]:

s(t) = q2 (t) + kq1 (t) (20)

The Gs can be derived based on Eqs. (14) and (20), as follows [20]:

G s = k Iα H (21)

where k denotes the gain matrix of the proposed KFSMC, and Iα denotes an unitary
matrix with dimension of α × α.
(B) Gain matrix of KFSMC
The k is computed based on the IP model dynamics for the minimum value of the
quadratic performance index j as specified in Eq. (22).

∞

j= q(t)T Q q q(t) dt (22)
0

 T
where Q q = H −1 Q 1 H −1 and Q 1 = CmT Cm . The Q1 signifies a positive semi-
definite weighted matrix of state. The matrix Qq can be decomposed with respect to
Kalman Filtering Sliding Mode Controller Design … 397

Eq. (19) as follows:


Qr N
Qq = (23)
N T R1

where the matrix Qr is the first subpart of the matrix Qq with size of (n −α)×(n −α).
The matrix N is the second subpart of the matrix Qq with size of (n − α) × α. The
N T is the third subpart of the Qq with dimension of α × (n − α). The R1 symbolizes
a positive definite weighted matrix of input, and it is the fourth subpart of the matrix
Qq with size of α × α. Equation (22) can be reformulated as follows:

∞

j= q1 (t)T Q r q1 (t) + 2q1 (t)T N q2 (t) + q2 (t)T R1 q2 (t) dt (24)
0

The system can achieve RC, only when s(t) = 0. Under RC, the system dynamics
can be formulated based on the full state feedback along with size of (n − α) × α as
follows [20]:

q̇1 (t) = Am1 q1 (t) + Am12 q2 (t) (25)

The k is computed based on the Eqs. (23) and (25) for the minimum value of the
quadratic performance index as specified in Eq. (26).
 T 
k = R1−1 Am12 P + N T (26)

where P denotes the controller algebraic Riccati equation (CARE) solution. The
CARE is specified as follows [20]:
T T
P(Am11 − Am12 R1−1 N T ) + (Am11 − N R1−1 Am12 )P
T
− P Ām12 R1−1 Am12 P + Q r − N R1−1 N T = 0 (27)

The sliding hyperplane parameter Gs can be computed by putting Eq. (26) in

Eq. (21). The system can achieve RC and enhanced robustness through the sliding
hyperplane parameter Gs . To satisfy the RC and chattering free control is chosen as
[20]:

u(t) = −(G s Bm )−1 diag(η)sat(s(t)) + u eq (t) (28)

where −(G s Bm )−1 diag(η)sat(s(t)) is the switching function u st (t) and diag(η) is
a diagonal matrix with ith diagonal element equal to a positive number ηi . The ith
element of sat(s(t)) is formulated as:
398 A. K. Patra et al.

sgn(si (t)) if |si (t)| > ρi
sat(si (t)) = (29)
si /ρi otherwise

where ρi denotes the boundary layer thickness nearby the ith hyperplane as shown
in Fig. 4.

3.1.2 Equivalent Control Input ueq (t)

To satisfy the RC, the u eq (t) can be formulated with the use of k f and k c . The states
of the system are estimated recursively based on the Kalman filtering approach to
enhance the system response. The estimated state-space equation of the system is
defined as [20]:

x̂˙m (t) = Am x̂m (t) + Bm u(t) + k f [r (t) − y(t) − ŷ(t)] (30)

ŷ(t) = Cm x̂m (t) (31)

where x̂m (t) signifies the estimated state and ŷ(t) signifies the estimated output.
The k f can be formulated as [20]:
T −1
kf = k C m R2 (32)

where k signifies the filter algebraic Riccati equation (FARE) solution. The FARE
is specified as follows [20]:
T −1
Am k + T
k Am + Bd Q 2 BdT − k C m R2 C m k =0 (33)

The sliding hyperplane vector s(t) can be formulated with the use of Gs and x̂m (t)
as follows [20]:

s(t) = G s x̂m (t) (34)

To satisfy the RC, the u eq (t) can be derived from Eqs. (30) and (34) as [20]:
  
u eq (t) = −(G s Bm )−1 G s Am − k f Cm x̂m (t) + G s k f [r (t) − y(t)] (35)

Under RC, the control law u(t) of suggested KFSMC is formulated based on
Eqs. (28), (29), and (35) as follows [20]:
  
u(t) = −(G s Bm )−1 G s Am − k f Cm + γ In x̂m (t) + G s k f [r (t) − y(t)] (36)

u(t) = −kc x̂m (t) − K s [r (t) − y(t)] (37)

Kalman Filtering Sliding Mode Controller Design … 399

Table 3 The optimal values

Q1 R1 Q2 R2
of control parameters
1000 ∗ CmT ∗ Cm 1 0.01 0.0025

 
where K s = (G s Bm )−1 G s k f and kc = (G s Bm )−1 G s Am − k f Cm + γ In . The I n
is an unitary matrix with size of n × n, and n is the order of the Am . The k c and γ
are denoted for the feedback gain and sliding parameter of the suggested KFSMC,
respectively. The γ can be estimated as:
ηi η
γ = = (38)
ρi ρ

where ρi and ηi are denoted for the thickness of the boundary layer and any positive
number of the ith sliding hyperplane, respectively, as illustrated in Fig. 4. Both ηi
and ρi are taken as identical value for each γ and set to η and ρ, respectively.

3.1.3 Transfer Function (TF) of KFSMC

The TF of the KFSMC is defined from Eqs. (30) and (37) as follows [20]:
 −1
K (s) = β1 s In − Am + Bm β1 + k f Cm β2 + (G s Bm )−1 G s k f (39)
 
where K (s) is denoted as the TF of KFSMC, β1 = (G s Bm )−1 G s Am + γ In − k f Cm
and β2 = k f − Bm (G s Bm )−1 G s k f . The packed matrix notation of K(s) is specified
as follows [20]:

(40)

For the design of suggested KFSMC, the optimal values of control parameters are
evaluated with help of MATLAB and represented as in Table 3.

4 Outcomes and Discussions

The response of the closed-loop IP with proposed KFSMC is described in detail

in this section. The proposed control approach (KFSMC) is compared with other
popular control algorithms to justify its enhanced performance.
400 A. K. Patra et al.

Fig. 5 a AP of the pendulum with the application of F(t) based on KFSMC; b CP with the
application of F(t) based on KFSMC

4.1 Performance Analysis of IP System with KFSMC

In this section, all physical activities of the closed-loop IP model with suggested
KFSMC are examined under different conditions and the huge deviation of applied
horizontal force. The closed-loop system dynamics with 1 N impulsive horizontal
force F(t) at the time of 1.0 s are displayed in Fig. 5a, b. The outcomes clearly
specify the pendulum achieves the zero AP with less settling time and cart also
attains the balance position where the IP system is absolutely steady. To achieve
the enhanced system response, the required control force u(t) is generated by the
suggested KFSMC and demonstrated in Fig. 6.

4.2 Robustness of the KFSMC

Figure 7 illustrates the AP and AV of the pendulum with suggested KFSMC under the
huge deviation of applied horizontal force F(t). The time-domain outcomes under
huge deviation of applied forces show the enhanced performance of the closed-loop
IP model with KFSMC. Overall in each case, the pendulum achieves finally zero
AP and zero AV with less settling time. As indicated by the results, the suggested
KFSMC robust performance under the huge deviation of applied horizontal forces
is much better.

4.3 Stability Investigation

Figure 8a, b illustrates the magnitude plots result of the open-loop and closed-loop
IP system to verify and analyze the stability conditions. From both the magnitude
Kalman Filtering Sliding Mode Controller Design … 401

Fig. 6 Controlled signal u(t) of the KFSMC

Fig. 7 AP of the pendulum with a deviation of ±40% applied F(t) based on KFSMC
402 A. K. Patra et al.

Fig. 8 a Magnitude plot of the IP system; b Magnitude plot of the closed-loop IP system

plots, it is observed a better smoothness referring to the wider steady-state stability

of closed-loop system (Fig. 8b) than the open-loop system (Fig. 8a). In other words,
the bandwidth is increased in case of a closed-loop system with KFSMC than the
open-loop system. This clearly indicates a faster dynamics, and also it results zero
AP and zero AV of pendulum with quick settling time in case of the closed-loop
system. This justifies better stability during system operation.

4.4 Comparative Study

The proposed KFSMC control approach is compared with other popular control
approaches such as PID, fuzzy, LQR, H ∞, FSM, and BLQG which justify its
enhanced performance as a controller. Figure 5a illustrates the effect of applied force
in the AP of the pendulum with the proposed KFSMC approach. Table 4 presents a
comparative analysis with respect to t s (sec), OMax (rad), UMax (rad), noise (%), and

Table 4 Comparative result analysis related to AP of pendulum

Controller PID [5] Fuzzy LQR [9] H∞ FSM BLQG KFSMC
[7] [15] [18] [19] (proposed)
Applied 1 1 1 1 1 1 1
force (N)
t s (s) 2.8 3.0 3.2 1.5 1.6 0.4 0.36
OMax 0.1 0.226 0.081 0.107 0.042 0.041 0.04
(rad)
U Maz (rad) 0.01 0.087 0.02 0.045 0.151 0.01 0.009
Noise (%) 10 10 5 5 5 1 1
ess (%) 0 0 0 0 0 0 0
Kalman Filtering Sliding Mode Controller Design … 403

ess (%). The effect of applied force in the AP of pendulum applying different control
approaches such as PID, fuzzy, LQR, H ∞, FSM, and BLQG is also demonstrated
in Table 4 based on the references [5, 7, 9, 15, 18, 19], respectively. Similar working
conditions are followed with the same level of sensor noise in all control techniques
application for comparison.
The AP of pendulum under 1 N impulsive horizontal force is tested. The cor-
responding results are presented for the various control approaches along with the
proposed KFSMC with respect to time-domain specifications such as OMax (rad),
U Max (rad), and t s (s). The results signify the better controllability of the KFSMC.
The simulation results also demonstrate the high noise and chattering elimination
capability with high robustness for the proposed approach. Overall, by looking to the
above comparative analysis, the findings of suggested approach advantages are the
higher accuracy and stability, more robustness, high noise and chattering elimination
capability, and better capability to handle uncertainty under various conditions and
huge deviation of road disturbance.

5 Conclusions

The manuscript proposed a novel control strategy (KFSMC) based on the Kalman
filtering approach to balance the pendulum. To justify its enhanced performance,
it has been applied and tested to control the system dynamics of IP system within
the stable range. Initially, the IP system is modeled as the fourth-order state-space
representation. Then, the proposed control approach (KFSMC) is designed. In sug-
gested KFSMC, state estimator is utilized to enhance the control performance. The
comparative results clearly reflect that the suggested KFSMC is arrived at better per-
formance than the other control approaches such as PID, fuzzy, LQR, H ∞, FSM, and
BLQG with respect to stability, reliability, and robustness under various abnormal
conditions and disturbances. The better performance of the suggested approach in
terms of improved accuracy and stability, enhanced robustness, high noise and chat-
tering elimination capability, and better ability to control uncertainty which justifies
its real-time application.

References

1. Iqbal, J., Islam, R.U., Syed, Z.A., Abdul, A.K., Ajwad, S.A.: Automating industrial tasks
through mechatronic systems—a review of robotics in industrial perspective. Tehnicki vjesnik
Tech. Gaz. 23, 917–924 (2016)
2. Ajwad, S.A., Asim, N., Islam, R.U., Iqbal, J.: Role and review of educational robotic platforms
in preparing engineers for industry. Maejo Int. J. Sci. Technol. 11, 17–34 (2017)
3. Bettayeb, M., Boussalem, C., Mansouri, R., Al-Saggaf, U.: Stabilization of an inverted
pendulum-cart system by fractional PI-state feedback. ISA Trans. 53, 508–516 (2014)
404 A. K. Patra et al.

4. Iqbal, J., Ullah, M., Khan, S.G., Khelifa, B., Ukovic, S.C.: Nonlinear control systems—a brief
overview of historical and recent advances. Nonlinear Eng. 6, 301–312 (2017)
5. Ghosh, A., Krishnan, T., Subudhi, B.: Brief paper—robust proportional-integral-derivative
compensation of an inverted Cart-Pendulum System: an experimental study. IET Control
Theory Appl. 6(8), 1145–1152 (2012)
6. Wang, C., Yin, G., Liu, C., Fu, W.: Design and simulation of inverted pendulum system
based on the fractional PID controller. In: IEEE 11th Conference on Industrial Electronics
and Applications (ICIEA), pp. 1760–1764 (2016)
7. Magana, M.E., Holzapfel, F.: Fuzzy-logic control of an inverted pendulum with vision feedback.
IEEE Trans. Educ. 41(2), 165–170 (1998)
8. Ozana, S., Pies, M., Slanina, Z., Hajovsky, R.: Design and implementation of LQR controller
for inverted pendulum by use of REX control system. In: IEEE International Conference on
Circuits and Systems, vol. 1, pp. 343–347 (2012)
9. Kumar, E.V., Jerome, J.: Robust LQR controller design for stabilizing and trajectory tracking
of inverted pendulum. Procedia Eng. 64, 169–178 (2013)
10. Prasad, L.B., Tyagi, B., Gupta, H.O.: Modelling and simulation for optimal control of nonlinear
inverted pendulum dynamical system using PID controller and LQR. In: Sixth Asia Modelling
Symposium (Ams), 138–143 (2012)
11. Pasemann, F.: Evolving neurocontrollers for balancing an inverted pendulum. Netw. Comput.
Neural Syst. 9, 1–4 (1998)
12. Deng, L., Gao, S.: The design for the controller of the linear inverted pendulum based on
backstepping. In: International Conference on Electronic and Mechanical Engineering and
Information technology (EMEIT), vol. 6, pp. 2892–2895 (2011)
13. Jörgl, M., Schlacher, K., Gattringer, H.: Passivity based control of a cart with inverted pendulum.
Appl. Mech. Mater. 332, 339–344 (2013)
14. Žilić, T., Pavković, D.: Modeling and control of a pneumatically actuated inverted pendulum.
ISA Trans. 48, 327–335 (2009)
15. Lambrecht, P., Vander, G.: H-infinity control of an experimental inverted pendulum with dry
friction. IEEE Contr. Syst. Mag. 13(4), 44–50 (1988)
16. Wai, R.J., Chang, L.J.: Adaptive stabilizing and tracking control for a nonlinear inverted-
pendulum system via sliding-mode technique. IEEE Trans. Ind. Electron. 53, 674–692 (2006)
17. Tao, C.W., Taur, J., Chang, J.: Adaptive fuzzy switched swing-up and sliding control for the
double-pendulum-and-cart system. IEEE Trans. Syst. Man Cybern. B Cybern. 40(1), 241–252
(2010)
18. Chen, C.S., Chen, W.L.: Robust adaptive sliding-mode control using fuzzy modelling for an
inverted-pendulum system. IEEE Trans. Ind. Electron. 45(2), 297–306 (1998)
19. Patra, A.K., Rout, P.K.: Backstepping linear quadratic gaussian controller design for balancing
an inverted pendulum. IETE J. Res. pp. 1–15 (2019)
20. Khalil, H.K.: (2002) Nonlinear Systems, 3rd edn. Prentice Hall, Upper Saddle River, New
Jersey
21. Patra, A.K., Rout, P.K.: Adaptive sliding mode gaussian controller for artificial pancreas in
TIDM patient. J. Process Control 58, 23–27 (2017)
22. Patra, A.K., Rout, P.K.: Backstepping sliding mode Gaussian insulin injection control for blood
glucose regulation in TIDM patient. J. Dyn. Sys., Meas., Control. 140 (9), 091006–091006-15
(2018)
23. Irfan, S., Mehmood, A., Razzaq, M.T., Iqbal, J.: Advanced sliding mode control techniques for
inverted pendulum: modelling and simulation. Eng. Sci. Tech. Int. J (2018). https://doi.org/10.
1016/j.jestch.2018.06.010
24. Ronquillo-Lomeli, G., Ríos-Moreno, G.J.: Nonnlinear identification of inverted pendulum
system using volterra polynomials. Mech. Based Des. Struct. Mach 44(1), 5–15 (2016)
Electromagnetic and Thermal Analysis
of Permanent Magnet BLDC Wiper
Motor

Aneesh Jategaonkar, P. Ramesh, Pratik Kochgabay, and N. C. Lenin

Abstract This paper compares the different permanent magnet materials for a wiper
application, especially front wiper system. Simulation-based study on electromag-
netic performance and thermal predictions are discussed. Initially, the design of a
brushless direct current (BLDC) motor is made for the required power of 52 Watts
within a pack size of 56 * 35 mm. Results obtained by using different types of per-
manent magnets in the same pack size are carefully studied and compared. Finally,
the thermal characteristics of the designed motors are predicted and compared.

Keywords Brushless DC motor · Finite element analysis · Thermal and

electromagnetic analyses

1 Introduction

Permanent magnet (PM) BLDC motors are gaining significant popularity in auto-
motive industries particularly for the past two decades. This is primarily due to their
compact size and lower weight as compared to commutator machines providing sim-
ilar power outputs [1–4]. Many problems are associated with traditional DC motors
like mechanical friction, sparks, radio interference and motor life shortening due to
the brushes which are used for commutation [5, 6]. Whereas PMBLDC offers high
efficiency, high power density, wide speed range, good position control and silent
operation [7–10].
The number of automobiles is growing day by day and more and more motors
are being used for various automobile applications. One such important application
is the windscreen wiper [11–14]. The movement of the wiper takes place due to the
windscreen wiper motor. A linkage is responsible for the conversion of the rotational
motion to the back and forth motion of the wiper [15]. In this paper, a 52 Watts
PMBLDCM is designed for wiper application. The design specifications and design

A. Jategaonkar · P. Ramesh · P. Kochgabay · N. C. Lenin (B)

Vellore Institute of Technology, Chennai Campus, Chennai 600127, India
e-mail: [email protected]

© Springer Nature Singapore Pte Ltd. 2020 405

G. Pradhan et al. (eds.), Advances in Electrical Control and Signal Systems,
Lecture Notes in Electrical Engineering 665,
https://doi.org/10.1007/978-981-15-5262-5_29
406 A. Jategaonkar et al.

Table 1 Specifications
Quantity Values
Output power 52 W
Voltage 48 V (DC)
Speed 700 rpm
Torque 0.71 Nm
Stack length 35 mm
Stator outer diameter 56 mm
Stator slot fill factor 45%
Shaft diameter 8 mm

approach are explained in Sect. 2. In Sect. 3, the results obtained from the electro-
magnetic analysis are discussed. Finally, the thermal characteristics is analyzed and
compared in Sect. 4.

2 Specifications and Design Approach

The design specifications of the wiper motor are listed in Table 1. A BLDC motor
is designed for the given specifications. The design procedure carried is depicted in
the flowchart shown in Fig. 1 [16, 17].
The designed BLDC motor has 15 stator slots with 14 pole configuration. In order
to avoid practical difficulties, the slot fill factor is to 45%. The designed slot shape
is depicted in Fig. 2, and the slot dimensions are listed in Table 2.
The stator and rotor core are designed with M36_29G silicon steel. Whereas, the
rotor poles are made of NdFeB magnets. The structure of the designed BLDC motor
is depicted in Fig. 3.
The windings are double layered concentrated type with copper as the winding
material. In double layer, two types of arrangements are possible, one is side by side
arrangement and the other is top-bottom arrangements, as shown in Figs. 4 and 5.

3 Electromagnetic Analysis and Results

The designed BLDC motor is analyzed in finite element platform. Figure 6 shows
the flux distribution in the designed BLDC motor, at the rated load condition. As
shown in Fig. 6, the stator core is designed such a way to operate mostly on the knee
region of the magnetic permeability of M36_29G silicon steel. Whereas, the rotor
has a very big back iron, in which the flux is flowing through lesser regions. The
rotor core can be effectively used by introducing ducts in the non-flux areas, also by
considering the mechanical strength.
Electromagnetic and Thermal Analysis of Permanent Magnet … 407

Start

Calculate the machine dimensions and

conductor size according to the required
specifications

Stop
Design the machine in finite element platform
Yes

Analyze the electromagnetic performance Temperature

of the designed machine rise is under
No
limit

No Obtained re- Yes

Predict the thermal characteristics
quired results?

Fig. 1 Design flowchart

Fig. 2 Shape of the stator

slot
408 A. Jategaonkar et al.

Table 2 Slot dimensions

Parameters Numerical value (mm)
Hs0 0.2
Hs1 0.2
Hs2 3
Bs0 0.6
Bs1 5.5
Bs2 8
Rs 2

Winding
Air gap

Rotor core

Permanent Stator core

Magnet

Fig. 3 Designed BLDC motor

In order to have an effective comparison, the designed BLDC has been analyzed
with both side by side and top-bottom configurations. Table 3 shows the results of both
winding configurations, while operating the motor at the rated load (speed—700 rpm,
torque—0.71 Nm).
Compared to side by side winding arrangement, the BLDC motor with top-down
winding arrangement delivers the required output power with 2.3% reduced losses.
So, the efficiency is also better in top-down winding arrangement.
Electromagnetic and Thermal Analysis of Permanent Magnet … 409

Fig. 4 Side by side winding arrangement

Fig. 5 Top-bottom winding arrangement

3.1 Performance Comparison for Different Hard Magnet

Materials

From the previous analysis, it is conformed that the designed BLDC motor can deliver
the required output power. But NdFeB (NdFe35) magnets are used, which is a rare
earth material. In order to study the performance difference of BLDC motor using
rare earth and non-rare earth magnets, two more BLDC motors are designed for
same specifications (given in Table 1) using non-rare earth magnets such as ferrite
(ceramic 10) and Alnico (Alnico 9).
410 A. Jategaonkar et al.

Fig. 6 Flux distribution

Table 3 Performance
Parameters Side to side Top-bottom
comparison—winding
winding winding
arrangements
Number of turns 50 50
Core loss (W) 1.2 1.2
Copper loss (W) 15.8 14.9
Total loss (W) 17.1 16.7
Output power (W) 53.2 53.1
Input power (W) 70.8 69.7
Efficiency (%) 75.12 76.04

The input current consumption of BLDC motor with ferrite magnet is 59% higher
than the BLDC motor with NdFeB. Whereas, BLDC motor with Alnico magnet is
6% higher. Due to this, the efficiency of BLDC motor with ferrite magnet is 37%
reduced than NdFeB magnet motor, as shown in Fig. 7.

Fig. 7 Efficiency 80
comparison
Efficiency (%)

60

40

20

0
NdFeB Alnico Ferrite
Electromagnetic and Thermal Analysis of Permanent Magnet … 411

Fig. 8 Power to weight 0.11

comparison 0.1
0.09

kW/kg
0.08
0.07
0.06
0.05
NdFeB Alnico Ferrite

Even though the BLDC motors with NdFeB and Alnico magnets provides bet-
ter performance, the mass density of these magnets are nearly two times of ferrite
magnet. So, the power to weight ratio is comparatively higher in BLDC motor with
ferrite magnets, depicted in Fig. 8. The cost of the machine can also be reduced by
using ferrite magnets. The material cost of all the discussed machines are listed in
Table 4.

Table 4 Material cost estimation

Component Material used Weight (kg) Approx. cost per kg Material cost (Rs.)
(Rs.)
BLDC with NdFeB
Stator core M36 29G 0.16 190 30.40
Winding Copper 0.07 900 63.00
Rotor core M36 29G 0.22 190 41.80
Magnet NdFe35 0.05 5337 278.06
Total cost 413.26
BLDC with Alnico
Stator core M36 29G 0.16 190 30.40
Winding Copper 0.06 900 54.00
Rotor core M36 29G 0.23 190 43.70
Magnet Alnico 9 0.05 4127 206.35
Total cost 334.45
BLDC with Ferrite
Stator Core M36 29G 0.15 190 28.5
Winding Copper 0.09 900 81
Rotor core M36 29G 0.21 190 39.9
Magnet Ceramic10 0.02 505 12.12
Total cost 161.52
412 A. Jategaonkar et al.

4 Thermal Prediction

The temperature rise plays a vital role in life time of the machines. The losses are the
source for temperature rise. In the designed machines, the losses are low in BLDC
motor with NdFeB and high in ferrite magnet motor. So, the thermal characteristics
of these motors are predicted, in order to ensure the safer operation.
A coupled electromagnetic and thermal analysis has been carried, in which the
instantaneous losses from electromagnetic analysis are directly fed to the thermal
prediction. In this, the motor is operated in rated load continuously for three hours,
with the ambient temperature of 30 °C. Figure 9 shows the temperature rise in various
parts of the BLDC motor with NdFeB.
After 3 h of continuous operation, the motor has reached the maximum temper-
ature of 101 °C at the windings. Similarly, the thermal analysis has been carried for

110

100

90
Temperature (oC)

80

70

60

50

40

30
0 10 20 30 40 50 60 70 80 90 100 110 120 130 140 150 160 170 180
Time (minutes)

Rotor Core (ºC) Rotor Magnet (ºC) Stator Back Iron (ºC)
Winding (ºC) End Winding (ºC)

Fig. 9 Temperature variation—BLDC with NdFeB

Electromagnetic and Thermal Analysis of Permanent Magnet … 413

190

170

150
Temperature (oC)

130

110

90

70

50

30
0 10 20 30 40 50 60 70 80 90 100 110 120 130 140 150 160 170 180
Time (minutes)

Rotor Core (ºC) Rotor Magnet (ºC) Stator Back Iron (ºC)
Winding (ºC) End Winding (ºC)

Fig. 10 Temperature variation—BLDC with ferrite

ferrite magnet motor and the temperature rise is depicted in Fig. 10. In this motor, the
winding temperature has been reached up to 187 °C. The high current requirement of
BLDC motor with ferrite magnet increases the temperature up to 85%, than NdFeB
magnet motor.

5 Conclusion and Future Work

In this paper, various BLDC motors are designed for wiper application. From the
analysis, it can be concluded that preferring top-down winding arrangement than
side by side winding arrangement, reduces the losses by 2.3%. The performance
of the BLDC motor using various magnets such as NdFeB, Alnico and ferrite are
also compared. Even though BLDC motor with NdFeB magnets provides better
414 A. Jategaonkar et al.

efficiency, the low mass density of ferrite enhances the power to mass ratio. The
thermal characteristics of the NdFeB and ferrite magnet machines are analyzed and
compared. The future work is planned to make the design changes in ferrite magnet
BLDC motor, in order to reduce the temperature rise.

References

1. Sun, L., Gyu, K., Jin, H., Byoung, K.: Stator and rotor shape designs of interior permanent
magnet type brushless DC motor for reducing torque fluctuation. IEEE Trans. Magn. 48(11),
4662–4665 (2012)
2. Baharudin, N., Ayob, M.: Brushless DC motor drive control using single input fuzzy PI con-
troller (SIFPIC). In: IEEE Conference on Energy Conversion (CENCON), pp. 13–18. IEEE
Explore, Malaysia (2015)
3. Shokri, G., Naderi, E.: Research on simulation and modelling of simple and cost effective
BLDC motor drives. Int. J. Model. Simul. 37(1), 15–24 (2016)
4. Moussa, K., Mourad, M., Ammar, M.: An inverse problem methodology for design and opti-
mization of an interior permanent magnetic BLDC motor. Int. J. Hydrogen Energy 42(28),
17733–17740 (2017)
5. Tae, L., Myung, S., Yong, K., Sang, J.: Motor design and characteristics comparison of outer-
rotor-type BLDC motor and BLAC motor based on numerical analysis. IEEE Trans Appl.
Supercond 26(4), (2016)
6. Abbas, M., Kh, K.: Thermal analysis of BLDC motor with propose new arrangement for
permanent magnets to magnet eddy current loss reduction. In: 24th Iranian Conference on
electrical engineering (ICEE), pp. 1769–1774. IEEE, Iran (2016)
7. Alireza, S., Armin, T., Massoud, P., Aminollah, K., Abdolhay, A.: Optimization of brushless
direct current motor design using an intelligent technique. ISA Trans. 57, 311–321 (2015)
8. Clark, K.: Electronic design of compact BLDC motor control. SAE Int. J. Passeng Cars
Electron. Electr Syst. 8(2), 229–239 (2015)
9. Parente, D., Villani, M., Canosa, A., Nanni, G.: Noise reduction in permanent magnet DC motor
for rear wiper. In: IEEE International Electric Machines and Drives Conference (IEMDC),
pp. 582–584. IEEE, USA (2001)
10. Liu, Y., Jing, Z., Lei, Y., Xiangdong, L., Lei, Z.: Investigation of a stator-ironless brushless DC
motor with non-ideal back-EMF. IEEE Access 7, 28044–28054 (2019)
11. Tayfun, G., Guven, K.: Design of permanent magnet machines with different rotor type. Int. J.
Electr. Comput. Eng. 4(10), 1510–1515 (2010)
12. James, K., Pyung, H.: Thermal management of electric vehicle BLDC motor. In: 16th Asia
Pacific Automotive Engineering Conference. SAE International, India (2011)
13. Yunbing, W., Wenzhong, Z., Genliang, L., Jianqin, F.: Mechanical properties test for wiper
motors using wavelet ridge. In: International Conference on Electrical Machines and Systems,
pp. 255–258. IEEE, China (2008)
14. Marco, P.: PSpice simulation of the power stage for DC brush motors using state of the art
power MOSFETs. In: IEEE Vehicle Power and Propulsion Conference, pp. 1567–1572. IEEE,
USA (2009)
15. Fazle, E., Mohammad, R.: Intelligent windshield for automotive vehicles. In: 17th International
Conference on Computer and Information Technology (ICCIT), pp. 392–396. IEEE, USA
(2014)
16. Hong, K., Byung, K.: Optimal design of motor shape and magnetization direction to obtain
vibration reduction and average torque improvement in IPM BLDC motor. IET Electr. Power
Appl. 11(3), 378–385 (2017)
17. Qinghu, Z., Siwei, C., Dong, W., Zhewu, J.: Multiobjective design optimization of high-power
circular winding brushless DC motor. IEEE Trans. Industr. Electron. 65(2), 1740–1750 (2018)
Improved Sector-Based DTC-SVM
for Induction Motor Drive Using Hybrid
Fuzzy-PI Controller

Abhimanyu Sahu, Kanungo Barada Mohanty, and Rabi Narayan Mishra

Abstract This paper focuses on the space vector modulation (SVM) strategy applied
to direct torque control (DTC) to overcome the high ripples content in torque and
improve the speed response using hybrid fuzzy-PI controller. A key feature of DTC
is its ability to control electromagnetic torque in direct by proper selection of voltage
vectors. In DTC-SVM, the division of six sectors of stator flux gives only limited
control over the voltage vector. So, further division of the stator flux locus from
six sectors to twelve sectors is proposed here which leads to better choice of volt-
age vector. Unlike conventional DTC-SVM, a more advanced fuzzy-PI controller is
introduced with the DTC-SVM in place of the hysteresis torque controller to reduce
the ripple content in torque. So a novel strategy for the selection of optimum voltage
vector from a hybrid fuzzy logic (FL) controller is developed here which has the
flux locus divided into twelve sectors would achieve minimized torque ripple and
improved speed response.

Keywords Induction motor (IM) · Hybrid fuzzy logic (FL) controller · Direct
torque control (DTC) · Space vector modulation (SVM)

1 Introduction

Induction motors are employed in industrial applications which require variable

speed control. These motors have several advantages such as they are inexpensive,
lightweight and have low maintenance cost in comparison with the dc motor. How-
ever, for variable speed applications of an induction machine, it requires control over
frequency, voltage and current. A controller should be employed in speed control of
the induction motor in order to achieve maximum torque with a minimum loss. DTC
seems to be a mature and viable method in doing so [1, 2]. In conventional DTC,

A. Sahu (B) · K. B. Mohanty

National Institute of Technology, Rourkela, India
e-mail: [email protected]
R. N. Mishra
Silicon Institute of Technology, Bhubaneswar 751024, India
© Springer Nature Singapore Pte Ltd. 2020 415
G. Pradhan et al. (eds.), Advances in Electrical Control and Signal Systems,
Lecture Notes in Electrical Engineering 665,
https://doi.org/10.1007/978-981-15-5262-5_30
416 A. Sahu et al.

switching table is used to acquire an optimum voltage vector among the inverter
generated feasible voltage vectors. The voltage vector selected should ensure that
the electrical torque and stator flux produced stay within the defined band limit. In
principle, both stator flux and torque are controlled separately in a DTC without using
any PWM algorithm, rotary coordinate conversion module or current feedback [3,
4]. Therefore, advantageous are when compared to the field-oriented control method.
However, conventional DTC suffers from fluctuation in torque and flux. It is
attributed to the fact that the generated six non-zero voltage vectors by the voltage
source inverter cannot meet the required torque demand at all times [5]. The required
torque is satisfied only a few instants of switching and in most of the cases the torque
produced is either more or less than the required, which generates ripples in both
torque and flux. Inverter switching frequency is increased by space vector modulation
technique in order to reduce the torque ripple [6, 7]. The switching frequency can
also be varied, with additional complexities, by the method of dead beat controller
[8]. More control over the voltage vector was achieved on division of stator flux
locus into twelve sectors, which provided a total of fourteen voltage vectors [9]. A
comparative study of conventional and twelve sectors performed well at almost the
majority of the performance studies [10]. A fuzzy control technique is applied to
DTC sensor less MRAS IM drive, especially to improve the low speed performance
and dynamic response [11]. In paper [12], a modified look up table for the selection of
switching vectors is proposed and a fuzzy controller system is applied which selects
proper voltage vector for minimum torque ripples. In paper [13] and [14], various
soft computing approaches are implemented to extensively reduce the torque ripples.
Replacing the hysteresis torque controller with hybrid fuzzy-PI controller showed
excellent result. The simulation result is compared for torque ripple, speed response
and stator current using conventional DTC-SVM and modified hybrid fuzzy-PI-based
DTC-SVM.

2 Conventional DTC-SVM

Torque equation in the synchronous reference frame is

3 P 
Te = ψαs i βs − ψβs i αs (1)
22
−
→
Consider a stationary reference frame having stator flux vector ψs and rotor flux
−
→
vector ψr rotating at synchronous speed in an anticlockwise direction as shown in
Fig. 1.
−
→ −
→
Let the stator flux ψs leads the rotor flux ψr by an angle γsr . At a particular instant
−
→ −
→
of time, the angle between ψs and the α-axis be θ f s and that of ψr and the α-axis
be θ f r . α-axis and β-axis vector components of flux and current are implemented in
torque (1). The modified torque equation is
Improved Sector-Based DTC-SVM for Induction Motor Drive … 417

Fig. 1 Stationary reference

frame having stator and rotor
flux vector

3 P L m − →−
→
Te = ψs ψr  sin(γsr ) (2)
2 2 σ Ls Lr

Since the inductance and the number of poles in a machine are fixed, the torque
−
→ −
→
can be controlled by changing the magnitude of ψs and ψr or by changing the angle
between them. In order to prevent the machine from operating at over saturation
condition or underutilized condition, the flux magnitude is kept constant. So, to
control the torque, the angle between the two fluxes is to be controlled using space
vector method [15]. The relation between the voltage vector and the stator flux while
neglecting the resistance drop is
−
→
d ψs
=−
→
us (3)
dt
Assuming the voltage vector to remain the same for an infinitesimal duration of
time, the equation can be linearized as
−
→ − −
→
ψs = →
u s t + ψ s|t=0 (4)

−
→
Hence, on proper choice of the voltage vector, the stator flux ψs can be varied.
−
→ −
→
Thus, the angle between ψs and ψr can be varied and hence the torque can be
controlled.
The voltage vectors are generated using a voltage source inverter, out of the
generated eight voltage vectors six are non-zero active voltage vectors (u1 , u2 ,…, u6 )
and the remaining two are zero voltage vector (u0 and u7 ). The spatial arrangement
of the voltage vector and the stator flux locus segmentation is shown in Fig. 2.
418 A. Sahu et al.

Fig. 2 Sector
representations along with u3 (0 1 0) u2(1 0 1)
voltage space vector
S3
u2
u3
S2
S4
uv u5 u6
(0 1 1)u4 u1(1 0 0)
S1
S5
S6
u5

Applying the active voltage vector will cause the stator flux to move at higher
rate, whereas it almost stops in the case of zero voltage vectors. Considering an
anticlockwise rotating stator flux lying in sector 1, the effect on it on the application
of various voltage vectors at that particular instant of time has been shown in Fig. 3.
The sector that contains the stator flux can be determined from the flux angle and
it is obtained as
ψβs
ρs = tan−1
ψαs

2.1 Basic Operation

The flux hysteresis controller is a two-level hysteresis controller and torque hysteresis
controller is three-level, shown in Fig. 4 [16]. A small flux hysteresis band demands
a high switching frequency. If the switching frequency is high, the stator flux locus
approaches a circle and a flux are controlled within the hysteresis flux error band.
When the actual flux linkage vector is less than the reference vector, the error becomes
positive. This positive value is then passed through the hysteresis controller. If the
positive error so obtained is greater than the upper tolerance value, the output from the
hysteresis controller is 1. If the negative error so obtained is more negative than the
lower tolerance value, the output from the hysteresis controller is −1. This instructs
the processor to decrease the flux linkage.
−   →
→ − 
ψs = 1 if ψs  ≤ ψs∗  − |hysteresis band|
−   →
→ − 
ψs = 0 if ψs  ≥ ψs∗  + |hysteresis band|
Improved Sector-Based DTC-SVM for Induction Motor Drive … 419

Fig. 3 Torque and flux variations to various voltage space vectors

The torque hysteresis controller is a three-level hysteresis controller and has values
1, 0 and −1. Small values of torque hysteresis band help in smooth generation of
torque. The error obtained is the difference between the actual torque vector and the
reference torque vector. The positive, negative and zero value of error gives an output
of 1, −1 and 0 is shown, Te∗ is the reference and Te is estimated torque.
420 A. Sahu et al.

Fig. 4 Classical DTC-SVM-based IMD

Table 1 Six sector switching table

Flux Torque S1 S2 S3 S4 S5 S6
P P u2 u3 u4 u5 u6 u1
ZE u0 u7 u0 u7 u0 u7
N u6 u1 u2 u3 u4 u5
N P u3 u4 u5 u6 u1 u2
ZE u7 u0 u7 u0 u7 u0
N u5 u6 u1 u2 u3 u4
P Positive, ZE Zero, N Negative, S x Sector, ux Voltage vector

Te = 1 if Te < Te∗ − |hysteresis band|

Te = 0 if Te = Te∗
Te = −1 if Te > Te∗ + |hysteresis band|

Based on the outputs of flux and hysteresis controller and sector to which the
flux lays an optimum voltage vector such that torque and flux errors are kept within
the limit. The voltage vector is generated by giving the proper command signal to
the inverter. The optimum voltage vector that is suited for the specific change in the
torque and flux in a given vector is tabulated in Table 1.

3 Hybrid Fuzzy-PI Controller

In the course of time, several methods have been implemented to reduce the ripples
in torque and flux and hence achieve better performance characteristics. The major
Improved Sector-Based DTC-SVM for Induction Motor Drive … 421

cause of ripple in the conventional DTC scheme is due to the hysteresis controller
ripples, hence, replacing the hysteresis controller with more advanced control meth-
ods will aid in reducing the ripple. As a strategy, the hysteresis torque controller is
replaced with hybrid fuzzy-PI controllers that create more accurate voltage vectors
thus obtaining desired torque response. For further improvement in speed tracking,
an additional fuzzy controller can be added, as shown in Fig. 5.
The overall performance is evaluated by the membership function shape, the
defuzzification method and the fuzzy reasoning logic used in the proposed method.
The output of the fuzzy controller is the desired voltage. The voltage is generated by
the voltage source inverter by converting the voltage controller output into the proper
trigger signal for the inverter. In the conventional DTC scheme, there are two states
per sector which creates an ambiguity in torque. Therefore, their usages are limited.
The novelty in dividing the stator flux into twelve sectors, instead of six, allowed the
use of all six active states. Which gave birth to the concept of twelve-sector modified
DTC. The modified stator flux locus is shown in Fig. 6. With the modified DTC, the
idea of small torque variation came, mainly attributed to the fact that the tangential
voltage vector component has very small amplitude and variation in the produced
torque is also small. Thus, it is necessary to define small and large variations. For
a stator flux linkage, space vector lying in the sector S 12. The variations in torque
on the application of different voltage vectors will clear the idea of small and large
torque changes. u1 will cause huge variation in flux magnitude and slight variation
in torque magnitude, whereas u2 will do the vice versa. This idea of twelve sectors
as shown in Fig. 6 is implemented through a fuzzy controller in the torque loop so
that appropriate voltage vectors are chosen that maintains the flux and torque ripples
to the minimum amount. With the division of the flux sector from six to twelve gives
more accurate choices in voltage vector for torque variations.
In general, a fuzzy logic controller (FLC) as shown in Fig. 7 contains four parts,

Fig. 5 Hybrid fuzzy-PI controller-based IM drive

422 A. Sahu et al.

Fig. 6 Modified 12 sector representations

Fig. 7 Fuzzy logic controller block diagram

two of them perform the transformation. A fuzzifier performs the task of transforming
the measured signal into fuzzy quantities, which implies all the crisp input quanti-
ties with well-defined value are converted into linguistic variables in literature. It
performs the functioning of measuring the input variable, scaling and fuzzification
(transformation 1). Transformations are performed using membership functions. A
membership value of 1 indicates that the quantity absolutely belongs to that set and
a membership value of 0 indicates that the quantity does not belong to that set. The
knowledge base is the memory block which contains the database and the linguistic
control rule base. The database provides the information which defines the linguistic
control rules and the fuzzy data manipulation in the fuzzy logic controller.
The inference engine is the kernel of a fuzzy logic controller and has the capability
of both performing fuzzy control action by using the fuzzy set and the fuzzy rule
base and of simulating a human decision-making based on fuzzy concepts.
The defuzzifier performs the defuzzification as well as scale mapping, which is
the second transformation in the FLC. The defuzzifier performs the reverse action
of a fuzzifier; it produces a crisp output value from the inferred fuzzy control action
by using membership function (MF).
The speed response is improved by using a fuzzy logic controller in the speed
loop, as shown in Fig. 5. Since the response has become fast, the reference is tracked
as quick as possible.
Improved Sector-Based DTC-SVM for Induction Motor Drive … 423

Hybrid torque controller consists of a fuzzy logic controller (FLC) along with PI
controller as shown in Fig. 5 is used in torque loop of the induction motor drive.
Even when the load changes the torque ripple of the IM is controlled by using a
hybrid torque controller. The inputs of the fuzzy controller are torque error (E T )
and torque error deviation (d E T ), whose output is multiplexed with the output of PI
which is connected in parallel with fuzzy controller. The multiplexed output is given
to the switching table to generate pulse for inverter switching. In hybrid fuzzy-PI,
the switching time is reduced and so that ripple contain in torque is reduced.

E T (K ) = Te∗ − Te
E T (K ) − E T (K − 1)
d E T (K ) =
Ts

where T s is the sampling period. The outcome of the FLC is given to the PI controller.
Torque reference T *e is obtained as the output from the PI controller, which is the
output of the hybrid speed controller. Fuzzy membership functions for torque error
and speed are shown in Fig. 8.
The speed fuzzy is designed for seven membership functions, these are negative
large (NL), negative (N), negative small (NS), zero (Z), positive large (PL), positive
small (PS) and positive (P), as shown in Fig. 8a. In torque control, the torque error
is presented by four MF’s, which makes the fuzzy variable small. These are either
trapezoidal or triangular MF’s specified by positive (P), positive small (PS), negative
(N) and negative small (NS), as shown in Fig. 8b.
The hybrid fuzzy-PI controller output is the calculated optimum voltage vector,
which is then used to generate the pulse signals for the inverter. The input variables

μ(ωe)

NL N NS ZE PS P PL

(a)

μ(Te)

N NS PS P

-0.4 -0.01 0.01 0.4

(b)
Fig. 8 a MFs of input speed error, b modified torque error MFs
424 A. Sahu et al.

are assigned as error in flux e , error in torque et , the position of flux θ s and the output
variable is assigned as voltage vector ‘u’. The voltage controller was used to track
the reference torque and flux. Ripples present in conventional DTC technique are
minimized by using a voltage controller, thus the efficiency of the machine increases.
The output of hybrid fuzzy-PI controller, hysteresis controller for flux error and
the position of flux θ s are given to the twelve-sector switching table to generate
pulses for inverter switching (Table 2).

4 Simulation Results and Analysis

A squirrel cage induction motor having parameters as shown in Table 3 is chosen

for the simulation with MATLAB–Simulink software. The simulation is done for a
target speed of 700 rpm and stator flux linkage of 0.8 Wb. The torque input given
has a step change from 0 to 10 Nm at 0.3 s. Simulations of both the conventional
and proposed methods are presented for comparative analysis and validation of the
proposed method.
Figures 9 and 10 show the responses for the conventional DTC and the proposed
twelve-sector hybrid fuzzy controlled DTC scheme running at a speed of 700 rpm.
The initial reference torque was 0 Nm for a period of 0.3 s and then a step change
from 0 to 10 Nm was applied to the machine. The runtime of the simulation is for
0.6 s in discrete mode.
The torque ripple magnitude has been reduced from 1.6 to 0.8 Nm by the proposed
method, as shown in Figs. 9a and 10a. The speed response is smooth in hybrid fuzzy
as compared to conventional method, as shown in Figs. 9b and 10b. The current
response is also smooth, less ripple and fast response in hybrid fuzzy as compared
to the conventional method.

5 Conclusion

Performance enhancement of the induction motor is achieved through an improved

twelve-sector fuzzy controlled DTC. The proposed scheme is a hybrid method that
combines the advantages of twelve-sector stator flux and hybrid fuzzy-PI controller
in order to reduce the ripple in flux and torque. The work began with explaining
basic DTC and twelve-sector fuzzy controlled DTC scheme. In twelve-sector fuzzy
controller, more switching pulse duration is given as compared to hybrid fuzzy-
PI controller. Hence, a proposed hybrid fuzzy-PI controller incorporated with an
improved twelve-sector-based DTC is much more effective in minimizing the ripple
and with the added proposed scheme, and the torque ripple has been reduced to a
much greater extent. Reduced torque ripple gives smooth operation for the induction
motor. The proposed method limits the high starting torque with a slight longer time
Table 2 Twelve-sector switching table
Flux Torque S1 S2 S3 S4 S5 S6 S7 S8 S9 S 10 S 11 S 12
P P u2 u3 u3 u4 u4 u5 u5 u6 u6 u1 u1 u2
sP *u2 u2 *u3 u3 *u4 u4 *u5 u5 *u6 u6 *u1 u1
sN u1 *u1 u2 *u2 u3 *u3 u4 *u4 u5 *u5 u6 *u6
N u6 u1 u1 u2 u2 u3 u3 u4 u4 u5 u5 u6
N P u3 u4 u4 u5 u5 u6 u6 u1 u1 u2 u2 u3
sP u4 *u4 u5 *u5 u6 *u6 u1 *u1 u2 *u2 u3 *u3
sN u7 u5 u0 u6 u7 u1 u0 u2 u7 u3 u0 u4
Improved Sector-Based DTC-SVM for Induction Motor Drive …

N u5 u6 u6 u1 u1 u2 u2 u3 u3 u4 u4 u5
P Positive, ZE Zero, N Negative, SP Small positive, SN Small negative, S x Sector, ux Voltage vector and *ux Next optimum voltage vector
425
426 A. Sahu et al.

Table 3 Machine parameters

Parameter Value
Power 3.7 kW
Stator resistance 3.32
Stator inductance 4.39 mH
Rotor resistance 2.11
Rotor inductance 4.39 mH
Mutual inductance 0.2373 H
Pole pair 2

Torque Vs Time
20

15
11
Torque(Nm)

10 10
9
5 0.4 0.45

-5
0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5 0.55 0.6
Time (seconds)
(a)

(b)

(c)

Fig. 9 Conventional DTC-SVM scheme of a torque, b speed, c stator current

Improved Sector-Based DTC-SVM for Induction Motor Drive … 427

Torque Vs Time
20

15 10.5
10
10
Torque(Nm)

9.5
5 0.4 0.45

-5

-10
0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5 0.55 0.6
Time (seconds)
(a)

(b)

(c)
Fig. 10 Twelve-sector DTFC-SVM scheme using hybrid fuzzy-PI in term of a torque, b speed,
c stator current
428 A. Sahu et al.

to reach the reference speed. The ripple in stator current is reduced with smooth
response in the proposed method as compared to the previous conventional method.

References

1. Takahashi, I., Noguchi, T.: A new quick-response and high-efficiency control strategy of an
induction motor. IEEE Trans. Ind. Appl. 22(5), 820–827 (1986)
2. Depenbrock, M.: Direct self-control (DSC) of inverter-fed induction machine. IEEE Trans.
Power Electron. 3(4), 420–429 (1988)
3. Casadei, D., Serra, G., Tani, K.: Implementation of a direct control algorithm for induction
motors based on discrete space vector modulation. IEEE Trans. Power Electron. 15(4), 769–777
(2000)
4. Casadei, D., Profumo, F., Serra, G., Tani, A.: FOC and DTC: two viable schemes for induction
motors torque control. IEEE Trans. Power Electron. 17(5), 779–787 (2002)
5. Kang, J., Sul, S.: Analysis and prediction of inverter switching frequency in direct torque
control of induction machine based on hysteresis bands and machine parameters. IEEE Trans.
Industr. Electron. 48(3), 545–553 (2001)
6. Lai, Yen-Shin, Chen, Jian-Ho: A new approach to direct torque control of induction motor
drives for constant inverter switching frequency and torque ripple reduction. IEEE Trans.
Energy Convers. 16(3), 220–227 (2001)
7. Habetler, T.G., Profumo, F., Pastorelli, M., Tolbert, L.M.: Direct torque control of induc-
tion machines using space vector modulation. In: Conference Record of the IEEE Industry
Applications Society Annual Meeting, Dearborn, MI, vol. 1, pp. 428-436 (1991)
8. Zhang Y., Yang, H., Li, Z, A simple SVM-based deadbeat direct torque control of induction
motor drives. In: International Conference on Electrical Machines and Systems (ICEMS),
Busan, pp. 2201–2206, (2013)
9. Kumar, B.S., Gupta, R.A., Kumar, R.: 12-Sector methodology of torque ripple reduction in a
direct torque controlled induction motor drive. In: SICE-ICASE International Joint Conference,
Busan, pp. 3587–3592 (2006)
10. Essaadi, M., Khafallah, M., Saad, A., Hamdoun, A., Chaikhy, H.: A comparative analysis
between conventional and new direct torque control strategies of induction machine. In: 2nd
World Conference on Complex Systems (WCCS), Agadir, pp. 350–354 (2014)
11. Zhao, T., Xiang, F., Wang, J., Zhang, G., Mao, J.: Research on speed sensorless direct
torque fuzzy control of induction motors. In: International Conference on Electronics and
Communication Systems, Coimbatore, pp. 1–5 (2014)
12. Gudhe, S., Pimple, B.B.: Improved torque response of induction motor drive using direct torque
control technique applying fuzzy logic control. In: IEEE 1st International Conference on Power
Electronics, Intelligent Control and Energy Systems (ICPEICES), pp. 1–6, (2016)
13. Mishra, R.N., Mohanty, K.B.: Design and implementation of a feedback linearization controlled
IM drive via simplified neuro-fuzzy approach. IETE J. Res. 64(2), 209–230 (2017)
14. Mishra, R.N., Mohanty, K.B.: Design and realization of an auto-tuned modified neuro-fuzzy
sliding-mode-based IM drive deploying feedback linearization. EPE J. 28(1), 28–42 (2018)
15. Lang, B., Miao, M., Liu, W., Luo, G.: Simulation and experiment study of space vector pulse
width modulation. In: 9th International Conference on Electronic Measurement & Instruments,
Beijing, pp. 1408–1412 (2009)
16. Vasudevan, M., Arumugam, R.: New direct torque control scheme of induction motor for
electric vehicles. In: 5th Asian Control Conference (IEEE Cat. No.04EX904), Melbourne,
Victoria, vol. 2, pp. 1377–1383 (2004)
Detection and Classification of Brain
Tumor Using Magnetic Resonance
Images

Limali Sahoo, Lokanath Sarangi, Bidyut Ranjan Dash,

and Hemanta Kumar Palo

Abstract The paper aims to provide a comparative study on the detection and
classification of brain tumors (BT) using different machine learning algorithms. In
the process, different popular and commonly BT image data sets such as the BRATS,
OASIS, and the NBTR have been used for the said purpose. The pre-processed BT
images are enhanced using the filtering approach and then segmented using the fuzzy
C-means (FCM) algorithm for the extraction of suitable and reliable features. The
multi-resolution capability of wavelet transform (WT) has been explored to extract
the detailed coefficients for simulation of the chosen classifiers. The recognition accu-
racy of the classification algorithms such as the K-nearest neighbor (KNN), decision
tree (DT), neural network (NN), discriminant analyzer (DA), support vector machine,
and Naive Bays’ (NB) have been compared for their applicability in classifying BT
images. The highest average recognition accuracy of 96.4% has been reported with
the KNN algorithms for the OASIS data set as revealed from our results.

Keywords Brain tumor · Segmentation · Magnetic resonance · Clustering · Image

enhancement · Classification

L. Sahoo · H. K. Palo (B)

Institute of Technical Education and Research, Siksha ‘O’ Anusandhan (Deemed to Be
University), Bhubaneswar, Odisha, India
e-mail: [email protected]
L. Sahoo
e-mail: [email protected]
L. Sarangi
College of Engineering, Bhubaneswar, Odisha, India
e-mail: [email protected]
B. R. Dash
Gandhi Polytechnic, Berhampur, Odisha, India
e-mail: [email protected]

© Springer Nature Singapore Pte Ltd. 2020 429

G. Pradhan et al. (eds.), Advances in Electrical Control and Signal Systems,
Lecture Notes in Electrical Engineering 665,
https://doi.org/10.1007/978-981-15-5262-5_31
430 L. Sahoo et al.

1 Introduction

Brain tumor (BT) is an abnormal growth of cells inside the brain cranium and can
occur in human being irrespective of their age. Its appearance limits brain function
and may cause death if not detected and treated in an early stage. The tumors in
the brain may be either malignant or benign [1]. While malignant BTs are rapidly
expanding and cancerous, the growth of benign BTs is rather slow and stagnant. Accu-
rate diagnosis of BTs requires accurate images using high resolution and advanced
techniques such as computed tomography (CT), ultrasonography (US), magnetic
resonance imaging (MRI), and positron emission tomography (PET) [2, 3]. In com-
parison with other methods, the MRI is a non-invasive method and has been most
preferred as it provides high-quality images of BTs and cancerous tissues [3–5].
The method uses a strong magnetic field as well as the radio waves of the human
body to generate BT images for better analysis. It can distinguish soft tissues with
high resolution and sensitivity to visualize or detect subtle changes that occur in
tissue density. The MRI images help clinicians to focus on precise lesion diagnosis
and provide numerous image acquisition schemes with multiple images of different
contrast for investigation of similar tissues [6]. Early identification and diagnosis of
cancerous brain tissues will benefit the patients and society in many ways. The anal-
ysis of BT requires the extraction of suitable features from segmented MR images
and equally efficient classifiers for recognition and detection. There have been many
machine learning algorithms successfully employed in this area each has its advan-
tages and limitations. One such algorithm is the unsupervised fuzzy C-means (FCM)
clustering attempted by a few researchers with excellent results. The algorithm has
outperformed the supervised feedforward cascade correlation (FFCC) in classifying
BTs using the segmented MR brain images [7]. Nevertheless, the support vector
machine (SVM) remains a better alternative than the FCM method of classifica-
tion [8]. The BT images enhanced using the mathematical morphological technique
before segmentation and clustered using the K-means algorithm is one such effort
in this direction. With linear kernel function, the reported average BT accuracy has
been 80% when the SVM classifier has been trained using the features of BT MR
images [9]. As compared to this, the accuracy has been 70%, 70%, and 65% under
similar conditions using the polynomial, RBF, and quadratic kernel, respectively [9].
The wavelets reduced using the independent component analysis (ICA) dimension
reduction tool in an SVM platform have resulted in enhanced accuracy [10]. The
authors have compared several classifiers such as the feedforward ANN (FFANN),
extreme learn machine (ELM), SVM, and an ensemble classifier in the recognition
of BT MRI images [11]. The fusion at the classification stage has provided the high-
est accuracy of 91.17% as compared to 84.44% with FFANN, 86% with ELM, and
89.67% using the SVM. Recently, the convolutional neural network (CNN) has been
a state-of-art algorithm applied in this field with excellent results. The investigation
of malignant and non-malignant tumors in the brain using the BT features extracted
from the CNN itself has resulted in low complexity with a very faster response in the
NN platform [12]. The reported classification accuracy has been 97.18% for cranial
Detection and Classification of Brain Tumor Using Magnetic … 431

MR images in the ELM-local receptive field platform as compared to 96.91% using

the AlexNet. The effectiveness of ELM-LRF is attributed to the random generation
of input weights and the connection as claimed by these authors.

2 Brain Tumor Database

The development of a suitable database of BT is an important step in its diagnosis

and detection. The database chosen is essential for the design and validation of the
segmented MR images. It must be chosen judiciously to ensure and authenticate
the chosen feature segmentation method, feature extraction technique, and the clas-
sification algorithm for accurate results. The three data sets used in this work for
comparison are OASIS, BRATS, and NBTR. While OASIS involves Alzheimer’s
disease (AD), the BRATS data set focuses on glioma BT and the NBTR data set is
covers brain cancer data. The studies show that the AD disease is correlated to brain
tumor and BT can lead to cancer [13, 14]. Initially, the work aims to classify BT using
the BRATS data set. However, due to a close correlation among the OASIS, BRATS
and the NBTR data sets, a comparison platform is set to determine the classification
accuracy using these three data sets. Arguably, the classification accuracy varies;
although a more accurate comparison can be made in case of data sets particularly
of BT can be used.
Observation shows that the BRATS and OASIS involve BT and AD disease,
respectively, which are correlated. Thus, there has been a little different in the accu-
racy level using these two data sets. However, the NBTR data set relates to brain can-
cer, which shows reverse characteristics than AD, i.e., persons with AD are less
risky to have brain cancer and vice versa. This observation leads to a sizable differ-
ence in recognition accuracy of NBTR when compared either with OASIS or BRATS.
Among the MRI modalities, the OASIS data set consists of T1-weighted MRI, T2-
weighted MRI, T1-weighted MRI with gadolinium contrast enhancement (T1-Gd),
and fluid attenuated inversion recovery (FLAIR). Similarly, the BRATS data set and
NBTR have the T1 modalities. To make the comparing platform uniform, we have
opted for T1 modalities of MR images in this work.
A few of the BT MR databases that have been used by most of the researchers in
this field in the characterization and classification of cancerous tissues are described
below.

2.1 BRATS Database

The BRATS is real MRI brain tumor data set involving 811 Children. It consists of 750
malignant, 50 benign, and 11 normal MRI tumor images in jpg format [BRATS] [15].
The BRATS data set provides segmentation ((both low-grade and high-grade) along
with the information on active tumor and ‘edema’ of the BT. The data division ratio
432 L. Sahoo et al.

Fig. 1 BRATS data set samples

is 70%/15%/15% for training, testing, and validation of the classifiers respectively.

The samples of the BRATS data set are shown in Fig. 1.

2.2 The NBTR Data Set

It comprises of primary malignant and benign brain cancer arising in a human central
nervous system in the skull and the pituitary gland [NBTR]. The data collection
has been permitted by the English Cancer Registries of the National Information
Governance Board [16]. NBTR covers both benign and malignant primary brain
cancer, arising in the central nervous system (CNS), the skull base, and pituitary
gland. The samples of the NBTR data set are shown in Fig. 2.

2.3 The OASIS Data Set

It has been collected by Dr. Randy Buckner at the Howard Hughes Medical Institute
of Harvard University. It consists of 416 BT images involving infants and children
up to 18 years old [17]. The samples of the OASIS data set are shown in Fig. 3. The
OASIS data set consists of T1-weighted MRI, T2-weighted MRI, and T1-weighted
MRI with gadolinium contrast enhancement (T1-Gd) and fluid attenuated inversion
recovery (FLAIR) modalities. The T1_w and T2_w are unable to segment the old
and new tumors from unaffected lesions. The analysis of BT using these modalities
Detection and Classification of Brain Tumor Using Magnetic … 433

Fig. 2 NBTR data set samples

Fig. 3 OASIS data set samples

requires advanced DSIP and computerized tools. In FLAIR modality, the signals of
water molecules are suppressed. This assists in identifying the edema region from
the cerebrospinal fluid (CSF)
434 L. Sahoo et al.

3 Approaches

Detection and classification of BTs are generally accomplished using the following
basic steps as shown in Fig. 4.
These steps include BT data acquisition (collection of MRI samples), pre-
processing/image enhancement, image segmentation, feature extraction, and clas-
sification. These steps are explained briefly below.

3.1 Pre-processing

It helps to remove the background noise and improves the quality of the MRI
BT images for further processing. The features extracted from pre-processed can
describe the desired BT parameters with clarity and effectiveness. Pre-processing of
the images is essential to improve the signal-to-noise ratio (SNR), get rid of the irrel-
evant noise or undesired background segments, maintenance of prominent edges,
etc. [1].

3.2 Image Enhancement

On receiving a BT image, it is desired to enhance the image to identify the suspicious

region. Different image enhancement techniques are employed to get finer details
of the BT image by eliminating noise or redundant information from the collected
image [18]. Image enhancement helps to provide a more accurate classification of

Original BT Image Enhanced BT Image Segmented

(MRI Data) (Pre-processed) BT Image

Classification Discriminating BT
Image Features

Fig. 4 Basic steps of brain tumor classification using image processing

Detection and Classification of Brain Tumor Using Magnetic … 435

Fig. 5 Enhancement of a brain image

BT as compared to clinical MRI images. Most of the image enhancement methods

use filtering approaches such as weighted median filter, anisotropic filter, or wavelet
de-noising method to eliminate salt and pepper noise, background noise, etc. [19].
Figure 5 provides a comparison between the original and enhanced brain images
using the median filtering approach. The objective is to remove the noise from the
MR images which has been accomplished using the median filter (nonlinear digital
filtering) to the gray-scale image of the original MR. The filter has an excellent
noise reduction capabilities and is very suitable to remove impulse or salt and pepper
noise. Although it blurs the image somewhat, however, the blurring is quite less as
compared to other filtering methods such as the mean. Unlike the mean filter, a single
unrepresentative pixel in a neighborhood can affect the mean significantly but not
the median. The median filter is a more robust method that can preserve the edges
during noise removal hence has been chosen in this work.

3.3 Segmentation

It is an important step to detect the BT in the MR image. In this step, the BT image
is segmented into smaller parts or mutually exclusive regions for better analysis and
to obtain finer detail of the image. It remains crucial to facilitate the delineation,
characterization, and visualization of affected regions in MRI images. However,
the segmentation of MRI images often poses a challenge due to the presence of
diverse image content, cluttered objects, image noise, occlusion, insufficient contrast
between adjacent brain tissues, non-uniform object texture, etc. [20].
The algorithms used to segment images are accomplished using two fundamen-
tal properties of image intensity values. These are (a) discontinuity (b) similarity
[19]. In the discontinuity or formal approach, the segmentation is accomplished by
partitioning the enhanced image based on a change in the level of intensity either
at the edges or at the corners. In the similarity approach, the segmentation is done
on some similarity criteria. A few of image segmentation methods popularly used
in this field are: Threshold or boundary approach, edge approach, artificial neural
436 L. Sahoo et al.

network-based segmentation, region approach (growing, region splitting, merging,

etc.), physical model-based approaches, pixel-based approach, deformation-based
approach, atlas-based approach, histogram-based methods, clustering approach (K-
means, fuzzy C-means, expectation-maximization, mean shift, etc.) [3, 5, 21–23].
The boundary approach is a commonly followed approach that uses a gray valve
remapping technique as observed in Eq. (1).

0 if v < t
I (v) = (1)
1 if v ≥ t

where t and v denote the threshold value and the gray valve, respectively. In this
approach, a conversion of gray to the binary image is done by segmenting the image
using the values 0 and 1 after thresholding. In the edge approach, the object bound-
aries are represented by the detected edges of a BT image for identification and
classification of the tumors. However, the method remains less popular due to the
need for very distinct and clear edges for closed boundaries to segment the image
directly. Further, there is a chance of false edge identification and the need to link
partial appearing edges to form a closed boundary during the segmentation process.
In the case of a region-based segmentation approach, the bordering pixels residing in
a region are assumed to have similar values. The approach aims to find the region of
an object rather than its edges. In this process, the neighboring pixels are compared to
the observed pixels to form clusters using congruence criteria for classification and
detection of BTs. A few of the clustering algorithms mostly used in this approach are
K-means clustering, fuzzy clustering, genetic algorithm (GA), and particle swarm
optimization (PSO) [24]. K-means is a hard clustering algorithm in which the chosen
features are partitioned into k-number of clusters based on some distance measure.
Although the clustering technique is faster and simple, it is difficult to obtain over-
lapping clusters or to cluster nonlinear data and noisy data. Fuzzy clustering is a
soft clustering approach in which a feature can be considered as a candidate either
in single or multiple clusters. The GA is an optimization algorithm used to tackle
higher nonlinearities, easy to implement, although it is computationally expensive.
The optimization takes place using three basic steps such as the recombination, muta-
tion, and selection candidate. Similarly, the PSO is another optimization algorithm
that is initiated using randomly selected particles or populations. In this, each par-
ticle has its fitness value and can be computed using a fitness function. There is no
direct recombination operator unlike GA and there is the possibility of the swarm
to converge permanently which is its major drawback. The integration of K-means
and fuzzy C-means (FCM) algorithm followed by thresholding and level set segmen-
tation has been approached to cluster BT from MRI images for accurate and faster
estimation [20]. An example of BT image segmentation performed in this work using
fuzzy C-means is shown in Fig. 6.
Detection and Classification of Brain Tumor Using Magnetic … 437

Fig. 6 BT image segmentation using FCM

3.4 Feature Extraction

The extraction of suitable and reliable features has been a major step in the recognition
of patterns as these parameters vary among different classes. Due to difficulties in
the extraction of suitable features from the complex human brain structure, most
researchers rely on the general shape, size, location, and composition of a BT image
for such purposes [3, 4, 22]. In this regard, the multi-resolution capability of wavelet
analysis in characterizing the BT images proves more accurate in describing the
cancerous tissues. The features of MR images extracted in the wavelet domain have
provided enhanced accuracy using different machine learners [1, 2]. The elimination
of the irrelevant information is the possible reason for the improvement in accuracy
and speed of response. Similarly, a few of the optimization tools such as the genetic
algorithm (GA) and the particle swarm optimization (PSO) often provide better BT
clustering as compared to the raw feature sets. These optimization techniques along
with FCM clustering detect BT more intelligently with less computation time than the
FCM when applied alone. However, the application of PSO with FCM has proved
more beneficial than that of GA and FCM in classification BT [6]. To extract the
desired BT features, the detailed coefficients of the wavelet transform have been
used in this work.
In wavelet analysis, the enhanced segmented BT image signal x(s) has been
decomposed as approximated coefficients a j,k and detailed coefficients d j,k cor-
responding to low and high pass filtering, respectively, as given by
⎧
⎪ 
∞
⎪
⎨ d j,k = x(s)h ∗ j (s − 2 jk)
−∞
DW T x(s) = 
∞ (2)
⎪
⎪
⎩ a j,k = x(s)g ∗ j (s − 2 jk)
−∞

where the variables j and k represent the scale and translation factors, whereas the
terms g(s) and h(s) correspond to the low and high pass filter coefficients, respectively.
438 L. Sahoo et al.

4 Classification Results

There have been many machine learning algorithms successfully employed in the
field of pattern recognition. Mostly the pattern recognition algorithms can be seg-
regated into (a) statistical techniques and (b) data comparison technique [25–28].
Statistical techniques like support vector machine (SVM), Naïve Bayes (NB), and
discriminative analysis (DA) remain complex in applying and computationally cum-
bersome. On the other hand, the data comparison techniques such as the nearest
neighbor, decision tree, and artificial neural network (ANN) are simple but time
consuming. For faster response and accurate results, most researchers often combine
these two algorithms that enhanced classification accuracy. Similarly, misdiagnosing
between a brain with a tumor and a healthy brain has been studied along with benign
tumors with malignant and malignant tumors with benign. The authors can detect
the BT using canny edge and adaptive threshold approach with excellent results. The
combination of K-means, FCM, and histogram thresholding tends to provide more
accurate classification accuracy than either of these methods when applied alone pro-
vides new insights in this field. The application of the expectation—maximization
algorithm is a simple algorithm applied in detection BT using MRI images as it pro-
vides better accuracy with less computation time. Similarly, a two-tier classification
scheme with an adaptive pillar K-means clustering algorithm in an SVM platform
using statistical features founds to be more accurate than single-stage classification
applied in the detection of BT [24]. The ANN and SVM are efficient classification
algorithms in the detection of BT; however, these are shallow architectures which
require a huge amount of nodes for effective classification. On the contrary, the deep
NN (DNN) models have superseded earlier algorithms by representing complex rela-
tionships without the use of a large number of nodes. The recent successes of the
CNNs to learn automatically representative complex features directly from the data
itself have been a breakthrough in the classification of BT from MRI images. The
network emphasizes on network architecture design by focusing on patches of MRI
images to extract a hierarchy of increasingly complex features. The authors have
claimed an excellent classification accuracy from MRI images by combining the
multi-resolution capability of the discrete wavelet transform (DWT) and efficient
DNN algorithm to classify the brain MRIs into normal or malignant brain tumors
(glioblastoma, sarcoma, and metastatic bronchogenic carcinoma).
The recognition accuracy of BT using different data sets with different classifica-
tion algorithms has been compared in Table 1. The accuracy enhancement is attributed
to a particular classifier is due to the method of feature extraction algorithm chosen in

Table 1 Recognition accuracy of BT using different machine learning algorithm

Data set NN DA SVM NB KNN DT
OASIS 93.6 94.4 92.5 95.2 96.6 96.4
BRATS 96.1 90.1 92.6 93.5 93.7 96.2
NBTR 91.4 75.9 92.6 76.1 82.3 87.0
Detection and Classification of Brain Tumor Using Magnetic … 439

Table 2 Percentage misclassification of BT using different machine learning algorithm

Data set NN DA SVM NB KNN DT
OASIS 6.4 5.6 7.5 4.8 3.4 3.6
BRATS 3.9 9.9 7.4 6.5 6.3 3.8
NBTR 8.6 24.1 7.4 23.9 17.7 13.0

this work. As shown in this table, the accuracy with the KNN classifier outperforms
all other algorithms in classifying the BT data for OASIS and BRATS data set. The
DT classifier remains the next better classification algorithm in this case. However,
better accuracy has been observed with the SVM followed by the NN classifier for
the NBTR data set. It can be concluded that the variation in accuracy is attributed to
the method of feature extraction techniques and the choice of data set.
Among the data sets, the BRATS and OASIS involve BT and AD disease, respec-
tively, which are correlated. Hence, there has been a meager difference in accuracy
level using these two data sets. However, the NBTR data set relates to brain cancer
which shows reverse characteristics than AD, i.e., persons with AD are less risk to
have brain cancer and vice versa. This observation leads to a sizable difference in
recognition accuracy of NBTR when compared either with OASIS or BRATS. The
percentage misclassification of BT using different machine learning algorithms has
been provided in Table 2.

5 Conclusion

The extraction of robust features for BT has been a major area of research since
long as the efficiency of the classification algorithm depends on the feature extrac-
tion technique employed. In this regard, the WT proves to provide the desired scope
as it describes both the time and frequency resolution of a signal using a varying
window. Similarly, the choice of a suitable data set, image enhancement, and seg-
mentation methods employed for the extraction of the desired feature sets are few
important parameters that improve the recognition accuracy as revealed from our
results. Although a series of classifiers have been simulated and test for their appli-
cability in enhancing the classification accuracy of BT images, there is still scope
for further improvement. The exploration of different enhancement, segmentation,
feature extraction, and classification algorithms can provide new research directions
for the emerging researchers in this area.
440 L. Sahoo et al.

References

1. Sawakare, S., Chaudhari, D.: Classification of brain tumor using discrete wavelet transform,
principal component analysis, and probabilistic neural network. Int. J. Res. Emerg. Sci. Technol.
1(6), 2349–7610 (2014)
2. Shree, N.V., Kumar, T.N.R.: Identification and classification of brain tumor MRI images with
feature extraction using DWT and probabilistic neural network. Brain Inform 5(1), 23–30
(2018)
3. Jalab, H.A., Hasan, A.: Magnetic resonance imaging segmentation techniques of brain tumors:
a review. Archives of Neuroscience, p. e84920 (2019, in press)
4. Mukaram, A., Murthy, C., Kurian, M.Z.: An automatic brain tumor detection, segmentation,
and classification using MRI image. Int. J. Electron. Electr Comput Syst. 6(5), 54–65 (2017)
5. Naik, J., Patel, S.: Tumor detection and classification using decision tree in brain MRI. Int. J.
Comput. Sci. Netw. Secur. (IJCSNS) 14(6), 87 (2014)
6. El-Dahshan, E.S.A., Mohsen, H.M., Revett, K., Salem, A.B.M.: Computer-aided diagnosis of
human brain tumor through MRI: A survey and a new algorithm. Expert Syst. Appl. 41(11),
5526–5545 (2014)
7. Hall, L.O., Bensaid, A.M., Clarke, L.P., Velthuizen, R.P., Silbiger, M.S., Bezdek, J.C.: A com-
parison of neural network and fuzzy clustering techniques in segmenting magnetic resonance
images of the brain. IEEE Trans. Neural Networks 3(5), 672–682 (1992)
8. Singh, A.: Detection of brain tumor in MRI images, using combination of fuzzy c-means and
SVM. In: 2015 2nd International Conference on Signal Processing and Integrated Networks
(SPIN), pp. 98–102 (2015). IEEE
9. Giraddi, S., Vaishnavi, S.V.: Detection of brain tumor using image classification. In:
2017 International Conference on Current Trends in Computer, Electrical, Electronics and
Communication (CTCEEC), pp. 640–644, (2017). IEEE
10. Sandhya, G., Giri, K., Savitri, S.A.: Novel Approach for the detection of tumor in MR images
of the brain and its classification via independent component analysis and kernel support vector
machine. Imaging Med. 9(3), 33–44 (2017)
11. Kumar, P., Kumar, B. V.: Brain tumor MRI segmentation and classification using ensemble
classifier. Int. J. Recent Technol. Eng, 8(1S4), 244–252 (2019)
12. Sharmila, A.A., Arun, D.C., Venkatesh, J., Sudarshan, S., Pranav, A.: Predicting survival of
brain tumor patients using deep learning. Int. J. Innov, Technol. Exploring Eng. 8(6), 1441–1448
(2019)
13. Lehrer, S.: Glioblastoma and dementia may share a common cause. Med. Hypotheses 75, 67–68
(2010)
14. Lehrer, S.: Glioma and Alzheimer’s disease. J. Alzheimer’s Dis. Rep. 2(1), 213–218 (2018)
15. Hamamci, A., Unal, G.: Multimodal brain tumor segmentation using the tumor-cut method on
the BraTS data set. Proc. MICCAI-BRATS, 19–23 (2012)
16. Akinyemi, R.O., Allan, L.M., Oakley, A., Kalaria, R.N.: Hippocampal neurodegenerative
pathology in post-stroke dementia compared to other dementias and aging controls. Front.
Neurosci. 11, 717 (2017)
17. Fernandez-Banet, J., Esposito, A., Coffin, S., Horvath, I.B., Estrella, H., Schefzick, S., Roberts,
P.: OASIS: web-based platform for exploring cancer multi-omics data. Nat. Methods 13(1), 9
(2016)
18. Dong, Y., Xu, S.: A new directional weighted median filter for removal of random-valued
impulse noise. IEEE Signal Process. Lett. 14(3), 193–196 (2007)
19. Bhateja, V., Verma, A., Rastogi, K., Malhotra, C., Satapathy, S.C.: Performance improvement
of decision median filter for suppression of salt and pepper noise. In Advances in Signal
Processing and Intelligent Recognition Systems, pp. 287–297, Springer, Cham (2014)
20. Uma, K., Suhasini, K., Vijayakumar, M.: A comparative analysis of brain tumor segmentation
techniques. Indian J. Sci. Technol. 9, 48 (2016)
21. Sharma, N., Mishra, M., Shrivastava, M.: Colour image segmentation techniques and issues:
an approach. Int. J. Sci. Technol. Res. 1(4), 9–12 (2012)
Detection and Classification of Brain Tumor Using Magnetic … 441

22. Sharma, N., Aggarwal, L.M.: Automated medical image segmentation technique. J. Med. Phys.
35(1), 3–14 (2009)
23. Zanaty, E.A., El-Zoghdy, S.F.: A novel approach for color image segmentation based on region
growing. Int. J. Comput. Appl. 39(3), 123–139 (2017)
24. Sheshathri, V., Sukumaran, S.: A Hybrid Clustering Based Color Image Segmentation using
Ant Colony and Particle Swarm Optimization Methods. Int. J. Innov. Technol. Exploring Eng.
8(7), 352–358 (2019)
25. Palo, H.K., Mohanty, M.N., Chandra, M.: Efficient feature combination techniques for
emotional speech classification. Int. J. Speech Technol. 19, 135–150 (2016)
26. Palo, H.K., Sagar, S.: Comparison of neural network models for speech emotion recognition.
In: 2018 2nd International Conference of Data Science and Business Analytics, pp. 127–131
(2018). IEEE
27. Palo, H.K., Mohanty, M.N.: Wavelet-based feature combination for recognition of emotions.
Ain Shams Eng. J. 9, 1799–1806 (2018)
28. Palo, H.K., Chandra, M., Mohanty, M.N.: Recognition of human speech emotion using variants
of mel-frequency cepstral coefficients. In: Advances in Systems, Control and Automation,
pp. 491–498, Springer, Singapore (2018)
Real-Time Transmission Line Situational
Awareness Using NI Phasor
Measurement Unit

Kunja Bihari Swain, Satya Sopan Mahato, Sushant K. Mandal,

and Murthy Cherukuri

Abstract Today’s power network is spread over a large geographical area.

Transmission line forms major part of it, which is exposed to the environment and
inherently affected by environmental weather conditions. Thus, situational aware-
ness of the transmission line is extremely important for protecting and restoring
power failure quickly. Situational awareness can be defined as the fault perception,
fault type comprehension and projection of possible remedial action for restoring the
transmission line from faults. Hence, the situational awareness of transmission line
draws the attention of recent researches. Particularly, today’s efficient equipment like
national instrument (NI) compact RIO (cRIO), equipped with the faster sampling rate
and high performance digital signal-processing capability enables the digital real-
time simulation expertly. In this paper, a laboratory transmission line prototype of
110 volts, 200 km has been used on which various types of faults were applied.
NI real-time data acquisition system along with LabVIEW has been incorporated
for fault perception and comprehension of fault type to predict possible situational
awareness to protect the transmission line from faults. Implementation in LabVIEW
tool support interactive visual display to enable perceiving and decision making even
for amateur users.

Keywords Phasor measurement unit · Situational awareness · Transmission line ·

Agglomerative hierarchical clustering

K. B. Swain · S. K. Mandal
Electronics and Communication Engineering, Centurion University of Technology and
Management, Paralakhemundi, Odisha, India
S. S. Mahato
Electronics and Communication Engineering, National Institute of Science and Technology,
Berhampur, Odisha, India
M. Cherukuri (B)
Electrical and Electronics Engineering, National Institute of Science and Technology, Berhampur,
Odisha, India
e-mail: [email protected]

© Springer Nature Singapore Pte Ltd. 2020 443

G. Pradhan et al. (eds.), Advances in Electrical Control and Signal Systems,
Lecture Notes in Electrical Engineering 665,
https://doi.org/10.1007/978-981-15-5262-5_32
444 K. B. Swain et al.

1 Introduction

Situational awareness (SA) of growing and operational complex modern power sys-
tem becomes more challenging due to addition of renewable energy sources, stress
level of deregulated and new additional transmission capacity. The power system and
transmission system operator often find it difficult to provide accurate and complete
picture of the system. These challenges restrict from achieving the acceptable level
of SA, which is necessary to make the right decision and effective control measure.
SA is a term originating a development of a knowledge-based approach for power
system network. SA has become an indispensable part of the modern power system.
SA can be enhanced by using PMU data. The PMU, which is capable of provid-
ing high-speed real-time measurement up to 120 samples per cycle along with GPS
time stamp and geographical locations, enables effective SA. Lack of secure SA
in operation, at critical points initiates severe stability problems. It is a great chal-
lenge to operate in secure level due to high nonlinearity and higher dimensionally
of the large-scale power system. It is essential to enhance SA of largely intercon-
nected power system. For post-event analysis, it takes several months together. By
using PMU, online event handling analysis and precaution measures can take at
appropriate time to save from cascaded breakdown. The instability in power system
network leads to cascading outage, which results in a blackout. Advent of computa-
tional tools like support vector machine (SVM), neural network (NN), deep learning
and machine learning provides a promising solution to handle these disturbances
efficiently [1–3] within very short time. There are numerous researches dedicated
for the transmission line fault detection and classification. A wavelet-based fault
detection presented in [4] along with nonstationary characteristics of the events.
In reference [5], the author presented frequency domain fault classification using
multi-resolution wavelet transform and shown the superiority of Db4 wavelet. In
reference [6], a wavelet transformation-based fault detection and classification is
presented using rule-based ANN. The algorithm is able to differentiate the faults
from power quality disturbance like voltage sag and oscillatory transient. In refer-
ence [7], a wavelet and fuzzy-based combined approach presented for classification
and location of the transmission line fault. A FFT-based fault detection and classifica-
tion is presented in [8] using PMU data. The fault detection being done based on the
variation of FFT coefficients of equivalent voltage phasor angle (EVPA) and the clas-
sification is based on the variation of the FFT coefficient of equivalent current phasor
angle (ECPA). A principal component analysis (PCA)-based early event detection
is reported in [9]. The detection was being done by comparing the error between
the projected PCA and the actual value of selected locations. In reference [10], a
K-NN-based event detection and classification is presented for distance protection
by comparing each sample with its fifth sample by taking a half cycle moving win-
dow. A fuzzy logic-based fault classification is presented in reference [11], which
is based on its three-phase line current for distance protection. In reference [12],
the author presented an event detection and classification based on S-transform. A
wavelet singular entropy-based fault detection and classification is presented in [13]
Real-Time Transmission Line Situational Awareness … 445

in extra high volt (EHV) transmission line. The author has shown its superiority and
reliability for a different fault type, fault resistance, fault location and fault inception
time. Full cycle Fourier and cosine phasor filtering system are typically numerical
distance relay implementations with the response time nearly one cycle [8]. The lit-
erature has proposed fast sub-cycle numerical relaying method such as a half cycle
Fourier method, phaselet, least square error and wavelet-based method.
In this paper, a real-time SA of transmission line implemented in a 200 km trans-
mission line prototype model by quick fault perception and fault type comprehen-
sion based on NI-based PMU in conjunction with virtual instrumentation tool. The
most significant contribution is the real-time implementation of the digital signal-
processing and bottom-up agglomerative hierarchical clustering computational tech-
nique for fault type comprehension. The next section explains the details about the
experimental setup used for the work. Third section describes the methodology used
for real-time fault perception. Fourth section explains the fault perception using the
agglomerative hierarchal clustering. Fifth section explains the result and discussion,
and the last section summarizes with a conclusion.

2 Experimental Setup

The experimental setup used for the SA of transmission line is depicted in Fig. 1.
It consists of a 200 km transmission line, a NI-based PMU, fault applying block, a
three-phase lamp load and a dell precision tower 3620 work station installed with
the LabVIEW virtual instrumentation tool. These blocks are connected as shown in
Fig. 2. Figure 3 shows the NI-based PMU configuration.
The details of each block are described as follows.

Laboratory Transmission line NI PMU

PC with LabVIEW
Pi-sections

Fault Block 3

Fig. 1 Experimental setup for SA of transmission line

446 K. B. Swain et al.

Fig. 2 Block diagram of SA transmission line

Fig. 3 NI PMU

2.1 NI Pmu

NI cRIO 9066 is an embedded real-time gate array (FPGA) controller for the C
series module. It is helpful for the sophisticated implementation of control and mon-
itoring. It includes a dual-core CPU with 667 MHz, 256 MB DRAM. NI-9246 is a
20 Arms, 30 A peak to peak, and 24 bits with three-channels C series present input
module. NI-9242 is an input voltage module of 250 V rms L-N, 400 V rms L-L,
50 kS/second/channel (kS/s/ch), 24-bit, three-channel C series The current module
and voltage module act as a current transformer (CT) and potential transformer (PT)
of the conventional PMU. NI-9467 is a GPS C series synchronization module. It
provides precise time synchronization with the geographical location (latitude and
longitude) to cRIO-9066. The GPS module also sends a pulse per second (PPS)
for necessary sampling clock pulse for synchronization. NI 9401 is a 5 V/TTL, 8
channels, 100 ns C series module, which can act as a digital input output module.
It creates a triggering signal to the relay associated with the faulty line, which in
turn helps to isolate only the faulty portion of the system from the rest of the healthy
portion when the system perceives the fault. These modules along with the cRIO in
conjunction with LabVIEW together perform as PMU providing interactive visual
representation.
Real-Time Transmission Line Situational Awareness … 447

2.2 Laboratory Transmission Line Prototype

200 km transmission line prototype, equivalent to 400 kV twin moose line carrying
110 v phase voltage consists of 4-pi sections. Each section is about 50 km. The
parameters of the transmission line are 2μF capacitance, inductance of 8.4 mH and
resistance of 0.2 .

2.3 Fault Block

To validate the real-time fault perception and comprehension, all types of fault (LG,
LL, LLG and LLL fault) generated by means of fault block and are applied at different
location (tap points provided at each pi section) of the transmission line.

2.4 Load

Two types of resistive load are considered to carry out the experiment. One is a three-
phase resistive lamp load of 200 W, 230 V and another is a three-phase1K, 300 V,
1.2 A resistive load.

3 Real-Time Fault Perception

Fault perception is achieved by acquiring real-time three-phase current using NI PMU

in LabVIEW platform as shown in Fig. 4. The signal is brought to LabVIEW platform
and FFT algorithm is applied to perceive the occurrence of fault in transmission line
in the frequency domain at DC (0 Hz), the fundamental frequency (50 Hz) and
at higher harmonics (100, 150, 200 Hz). The FFT has been applied to equivalent
current phase angle (ECPA) of three-phase current, which was derived using Park’s
transformation [14] as per Eqs. (1), (2), (3) and (4) presented in Fig. 5.

Fig. 4 Data flow in NI-based PMU

448 K. B. Swain et al.

Fig. 5 Fault perception

algorithm

 ⎤  ⎡
  IR
−1
Iα 2 1 − 1
= × √2× ⎣ IY ⎦
√
2
− 3 (1)
Iβ 3 0 23 2
IB
     
Id cos θ sin θ I
= × α (2)
Iq − sin θ cos θ Iβ
Iq
ECPA(ϕi ) = tan−1 (3)
Id
N −1 N −1
1 −2πnk
FFT of ECPA(ψ f ) = . ϕi (n). e N (4)
N k=0 n=0

For f = 0, 50, 100, 150, 200 Hz.

where I R , IY and I B are three-phase currents, Iα and Iβ are fictitious and qudrature-
phase (two-phase) current components and Id and Iq are the direct and quadrature axis
Real-Time Transmission Line Situational Awareness … 449

currents. The real-time fault detection performed in this paper as depicted in Fig. 5.
The three-phase currents (I R , I Y and I B ) are converted to the direct and quadrature
components. Then, the ECPA is evaluated. Based on a variation of the ECPA, fault
is perceived.

4 Fault Type Comprehension Using Agglomerative

Hierarchal Clustering

Agglomerative hierarchal clustering (AHC) is an unsupervised bottom-up computa-

tional method, which needs no pre-specified number of clusters. In this paper, AHC
has been employed for transmission line fault type comprehension. This process is
an iterative procedure, which is based on the following steps:
(a) Find the closest cluster according to Euclidian distance
(b) Agglomerate then to form of a new cluster.
It begins with a set of individual entities, and then grouped into a cluster pairwise
until single cluster is formed. It is a simple technique in which a cluster formed
based on the Euclidian distance between each pair of entities. Equations (5) and
(6) show the linkage function of Euclidian distance. Since a huge amount of data
is collected, Park’s transformation has been used to convert the phase current data
into feature vector, prior to the application of the proposed clustering technique. The
mathematical background behind the concept of Park’s transformation is shown in
Eqs. (1) and (2) in Sect. 3.
Mathematically the linkage function and the Euclidean distance can be evaluated
as per Eqs. (5) and (6)

D(x, y) = min d(x, y) (5)

x∈X,y∈Y

min{d(x, y)} : x ∈ X, y ∈ Y

X − Y 2 = (xi − yi )2 (6)
i

5 Results and Discussion

The real-time fault perception is accomplished by employing FFT on ECPA. The

variation of FFT coefficients at different frequency components is shown in Fig. 5.
The three-phase current phasors measured at the beginning of first pi section through
450 K. B. Swain et al.

NI PMU with LabVIEW virtual tool. The fault perception and fault comprehen-
sion algorithms are deployed in cRIO-9063. The interactive visualization presented
through the LabVIEW front panel to indicate fault instance, in three-phase current
waveform as shown in Fig. 6 based on the deviation observed of the FFT coefficients
of ECPA at DC, fundamental frequency and different harmonics. Figure 6a shows
the waveform of three-phase current under LG fault. It can be inferred that before the
fault has occurred, the system is balanced. Figure 6a shows the waveform of three-
phase current under LG fault. It can be inferred that before the fault has occurred, the
system is balanced and the peak amplitude is 0.7 A. When the ground fault occurs in
R-phase, the current of R-phase current increased suddenly to 7.3 A without change
in Y and B phase current amplitude. Similarly from Fig. 6b, c, d, it can be inferred
that when a fault occurs on a particular phase, there is an increase in current in that
particular phase, this deviation in waveform from the balanced condition. Figure 7a-d
shows the phasor for the same faults corresponding to Fig. 6, respectively. It can be
inferred from Fig. 7 that when the fault has occurred, the phase angle as well as the
amplitude gets disturbed from its balanced condition.
The fault type comprehension or fault classification is accomplished with the
help of agglomerative hierarchical clustering (AHC) algorithm in LabVIEW tool.
Figure 8 shows LabVIEW front panel of agglomerative clustering, which is formed

Fig. 6 Current waveform of fault perception under a LG fault, b LL fault, c LLG fault and d LLL
fault
Real-Time Transmission Line Situational Awareness … 451

Fig. 7 Phasor diagram of three-phase current under a LG fault, b LL fault, c LLG fault and d LLL
fault

using the FFT coefficients (ψ f ) of ECPA of the three-phase current phasors at dif-
ferent frequencies. Each cluster corresponds to one fault class distinct from another
class. Class0 corresponds to normal (no fault) condition. Class1, class2, class3 and
class4 correspond to line-ground (LG) fault, Line-Line fault (LL), Line-Line-ground
fault (LLG) fault and Line-Line-Line (LLL) fault, respectively. Figure 9 shows the
dendrogram drawn using MATLAB taking their FFT coefficients of ECPA of current
phasors for each fault, it can be inferred from Fig. 9. That each cluster are separated
according to their Euclidian distances. As hierarchal clustering, it forms a single
cluster at the end of the clustering process.
The required clusters are obtained based on their distance. A dotted line is drawn
to get required no of clusters.
452 K. B. Swain et al.

Fig. 8 Agglomerative hierarchical clustering of fault comprehension

Fig. 9 Dendrogram of fault clusters

Real-Time Transmission Line Situational Awareness … 453

Once the fault perception and fault type comprehension are done based on the
AHC algorithm, a trigger signal has been generated and sent to the NI 9401 (a digital
I/O module), which in turn activate the relay module to disconnect the faulty line
from other healthy lines to protect the system, and thus SA projection is achieved.
When the fault was cleared, the relays are deactivated by taking out the trigger signal
automatically by LabVIEW.

6 Conclusion

The modern power system is becoming complex to cater the needs of various enti-
ties efficiently. In order to protect the power system from unanticipated faults, it
is of utmost importance to have a data acquisition system whose sampling rate is
of 50 kSamples/Sec. This paper primarily focused on the situational awareness of
transmission line. The fault perception is accomplished with the help of FFT, fault
comprehension with the help of AHC. The whole analysis is performed on LabVIEW
platform, which gives a better interactive visualization.

Acknowledgements The authors would like to thank Science and Technology-Science and Engi-
neering Research Board, (SERB), India for providing the research funding under Early Career
Research Award category to carry out the research work. [Grant No. - ECR/2017/000812]

References

1. Zheng, G., Craven R.: Multiclass support vector machines for power system disturbance classi-
fication based on wide-area frequency measurements. In: Conferences on Proceedings—IEEE
SOUTHEASTCON, pp. 68–72, 2011
2. Rigatos, G., Piccolo A., SianoP.: Neural network-based approach for early detection of
cascading events in electric power systems. IET Gener. Trans. Distrib., 3(7), 650–665 (2009)
3. Zhang, N.: Kezunovic:Coordinating fuzzy ART neural networks to improve transmission line
fault detection and classification. 2005 IEEE Power Eng. Soc.Gen. Meet., Vol. 1, pp. 734–740,
2005
4. Kim, D.I., Chun, T.Y., Yoon, S.H., Lee, G., Shin, Y.J.: Wavelet-based event detection method
using PMU data. IEEE Trans. Smart Grid 8(3), 1154–1162 (2017)
5. Reddy, M.J., Mohanta, D.K.: A DSP based frequency domain approach for classification of
transmission line faults. Digit. Signal Process. A Rev. J. 18(5), 751–761 (2008)
6. Silva, K. M., Souza B.A., Brito, N.S. D.: Fault detection and classification in transmission lines
based on wavelet transform and ANN. IEEE Trans. Power Deliv. 21(4), 2058–2063 (2006)
7. Jayabharata, Reddy M., Mohanta, D.K.: A wavelet-fuzzy combined approach for classification
and location of transmission line faults. Int. J. Electr. Power Energy Syst. 29(9), 669–678 (2007)
8. Gopakumar, P., Reddy, M.J.B., Mohanta, D.K.: Transmission line fault detection and locali-
sation methodology using PMU measurements. IET Gener. Trans. Distrib. 9(11), 1033–1042
(2015)
9. Chen, Y., Xie, L., Kumar, P.R.: Dimensionality reduction and early event detection using online
synchrophasor data. IEEE Power Energy Soc. Gen. Meet. 1–5 (2013)
454 K. B. Swain et al.

10. Asadi, Majd, A., Samet, H., Ghanbari, T.: K-NN based fault detection and classification methods
for power transmission system. Prot. Control Mod. Power Syst. 2(1), (2017)
11. Das, B., Reddy, J.V.: Fuzzy-logic-based fault classification scheme for digital distance
protection. IEEE Trans. Power Deliv. 20(2 I), 609–616 (2005)
12. Das, S., Moirangthem, J., Dash, P.K.: Distance protection of shunt compensated transmission
line using a sparse S-transform. IET Gener. Trans. Distrib. 9(12), 1264–1274 (2015)
13. He, Z., Fu, L., Lin, S., Bo, Z.: Fault detection and classification in EHV transmission line based
on wavelet singular entropy. IEEE Trans. Power Deliv. 25(4), 2156–2163 (2010)
14. Gopakumar, P., Reddy, M.J.B., Mohanta, D.K.: Adaptive fault identification and classification
methodology for smart power grids using synchronous phasor angle measurements. IET Gener.
Trans. Distrib. 9(2), 133–145 (2015)
Photovoltaic Cell with Shunt Active
Power Filter for Harmonic Cancelation
Using Modified PSO-Based PI Controller

Amba Subhadarshini Nayak, Devi Prasad Acharya,

and Subhashree Choudhury

Abstract This chapter explains excellent performance of shunt active power fil-
ter (SAPF) integrated with photovoltaic source, for considerable energy manage-
ment, harmonic alleviation, and reactive power recompense. The solar plant produces
extreme or equal power essential to the load requirement, hence the coordination with
grid is complex. The PV module is designed with INC-MPPT technology, and the
shunt active power filter is connected at the AC side of the inverter integrated with
a nonlinear load. Here, the shunt active power filter design depends on PI controller
whose gains are selected arbitrary. In the first study, the total harmonic distortion
(THD) is calculated with no filter and nonlinear load. Then, THD analysis is per-
formed with SAPF. Further to pick up the recital of shunt active power filter, a new
PSO optimization technique is applied to select the proper value of PI gains R and L
of nonlinear load. The simulation result depicts that the optimized shunt active filter
reduces harmonics to a great extent. The results of SAPF, PSO-SAPF, APSO-SAPF,
and TCPSO-SAPF are compared.

Keywords SAPF · Voltage-source inverter (VSI) · INC-MPPT · THD · PSO ·

APSO and TCPSO

1 Introduction

The use of receptive loads like computers, medical instruments, sensors, forecasting
devices, some nonlinear loads, and power electronic devices provides a challenging

A. S. Nayak
IIIT, Bhubaneswar, India
e-mail: [email protected]
D. P. Acharya · S. Choudhury (B)
Department of EEE, Siksha ‘O’ Anusandhan (Deemed to be University), Bhubaneswar 751030,
Odisha, India
e-mail: [email protected]
D. P. Acharya
e-mail: [email protected]

© Springer Nature Singapore Pte Ltd. 2020 455

G. Pradhan et al. (eds.), Advances in Electrical Control and Signal Systems,
Lecture Notes in Electrical Engineering 665,
https://doi.org/10.1007/978-981-15-5262-5_33
456 A. S. Nayak et al.

task for reduction of power quality issues like harmonics. 24-hour operations of the
equipment requires consistent of power supply. Due to the supply of inconsistent
power, this equipment causes tremendous loss to the home and business customers,
power loss, which can cause harmonics in the line currents. Thus, for reliable opera-
tion of electrical power system and to achieve consumer satisfaction, power quality
issues must be taken care. Power system harmonics is one of the important power
quality issues which are to be minimized for better system operation.
The proposed work is to reduce the harmonic distortion of a PV-based shunt
active power filter connected to a rectifier load. In general, the SAPF supplies an
equal and opposite to the source current. The converter voltage reduces and filter
capacitor consumes more power from the nonlinear load during compensation. Thus,
to overcome those reverse effects, design of an optimized shunt active power filter
is suggested in this work [1–5].
Many literatures included the study of power quality issues in PV-based system
with shunt active power filter (SAPF). The third harmonic suppression in a three-
phase four-wire distribution system with implementation of different current control
methods is proposed in [5], however the structure of the filter and the parameter
modification has not taken care by the researchers. The enhanced current control
approach with MPPT controller has been implemented in order to mitigate sys-
tem harmonic with the help of grid connected inverter in [6–7]. Different filtering
techniques like self-tuning PI controller have been used to suppress the harmonic
distortions in a grid connected with a 3-phase, 4-leg VSI, a dc-link capacitor, and
a hysteresis current controller with various practical applications [8]. The harmonic
suppression and the VAR compensation of a SAPF are controlled by a two-loop
cascade control strategy which consists of back-stepping technique as inner loop
design and to compensate an outer-loop designed to control voltage of capacitor.
The controller efficiency is proved, using the averaging theory [9]. The source and
load currents are calculated, and their harmonics are reduced followed by reactive
power compensation of the nonlinear load, and to enhance power quality. A type-2
fuzzy logic controller is applied to the control system of the voltage-source SAPF to
assess its recital in this system and compared with d-q control theory. In reference [10,
11], the authors have implemented shunt active power filter (SAPF) using a control
hardware. By the use of a SAPF, the harmonic distortion is eliminated in distribu-
tion network which connected power electronics-based energy storage equipment.
Current harmonics extraction algorithm has been implemented which improved the
compensation current of SAPF, with the help of adaptive linear neuron (ADALINE)
technique. The selection of filter parameters and PI controller gains affects the per-
formance of the SAPF. Thus, the different optimization techniques like PSO and ant
colony optimization modification are applied in the filter to reduce harmonic effects
(THD) [12].
Here, in this paper, a PV model with nonlinear load is designed as shown in
Fig. 1. The nonlinear load produces harmonics which affects the power quality. A
shunt active power filter is designed to reduce the harmonic effects in terms of total
harmonic distortion (THD).
Photovoltaic Cell with Shunt Active Power Filter … 457

Fig. 1 Simulink diagram of PV-based SAPF model

The main contribution to this work is the shunt active filter is designed for both
diode and RL load. The act of the load is enhanced by selecting the PI controller
gain with application of PSO and modified PSO.
The rest part of the paper is arranged as follows: The Simulink model is discussed
in the Sect. 2. In Sect. 3, the shunt active filter model and total harmonic distortion is
discussed. The optimization technique and its modification are explained in Sect. 4.
Finally, the simulation results are presented in Sect. 5, and the conclusion is given
in Sect. 6.

2 Description of the Model

The above system consists of a photovoltaic cell with rectifier load through a shunt
active power filter. The SAPF is the combination of switching components and filter
network, and it is a bidirectional current converter. The major component of a SAPF
is a VSI with DC energy storage capacitor and a diode rectifier bridge along with a RL
load. To achieve bidirectional current flow, the SAPF diodes are connected in anti-
parallel with IGBT. For reactive power adjustment, the SAPF injects PV power to
the line at PCC. The SAPF injects an opposite compensating current which produces
same amount of harmonics to the line so that the harmonic current caused by the
nonlinear load is minimized. The SAPF is integrated with the line at PCC through a
three-phase breaker, which takes care of any fault/over current in the neighborhood.
The timing of the breaker is 0.02 s.
458 A. S. Nayak et al.

2.1 Mathematical Design of Shunt Active Power Filter

The selection of shunt active power filter parameters like coupling inductance L f
and dc side capacitor Cdc and selection of reference dc voltage requires following
assumptions.
• Source current must be sinusoidal.
• The system current operates within 5% distortion.
• Converter works in a linear mode.
The L f and Vc can be obtained from the following two equations.
 
Vc Vs
Q c = 3Vc Ic = 3Vs 1− (1)
ωL f Vc
Vh
Ih = (2)
m f ωL f

The Vdcref is set depending upon the rated voltage. The choice of Cdc is based on
the following equation for a specified permissible peak–peak voltage ripple.

π ∗ I rated
Cdc = √ c ripple (3)
3 ωV pp

where
Vc Voltage injected by the filter.
Lf Coupling inductance of SAPF.
Cdc DC side capacitor of the filter.
Vs SAPF voltage.
Ic Compensated current.
Vh Harmonic voltage.

3 Optimization Technique

3.1 Basic PSO

The PSO algorithm consists of updating expression of both position and velocity.
Every individual particle is computed with its previous best value. The pbest and gbest
are two best values which changes location and speed information. The fundamental
PSO have iteration formulae [13–14].

in = velin + k1 r 1 ( pbest − x in ) + k2 r 2 (gbest − x in )

velt+1 t t t t t
(4)
Photovoltaic Cell with Shunt Active Power Filter … 459

Fig. 2 Position and velocity

diagram of PSO

xint+1 = xint + vin

t+1
(5)

The particles with ‘m’ population searching the optimum value in an ‘n’-
dimensional space. xin Is the i th particle position in ‘n’-dimensional space with the
current velocity? velin . The pbest and gbest are the position best and global best value,
and ‘t’ is the present iteration. K1 and K2 are two cognitive social factors. r 1 and r 2
are two arbitrary numbers selected in between [0, 1].
As shown in Fig. 1, U is the search space of a function, and ‘o’ is the theoretical
optimum value to be calculated. xit is the position of the initial particle. velit is the
velocity of current individual member. velit+1 is the velocity of next iteration that
is disturbed by different aspects. The element memory and swarm influence are
represented by two lines which are parallel to the xt connected lines to pbest and
gbest . In this computational step, the particle i is influenced by vit first. In the second
step after element memory and swarm affected, i reaches at xit+1 from xit and at the
velocity vit+1 . Likewise in the next iteration the element will proceed from xit+1 to
a new position and the process will continue till to achieve the theoretical optimal
solution or it closer value (Fig. 2).
The updated velocity formula is expressed as

in = ω velin + k1 r 1 ( pbest − x in ) + k2 r 2 (gbest − x in )

velt+1 t t t t t
(6)

The equilibrium between local and global search is maintained effectively by

inserting an inertia weight ω capability. The strength of global search increases with
larger value selection of inertia weight and simultaneously the local search capacity
is more visible.
The steps of the algorithm are
Step 1: Select the initial population arbitrarily and set maximum of iteration count,
population size, ω, velocity, and position.
Step 2: The fitness value of the particle as per the fitness function is to be calculated.
Step 3: The present fitness value of each particle is compared with the past solution
pbest If the fitness value is less than the pbest , then set the calculated value;
otherwise, pbest will remain same.
In the same way, fitness of global best gbest is calculated.
460 A. S. Nayak et al.

Step 4: By Eqs. (2) and (3) position and velocity are modified within their maximum
limit.
Step 5: Theoretical optimum value should be achieved; the operation should be
stopped. Otherwise, return to Step 2.

3.2 Accelerated Particle Swarm Optimization

The accelerated particle swarm optimization (APSO) is alike to that of PSO with a
difference that it is not able to calculate the individual best value. In each step, it
updates acceleration along with velocity and position by which convergence speed
increases. This new APSO technique overcomes all the drawback of PSO during
operating for a multimodal function. In PSO, the consecutive position and velocity
of the particle are determined by adopting the following equation.
   
velk+1
n = velkn + α a1 gbest − xnk + β a2 pbest − xnk (7)

xnk+1 = xnk + vnk+1 (8)

where Gbest is the global best, I best is the individual best, and x kn and vkn are the current
position and velocity of the particle n, respectively. a1 and a2 are the random number
that ranges from 0 to 1, α and β are the acceleration constant, and k + 1 is the iteration
number. In case of APSO, Eq. 13 becomes more simpler and is illustrated below
 
vnk+1 = vnk + α a + β G best − xik (9)

The upgradation of the position can be achieved in a more simplified manner; by

combining Eqs. 14 and 15, we get

xnk+1 = (1 − β)xnk + β G best + α a (10)

The range of α lies between 0.1 to 0.5, and β ranges from 0.1 to 0.7. In this paper,
the values of α and β are taken as 0.5 and 0.7, respectively.

3.3 Two-Swarm Cooperative PSO (TCPSO)

Here, in this work, a modified PSO known as two-swarm cooperative particle swarm
optimization (TCPSO) [15] is applied to optimize the filter and PI parameters. In this
PSO two swarms, one acts as master and other one as slave worm which overcomes
the drawbacks like diversity, slow convergence rate, etc.
Step 1: Assign randomly the initial values of the master and slave swarms.
Photovoltaic Cell with Shunt Active Power Filter … 461

Step 2: Calculate Pbest , Psbest , gbest , and Pmbst are fitness values of two swarms.
The first two particles are from slave group, and the last two swarms are
from master group.
Step 3: Imitation and modifying.
Step 3.1: Eqs. (11) and (12) may be used for updating of slave swarm such that the
velocity and positions are within the limits.

vels,t+1
in
s,t
= k1r1 (1 − r2 )(xkn − xins,t ) + k2s (1 − r1 )r2 (gbest
t
− xins,t ) (11)

x p is randomly chosen
From the neighborhood of the xi according to Eq. (11):
  
i− i
2
+ 1, i + 2l if l  is even.
p∈ (l=1) (12)
i− 2
, i + (l−1)
2
if l  is odd.

‘l’ is the size of neighborhood whose experimental value is 2. Best in their

experiments.
Step 3.2: The master swarms are updated as per the following equations.

velinM,t+1 = w M velinM,t + k1M r1 (1 − r2 )(1 − r3 ) pbest

M
− xinM,t

+ k2M r2 (1 − r1 )(1 − r3 ) gbest

S
− xinM,t

+ k3M r3 (1 − r1 )(1 − r2 ) gbest − xinM,t (13)

xinM,t+1 = xinM,t + velinM,t+1 (14)

‘M’ stands here master swarm. gbest will be obtained for the next iteration.
Step 4: Get the optima if it meets the termination condition; otherwise, go to Step
2.
Table 1 denotes the optimized parameters obtained through various optimization
techniques discussed.

Table 1 Optimized parameters values

Parameter kp ki Cdc (μF) L f (m H ) R f ()
PSO 2.114 7.867 3000 0.0323 0.1621
APSO 1.921 8.012 3182 0.0172 0.3211
CPSO 3.921 6.832 2921 0.0121 0.1721
TPSO 2.831 5.321 2721 0.0215 0.2631
462 A. S. Nayak et al.

4 Result Analysis

The objective of this work is harmonic elimination of shunt active power filter in
a photovoltaic generation-based microgrid. A microgrid is designed with PV cell
as source, inverter, MPPT controller, a short transmission line, and nonlinear load.
The output of solar cell is intermittent, and use of power electronics device and
interconnection of nonlinear loads produces harmonics in the system voltage and
currents. Here, a three-phase fault is initiated at the load end, and studies have been
investigated. The FFT analysis is done with nonlinear load with a three-phase fault.
The switching time of the simulation is 0.05 s. However, in this work, the fault
analysis has not been taken into account.
Further, the gains of the PI controller, parameters of SAPF, and nonlinear load
parameters are selected arbitrary in trial and error methods which affect the perfor-
mance of the SAPF. Thus, to improve the SAPF capability, the system parameters
are optimized by PSO, APSO, CPSO, and TCPSO, and the results are compared. It
is observed that TCPSO-SAPF performance is better than other techniques (Fig. 3).
Case-1: Harmonic Analysis of PV-based SAPF.
In this case study, a three-phase LLL-G fault is initiated at the load side of the
PV-based microgrid with nonlinear load. The system is simulated for 2 s without
any filter, and the total harmonic distortion(THD) is calculated. The FFT analysis
shown in Fig. 4 depicts that without SAPF the total harmonic distortion (THD) of
the load voltage is 11.32%. The THD of the load current is 5.39%. That of the source
current is 4.29%. However with integration of shunt active power filter (SAPF), the
corresponding THDs are 9.33%, 4.48%, and 3.34% respectively. Thus, the shunt
active power filter is designed in such a way that the current harmonics is within 5%
(Fig. 4).
Case-2: Harmonic Analysis of PV-based PSO-SAPF.
With same system condition, a short-circuit fault is created in the load end and the
system is simulated for 2 s. Now, the particle swarm optimization is used to optimize

Fig. 3 Simulation of three-phase load voltage, load current, and filter source current
Photovoltaic Cell with Shunt Active Power Filter … 463

Load voltage Without SAPF Load current with no SAPF

Fig. 4 FFT analysis of load voltage, current, and source current with SAPF

the SAPF and load parameters and then the model is simulated. The FFT analysis of
the different voltage and currents with optimized parameters is shown in Fig. 5. It is
seen from the results of FFT analysis that the total harmonic distortion (THD) of the
load voltage with no SAPF is 7.88%. The THD of the load current is 3.19%. That
of the source current is 3.32%. However with PSO-based shunt active power filter
(PSO-SAPF), the corresponding THDs are 6.18%, 2.64%, and 2.14%, respectively.
The harmonics of load voltage are reduced by 4.9%; other harmonics reduction orders
are obtained.
Case-3: Harmonic Analysis of PV-based APSO-SAPF.
The next study includes the FFT analysis by an accelerated PSO-based microgrid,
and the THD of various voltage and currents is calculated. Under same operating
condition, a three-phase short-circuit fault is created and then the model is simulated
for 2 s. The THD of load voltage with no SAPF is 7.02% which is small from the
previous case. The THD of the load current is 2.12%. That of the source current is
464 A. S. Nayak et al.

Load volage Load current

Fig. 5 FFT analysis of load voltage, current, and source current with PSO-SAPF

1.67%. However, with accelerated particle swarm optimization-based shunt active

power filter (APSO-SAPF), the results are better than the previous cases. The cor-
responding THDs are 6.83%, 1.93%, and 1.12% respectively, which satisfies the
standard value of THD, i.e., less than 5%. Figure 6 represents the APSO based SAPF
results.
Case-4: Harmonic Analysis of PV-based TCPSO-SAPF.
The last study of this paper includes the FFT analysis by a two-swarm coordinated
particle swarm optimization (TCPSO)-based SAPF and the THD of various voltage
and currents is calculated. Under the same operating condition, a three-phase short-
circuit fault is created and then the model is simulated for 2 s. The THD of load
voltage without SAPF is 5.65% which is small from the previous case. The THD of
the load current is 1.72%. That of the source current is 1.37%. However, with accel-
erated particle swarm optimization-based shunt active power filter (APSO-SAPF),
the results are better than the previous cases. The corresponding THDs are 4.83%,
1.03%, and 0.92%, respectively, which satisfies the standard value of THD, i.e., less
than 5%. Figure 7 denotes the TCPSO based SAPF results.
Photovoltaic Cell with Shunt Active Power Filter … 465

Load voltage Load current

Fig. 6 FFT analysis of load voltage, current and source current with APSO-SAPF

4.1 Comparison of Results

Table 2 gives a comparative analysis of load voltage, load current and source cur-
rent THD obtained both with and without SAPF with the discussed optimization
techniques.

5 Conclusion

In this paper, the shunt active power filter integrated with PV cell has been designed
with nonlinear load and simulated for 2 s. The harmonics in load voltage, load current,
and filter source current are measured with different SAPF design. During design
of filter, some parameters are randomly selected which affects the performance of
SAPF. Thus, a new optimization technique like two-swarm coordination particle
swarm optimization (TCPSO) has been used to optimize the parameters like k p , ki ,
Cdc , L f , and R f , and the system is simulated. The total harmonic distortion (THD)
is calculated and compared with PSO-SAPF and APSO-SAPF, and it is observed
that the performance of TCPSO-SAPF is better than other optimized active power
filters. In all the cases, the effective simulation achieved better harmonics as per IEEE
standard (within 5%).
466 A. S. Nayak et al.

Load voltage Load current

Fig. 7 FFT analysis of load voltage, current, and source current with TC PSO-SAPF

Table 2 THD values for different techniques without and with SAPF
Cases THD in percentage (%)
Without SAPF With SAPF
Parameters VLoad Iload Isource VLoad Iload Isource
No optimization 11.32 5.39 4.29 9.33 4.48 3.34
PSO-SAPF 7.88 3.19 3.32 6.18 2.64 2.14
APSO-SAPF 7.03 2.12 1.67 6.83 1.93 1.12
TCPSO-SAPF 5.68 1.72 1.37 4.83 1.03 0.92

References

1. Wagner, V.E., Balda, J.C., Griffith, D.C., et al.: Effects of harmonics on equipment. IEEE Trans.
Power Delivery 8(2), 672–680 (1993)
2. Raju, N.R., Venkata, S.S., Kagalwala, R.A., Sastry, V.V.: An active power quality conditioner
for reactive power and harmonics compensation. In: 1995 PESC ‘95 Record, 26th Annual IEEE
Power Electronics Specialists Conference, vol. 1, pp. 209–214, Atlanta, GA, USA, June (1995)
3. Inoue, S., Shimizu, T., Wada, K.: Control methods and compensation characteristics of a series
active filter for a neutral conductor. IEEE Trans. Industr. Electron. 54(1), 433–440 (2007)
Photovoltaic Cell with Shunt Active Power Filter … 467

4. da Silva, S.A.O, Sampaio, L.P., Campanhol, L.B.G.: Single-phase grid-tied photovoltaic system
with boost converter and active filtering. In: 2014 IEEE 23rd International Symposium on
Industrial Electronics (ISIE), pp. 2502–2507, Istanbul, Turkey, June (2014)
5. Méndez, I., Vázquez, N., Vaquero, J., Vázquez, J., Hernández, C., López, H.: Multifunctional
grid-connected photovoltaic system controlled by sliding mode. In: IECON 2015—41st Annual
Conference of the IEEE Industrial Electronics Society, pp. 1339–1344, Yokohama, Japan,
November (2015)
6. Wajahat, U.T., Mekhilefa, S., Seyedmahmoudianb, M., Horanb, B.: Active power filter (APF)
for mitigation of power quality issues in grid integration of wind and photovoltaic energy
conversion system. Renew. Sustain. Energy Rev. 70, 635–655 (2017)
7. Jayasankar, V.N., Vinatha, U.: advanced control approach for shunt active power filter inter-
facing windsolar hybrid renewable system to distribution grid. J. Electr. Syst. 4(2), 88–102
(2018)
8. Imani Jajarmi, H.R., Mohamed, A, Shareef, H., Subiyanto, S.: A new method for calculating
the reference current of shunt active power filters based on recursive discrete Fourier transform.
Int. Rev. Model. Simul. 5(4), 1439–1446 (2012)
9. Pedapenki, K.K., Gupta, S.P. Pathak, M.K.: Experimentation of shunt active filters. Int. J. Pure
Appl Math. 114(7), 65–75 (2017)
10. Tareen, W.U., Mekhilef, S.: Transformer-less 3P3W SAPF (three-phase three-wire shunt active
power filter) with line-interactive UPS (uninterruptible power supply) and battery energy
storage stage. Energy Elsevier 109(C), 525–536 (2016)
11. Tareen, W., Mekhilef, S, Seyedmahmoudian, M.: Active power filter (APF) for mitigation of
power quality issues in grid integration of wind and photovoltaic energy conversion system.
Renew. Sustain. Energy Rev. (70), 635–655 (2017)
12. Sakthivel, A., Vijayakumar, P., Senthilkumar, A.: Design of ant colony optimized shunt active
power filter for load compensation. Int. Rev. Electr. Eng. Praiz Worthy 9(04), 2182–2192 (2014)
13. Kennedy, J. Particle Swarm Optimization. In Proceedings of the 1995 International Confer-
ence on Neural Networks, Perth, WA, Australia, 27 November–1 December 1995; Volume 4,
pp. 1942–1948.
14. Shi, Y.H., Eberhart, R.C. Empirical study of particle swarm optimization. In Proceedings of
the 1999 Congress on Evolutionary Computation-CEC99 (Cat. No. 99TH8406), Washington,
DC, USA, 6–9 July 1999; pp. 1945–1950.
15. Sun, S., Li, J. A two-swarm cooperative particle swarms optimization. Swarm Evol. Comput.
15, 1–18 (2014).
Comparative Study of Segmentation
of Noisy Color Synthetic Image Using
FCM and PFCM

P. Ganesan, B. S. Sathish, and L. M. I. Leo Joseph

Abstract In most of the image processing applications, noise is the major concern
for the degradation of the feature of the image. The noise arises due to various factors
such as sensor and faulty circuits. Segmentation is the procedure of detachment of
the whole image into many considerable clusters. Every cluster should be of unique
in terms of any one of the image attributes such as texture, intensity or color. The
accomplishment of image analysis primarily based on the upshot of the segmentation
process. The proposed work explains the impact of noise in the segmentation process.
The proposed work analyzed and compared the segmentation of noisy color synthetic
using fuzzy C-means clustering (FCM) and possibilistic fuzzy C-means clustering
(PFCM).

Keywords Segmentation · Noise · Clustering · FCM · PFCM

1 Introduction

In image processing and computer vision, the role of the segmentation process is
inevitable. The outcome of the segmentation process is the basis for image analysis.
In segmentation, the complete image is clustered into number of sub-images in
meaningful manner [1–3]. In a cluster, there should be some uniqueness among

P. Ganesan (B)
Department of Electronics and Communication Engineering, Vidya Jyothi Institute of
Technology, Aziz Nagar, C.B.Road, Hyderabad, India
e-mail: [email protected]
B. S. Sathish
Department of Electronics and Communication Engineering, Ramachandra College of
Engineering, Eluru, Andhra Pradesh, India
e-mail: [email protected]
L. M. I. Leo Joseph
Department of Electronics and Communication Engineering, S.R. Engineering College,
Warangal, Telangana, India
e-mail: [email protected]

© Springer Nature Singapore Pte Ltd. 2020 469

G. Pradhan et al. (eds.), Advances in Electrical Control and Signal Systems,
Lecture Notes in Electrical Engineering 665,
https://doi.org/10.1007/978-981-15-5262-5_34
470 P. Ganesan et al.

all the image elements (pixels). The segmentation is based on the image attributes
such as color and texture [4]. Most of the segmentation techniques are application-
oriented, i.e., it is very hard to get the meaningful result for more than one applications
using same segmentation method [5]. This is major drawback of the segmentation
techniques. There is lot of image segmentation techniques based on threshold, edge,
region, clustering and so on. In this work, cluster-based methods FCM and PFCM
are utilized to segment the noisy images. Noise is the part and parcel of the image
acquisition and transmission. It is major concern for the degradation of the images [6].
There are many reasons for noise added to the original images. For example, the more
temperature on image sensor can excite thermally generated electrons (thermal noise)
which gets merged with the real photoelectrons [7–10]. It is necessary but difficult
task to differentiate the image pixels from the noisy one. In this work, it investigates
the impact of two significant image noises salt and pepper (noise density of 0.05)
and Gaussian (zero mean and variance of 0.020) on the color image. The efficiency
of the segmentation methods is measured in terms of PSNR and computational cost.

2 Noise in Image Processing

In image processing, noise, the unwanted disturbance, is considered as a unsystematic

fluctuation of image attributes such as color or intensity [11]. This may due to the
faulty circuit of digital camera (scanner) or sensor problem. This electronic noise
tries to corrupt the original information content of image. Mathematically, noise can
be characterized as

f  (x, y) = f (x, y) + n(x, y)

where f (x, y) is the original image,

n(x, y) is the noise term and
f  (x, y) is the resulting noisy image
Gaussian noise is an additive noise caused by random fluctuations in the sig-
nal. This noise has a probability density function (PDF) of the normal (Gaussian)
distribution.
1
e−(z−μ) /2σ
2 2
p(z) = √
2π σ

z Gaussian random variable (gray level)

μ mean gray level
σ standard deviation of gray level
σ2 variance of gray level

Salt and pepper noise, sometimes referred as shot or impulse noise, looks like
indiscriminately disseminated white (salt) and black (pepper) pixels over the image.
Comparative Study of Segmentation of Noisy Color Synthetic … 471

The PDF of impulse noise is given by

⎧
⎨ Pa for z = a
p(z) = Pb for z = b
⎩
0 otherwise

p(z) distribution of salt and pepper noise in image

Pa , Pb the probabilities density function (PDF).

3 Methodology

Fuzzy C-means clustering is the standard algorithm based on cluster centers and
membership to partition a complete dataset into number of clusters [12–14]. This
algorithm utilizes the Euclidean distance to compute the space between the pixels
(data points) and cluster center [15]. The major function of the algorithm is to lessen
its objective function provided in Eq. (1) [18].


c 
n
Fm (U, V ) = μik
m
xk − vi2 (1)
i=1 k=1

where m = the weighting exponent (fuzziness) parameter. Most cases, m = 2.

vi = cluster centers and μik = membership
n
k=1 μik X k
m
vi =  n (2)
k=1 μik
m

⎧ ⎫−1
⎪  2/ ⎪
⎨ c
xk − vi  (m − 1) ⎬
μik = (3)
⎪
⎩ xk − v j ⎪
⎭
j=1

Possibilistic C-means (PCM) clustering gave the solution for the column sum
constraint of FCM [16]. The objective function of PCM is illustrated in Eq. (4)


n 
c 
c 
n
Pm (T, V ; X, γ ) = tikm dki2 + γi (1 − tki )m (4)
i=1 k=1 i=1 k=1

where γ = weighting exponent parameter.

When γ = o and m = 1, PCM behaves as hard C-means (HCM) clustering. In
general, m = 2 and γ > 0 is the preferable choice. The optimum (minimum) can be
achieved if it satisfies the following conditions.
472 P. Ganesan et al.

Fig. 1 Synthetic image

 1
dik / m − 1
tki = 1/ 1 + , 1 ≤ i ≤ c;1 ≤ k ≤ n (5)
γi
n
xk tkim
vi = k=1
n m (6)
k=1 tki

The noise sensitivity problem of FCM and coincident clusters issue of PCM is
addressed by possibilistic fuzzy C-means (PFCM) clustering [17]. The objective
function of PCM is shown in Eq. (7).


n 
c
 m η
P Fm (T, V, U ; X, γ ) = aμik + btik dki2
i=1 k=1
c 
n
+ γi (1 − tki )η (7)
i=1 k=1

where m, > 1 and a, b, γ > 0. The optimum (minimum) can be achieved if it satisfies
the following conditions.
n  m η
k=1 aμik + btik x k
Vi = n  m η (8)
k=1 aμik + btik
1
tik =  2
1/(m−1) (9)
dik
1+ γi

1
μik =  2/m−1 (10)
c dik
j=1 d jk

4 Experimental Result and Discussion

Figure 1 illustrates the synthetic fiber test image (image without noise) to evaluate
the efficiency of proposed techniques. The competence of the segmentation methods
Comparative Study of Segmentation of Noisy Color Synthetic … 473

is investigated against the salt and pepper (noise density of 0.05) and Gaussian (zero
mean and variance of 0.020) noise. This is illustrated in Fig. 2.
The outcome of the proposed method for the test image without noise is depicted
in Fig. 3. Figure 3a is the result of the FCM method and outcome of the PFCM
is portrayed in Fig. 3c. The error image of the corresponding method is displayed
in Figs. 3b, d respectively. The result of the proposed method for the test image
with added Gaussian noise is depicted in Fig. 4. Figure 4a is the result of the FCM
method and outcome of the PFCM is portrayed in Fig. 4c. The error image of both
the methods is displayed in Fig. 4b, d respectively.

(a) Gaussian noise (b) salt and pepper noise

Fig. 2 Noisy images

Fig. 3 Outcome of the proposed method for the test image without noise
474 P. Ganesan et al.

Fig. 4 Outcome for the synthetic image with gaussian noise

The result of the proposed method for the test image with added salt and pepper
noise is depicted in Fig. 5. Figure 5a is the result of the FCM method and outcome of
the PFCM is portrayed in Fig. 5c. The error image of both the methods is displayed
in Fig. 5b, d respectively.

Fig. 5 Outcome for the synthetic image with salt and pepper noise
Comparative Study of Segmentation of Noisy Color Synthetic … 475

The experimental result of the proposed approach using FCM is tabulated on

Table 1. The highlighted outcome for the PFCM is illustrated in Table 2.
The pictorial representation of the comparison of segmentation of noisy color
synthetic image using FCM and PFCM is depicted in Figs. 6 and 7.

Table 1 Experimental result using FCM

Type of noise Iteration Cluster Cluster centers PSNR Computational cost
(s)
No noise 3 3 200.8 160.7 107.1 18.21 7.2311
200.8 160.7 107.1
99.32 38.53 30.14
Gaussian 15 8 98.51 46.64 45.19 27.23 9.0870
230.0 136.3 45.28
150.3 153.2 133.5
215.5 194.2 97.31
215.3 68.88 22.00
135.7 77.74 58.40
223.4 232.9 221.6
52.54 17.82 16.77
Salt and pepper 6 3 200.4 160.6 107.2 18.10 8.6792
200.4 160.6 107.2
99.17 38.52 30.18

Table 2 Experimental result using PFCM

Type of noise Iteration Cluster Cluster centers PSNR Computational cost
(s)
No noise 3 3 171.2 162.3 120.3 19.35 1.720
171.2 162.3 120.3
96.17 30.17 32.14
Gaussian 3 8 98.54 44.77 43.28 27.34 2.678
236.9 134.1 44.69
151.3 153.1 130.7
219.3 194.4 95.23
217.0 70.80 16.66
133.2 76.54 58.48
227.9 239.9 223.4
55.16 11.97 10.15
Salt and pepper 3 3 170.9 162.0 120.0 19.24 3.110
170.9 162.0 120.0
96.03 30.11 32.07
476 P. Ganesan et al.

Fig. 6 Assessment of the 30

outcome based on PSNR
25
20

PSNR
No Noise
15
Gaussian
10
Salt &Pepper
5
0
FCM PFCM

Fig. 7 Assessment of the 10

outcome based on

ComputaƟonal Cost
computational cost 8

6 No Noise
(in sec)
4 Gaussian
Salt and Pepper
2

0
FCM PFCM

5 Conclusion

The impact of noise in the synthetic color image segmentation process is enlightened
in the proposed work. The source and mathematical model of the salt and pepper
and Gaussian noise is analyzed. The synthetic noisy image is clustered segmented
using fuzzy FCM and PFCM. The experimental analysis clearly explained that the
amount of noise present in the image has a greater role in the segmentation process.
In this comparative study, PFCM outperformed FCM in both noisy and noise-free
condition.

References

1. Pal, N.R., Pal, K., Bezdek, J.A.: A possibilistic fuzzy c-means clustering algorithm. IEEE
Trans. Fuzzy Syst. 13(4), 517– 530 (2005)
2. Ganesan, P., Palanivel, K.., Sathish, B.S., Kalist, V., Shaik, K.B.: Performance of fuzzy
based clustering algorithms for the segmentation of satellite images—a comparative study. In:
IEEE Seventh National Conference on Computing, Communication and Information Systems
(NCCCIS), pp. 23–27 (2015)
3. Lei, T., Jia, X., Zhang, Y., He, L., Meng, H., Nandi, A.K.: Significantly fast and robust fuzzy
c-means clustering algorithm based on morphological reconstruction and membership filtering.
IEEE Trans. Fuzzy Syst. 26(5), 3027–3041 (2018)
4. Sathish, B.S., Ganesan, P., Shaik, K.B.: Color image segmentation based on genetic algorithm
and histogram threshold. Int. J. Appl. Eng. Res. 10(6), 5205–5209 (2015)
Comparative Study of Segmentation of Noisy Color Synthetic … 477

5. Krishnapuram, R., Keller, J.: The possibilistic c-means algorithm: Insights and recommenda-
tions. IEEE Trans. Fuzzy Syst. 1(2), 98–110 (1993)
6. Ganesan, P., Sajiv. G.: User oriented color space for satellite image segmentation using fuzzy
based techniques. In: International Conference on Innovations in Information, Embedded and
Communication Systems (ICIIECS) pp. 1–6 (2017)
7. Correa, C.: A comparison of fuzzy clustering algorithms applied to feature extraction on
vineyard. In: Lecture Notes in Computer Science, Springer, Berlin, pp. 56–65 (2012)
8. Ganesan, P., Sajiv, G.: Unsupervised clustering of satellite images in CIELab color space
using spatial information incorporated FCM clustering method. Int. J. Appl. Eng Res. 10(20),
18774–18780 (2015)
9. Parker, J.K.., Hall, L.O., Bezdek, J C.: Comparison of scalable fuzzy clustering methods.
Proceedings of IEEE International Conference on Fuzzy Systems. pp. 1–9 (2012)
10. Ganesan, P., Rajini, V.: Assessment of satellite image segmentation in RGB and HSV color
space using image quality measures. In: International Conference on Advances in Electrical
Engineering (ICAEE), pp. 1–5 (2014)
11. Wang, Z., Song, Q., Soh, Y.C., Sim, K.: An adaptive spatial information-theoretic fuzzy clus-
tering algorithm for image segmentation. Comput. Vis. Image Understand. 117, 1412–1420
(2013)
12. Kalist, V., Ganesan, P., Sathish, B.S., Jenitha, J.M.M.: Possiblistic- Fuzzy C-Means Clustering
Approach for the Segmentation of Satellite Images in HSL Color Space. Procedia Comput.
Sci. 57, 49–56 (2015)
13. Havens, T., Bezdek, J.C., Leckie, C., Hall, L.O., Palaniswami, M.: Fuzzy c-means algorithms
for very large data. IEEE Trans. Fuzzy Syst. 20(6), 1130–1146 (2012)
14. Sajiv, G., Ganesan, P.: Comparative study of possiblistic fuzzy c-means clustering based image
segmentation in RGB and CIELuv color space. Int. J. Pharm. Technol. 8(1), 10899–10909
(2016)
15. Jia, S., Zhang, C.: Fast and robust image segmentation using an superpixel based FCM
algorithm. In: IEEE International Conference on Image Processing (ICIP), pp. 947–951 (2014)
16. Ganesan, P., Sathish, B.S., Sajiv, G.: A comparative approach of identification and segmentation
of forest fire region in high resolution satellite images. In: 2016 World Conference on Futuristic
Trends in Research and Innovation for Social Welfare (Startup Conclave), pp. 1–6 (2016)
17. Krinidis, S., Chatzis, V.: A robust fuzzy local information c-means clustering algorithm. IEEE
Trans. Image Process. 19(5), 1328–1337 (2010)
18. Chuang, K.S., Tzeng, H.L., Wu, J., Chen, T.J.: Fuzzy c-means clustering with spatial
information for image segmentation. Comput. Med. Imag. Graph. 30(1), 9–15 (2006)
FPGA Implementation of 16-Bit
and 32-Bit Heterogeneous Adders

Salony Mahapatro and Shasanka Sekhar Rout

Abstract This brief presents the design of 16-bit and 32-bit heterogeneous adders.
For designing efficient adder architecture in terms of power, area and speed, a number
of research works have been implemented to design an optimized architecture. This
paper gives a comparative analysis of previously presented designs and the proposed
heterogeneous adder architectures to validate a trade-off between power, area and
delay. The proposed heterogeneous adder designs consist of cascading of ripple
carry adder (RCA) and carry-lookahead adder (CLA) and are implemented in Xilinx
Vivado 2017.1 design tool and FPGA-Kintex7 (xc7k70tfbv676-1) device without
any constraints. These proposed adders are coded with VHDL language for obtaining
better performance parameters in comparison with other reported adders for different
application points of view.

Keywords Constraints · Heterogeneous adder · Ripple carry adder · VHDL ·

Vivado

1 Introduction

The electronic applications mostly use adders for performing various algorithms.
Major issue in VLSI design is to propose architectures with reduction of power
consumption, area utilization and increase of operation speed. Glitching causes a lot
of unwanted circuit behavior which may be brought out at gate level. Both balanced
tree structure and cascaded gate structure are some of such implementations. If
equal input arrival times and gate delays are assumed as in [1], we can observe
that the cascaded structure experiences more transitions as compared with the tree
structure before settling at its steady-state value. The input signals arriving may
trigger transitions at the output of each gate where such output transitions may

S. Mahapatro
Asiczen Technologies, Bhubaneswar 751024, India
S. S. Rout (B)
GIET University, Gunupur 765022, India
e-mail: [email protected]
© Springer Nature Singapore Pte Ltd. 2020 479
G. Pradhan et al. (eds.), Advances in Electrical Control and Signal Systems,
Lecture Notes in Electrical Engineering 665,
https://doi.org/10.1007/978-981-15-5262-5_35
480 S. Mahapatro and S. S. Rout

trigger back additional transitions for gates within their fan-out. The path delays in
the balanced tree structure are balanced in nature, and henceforth, each node makes
a single transition, and no power is unused. Glitching can be eliminated in static
circuits to reduce power consumption.
Many research works have been carried out previously proposing various designs
for adder architecture on different platforms to give least power dissipation, area
utilized and worst-case delay [1–6]. In this paper, the performance parameters of
the proposed 16-bit and 32-bit heterogeneous adder have been compared with other
reported works. The Xilinx Vivado 2017.1 design tool software is used for synthesis
of proposed designs, and FPGA-Kintex7 (xc7k70tfbv676-1) device is selected to get
area report without any constraints.
The paper is planned as follows. The performance analysis of proposed designs
is discussed in Sect. 2. The synthesis results are reported in Sect. 3. Section 4 shows
the comparison and observation of the proposed work. Finally, Sect. 5 concludes it.

2 Performance Analysis of the Proposed Design

Here, homogeneous adders like ripple carry adder, carry-lookahead adder and het-
erogeneous adder are explained briefly with power dissipation and delay estimation.

2.1 Homogeneous Adders: Ripple Carry Adder

Cascading full adder blocks in a series manner is done for constructing ripple carry
adder (RCA). Since the carry out of each full adder phase is considered as an input
in subsequent full adder stage, so the RCA is plagued with slow speed [2]. Figure 1
presents the building block of ripple carry adder, where A0 − An and B0 − Bn are the

Fig. 1 n-bit ripple carry adder

FPGA Implementation of 16-Bit and 32-Bit Heterogeneous Adders 481

two n-bit inputs, C 0 is carry in and S 0 − S n is the n-bit sum output, while Cn is carry
out. The remaining C 1 , C 2 … C n − 1 are the intermediate carry.
The major disadvantage of RCA is that when the bit length is going to increase,
there is an enhancement in delay. The major delay occurs here due to the carry
propagation. Hence, RCA is not preferable for addition of large number bits.

2.2 Homogeneous Adders: Carry-Lookahead Adder

By reworking the expressions of full adder, the operation of the CLA can be explained.
The propagate  Pi and generate  G i of a full adder circuit are written as follows:

Pi = xin xor yin Carry Propagate (1)

G i = xin and yin Carry Generate (2)

So, we can see that generate and propagate bits stand only on input bits and
are thereby applicable after single gate delay. Thus, the sum and carry out may be
expressed as:

Sum = Si = Pi xor Ci−1 (3)

Carryout = Ci+1 = G i + Pi and Ci (4)

The Eqs. (3) and (4) represent two cases where the carry signal generates.
(a) If both bits A and B are 1
(b) If either A or B is 1 and the C 0 is 1
Applying the above equations in a 4-bit adder:

C1 = G 0 + P0 C0
C2 = G 1 + P1 C1 = G 1 + P1 [G 0 + P0 C0 ]
= G 1 + P1 G 0 + P1 P0 C0
C3 = G 2 + P2 C2 = G 2 + P2 G 1 + P2 P1 G 0 + P2 P1 P0 C0
C4 = G 3 + P3 C3 = G 3 + P3 G 2 + P3 G 2 + P3 P2 G 1 + P3 P2 P1 G 0
+ P3 P2 P1 P0 C0

Similarly, the general expression is written in Eq. (5)

Ci + 1 = G i + Pi G i−1 + Pi Pi−1 G i−2 + . . . Pi Pi−1 . . . P2 P1 G 0

+ Pi Pi−1 . . . P1 P0 C0 (5)
482 S. Mahapatro and S. S. Rout

Fig. 2 Carry-lookahead adder

CLA adder configuration divides into three parts: sum generator, propagate or
generate generator and carry generator [7, 8]. The structural design of n-bit CLA
adder is presented in Fig. 2.

2.3 Heterogeneous Adders

Various 16-bit and 32-bit heterogeneous adder designs are proposed here which
consist of two sub-adders RCA and CLA of different orders. All these two sub-
adders are concatenated to form a 16-bit and 32-bit heterogeneous adder. The order
(bit width) of individual sub-adder has a strong impact on the performance of the
heterogeneous adders. The complete structure of the proposed model is shown in
Fig. 3.

Fig. 3 Inner structure of the proposed model

FPGA Implementation of 16-Bit and 32-Bit Heterogeneous Adders 483

2.4 Power Dissipation

In idle state, nearly zero power is dissipated by the static CMOS gates. Total power
dissipation is the net sum of static dissipation and dynamic dissipation components
[5].

Ptotal = Pstatic + Pdynamic (6)

Power dissipation in CMOS circuits is available as two components:

(a) Static Dissipation: It is the product of total leakage static current and supply
voltage.

Pstatic = Istatic VD D (7)

(b) Dynamic Dissipation: It is defined in Eq. (8).

Pdynamic = αC VDD
2
(8)

Activity factor, α = 1 of a clock, since it rises and falls in every cycle.

In Xilinx Vivado 2017.1 design tool, the dynamic power can be calculated as
follows:

PDynamic = PSignals + PLogic + PIO (9)

2.5 Delay Estimation

There are a number of paths, called critical paths giving worst-case delay which need
attention to adjust timing parameters.
For a RCA adder:

tripple = tpg + (N − 1)tAO + txor (10)

where t pg is 1-bit propagate or generate gates delay, t AO is the delay of AND_OR

gate and t xor is the delay of the final sum XOR.
For a CLA adder using k-groups of n-bits, individual has a delay:

tCLA = tPG + tPG(n) + [(n − 1) + (k − 1)]tAO + txor (11)

where tPG(n) is AND_OR_AND_OR….AND_OR gate delay computing valence n

generate signal [9].
484 S. Mahapatro and S. S. Rout

3 Simulation and Synthesis Results

The proposed 16-bit and 32-bit homogeneous adders and heterogeneous adders are
synthesized in Xilinx Vivado 2017.1 design tool to obtain power utilization summary,
lookup tables (LUTs) and maximum combinational path delay. The functional sim-
ulation waveforms of 16-bit and 32-bit heterogeneous adder variants for the FPGA
implementation are shown in Figs. 4 and 5, respectively.
The post-implementation simulation waveforms of 16-bit and 32-bit heteroge-
neous adder variants are displayed in Figs. 6 and 7, respectively. The power utiliza-
tion summary of 16-bit and 32-bit adders is given in Tables 1 and 2, respectively. It
demonstrates the detailed information regarding the total power distribution which
is consumed by each design.

Fig. 4 Functional simulation for 16-bit heterogeneous adder

Fig. 5 Functional simulation for 32-bit heterogeneous adder

FPGA Implementation of 16-Bit and 32-Bit Heterogeneous Adders 485

Fig. 6 Post-implementation simulation for 16-bit heterogeneous adder

Fig. 7 Post-implementation simulation for 32-bit heterogeneous adder

The area and timing report summary of 16-bit and 32-bit adders are given in
Tables 3 and 4, respectively. It briefs the maximum combinational path delay and
total number of path details.

4 Comparison and Observation

Tables 5 and 6 represent the comparison of delays and area utilization, respectively,
of the proposed work with other previously reported works. It can be observed that
CLA requires lesser area taking considerations from the utilization summary of slices,
LUTs with all counted gates and the total path delay as comparison in [4].
486 S. Mahapatro and S. S. Rout

Table 1 Power analysis of 16-bit adders

Adder Total Dynamic power PD (in W) PD = PS Static
variants power (in Signal Logic IO power + PL + power (in
W) power (in power (in (in W) PI PI (in W) W)
W) PS W) PL
16CLA 10.970 0.423 0.117 10.318 10.849 0.121
2RCA + 10.848 0.410 0.108 10.218 10.736 0.112
14CLA
4RCA + 10.827 0.408 0.104 10.202 10.715 0.112
12CLA
8RCA + 10.833 0.415 0.106 10.200 10.721 0.112
8CLA
10RCA + 10.828 0.412 0.105 10.200 10.716 0.112
6CLA
12RCA + 10.832 0.416 0.104 10.199 10.720 0.112
4CLA
16RCA 10.962 0.419 0.113 10.318 10.849 0.113

Table 2 Power analysis of 32-bit adders

Adder Total Dynamic Power PD (in W) PD = PS Static
Variants Power (in Signal Logic IO Power + PL + Power (in
W) Power (in Power (in (in W) PI PI (in W) W)
W) PS W) PL
32CLA 22.362 1.540 0.248 20.384 22.171 0.191
2RCA + 21.627 0.941 0.218 20.284 21.444 0.183
30CLA
4RCA + 21.605 0.936 0.216 20.269 21.421 0.184
28CLA
8RCA + 21.613 0.945 0.218 20.266 21.43 0.183
24CLA
12RCA + 21.620 0.952 0.219 20.266 21.437 0.183
20CLA
16RCA + 21.621 0.947 0.225 20.266 21.438 0.183
16CLA
20RCA + 21.620 0.948 0.222 20.266 21.436 0.184
12CLA
24RCA + 21.634 0.964 0.221 20.266 21.451 0.183
8CLA
28RCA + 21.632 0.961 0.222 20.266 21.449 0.183
4CLA
32RCA 21.749 0.946 0.235 20.384 21.564 0.185
FPGA Implementation of 16-Bit and 32-Bit Heterogeneous Adders 487

Table 3 Area and timing report of 16-bit adders

Adder LUT Slice Cells Nets Max Total no.
variants (41,000) (10,250) combinational of paths
path delay
(ns)
16CLA 16 7 37 129 5.196 321
2RCA + 15 7 74 120 5.031 261
14CLA
4RCA + 14 7 83 129 4.479 217
12CLA
8RCA + 14 7 50 96 3.260 177
8CLA
10RCA + 14 7 42 88 3.383 181
6CLA
12RCA + 14 7 34 80 3.935 201
4CLA
16RCA 16 6 16 65 5.047 321

Table 4 Area and timing report of 32-bit adders

Adder LUT Slice Cells Nets Max Total no.
variants (41,000) (10,250) combinational of paths
path delay
(ns)
32CLA 32 15 69 257 9.036 1153
2RCA + 31 16 154 248 9.447 1029
30CLA
4RCA + 30 14 146 240 8.895 921
28CLA
8RCA + 30 14 130 224 7.676 753
24CLA
12RCA + 30 15 114 208 6.151 649
20CLA
16RCA + 30 14 98 192 5.047 609
16CLA
20RCA + 30 14 82 176 6.143 633
12CLA
22RCA + 30 15 74 168 6.695 669
10CLA
24RCA + 30 15 66 160 7.247 721
8CLA
26RCA + 30 14 58 152 7.799 789
6CLA
28RCA + 30 14 50 144 8.351 873
4CLA
32RCA 32 16 32 129 9.463 1153
488 S. Mahapatro and S. S. Rout

Table 5 Comparison of
Adder variant in Delay (in ns)
delays
References Proposed As per As per
reference proposed
[2] 16-bit RCA 24.68 5.047
[2] 8RCA + 24.6 3.260
8CLA
[2] 12RCA + 23.732 3.935
4CLA
[3] 16-bit RCA 24.1 5.047
[3] 16-bit CLA 20.7 5.196
[3] 8RCA + 22.1 3.260
8CLA
[5] 16-bit RCA 30.59 5.047
[5] 32-bit RCA 63.41 9.463

Table 6 Comparison of area

Adder variant in Area (in terms of LUTs)
utilization
References Proposed As per As per
reference proposed
[2] 16-bit RCA 32 16
[2] 16-bit CLA 27 16
[2] 8RCA + 32 14
8CLA
[2] 12RCA + 31 14
4CLA

Observing the data from [1–6], we can find that the proposed 16-bit heterogeneous
adders operate faster and occupy less area in terms of LUTs. All the proposed 16-bit
heterogeneous adders have maximum combinational path delay less than 6 ns and
occupy less area comparatively. We can also observe that the delays of 32-bit adder
variants are faster as compared with the adders in [5].

5 Conclusion

In this paper, we successfully designed the 16-bit and 32-bit heterogeneous adders
of various possible combinations using m-bit RCA and n-bit CLA to analyze the
power and area performances of each adder architecture. The simulation, synthesis
and implementation of each model are carried out. The model with least power
consumption is selected based on least dynamic power. Similarly, the model with
best area utilization is selected based on least LUT count. It can be concluded that the
FPGA Implementation of 16-Bit and 32-Bit Heterogeneous Adders 489

area and speed are inversely proportional to each other. The application requirement
of area, power and speed decides the type of adder to be selected. Also, we can
say using cascaded adders or heterogeneous adders have great advantages over the
homogeneous adders, or balance tree structured adders [10] like leakage power and
dynamic power is reduced, area utilized is less and operation speed is fast enough.
The power consumed by adders proposed in this paper is very high as we have
not applied any constraints, so this work can be extended for reducing the power
consumption using constraints.

References

1. Nagendra, C., Mehta, U., Owens, R., Irwin, M.: A Comparison of the power-delay characteris-
tics of CMOS adders. In: International Workshop on Low Power Design, pp. 1–5, Napa Valley
(1994)
2. Gowthami, K., Yamini-Devi, Y.: Design of 16-bit heterogeneous adder architectures using
different homogeneous adders. Int. J. Adv. Res. Electr Electron. Instrum. Eng. 5(10), 7843–
7849 (2016). https://doi.org/10.15662/ijareeie.2016.0510007
3. Singh, R., Chaturvedi, A., Singh, O.: Trade-offs in designing high-performance digital adder
based on heterogeneous architecture. Int. J. Comput Appl. 56(13), 12–16 (2012). https://doi.
org/10.5120/8950-3132
4. Kumar, R., Dahiya, S.: Performance analysis of different bit carry look ahead adder using
VHDL environment. Int. J. Eng. Sci. Innov. Technol. 2(4), 80–88 (2013)
5. Nagendra, C., Irwin, MOwens, R.: Area-time-power tradeoffs in parallel adders. IEEE Trans.
Circ. Syst II Analog Digital Signal Proc. 43(10), 689–702 (1996). https://doi.org/10.1109/82.
539001
6. Singh, R., Chaturvedi, A.: VLSI Design and implementation of heterogeneous adder for perfor-
mance optimization. Int. J. Comput. Appl. 51(7), 37–40 (2012). https://doi.org/10.5120/8057-
1405
7. Pai, Y., T., Chen, Y., K.: The fastest carry-lookahead adder. In: The Second IEEE International
Workshop on Electronic Design Test and Applications (DELTA), pp. 1–3, Australia (2004).
https://doi.org/10.1109/delta.2004.10071
8. Wang, Y., Pai, C., Song, X.: The design of hybrid carry-lookahead/carry-select adders. IEEE
Trans. Circ Syst. II Analog Digital Signal Proc. 49(1), 16–24 (2002). https://doi.org/10.1109/
82.996053
9. Mahapatro, S., Bhuyan, K., C., Acharya, S., Mishra, A.: Simulation and synthesis of heteroge-
neous adder using VIVADO. Int. J. Res. Appl Sci. Eng. Technol. 6(3), 964–976 (2018). https://
doi.org/10.22214/ijraset.2018.3154
10. Alisha, Raj., T.: Tracing of delay estimation in heterogeneous adder with FPGA. Int. J. Technol
Res. Eng. 4(11), 2323–2326 (2017)
Load Frequency Control of Multi-source
Interconnected Power System Controlled
by Cascaded Controller

Subhranshu Sekhar Pati and Saroj Kumar Mishra

Abstract To realize the smart grid, distributed generation becomes essential and
needs to be highlighted. The loading to such generations may hinder the system
stability as the size as well as complexity increases. The study describes the load
frequency control of multi-area system comprising of conventional thermal power
plant and renewable-based small hydro plant (SHP) and solar thermal plant (STP).
The frequency stability is maintained through recently used cascaded controller in
which two controllers are connected in cascade fashion for achieving an enhanced
result. To validate system stability, a time-varying load disturbance has been applied
to the system. The gain parameters of the cascaded controller are tuned through
Jaya algorithm. The dynamic interaction of the cascaded controller with the test
system provides better results than the other controller schemes such as proportional–
integral–derivative with filter (PIDF) and proportional–integral (PI)-based controller.
A significant enhancement in the convergence time is perceived due to the efficient
optimization of the controller for maintaining system stability.

Keywords Load frequency control (LFC) · Renewable energy · Cascaded

controller · Jaya algorithm

1 Introduction

A successfully organized power network is epitomized by its concomitant system

constraints as well as associated control strategies. With regards to above, LFC is
regarded as one of the effective control schemes that restrict the frequency deviation
and not to allow the deviation to go beyond permissible limits. In addition to that, it
also controls the traction of power within diverse control area known as tie line power

S. S. Pati
Department of Electrical and Electronics Engineering, VIT, Balasore, Odisha 756043, India
e-mail: [email protected]
S. K. Mishra (B)
Department of Electrical Engineering, IIIT, Bhubaneswar, Odisha 751003, India
e-mail: [email protected]

© Springer Nature Singapore Pte Ltd. 2020 491

G. Pradhan et al. (eds.), Advances in Electrical Control and Signal Systems,
Lecture Notes in Electrical Engineering 665,
https://doi.org/10.1007/978-981-15-5262-5_36
492 S. S. Pati and S. K. Mishra

[1]. In the recent scenario, most of the power is derived from the conventional thermal
and hydropower plant. However, these conventional plants, particularly thermal plant,
put a negative impact on climate and cause global warming which should be avoided
as quickly as possible [2]. Hence, power derived from conventional power plant
is shifted to renewable-based power unit like SHP, wind, and STP. Moreover, it is
observed from the past literatures that most of the LFC study deals with classical
thermal, hydro, and gas plant [3, 4]. So, analysis with recent type of renewable plants
such as STP and SHP opens a new perspective in LFC study.SHP is a low investment
renewable power plant that is regarded as distributed generations with low operating
cost [5, 6]. One of the other renewable plants named as STP is gaining popularity
day by day and laid its prominent footprint in renewable power sector. STP can be
operated at island or grid-connected mode [7, 8].
For effective monitoring and controlling of highly complex power system, well
operative controller should be used. Various researchers used the diverse controller
in their respective study. Some of these are proportional–integral–derivative-based
controller, fractional-order controller, and two-degree-of-freedom controller [9]. In
the recent scenario, cascaded controller is used in complex system and successfully
stabilizes the response from external disturbances [10]. Keeping the view of above,
this study utilizes a new combination of cascaded controller for controlling and mon-
itoring the system response. Generally, the controller parameters are tuned through
optimization algorithm. Researchers use various algorithms like artificial bee colony,
bat algorithm, and ant colony algorithm [3–12]. In this paper, parameters of cascaded
controller are tuned through Jaya algorithm, which is a recent but robust optimiza-
tion algorithm [12, 14]. The key points of this study are highlighted by the following
points.
• To develop a renewable-based two-area system incorporating STP and SHP.
• To design the Jaya algorithm optimized cascaded controller and utilized it in the
above said system.
• To evaluate the efficiency of the cascaded controller with dynamic loading
and establishing the supremacy of the proposed controller with that of other
controllers.

2 System Study

The model taken for investigation has two areas, and both areas are interconnected
through tie line. Either area contains thermal, SHP, and STP power sources. The
thermal plant contains a hydraulic amplifier and non-reheat turbine. The general-
ized structure and transfer function model of two-area interconnected system are
presented in Figs. 1 and 2, respectively. With advancement of technology and stor-
age devices, the dependency on fossil fuel burning plant is significantly decreasing
[3–5]. Moreover, in the recent scenario, most of the power generation is expected
from renewable energy such as wind and solar. This can also be useful for achieving
Load Frequency Control of Multi-source Interconnected Power … 493

Area-1 Area-2

Thermal Thermal

Tie line 1-2

SHP SHP

STP STP

Fig. 1 Proposed test model considered for investigation

Fig. 2 Block diagram representation of the multi-source test system

smart grid where distributed renewable generations play a vital role. SHP is one of
the emerging technologies whose popularity is increasing in an incremental manner
due to its economic viability and availability. Normally, SHP contains a generator,
turbine, and regulator for speed control. The running parameter of SHP is taken from
[5, 6].
Due to the advancement of solar technology and supports from government pol-
icy, the power generation from the solar sector increased multiple times in the last
494 S. S. Pati and S. K. Mishra

decades. The solar power is not only used for powering purpose in residential area
but also used in satellites [7]. In STP, large area collector is used for focusing the solar
beam into number of tubes in which heat absorption substances like water and molten
salts are present [8]. The STP used in the study has a solar field, turbine, and governor
as shown in the figure. For simulating the practical scenario, 1 s delay is considered
in either area STP plant. In this study, for effective monitoring and controlling of
complex the power system, Jaya-tuned cascaded controller is considered.

3 Controller and Objective Function

With the advancement of technology in the renewable power sector, the power system
becomes highly complex. The system stability and reliability should be maintained
through proper selection of controller. In concerning to the above, PID controller
mostly used in LFC study, which is a simple structure controller yet practical and easy
to operate [9]. However, PID controller is easily implementable, but the response is
decorated at transient as well as steady state. Furthermore, proportional gain which
is accountable for improving the steady-state error may degrade the response in
transient state. In addition to that, the high value of proportional gain delays the
system speed. Hence, the proportional gain effect should be minimized according to
system constraints. This problem can overcome by adopting a cascaded controller
that is (PD) − (1 + PI) controller, in which the first phase comprises PD gain term
and the second stage holds (1 + PI) controller [10]. The transfer function of the
projected controller is presented in Eq. (1), and the diagram is shown in Fig. 3.

T . Fmultistage = (K p + s K D ) × (1 + K PP + K I /s) (1)

Cascaded control is an optimal controlling technique for achieving the enhanced

response in complex interconnected power system. It provides superior stability of
the system during the disturbances caused by either externally or internally. More-
over, it has more number of tuning regulators than the other controllers. Adhering
such benefits of the cascaded controller, various researchers considered this type

Fig. 3 Schematic diagram

of cascaded controller 1
KP

ACE +
+
U
Σ K PP Σ
+
+ +

sK D 1
KI
s
Cascaded controller
Load Frequency Control of Multi-source Interconnected Power … 495

of arrangement in their respective study [11]. Thus, this study also considered the
cascaded controller for regulating the frequency deviation. There are several cost
functions sometimes known as objective function which is associated with the pro-
cess of tuning of controller parameter [3–6]. In context to that, this study uses integral
time absolute error (ITAE), owing to rapid realization of targeted controlling action,
and the same is depicted in Eq. 2 [13].

tsim
J = ITAE = (|F1 | + |F2 | + |Ptie |) × t × dt (2)
0

In the above equation, F 1 and F 2 are referred as fluctuations of frequency in

the first area and second area, respectively and P tie is the change of power in tie
line.

4 Jaya Optimization Algorithm

To derive the solution of any maximization or minimization problem, an optimiza-

tion algorithm is needed. In this context, Rao et al. proposed a simple and influential
algorithm known as Jaya algorithm, which can be used for both constrained and
unconstrained problem [12]. The formulation of Jaya algorithm involves simple
steps, and these are designed in such a way that after completion of each iteration,
the final solution always moves closer to the best solution and not diverted into worst
solution, and at the same time, the algorithm does not have algorithm-specific param-
eters. As a result, the speed of convergence increases and makes it as a perfect choice
for optimization problem. It is observed that the working performance of Jaya algo-
rithm is superior to the evolutionary-based algorithm and swarm intelligence-based
algorithm. The equation responsible for updating the solution is easy to implement,
and the rate of accurateness is also very high, and the same is stated in Eq. 2 in which
X j, k, i is called as present set of parameters, while r1 and r2 are the constants which
lie in between 0 and 1 [14]. The revised solutions are refereed as X j,k,i which is
attained through the modification of X j, k, i with X j, best, i and X j, worst, i .
   
X j,k,i = X j,k,i + r1, j,i X j,best,i −  X j,k,i 
   
− r2, j.i X j,worst,i −  X j,k,i  (3)

where ‘j’ and ‘i’ denote the controller parameter and population space and ‘k’ stands
for cycle number. The basic working code of Jaya algorithm is presented in the
succeeding stage.
Step 1 Initialization
Step 2 The best and worst solution is evaluated from population size.
Initialize j, i, and k. The termination rule is also set.
496 S. S. Pati and S. K. Mishra

Step 3 Utilizing Eq. 3, ultimate solution is updated by using the best and worst
solution.
Step 4 Check the solution reported by X j,k,i is better than X j, k, i .
If yes: Ultimate solution is replaced by the previous solution.
If no: No need of chage.
Step 5 Checking of termination criteria
If yes: Print the optimum solution and algorithm stops.
If no: Go to step 2 and repeat.

5 Result and Analysis

To make the analysis more realistic, conventional power plant such as thermal along
with recent STP and SHP plants incorporated in this study. As the system is highly
complex and traditional controller cannot provide better response, recently used
cascaded controller is used in the study. The higher and lower limit should lie between
−3 and +3. To access the cascaded controller performance, it is compared with the
PIDF and PI controller and observed that cascaded controller works better in complex
system than the other controllers employed. The Jaya-tuned cascaded controller
parameters for two areas are presented in Table 1.
To check the system stability, a time-varying load disturbance is applied in the test
system as it perfectly simulates real-time load disturbances. The applied dynamic
loading is shown in Fig. 4. It can be easily inferred from Fig. 4 that +2%, −1%,
−2%, −1%, and +1% are applied at time interval of 10 s, 40 s, 65 s, 115 s, and
150 s., respectively. The graphical demonstration of the outcome derived from the
model with application of time-varying load distribution is shown in Fig. 5 and the
required measured mathematical data such as under/overshoot for the test system is
represented in Table 2. The response showcases the superiority of cascade controller,
and its comparative assessment that of PIDF and PI controller is also shown in the
same figure.

Table 1 Parameters of cascaded controller tuned by Jaya algorithm

Controller/optimization technique Optimized controller parameters
Kp KD K pp Ki
Area-1 PD-(1 + PI): Jaya 1.171 1.981 2.811 2.439
Area-2 PD-(1 + PI): Jaya 1.558 1.489 2.674 2.159
Load Frequency Control of Multi-source Interconnected Power … 497

-1

0 20 40 60 80 100 120 140 160 180 200

Time (sec)

Fig. 4 Time-varying load disturbances applied to the test system

6 Conclusion

The study offers a critical discussion of renewable-based multi-source two-area

power system. The study includes a recent renewable-based plant that is SHP and
STP in either area. These plants have greater impact on the recent scenario for the
realization of smart grid. For maintaining the frequency stability, cascaded con-
troller is employed in this study. The suggested controller parameters are optimized
through Jaya algorithm. With the use of Jaya-optimized cascaded controller, it is
inferred from the response that the system is not only stabilized as quick as possible
with time-varying load disturbances but also reached the steady state with minimum
time. Moreover, the optimum result of efficient Jaya algorithm is calculated with the
minutest time due to its parameter-free operation, thus plummeting the time delay.
498 S. S. Pati and S. K. Mishra

-3
10
5
PI
PI`DF
Cascaded

-5
0 20 40 60 80 100 120 140 160 180 200
Time(sec)
-3 (a)
10

-2

-4 PI
PIDF
Cascaded
-6
0 20 40 60 80 100 120 140 160 180 200
Time(sec)
10
-3 (b)

0.5

-0.5
PI
-1 PIDF
Cascaded
0 20 40 60 80 100 120 140 160 180 200
Time(sec)
(c)

Fig. 5 System response for a time-varying load variation applied to system

Load Frequency Control of Multi-source Interconnected Power … 499

Table 2 System response parameters f 1 , f 2 , and Ptie for time-varying load disturbances
Loading/controller f 1 f 2 Ptie12
Under/overshoot Under/overshoot Under/overshoot
(× 10−3 ) (× 10−3 ) (× 10−3 )
PD = +2% at Cascaded 3.054 3.018 0.778
10 s PIDF 3.908 4.517 0.962
PI 4.019 4.891 1.258
PD = −1% at Cascaded 1.512 1.518 0.367
40 s PIDF 1.834 1.840 0.418
PI 1.935 2.221 0.507
PD = −2% at Cascaded 3.054 3.165 0.874
65 s PIDF 3.948 4.671 0.981
PI 4.025 5.018 1.236
PD = −1% at Cascaded 1.818 1.934 0.511
115 s PIDF 2.108 2.657 0.648
PI 2.291 2.913 0.699
PD = +1% at Cascaded 0.618 0.781 0.198
150 s PIDF 0.851 0.828 0.204
PI 0.982 0997 0.225

References

1. Bevrani, H.: Robust power system frequency control. Springer (2014)

2. Elgerd, O.I.: Electric Energy Systems Theory: An Introduction. McGraw-Hill Book Company,
New York, NY (2000)
3. Debbarma, S., Dutta, A.: Utilizing electric vehicles for LFC in restructured power systems
using fractional order controller. IEEE Trans. Smart Grid 8(6), 2554–2564 (2017)
4. Izadkhast, S., Garcia-Gonzalez, P., Fras, P., Bauer, P.: Design of plug-in electric vehicle’s
frequency-droop controller for primary frequency control and performance assessment. IEEE
Trans. Power Syst. 32(6), 4241–4254 (2017)
5. Doolla, S., Bhatti, T.S.: Load frequency control of an isolated small-hydro power plant with
reduced dump load. IEEE Trans. Power Syst. 21(4), 1912–1919 (2006)
6. Sarma, U., Saikia, L.C., Sana, A., Dash, P.: Load frequency control of a multi-area, multi-
source system using firefly algorithm optimized F2DOFIDD controller. In: IEEE Region 10
Conference (TENCON), Singapore, pp. 1475−1479 (2016)
7. Das, D.C., Sinha, N., Roy, A.K.: GA based frequency controller for solar thermal–diesel–
wind hybrid energy generation energy storage system. Int. J. Elect. Power Energy Syst. 43(1),
262–279 (2012)
8. Bayon, A., Bader, R., Jafarian, M., Fedunik-Hofman, L., Sun, Y., Hinkley, J., Miller, S.,
Lipiński, W.: Techno-economic assessment of solid–gas thermochemical energy storage
systems for solar thermal power applications. Energy J. 149, 473–484 (2018)
9. Pati, S.S., Mishra, S. K.: Automatic generation control of a wind turbine integrated hydro-
thermal plant with redox flow batteries for 2-area interconnected power system. Springer Nat.
Smart Intell. Comput. Appl. Smart Innovation, Syst. Technol. 159, 411–418 (2019). https://
doi.org/10.1007/978-981-13-9282-5_39
500 S. S. Pati and S. K. Mishra

10. Dash, P., Saikia, L.C., andSinha, N.: Automatic generation control of multi area thermal system
using Bat algorithm optimized PD–PID cascade controller. Int. J. Elect. Power Energy Syst.
364–372 (2015)
11. Pati, S.S., Mishra, S.K.: A PSO based modified multistage controller for automatic generation
control with integrating renewable sources and fact device. Int J. renew. Energy Res. 9(2),
673–683 (2019)
12. Rao, R.V.: Jaya: a simple and new optimization algorithm for solving constrained and
unconstrained optimization problems. Int. J. Ind. Eng. Comput. 7, 19–34 (2016)
13. Mishra, S.K., Pati, S.S.: A solar-hydro based frequency regulation in two-area power system
incorporating unified power flow control. Springer Nat. Smart Intell. Comput. Appl. Smart
Innov Syst. Technol. 159, 485–493 (2019). https://doi.org/10.1007/978-981-13-9282-5_46
14. Rao, R.V., Saroj, A.: A self-adaptive multi-population based Jaya algorithm for engineering
optimization. Swarm Evol. Comput. J. 37, 1–26 (2017)
A New Active Power Loss Allocation
Method for Radial Distribution Networks
with DGs

Ambika Prasad Hota, Sivkumar Mishra, and Debani Prasad Mishra

Abstract In this paper, a new active power loss allocation (LA) scheme is devel-
oped by eliminating the influence of cross-term mathematically from loss equation
for allocating losses to the network participants with/without distributed generators
(DGs). The unique bus identification scheme introduced performs load flow easily
and efficiently by considering DGs as negative constant power loads. This method
assigns losses to the consumers/DG units with due consideration to their power
consumptions/injections and physical locations in the radial distribution network
(RDN). The effectiveness of the proposed LA scheme has been investigated against
the quadratic scheme of LA using a 33-bus RDN with/without DGs.

Keywords Radial distribution networks · Loss allocation · Distributed

generation · Load flow

1 Introduction

Electrical power sector in many countries has been either deregulated or is in the
process of getting deregulated. Several issues are coming up due to this as separate
generation, transmission, and distribution companies are being set up in place of
single state-owned vertically integrated structure, with the main objective to introduce
competition for bringing down the cost of electricity and to enhance the service
quality. So it becomes extremely important that various wheeling activities must
be clearly defined and the corresponding cost must be recovered. Transmission and
distribution power loss is an important component of the cost of electricity, and

A. P. Hota (B) · D. P. Mishra

IIIT, Bhubaneswar, Odisha, India
e-mail: [email protected]
D. P. Mishra
e-mail: [email protected]
S. Mishra
CAPGS, BPUT, Rourkela, Odisha, India
e-mail: [email protected]
© Springer Nature Singapore Pte Ltd. 2020 501
G. Pradhan et al. (eds.), Advances in Electrical Control and Signal Systems,
Lecture Notes in Electrical Engineering 665,
https://doi.org/10.1007/978-981-15-5262-5_37
502 A. P. Hota et al.

hence, it is to be recovered from the network users by suitably and fairly allocating it
among the users. The nonlinear relationship between the power loss and the injected
power makes the loss allocation process difficult and complicated. It can be revealed
from the literature that most of the LA methods developed are allocating losses in the
transmission lines [1]. Similarly, LA schemes exclusively for RDNs are also proposed
[2–4], where the consumers at various buses are paying for the active power loss in
the RDN, and the substation bus, which is the natural slack bus, is not allocated with
any loss.
The electrical power distribution network recently has gained focus due to large
penetrations of DGs into the system. The injection of DG power alters the basic con-
figuration of the RDN from a passive system to an active one. Hence, loss allocation
methods are also need to be modified to include DGs in the loss allocation process.
Mutale et al. [5] considered DGs in RDNs for LA using two schemes; one based
on marginal method, which required reconciliation due to over recovery of loss and
the other based on direct loss coefficients. Costa and Matos [6] proposed a branch-
based LA scheme for distribution networks with DGs. Carpaneto et al. [7] proposed
a branch current decomposition-based LA scheme (BCDLA) for RDNs with DGs. In
[8], adaptions of several transmission system-based LA schemes to distribution net-
works with DGs are critically compared. Carpaneto et al. [9] extended the BCDLA
scheme to three-phase unbalanced RDNs with DGs. Atanasovski and Taleski [10]
presented a power summation-based LA method (PSMLA) for RDNs with DGs using
a quadratic relation between losses and power flows. Again in 2012, Atanasovski
and Taleski [11] proposed an energy summation method of LA (ESMLA) for RDNs
with DGs utilizing a statistical representation of daily load and generation curves.
Ghofrani-Jahromi et al. [12] proposed a new LA method for RDNs with DGs, but
a normalization process is required to make the total loss allocated equal to the
actual loss. Jagtap and Khatod allocate losses in a RDN with DGs using the geo-
metric scheme of allocation for various load models in [13], using a branch-oriented
approach with backward sweep network reduction in [14, 15], using an exact method
in [16] and using the logarithmic scheme of allocation in [17]. Kashyap and De [18]
proposed a two-step proportional sharing-based LA approach for RDNs with DGs.
Recently, Kumar et al. [19] developed a cross-term decomposed technique for RDNs
with DGs.
In light of above developments, a new approach for active power loss allocation
of RDNs with DGs is proposed in this paper. A unique bus identification technique
[20, 21] is used to formulate and implement the proposed LA method for RDNs with
DGs, where DGs are considered as negative constant power loads. In this proposed
LA method, the active power loss in a branch of RDN is represented as a summation
of separate terms, and each term can be related to the active and reactive power
injections at the subsequent buses of a branch. Thus, the contribution of all the buses
(or consumers) to the total active power loss is determined so that losses will be
allocated to all the buses in a fair way. The LA results of a 33-bus RDN with three
embedded DGs are presented. These results are also compared with the results as
obtained by ‘Quadratic Allocation’ method [22, 23].
A New Active Power Loss Allocation Method … 503

2 Methodology

To understand the proposed LA scheme, a 12-bus RDN is considered (Fig. 1), where
a simple bus numbering scheme is followed. The branch numbers are one unit less
than that of its receiving end node numbers (Fig. 1). To avoid the conflict between
the complex operator ‘j’ and branch-j, the branch-j is represented as branch-jj in this
paper. Here, total number of nodes (NB) and branches (NBR) of RDN are related
as:

NBR = NB − 1 (1)

2.1 Proposed Bus Identification Scheme

To identify the adjacent buses and the subsequent buses in a RDN, several arrays
are proposed by the authors in [20, 21]. These arrays make the load flow faster due
to its reduced search time. This bus identification is also adaptable to various other
applications. Adjacent buses of all nodes are investigated and placed in an array
adb of dimension twice that of NBR. Two pointer arrays, namely mf and mt, are
proposed to locate the neighboring buses in the array adb. These pointer arrays have
the dimensions equal to NB, and they locate to the starting and ending memory
locations of the adjacent buses, respectively. In the same manner, an array pb of
dimension NB is used to store the corresponding previous buses in a RDN. The
formations of these arrays are made using the network data and simple programming
techniques. The contents of mf, mt, and adb arrays of the RDN corresponding to
Fig. 1 are shown in Fig. 2 to explain the storage and pointer operation of the arrays.
In the same manner, nsb and sb are the two other arrays proposed to store the
number of subsequent buses and the subsequent buses relating to all branches of the
RDN, respectively. For pointing the initial and final locations of the subsequent buses

Fig. 1 A sample 12-node

RDN
504 A. P. Hota et al.

Pb(a) 0 1 2 3 4 3 6 7 7 2 10 4
Bus(a) 1 2 3 4 5 6 7 8 9 10 11 12

adb(s) 2 1 3 10 2 4 6 3 5 12 4 3 7 6 8 9 7 7 2 11 10 4
s 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22

adb(8) adb(10)

mf(a) 1 2 5 8 11 12 14 17 18 19 21 22 mt(a) 1 4 7 10 11 13 16 17 18 20 21 22

(a) 1 2 3 4 5 6 7 8 9 10 11 12 (a) 1 2 3 4 5 6 7 8 9 10 11 12

mf(4)=8 mt(4)=10
Bus
1 2 3 4 5 6 7 8 9 10 11 12
(a)

Fig. 2 Contents of mf, mt and adb arrays of the RDN (Fig. 1)

nsb(3)=3

sb(x) 2 3 4 5 6 7 8 9 10 11 12 3 4 5 6 7 8 9 12 4 5 12 5 6 7 8 9 7 8 9 8 9 10 11 11 12

x 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36

sb(22)=12
sb(20)=4
nsb(jj) 11 8 3 1 4 3 1 1 2 1 1
mfs(3)=20 mts(3)=22
jj 1 2 3 4 5 6 7 8 9 10 11

mfs(jj) 1 12 20 23 24 28 31 32 33 35 36 mts(jj) 11 19 22 23 27 30 31 32 34 35 36

jj 1 2 3 4 5 6 7 8 9 10 11 jj 1 2 3 4 5 6 7 8 9 10 11

Fig. 3 Contents of mfs, mts, sb, and nsb arrays of the 12-bus RDN

in the sb array, again two pointer arrays mfs and mts are selected. Figure 3 presents
the contents of mfs, mts, nsb, and sb arrays of the RDN (Fig. 1). The subsequent
bus information relating to branch-3 of the RDN is explained in this figure for easy
understanding. The formations of these arrays are made using the network data and
simple programming techniques.

2.2 Proposed Loss Allocation Scheme

The equivalent current injection (ECI) at a node ‘a’ with constant complex power
load SLa = PLa + j Q La and node voltage Va can be computed as:

PLa − j Q La
ILa = , a = 2, 3, 4, . . . NB (2)
(Va )∗
A New Active Power Loss Allocation Method … 505

The branch-jj current of a RDN is expressed using mfs, mts, and sb arrays as:


mts( j j)
I ( j j) = I L(sb(a)) (3)
a=mfs( j j)

Using (3) and rewriting (2), the branch-jj current is expressed as


mts( j j)
PL(sb(a)) − j Q L(sb(a))
I ( j j) = (4)
a=mfs( j j)
(V(sb(a)) )∗

Branch-jj power loss with sending end voltage (Vs), receiving end voltage (Vr),
and branch current I(jj) is expressed as
 
Ploss ( j j) =  (Vs − Vr )∗ . I ( j j) (5)

Substituting (4) in (5), we get

⎡ ⎤

mts( j j)
PL(sb(a)) − j Q L(sb(a)) ⎦
Ploss ( j j) = ⎣(Vs − Vr )∗ .  ∗ (6)
a=mfs( j j)
V(sb(a))

Further arranging,
⎡ ⎤

mts( j j)
Vs − Vr ∗

Ploss ( j j) = ⎣ · PL(sb(a)) − j Q L(sb(a)) ⎦ (7)
a=mfs( j j)
V(sb(a))

Let
∗
Vs − Vr
= X (sb(a)) + jY (sb(a))
V(sb(a))

Branch-jj power loss is then evaluated as:


mts( j j)
Ploss ( j j) = X (sb(a)) · PL(sb(a)) + Y (sb(a)) · Q L(sb(a)) (8)
a=mfs( j j)

Using (8), branch-jj power loss can be assigned to load points at buses beyond
branch-jj. So, the consumers connected at buses subsequent to branch-jj
i.e., to buses sb(a), for a = mfs( j j) to mts( j j) are allocated with losses as:

Ploss (sb(a)) = X (sb(a)) · PL(sb(a)) + Y (sb(a))Q L(sb(a)) (9)

506 A. P. Hota et al.

Total loss allocation for a consumer at bus ‘a’ can be calculated as the sum of the
corresponding loss allocations from each of the branch-jj of the RDN using (8) as:


NB−1
Tploss (a) = Ploss (sb(a)) (10)
j j=1

Thus, total loss of the RDN is computed as:


NB
TPloss = Tploss (a) (11)
a=1

2.3 Algorithm of the Proposed LA Scheme

It can be viewed from (9) that the allocated losses to various buses are expressed
in terms of bus voltages of the RDN, which are the node voltages obtained from
power flow. So, load flow must be executed for the network before allocation of
losses. A load flow based on forward–backward sweep (FBS) [24] is considered to
execute load flow and subsequent loss allocation. The algorithm for the proposed
loss allocation using FBS-based load flow is presented here.
STEP 1: All arrays (mf, mt adb, mfs, mts, nsb, pb and sb) are formed using
network data.
STEP 2: All bus voltage magnitudes and bus angles are set to a flat start of ‘1’ p.u
and ‘0’ radian. Also, set all branch losses to zero.
STEP 3: Initialize iteration = 0 and tolerance = 0.0001.
STEP 4: Eq. (2) is used to compute all ECIs of the buses.
STEP 5: Arrays (mfs, mts, and sb) and (3) are used to compute to all the branch
currents of the RDN.
STEP 6: Set i = 2.
STEP 7: Bus voltage updating is performed through a forward sweep. Array pb(i)
finds the previous bus-k of bus-i.
STEP 8: Value of ‘i’ is incremented.
STEP 9: Is i = NB, if yes go to step-10 else step-7 is executed.
STEP 10: Iteration count is incremented.
STEP 11: Convergence of voltages is checked by relating it with the results of the
previous iteration. In case of convergence, proceed to Step 12 else execute
Step-4.
STEP 12: Active power losses in all the branches as well as the total power loss of
the system are evaluated.
STEP 13: Total loss allocated to each of the consumers (buses) are initialized as
zero, i.e., TPloss (a) = 0, a = 2,….NB.
A New Active Power Loss Allocation Method … 507

STEP 14: Initialize a = 2.

STEP 15: The previous bus of bus a is found as k = pb(a) and V (k) − V (a) is
computed.
STEP 16: Using sb[] array, subsequent nodes to the branch a-1 are identified, and
loss of this branch is allocated to the subsequent buses using (9).
STEP 17: The value of ‘a’ is incremented. Is a = NB?, If yes, Step 18 is executed,
else go to step-15.
STEP 18: Using (10) the total LAs to all the consumers are calculated.

3 Loss Allocation with DGs

In order to extend the proposed LA method for radial distribution networks with
embedded DGs, it is important to model the DGs suitably and perform the load flow.
Based on the DG technology, the bus where a DG is connected can be considered to
be a PQ type or PV (voltage controlled) type.

3.1 PQ Models of DGs

Owing to small size of DGs compared to the conventional power sources, the constant
power model is considered accurate for load flow analysis [25]. Moreover, DGs are
not allowed for voltage regulation but rather they regulate power and power factor,
so negative load modeling is followed [26]. Automatic voltage regulators (AVRs)
are available in most of the DGs and these DGs operate in constant power output
mode. As a result, the system voltage and the output of the DGs are at same level.
Hence, it is preferable to handle the interconnection nodes of DGs as the PQ node
model rather than the PV node model as pointed out in [27] is followed in this paper.

3.2 DSLF and Loss Allocation with DGs

Considering negative load modeling of DGs, the net power injection at node-a can
be computed as:

PNa = PLa − Pga (12-a)

Q Na = Q La − Q ga (12-b)

where PLa + j Q La and Pga + j Q ga are the complex load power and DG injected
complex power at node-a, respectively. With the net injected power, the ECIs can
508 A. P. Hota et al.

be evaluated at all the buses using Eq. (2) and losses can be allocated to network
participants as per the algorithm presented in subsection 2.3.

4 Results and Discussion

A 12.66 kV, 33-bus RDN [30] with three DGs connected at bus 17, 18, and 33 is
considered for the implementation of the proposed LA scheme (Fig. 4). The DGs are
placed at these buses using the loss sensitivity analysis as discussed in the paper [28].
The DGs at these buses inject active powers (i.e., DG at node-17 injects 572.4 kW, DG
at node-18 injects 107 kW, and DG at node-33 injects 1046.2 kW) into the system. The
entire calculation has been performed by considering DG units as negative constant
loads (PQ type). The corresponding line data of the RDN is collected from [30].
A load flow based on forward–backward sweep (FBS) [24] has been carried out to
get converged values of node voltages. The entire active power loss of the RDN is
allocated to various consumers of the network assuming that they will pay for the
losses.

4.1 Loss Allocation Without DGs

Using FBS-based DSLF [21, 24] and proposed LA technique, the total active power
loss of the 33-bus test system is calculated to be 202.65 kW when no DGs are
connected to the network. This loss is assigned to all the end users connected at
several nodes of the test system excluding the bus-1 using the proposed loss allo-
cation scheme. No loss is assigned to the substation node-1. These allocated losses
are presented in Table 1. Allocated losses using the ‘Quadratic Scheme’ [22] are
also presented in Table 1 for comparison. It can be verified that the consumers are
allocated with losses depending on their loads (both active and reactive) as well as
their geographical locations with respect to the substation bus.

Fig. 4 A 33-bus RDN with three embedded DGs

A New Active Power Loss Allocation Method … 509

Table 1 LA to various consumers of the RDN

Bus No. Load at node LA (proposed method) LA (quadratic method)
(kW) (kW)
Load (kW) Load Without With DGs Without With DGs
(kVAR) DGs DGs
2 100.00 60.00 0.3129 0.2350 0.2164 0.1937
3 90.00 40.00 1.6303 1.1260 0.8743 0.7456
4 120.00 80.00 3.2501 2.2840 2.7100 2.3271
5 60.00 30.00 2.1026 1.3425 0.8323 0.6874
6 60.00 20.00 3.2295 1.9020 1.1785 0.9235
7 200.00 100.00 11.1991 7.3838 12.5262 9.9714
8 200.00 100.00 12.3529 7.8807 14.3433 11.1942
9 60.00 20.00 4.1167 2.3152 1.7627 1.3177
10 60.00 20.00 4.4929 2.4260 2.0444 1.4910
11 45.00 30.00 3.3631 2.1828 1.4354 1.0941
12 60.00 35.00 4.6137 2.8040 2.5991 1.9492
13 60.00 35.00 4.9818 2.8979 2.9232 2.1123
14 120.00 80.00 10.1125 6.2136 11.4785 8.3717
15 60.00 10.00 5.3473 2.1246 2.4212 1.5759
16 60.00 20.00 5.3627 2.3630 2.6826 1.7313
17 60.00 20.00 5.4775 −11.3516 2.7688 −23.8212
18 90.00 40.00 8.1811 1.2471 6.6916 0.8858
19 90.00 40.00 0.3183 0.2411 0.1976 0.1791
20 90.00 40.00 0.5966 0.5186 0.4760 0.4568
21 90.00 40.00 0.6472 0.5692 0.5270 0.5077
22 90.00 40.00 0.6911 0.6130 0.5714 0.5519
23 90.00 50.00 1.9553 1.4785 1.0112 0.8797
24 420.00 200.00 11.7106 9.3290 15.0305 12.8528
25 420.00 200.00 13.0342 10.6275 16.3952 14.1860
26 60.00 25.00 3.3886 2.0584 1.2892 1.0166
27 60.00 25.00 3.5863 2.1420 1.3421 1.0524
28 60.00 20.00 4.3744 2.4351 1.4385 1.0892
29 120.00 70.00 10.1704 6.5040 7.3949 5.8073
30 200.00 600.00 22.5517 30.4091 54.4025 48.9239
31 150.00 70.00 13.9789 7.6670 11.1662 8.3675
32 210.00 100.00 19.7721 10.75 19.6977 14.6374
33 60.00 40.00 5.7481 −3.9828 2.2209 −36.5229
Total 3715.00 2300.00 202.65 96.73 202.65 96.73
510 A. P. Hota et al.

Table 2 LA to some categorized consumers of the RDN without DGs

Bus Loads connected at Relative Loss allocated to the bus Remark
No. the bus location of
PLa QLa the bus Proposed Quadratic
(kW) (kVAR) method scheme
24 420 420 4 br. away 11.71 15.03 Mostly for
25 420 200 5 br. away 13.034 16.39 heavily
loaded buses
32 210 100 12 br. away 19.77 19.69 LAproposed
30 200 600 10 br. away 22.55 54.40 < LAQuadratic
7 200 100 6 br. away 11.199 12.52
2 100 60 1 br. away 0.312 0.216 Buses Near
3 90 40 2 br. away 1.630 0.874 to S/S
LAproposed
19 90 40 2 br. away 0.318 0.197 > LAQuadratic

To compare the proposed method with ‘Quadratic LA’, loss allocations of some
special category of buses are presented in Table 2, which reveals some of the important
characteristics of the present LA scheme. In the first category of heavily loaded buses
(bus 24,25 32,30 and 7; ordered in the decreasing active power load criteria) bus-
30 which is quite far from the substation (S/S) bus is allocated with the highest
loss; 22.55 kW which is 15% of the total loss by proposed method and 54.4 kW,
i.e., 27% by quadratic scheme. It is observed that both methods allocate losses to the
consumers based on their loads connected (both active and reactive) and their relative
locations with respect to the S/S bus. But, mostly, for heavily loaded buses, LAproposed
< LAQuadratic ; although, it cannot be generalized. For relatively lighter loaded buses
(both near and far to S/S), LAproposed > LAQuadratic .

4.2 Loss Allocation with DGs

The total active power loss with three connected DGs comes out to be 96.73 kW,
which matches with the result (96.76 kW) as reported in [28]. The allocated losses
both by proposed method and ‘Quadratic Scheme’ are presented in Table 1. When
DGs are connected to the system, the overall loss reduces significantly (reduction of
53% from the case with no DGs connected). It is also observed that all customers
get benefited due to reduced loss allocation except the customer at bus 30. Table 3
analyzes the loss allocated to some categorized buses by the two methods with and
without DGs. Except bus 18, for the two other buses, where the active power load is
much less than the active power generation injected, the loss allocated is negative;
i.e., the consumers at these buses are rewarded for reducing the overall power loss.
The loss allocation of the consumer at bus-18 although nonnegative but has reduced
substantially. On the other hand, the consumer at bus-30 is allocated with highest loss
Table 3 LA to some categorized consumers of the RDN with DGs
Bus Loads connected at the Relative location of the bus Loss allocated to the bus Remark
No. bus
PLa QLa From the S/S From Proposed method Quadratic scheme
(kW) (kVAR) DGs Without With DGs Without With DGs
DGs DGs
18 90 40 17 branch DG connected 8.1811 1.2471 6.6916 0.8858 DG
away (107 kW) connected
17 60 20 16 branch DG connected 5.4775 −11.3516 2.7688 −23.8212 buses
away 572.4 kW customers at
these buses
get incentive
Quadratic
scheme
gives higher
A New Active Power Loss Allocation Method …

incentives to
DG buses
33 60 40 13 branch DG connected 5.7481 −23.9828 2.2209 −36.5229
away 1046.2 kW
24 420 420 4 branch far from all DGs 11.7106 9.3290 15.0305 12.8528 For heavily
25 420 200 5 branch far from all DGs 13.0342 10.6275 16.3952 14.1860 loaded buses
LAproposed
32 210 100 12 branch 1 branch 19.7721 10.75 19.6977 14.6374 < LAQuadratic
30 200 600 10 branch 3 branch away 22.5517 30.4091 54.4025 48.9239 Bus 30 gets
away. penalized by
7 200 100 6 branch Almost equidistant 11.1991 7.3838 12.5262 9.9714 the proposed
away from DGs method
(continued)
511
Table 3 (continued)
512

Bus Loads connected at the Relative location of the bus Loss allocated to the bus Remark
No. bus
PLa QLa From the S/S From Proposed method Quadratic scheme
(kW) (kVAR) DGs Without With DGs Without With DGs
DGs DGs
2 100 60 1 branch Far from all DGs 0.3129 0.2350 0.2164 0.1937 For lightly
away loaded buses
3 90 40 2 branch Far from all DGs 1.6303 1.1260 0.8743 0.7456 LA proposed
away > LAQuadratic
19 90 40 2 branch Far from all DGs 0.3183 0.2411 0.1976 0.1791
away
A. P. Hota et al.
A New Active Power Loss Allocation Method … 513

(30.4 kW, 32% of total load by proposed method and 48.92 kW, 52% by Quadratic
method). As shown in Table 3, mostly, for heavily loaded buses, LA proposed < LA
Quadratic and bus 30 gets penalized for being the highest load consumer (in terms of
total load). The proposed method provides lesser benefit to the lightly loaded buses.

5 Conclusion

In this paper, a new active power loss allocation method for RDN with embedded
DGs has been proposed. This LA method allocates losses to the customers at various
buses based on their power ratings and geographical locations. The DGs are modeled
as negative power injections and are included in the LA process. Further, the proposed
LA method is not involved in any over recovery of losses, so there is no requirement
of further normalization of the allocated losses. The unique bus identification scheme
makes load flow simple and faster. The comparison result reveals that in both methods
of LA most of the consumers get benefit due to DG placement. However, the proposed
method is more accurate as it is derived without any assumption, and can be extended
further to unbalanced RDNs [29] with DGs.

References

1. Conejo, A.J., Arroyo, J.M., Alguacil, N., Guijarro, A.L.: Transmission loss allocation: a
comparison of different practical algorithms. IEEE Trans. Power Syst. 17, 571–576 (2002)
2. Savier, J.S., Das, D.: Energy loss allocation in radial distribution systems: a comparison of
practical algorithms. IEEE Trans. Power Delivery 24, 260–267 (2009)
3. Savier, J.S., Das, D.: An exact method for loss allocation in radial distribution systems. Electr.
Power Energy Syst. 36, 100–106 (2012)
4. Mishra, S., Das, D., Paul, S.: Active power loss allocation schemes in radial distribution systems.
Cogeneration Distrib. Gener. J. 25, 26–43 (2010)
5. Mutale, J., Strabac, G., Curcic, S., Jenkins, N.: Allocation of losses in distribution systems with
embedded generation. IEE Proc. Gen. Trans. Distrib. 147, 7–14 (2000)
6. Costa, P.M., Matos, M.A.: Loss allocation in distribution networks with embedded generation.
IEEE Trans. Power Syst. 19, 384–389 (2004)
7. Carpaneto, E., Chicco, G., Akilimali, J.S.: Branch current decomposition method for loss
allocation in radial distribution systems with distributed generation. IEEE Trans. Power Syst.
21, 1170–1179 (2006)
8. Carpaneto, E., Chicco, G., Akilimali, J.S.: Characterization of the loss allocation techniques
for radial systems with distributed generation. Electr. Power Syst. Res. 78, 1396–1406 (2008)
9. Carpaneto, E., Chicco, G., Akilimali, J.S.: Loss partitioning and loss allocation in three phase
radial distribution systems with distributed generation. IEEE Trans. Power Syst. 23, 1039–1049
(2008)
10. Atanasovski, M., Taleski, R.: Power summation method for loss allocation in radial distribution
networks with DG. IEEE Trans. Power Syst. 26, 1747–1755 (2011)
11. Atanasovski, M., Taleski, R.: Energy summation method for loss allocation in radial distribution
networks with DG. IEEE Trans. Power Syst. 27, 1433–1440 (2012)
12. Ghofani-Jahromi, Z., Mahmoodzadeh, Z., Ehsan, M.: Distribution loss allocation for radial
systems including DGs. IEEE Trans. Power Delivery 29, 72–80 (2014)
514 A. P. Hota et al.

13. Jagtap, K.M., Khatod, D.K.: Loss allocation in radial distribution networks with different load
models and distributed generations. IET Gener. Transm. Distrib. 9, 1275–1291 (2015)
14. Jagtap, K.M., Khatod, D.K.: Loss allocation in distribution network with distributed genera-
tions. IET Gener. Transm. Distrib. 9, 1628–1641 (2015)
15. Jagtap, K.M., Khatod, D.K.: Loss allocation in radial distribution network with various
distributed generators and load models. Int. J. Electr. Power Energy Syst. 75, 173–186 (2016)
16. Jagtap, K.M., Khatod, D.K.: Allocation of Distribution Network Losses with Different Types of
Distributed Generations. Annual IEEE India conference (INDICON), New Delhi, India (2015)
17. Jagtap, K.M., Khatod, D.K.: Novel approach for loss allocation of distribution networks with
DGs. Electr Power Syst. Res 143, 303–311 (2017)
18. Kashyap, S.S., De, M.: A novel loss allocation method for radial distribution system with
distributed generations. In: 1st IEEE International Conference on Power Electronics. Intelligent
Control and Energy Systems (ICPEICES-2016), Delhi, India, pp. 1–6 (2016)
19. Kumar, P., Gupta, N., Niazi, K.R., Swarnkar, A.: A circuit theory-based loss allocation method
for active distribution systems. IEEE Trans. Smart Grid 10, 1005–1012 (2017)
20. Mishra, S., Das, D., Paul, S.: A simple algorithm to implement active power loss allocation
schemes in radial distribution systems. J. Inst. Eng. (India), Series- B 93, 223–232 (2012)
21. Mishra, S., Das, D., Paul, S.: A simple algorithm for distribution system load flow with
distributed generation, pp. 1–5. Jaipur, India, Proc. IEEE ICRAIE (2014)
22. Savier, J.S., Das, D.: Impact of network reconfiguration on loss allocation of radial distribution
systems. IEEE Trans. Power Delivery 22, 2473–2480 (2007)
23. Savier, J.S., Das, D.: Loss allocation to consumers before and after reconfiguration of radial
distribution networks. Electr. Power Energy Syst 33, 540–549 (2011)
24. Mishra, S.: A simple algorithm for unbalanced radial distribution system load flow, pp. 1–6.
Hyderabad, India, Proc. IEEE TENCON (2008)
25. Eminoglu, U., Gozel, T., Hocaoglu, M.H.: DSPFAP: distribution systems power flow analysis
package using MATLAB graphical user interface (GUI). Comp. Appl. Engg. Edu. 18, 1–13
(2010)
26. Dugan, R.C.: Challenges in considering distributed generation in the analysis and design of
distribution systems. Proc. IEEE PES Gen. Meet. 1–8 (2008)
27. Choi, J.H., Kim, J.C.: Advanced voltage regulation method at the power distribution systems
interconnected with dispersed storage and generation systems. IEEE Trans. Pow. Deliv. 15,
691–696 (2000)
28. Rao, R.S., Ravindra, K., Satish, K., Narasimham, S.V.L.: Power loss minimization in distri-
bution system using network reconfiguration in the presence of distributed generation. IEEE
Trans. Power Syst. 28, 317–322 (2013)
29. Mishra, S., Das, D.: A novel active power loss allocation scheme for unbalanced radial power
distribution system. Int. Rev. Electr. Eng. 3, 543–550 (2008)
30. Baran, M., Wu, F.F.: Network reconfiguration in distribution systems for lossreduction and
load balancing. IEEE Trans. Power Del. 4, 1401–1407 (1989)
A GPS-Gradient Mapped
Database-Based Fuzzy Energy
Management System
for a Series—Parallel Hybrid Electric
Vehicle
Jose Thomas, Allen Thomas, Akhil Biju, Aswin Mathew, C. Parag Jose,
and K. M. Haneesh

Abstract The Energy Management System developed for the hybrid electric
vehicle operates using a database with GPS co-ordinates and corresponding alti-
tudes mapped, thereby giving a predictive control to optimize the operation of the
series–parallel hybrid system. The system aims at extracting the maximum potential
of the series–parallel hybrid power train architecture. The mapping of the latitude and
longitude obtained from a global positioning system (GPS) to the altitude measured
to create a database which generates a predefined driving cycle prior to the actual
motion of the vehicle. The created database is then used in a MATLAB/Simulink
model to simulate the operation of the series–parallel hybrid system and implement
the Energy Management System. The validated data is then tested in a Raspberry
Pi (RPi)-based prototype. The Energy Management System regulates the vehicle
dynamics based on the input drive cycle. The fuzzy logic-based control mechanism
is implemented in the RPi to optimize the load sharing between the IC engine and
the brushless DC motor.

Keywords GPS · RPi · Planetary gears · ICE · SoC

1 Introduction

The conventional automotive industry today is facing a major challenge of developing

a system with improved performance and higher fuel economy. This is necessary for
better acceptance among the masses, in the context of rising fuel prices and depleting
fuel resources. The increasing number of vehicles on the road has led to increasing
level of emissions. Lung cancer and breathing disorders are becoming common,
thereby affecting the health and economy of the country. The increasing level of air
quality index (AQI) in metro cities like capital Delhi, where the AQI was recorded
at 471, [1] is an indication for adoption of urgent measures for controlling and

J. Thomas (B) · A. Thomas · A. Biju · A. Mathew · C. P. Jose · K. M. Haneesh

Faculty of Engineering, CHRIST (Deemed to Be University), Bengaluru, India
e-mail: [email protected]

© Springer Nature Singapore Pte Ltd. 2020 515

G. Pradhan et al. (eds.), Advances in Electrical Control and Signal Systems,
Lecture Notes in Electrical Engineering 665,
https://doi.org/10.1007/978-981-15-5262-5_38
516 J. Thomas et al.

reducing the emissions. These high values of AQI recorded in major cities across the
country indicate how severe the situation is and how unhealthy the conditions are
for people. However, there exists a paradigm of maintaining profit and at the same
time ensuring the desirability of the vehicles. Compromising in cost would lead to
lower performance and vice versa. But today, the hybrid electric vehicle industry is
emerging with better controllers ensuring higher fuel economy and reduced emissions
and could prove to be a viable alternative when compared to conventional vehicles.
However, the process of integration of an electric motor drive into the conventional
power train is a challenge with respect to packaging the electrical and mechanical
subsystems and integrating it for control of vehicle dynamics.
Hybridization helps in downsizing the IC Engine for similar vehicle performances,
while delivering improved fuel economy. The hybrid control strategy primarily
aims at improving fuel economy and reducing level of emissions. But in real-time
implementation, the improvement of one objective might affect the other objective
adversely. Therefore, for implementing the control strategy proper weighing func-
tions are to be defined which provides priorities to the objectives. But the main issue
that occurs is the non-causal nature of the strategy. The driver’s future action cannot
be predicted as it varies based on driving conditions. The driver can accelerate at
any time, and the control strategy should withstand this parameter while defining
the objective function. Therefore, charge sustaining control strategy is also to be
considered. The control strategy would determine the operating points of an engine
or a motor based on efficiency and emission characteristics while maintaining the
battery SoC. The novel methods for Energy Management System wherein the trip
schedule-based intelligent control is devised are being researched [2, 3].
The series–parallel architecture makes use of a motor, a generator, and an IC
Engine to propel the vehicle. A power split is provided to enable both series and
parallel operations as per requirement of the system [4]. The implementation of the
series–parallel hybrid architecture is usually performed either by using a mechanical
coupling device or through an electronic coupling mechanism. In the mechanical
power split mode, the fuel flow rate is determined by the power required for the
propulsion. The power requirement corresponds to the accelerator pedal input. The
load torque requirement will be monitored and is compared with the vehicle opti-
mum performance range and the torque control mechanism is selected [5]. During
electronic control mode, the electrical generator is coupled to the IC engine through
a fixed gear mechanism. The motor is coupled directly to the wheels through another
fixed gear mechanism. The utilization of a planetary gear system in the series–par-
allel architecture provides more flexibility. It operates in series mode while in low
speed region, behaving like a parallel system while in high speed region. When the
vehicle climbs up an inclination where more torque is required, the motor acts up
as the main propeller and provides majority of the power through the battery and
generator. During EV mode, the engine is kept off. In HEV energy management,
two main approaches are utilized for optimizing the operation. It includes offline and
online optimization approaches. In offline optimization approach, the road profile is
determined beforehand and an optimized path is charted for operation while in online
condition the data is taken in real time and the operation optimized [6]. But online
A GPS-Gradient Mapped Database-Based Fuzzy Energy Management … 517

optimization will be complex and difficult as the power train cannot be so dynamic in
responding to the optimization profile designed by the Energy Management System
[7].
GPS-based Energy Management Systems are discussed in different literatures,
[8–10]; however, the utilization of altitude to create a driving profile is new. Egor
et al. presented a novel method of GPS-track data processing to obtain the tangential
and normal acceleration and the elevation angle data of a vehicle based on its travel
route. The paper presents a new algorithm that makes the GPS data acquisition
more accurate [11]. Liu et al. have prepared a controller design based on dynamic
Programming (DP) involving an optimization algorithm based on state of charge
[12].
The rest of the work is structured as follows: Sect. 2 describes the creation and
utilization of the database, while in Sect. 3, the detailed operation of the proposed
Energy Management System is described. Section 4 explains the designed series—
parallel power train in a MATLAB environment, while Sect. 5 discusses the results
and the prototype development.

2 GPS-Gradient Database

The creation of a GPS-gradient database follows an offline optimization pattern. A

database would be created where the latitude and longitude of the path to be taken
are mapped with the gradient of the road with minute samples. This would enable to
obtain a database using which the altitude of the road can be predetermined. When the
driver enters the destination in the GPS interface, fixed in the vehicle, the coordinates
of the starting point and the coordinates of the destination would be extracted and
searched in the database. The most optimum path would be decided to travel between
the source and the destination. The path charted would also contain the altitudes of
the path between the source and destination. Therefore, before the beginning of
the journey the Energy Management System would have the information regarding
the most optimum path and also about the gradient of the road. Based on the road
gradient feature the Energy Management System would decide upon managing of
power flow from the IC engine and the motor. The advantage of this approach is
that it prepares an offline optimized operating mode even before the journey and so
the Energy Management System could bring about changes as planned. The only
factor that would be effective is the road traffic conditions which would be viewed
as disturbances in the charted path.
The Questar model G702-001UB GPS unit transmits the data through the ESP12-
NodeMCU to the cloud platform. The algorithm for receiving the data in reference
with the co-ordinates and also the variation in retrieval of data is as described:
• The main header file for the entire process to be solved as codes are mentioned
and described.
518 J. Thomas et al.

• A function is declared to enable communication of the digital pins of the

microprocessor (software serial NSS (arguments)).
• A function declared to fetch the data from the microprocessor and also store it
safe in the cloud for the reference.
• The objects are created for the function, and also pin 13 of the microprocessor is
declared as output.
• Baud rate of the microprocessor is mentioned.
• Readings are taken, analysed, and stored.
The data thus extracted is also interfaced with MATLAB which is then used
to simulate the setup and analyse its operation using a MATLAB/Simulink-based
model. The control signals generated uses a fuzzy-based controller which regulates
the load sharing between the brushless DC Motor and the IC engine.

3 Energy Management System

With increase in altitude, there occurs variation in atmospheric pressure which causes
reduction of air density. This would affect the air/fuel ratio of the engine. Conse-
quently, there is an enrichment effect to the combustion mixture with an increase
in altitude. If an engine tuned at sea level is operated at high altitude, there will
be a reduction in power and fuel economy. Moreover, severe carbon monoxide and
hydrocarbons exhaust emissions are expected. Various tests conducted on a vehi-
cle equipped with a sea-level carburettor would experience some 6% enrichment in
air/fuel ratio upon driving from sea level to an altitude of 1200 m [13]. The enrichment
in air/fuel mixture at altitudes substantially increased the BSFC of the engine.
Figure 1 shows the proposed controller where the load sharing is performed
between the IC engine and the brushless motor. The series–parallel mode has both
the propulsion sources integrated to the transmission, thereby enabling load sharing
between them easily. However, based on the level of SoC of the traction battery, the
EMS could activate the operation of the alternator coupled to the IC engine, thereby
enabling charging of the traction battery along with the motion of the vehicle. The
proposed model is a pseudo-dynamic system where, when the driver feeds in the des-
tination in the vehicle GPS monitor, the controller charts the path and also gathers
the altitude in the path through the cloud database. The Questar model G702-001UB
GPS unit captures the GPS co-ordinates and maps it to the cloud database through
the ESP12-NodeMCU. The ESP connects through Wi-fi and sends the sensor data
received from its GPIO pins to the dedicated cloud server. The data is then stored
in a database like MySQL or Excel. These coordinates obtained are mapped in the
database already created to identify the expected load sharing between IC engine and
the traction motor based on the predictive mechanism. Once the load sharing is deter-
mined for the current terrain condition, the Raspberry Pi-based Energy Management
System regulates the throttle of the IC engine using a 20 kgf DC servo motor and the
A GPS-Gradient Mapped Database-Based Fuzzy Energy Management … 519

Fig. 1 Load sharing controller

traction motor using the motor controller. This enables the Energy Management Sys-
tem to predict the gradient of the road that is expected and thus enables a predictive
model, thereby making the system to have more dynamic response. The response
would only be hindered by unexpected traffic which can disrupt the expected load
sharing. This could be rectified by performing the modelling with respect to the
traffic conditions as well.
The fuzzy-based controller regulates the load sharing between the brushless DC
motor and the IC engine. Figure 2 shows the implemented fuzzy logic controller. The
live data which is collected using the GPS unit is interfaced with MATLAB and the
altitude is loaded to the workspace and the signal is built on which the controller has
520 J. Thomas et al.

Fig. 2 Fuzzy controller for load sharing between motor and IC engine

to operate. The rules are defined such that the load sharing is done based on specific
variation in altitude. The controller analyses the current sample altitude obtained
from the database with the altitude at the previous instant thereby, obtaining a slope
which is then given to the fuzzy controller to decide upon the load sharing percentage.
The data obtained from the database is then given as the input to the fuzzy con-
troller as shown in Fig. 3. The road grade profile shows that the altitude is varying
with time.

4 Hybrid Electric Power Train Regulation

The proposed system is implemented in series—parallel hybrid electric vehicle. The

system has a brushless DC motor, an IC engine, and a generator. The specifications are
shown in Table 1. The model developed considers only longitudinal dynamics. The
indirect coupling effects due to vertical and lateral motions are neglected. Simplified
versions of the transient characteristics of the power train actuators are represented.
This is important because the actuators’ internal controls which have a considerable
impact on the transient behaviours of the actuators are not accurately known. The
torsional flexibilities of all the shafts and gears are considered to be very high. The
frictional properties due to the road surface are assumed to be acting uniformly on all
tires. The drivetrain losses are represented by lumped efficiency and friction models.
Frictional losses of driveline components due to effects such as gear meshing and
A GPS-Gradient Mapped Database-Based Fuzzy Energy Management … 521

Fig. 3 Altitude-based control signal generated from the fuzzy controller for load sharing between
motor and IC engine

Table 1 Simulation
Parameter Specifications
specifications
IC engine 57 kW, 5000 rpm
Brushless DC motor 50 kW
Generator 30 kW
Gear system Planetary 4 gear system
Battery pack 200 V, 6.5 Ah

bearing friction are not modelled individually. The impacts of environmental factors
such as temperature are not taken into consideration in the component models.
The fuzzy controller signal is transferred to the Energy Management System. This
system performs the control operation which also includes load sharing as well as
battery recharging action. The power drive train is divided into two sections which are
the electric drive train and the IC engine drive train. The electrical drive train consists
of the generator and the brushless motor. The Energy Management System based
on the information from the fuzzy controller decides the share of generator, motor,
and the IC engine. Torque actuators are used to monitor the power transferred by the
electric and the IC engine drive train. Based on the altitude, the Energy Management
System develops the reference torque which has to be maintained by the generator
and the motor. In the case of the IC Engine, the requirement is in terms of the throttle
control of the IC engine. The operation modes depend upon various factors including
522 J. Thomas et al.

the SoC of the battery pack, the altitude gradient, and the torque requirement. The
produced power is fed to the wheels through a planetary gear arrangement. In the
proposed system, a torque coupling mechanism is used and so the torque from the IC
engine and the electric power train can be independently controlled, but the speeds
are dependent. To monitor the speed of the motor, generator and the IC engine speed
sensors are used at the coupling point of these units with the drive shaft. The wheels
are modelled based on the nonlinear tire model proposed by Pacejka [6]. This model
uses the semi-empirical method known as the “Magic Formula” to compute the
tractive forces (F x,i ) generated by the tires:

Fx,i = μwh Fz,i sin(Cwh,1 arctan(Cwh,2 K i − Cwh,2 K i − arctan(Cwh,2 K i )))

where x denotes the linear force at an instant i.

The vertical forces acting on the front and rear wheels (F z,f and F z,r ) are a factor
of the vehicle acceleration, and the resulting longitudinal load transfer is given as
 
Mveh g aveh
Fz, f = 1 + K veh,1 + K veh,2
4 g
 
Mveh g aveh
Fz,r = 1 − K veh,1 − K veh,2
4 g

It is found that the transient tire slip, κ i , has an effect on the longitudinal tire force.
The vehicle dynamics includes detailed modelling of the wheels based on the pro-
posed equations. For the integration of the two different propulsion sources, planetary
gear system is used with the IC engine coupled to the sun gear, while the traction
motor is coupled to the ring gear. The planets are coupled to the transmission through
a connecting rod. This provision enables to lock the sun or the ring gear operation
as per the performance requirement (Fig. 4).

5 Controller Evaluation and System Modelling

The control of the hybrid electric vehicle is on two levels. The initial level includes
determination of the altitude of the road based on which the fuzzy-based controller
decides upon the load sharing. Figure 5 shows the fuzzy interface depicting the
control signals generated for load sharing. These control signals are then given to
the Energy Management System where the second level of control is performed in
terms of torque coupling and load sharing between the electric and IC engine power
trains.
The vehicle when goes upon a road with a gradient which keeps on varying whose
response of the system is shown in Fig. 6. The simulation represents the starting of
the vehicle at a gradient, and so the initial torque required is very high. The battery
pack is also checked for its SoC. Based on the load sharing control signal from the
A GPS-Gradient Mapped Database-Based Fuzzy Energy Management … 523

Fig. 4 Series—parallel hybrid power train

Fig. 5 Fuzzy Interface depicting the control signals generated for the load sharing

fuzzy controller, the Energy Management System shares the load between the motor
and the IC engine. But considering the level of SoC, the IC engine either powers the
generator to recharge the battery pack along with providing the propulsion power or
remains off. The characteristics in the figure therefore clearly show the load sharing
524 J. Thomas et al.

Fig. 6 Load sharing on a gradient

between the motor and the IC engine. It also shows an initial period where the
contribution for IC engine is zero. This is because the IC engine during this period is
not contributing to the propulsion action but can be used to operate the generator to
recharge the battery pack if necessary. The IC engine would not be operated during
an initial region as the majority of the torque requirement is met by the traction motor.
When the torque requirement is more than a prescribed load torque requirement, T r ,
chosen as 5 Nm for the prototype, the IC Engine operates meeting the remaining
torque requirement. If the SoC of the traction battery is noted to be below 40% SoC,
the IC engine charges the battery while also providing the propulsion power, thereby
turning off the traction motor. Once the battery pack retains its stable SoC, the IC
engine also helps in the vehicle propulsion based on the load sharing calculated by
the controller.
Figure 7 shows the situation when the vehicle is cruising on a level road where
there is no gradient. The initial torque requirement is met by the traction motor itself,
and when the load torque requirement T r goes beyond the prescribed 5 Nm, the IC
engine starts operating. Similar to the previous mode of operation during the period
from start to 5 Nm torque requirement, the IC engine could either remain inactive
or based on the traction battery SoC level could engage the generator through the
planetary gear arrangement to charge the traction battery. Based on the level of the
traction battery’s SoC, the load sharing is performed. At simulation time of 2 s, it
can be seen from the load sharing curve that the IC engine attached sun gear lock

Fig. 7 Load sharing on a level road—cruising mode

A GPS-Gradient Mapped Database-Based Fuzzy Energy Management … 525

is disengaged and the IC engine starts to meet the remaining torque requirement for
the vehicle propulsion
The net propulsion power provided by the motor and the IC engine for a road
with a gradient is shown in Fig. 8. The net power is sufficient for meeting the tractive
power required for the vehicle propulsion considering the various resistive forces
acting on the vehicle while moving up a gradient.
The Energy Management System was implemented and tested in a Raspberrry Pi-
based Controller. Figure 9 shows the implemented system. In the prototype model,
the two power sources were made to power the front-end and rear-end wheels. The
Questar model G702-001UB GPS unit transmits the altitude data through the ESP12-
NodeMCU which is then transferred to the RPi. The RPi calculates the slope between
consecutive samples of altitude and then regulates the load between the two power
sources based on the algorithm developed. The prototype development would be
followed by the hardware implementation of the system using a 150 cc 13 kW
single-cylinder IC engine and a 19 kW three-phase induction motor as the traction

Fig. 8 Net propulsion power together from motor and IC engine

Fig. 9 Implemented prototype model

526 J. Thomas et al.

motor. The traction battery used would be a 48 V, 200 Ah battery pack. The developed
model would be an easy conversion model for any IC engine vehicle into a plug-in
hybrid electric vehicle.

6 Conclusion

The paper introduces a novel Energy Management System where the load is shared
between the rear and front power sources based on the gradient of the road. It is
an attempt to develop a pseudo-predictive Energy Management System where the
information regarding the altitude at each location is made available to the con-
troller prior to the beginning of the journey so as to be prepared for the load sharing.
The GPS-based data logging and RPi-based regulation of the load enable to make
the system more optimized in its operation. The fuzzy-based controller enables a
smooth transition of control signals based on the altitude samples, thereby facilitat-
ing the vehicle to smoothly transition during dynamic changes. The series–parallel
hybrid also has a generator unit which has been effectively utilized by the Energy
Management System to charge the battery pack based on the SoC level.

Acknowledgements We are thankful to E-Mobility Research & Development Center at Faculty

of Engineering, CHRIST (Deemed to be University), Bengaluru, for the facilities provided in the
development of this project.

References

1. http://aqicn.org/map/india/
2. Gong, Q., Li, Y., Peng, Z.: Optimal power management of plug-in HEV with intelligent
transportation system. In: Proceeding of IEEE/ASME International Conference on Advanced
Intelligent Mechatronics, pp. 1–6 (2007)
3. Gong, Q., Li, Y., Peng, Z.: Trip-based optimal power management of plug-in hybrid electric
vehicles. IEEE Trans. Veh. Technol. 57(6), 3393–3401 (2008)
4. Dai, Y., Luo, Y., Li, K.: Longitudinal and lateral coordinated motion control of four-wheel-
independent drive electric vehicles. In: 2013 World Electric Vehicle Symposium and Exhibition
(EVS27), pp. 1–9 (2013). https://doi.org/10.1109/evs.2013.6914736
5. Veillette, R.J.: Course Lecture Notes on Design of Hybird Electric Vehicles (2005)
6. He, J.-H., Yang, L., Qiang, J.-X., Chen, Z.-Q., Zhu, J.-X.: Novel flexible hybrid electric system
and adaptive online-optimal energy management controller for plug-in hybrid electric vehicles.
J. Central South Univ. 19(4), 962–973 (2012)
7. Ramadan, H.S., Becherif, M., Cluade, F.: Energy management improvement of hybrid electric
vehicle via combined GPS/Rule based methodology. IEEE Trans. Autom. Sci. Eng. 14(2),
(2017)
8. Kulik, E, Tran, X.T., Anuchin, A., Vagapov, Y.: GPS-track data processing for the optimization
of the powertrain for hybrid electric vehicles. In: 58th International Scientific Conference on
Power and Electrical Engineering of Riga Technical University (RTUCON), (2017)
A GPS-Gradient Mapped Database-Based Fuzzy Energy Management … 527

9. Wallace, B., Rockwood, M., Goubran, R., Knoefel, F., Marshall, S., Porter, M.: Measurement
of vehicle acceleration in studies of older drivers from GPS position and OBDII velocity
sensors. In: 2015 IEEE International Symposium on Medical Measurements and Applications
(MeMeA) Proceedings, pp. 391–396 (2015)
10. Moshirvaziri, M., Malherbe, C., Moshirvaziri, A., Trescases, O.: Power-mix optimization for a
hybrid ultracapacitor/battery pack in an electric vehicle using real-time GPS data. In: IECON
2013—39th Annual Conference of the IEEE Industrial Electronics Society, pp. 4666–4671
(2013). https://doi.org/10.1109/iecon.2013.6699888
11. Kulik, E, Tran, X.T., Anuchin, A.: Estimation of the requirements for hybrid electric powertrain
based on analysis of vehicle trajectory using GPS and accelerometer data. In: 25th International
Workshop on Electric Drives: Optimization in Control of Electric Drives (IWED), Moscow,
Russia. Jan 31–Feb 02 (2018)
12. Liu, B., Li, L., Wang, X., Cheng, S.: Hybrid electric vehicle downshifting strategy based
on stochastic dynamic programming during regenerative braking process. IEEE Trans. Veh.
Technol. 75(8), 1091–1108 (2008)
13. He, J.-H., Yang, L., Qiang, J.-X., Chen, Z.-Q., Zhu, J.-X.: Novel flexible hybrid electric system
and adaptive online-optimal energy management controller for plug-in hybrid electric vehicles.
J. Central South Univ. 19(4), 962–997 (2012)
Optimal Placement of PMU: A Brief
Review

Sandeep Kumar Mishra and Murthy Cherukuri

Abstract For monitoring and controlling the large modern power system, a mea-
suring device such as phasor measurement unit (PMU) provides an important role.
APMU provides time-synchronized voltage and current phasor values relevant buses
where these PMUs are placed. As the PMUs and communication devices are expen-
sive, it is not possible economically as well as practically to install PMU on each bus
to measure the states. This paper describes the advantages of PMU inclusion on the
power network and discusses various optimum positioning methods which used to
determine the minimum amount of PMUs to obtain full power system observability.

Keywords Optimal placement of PMU · Power system observability

1 Introduction

Electric energy is being the most basic raw material for human beings in most activ-
ities. By increasing its role and demand in country’s economies, it must be delivered
with security and high reliability. Traditionally, the monitoring and controlling have
been done by supervisory control and data acquisition (SCADA) system [1]. In cur-
rent day’s power system security monitoring, protection and control are managed
by synchronized phasor measurement system that gives notable improvements in
system. PMU is able to measure voltage and branch current of the bus in terms of
phasor. Global Positioning Satellite (GPS) provides a common time synchronization
between PMUs [2]. Reference [3] performed a brief review on optimal placement
of PMU, applications of PMU and reliability analysis of PMU up to the year 2015.
This paper is targeted to review the papers from the year 2015 till date on optimal

S. K. Mishra
Electrical Engineering, National Institute of Science and Technology, Berhampur, India
e-mail: [email protected]
M. Cherukuri (B)
Electrical and Electronics Engineering, National Institute of Science and Technology,
Berhampur, Odisha, India
e-mail: [email protected]

© Springer Nature Singapore Pte Ltd. 2020 529

G. Pradhan et al. (eds.), Advances in Electrical Control and Signal Systems,
Lecture Notes in Electrical Engineering 665,
https://doi.org/10.1007/978-981-15-5262-5_39
530 S. K. Mishra and M. Cherukuri

placement of PMU. The following is the structure of this paper. Section 2 gives a
brief description about PMU. Section 3 speaks about power system observability.
Section 4 presents different optimization algorithms used in the literature for the opti-
mal placement of PMU. Section 5 describes the future scope of this work. Finally,
conclusions are given in Sect. 6.

2 Phasor Measurement Unit

PMU is also known as “synchronized phasor measurement unit,” which is a measur-

ing device for monitoring the transmission and distribution network status in modern
power system. PMU provides time-synchronized data up to 50 samples per second
[4]. The mismatch between investment economic justification and calculation accu-
racy resulted in a new design problem called optimum PMU placement (OPP) [5]
(Fig. 1).

Fig. 1 Phasor measurement unit system

Optimal Placement of PMU: A Brief Review 531

3 Power System Observability

Observability of the system refers to the fact that measurement sets and their distri-
bution are adequate to solve the current condition of the network. Whatever approach
is used after placing a new PMU, the power system’s observability must be checked.
There are two different approaches for the analysis; these are as follows:
(a) Topological observability
(b) Numerical observability
(a) Topology Observability: This uses the concept of decoupled calculation and the
principle of graphs where judgment is predicted on logical operations. Therefore,
it solely needs information regarding network connectivity, measuring types and
their locations to make full rank spanning tree with the present measurement set, for
system observability.
(b) Numerical Observability: If the information about all the buses of the system
is known, then the system is said to be fully observable.

4 Review on Optimal PMU Placement Algorithms

The main objective of OPP problem is to find the minimum number of PMUs as
well as their positions to make the power network topologically observable. Many
approaches have been proposed as an optimization problem to solve the OPP problem.
Some methods are as follows:

4.1 Simulated Annealing(SA)

In [6], the author proposed a customized simulated annealing with customized

exhaustive search method to reduce the time solving significant without affecting
the optimal solution, which can improve by increasing in the system scale. Refer-
ence [7] proposed simulated annealing (SA) technique for estimating the optimum
location of PMU.

4.2 Genetic Algorithm (GA)

In [8], author modeled genetic algorithm (GA) with zero-injection constraints which
is calculated with help of bus observability index for minimizing the number of
locations without losing total observability. In [9], genetic algorithm (GA) is for
optimizing the PMU placement for full system observability which applied on Alge-
rian network by considering with ZIB where it gives quick ideal arrangement and
532 S. K. Mishra and M. Cherukuri

finding the ideal number of PMUs for complete observability assessment of power
system and also has reliable process. In [10], genetic algorithm (GA) is for solving
the PMU placement problem to detect reverse power flow which has the ability to
find the optimal solution at the same time maintaining good convergence speed and
simplicity of implementation. In [11], multi-objective genetic algorithm (MOGA)
is proposed with multi-Prony analysis for solving the PMU placement problem by
minimizing the number of PMUs and maximizing the signal-to-noise ratio (SNR)
which is based on (n − 1) observability constant.

4.3 Particle Swarm Optimization(PSO)

In [12], binary particle swarm optimization (BPSO) has been used for finding the
optimum location of PMUs according to observability topology method and maxi-
mizing the efficiency of measurement in system buses. In [13], the efficiency and the
robustness of the binary particle swarm optimization (BPSO) method help to cover
the full system observability with minimum number of PMUs at minimum installa-
tion costs. In [14], the author uses binary particle swarm optimization (BPSO) by
considering the power flow measurement to solve the OPP problem and number of
PMU required to maintain the full system observation. In [15], the author introduces
improved binary particle swarm optimization (IBPSO) to achieve rapid convergence
without failing the full observability of the power system by increasing the flying par-
ticle search space in order to obtain more feasible solutions during the initialization
process; it also considers zero-injection bus. In [16], the author introduced exponen-
tial binary particle swarm optimization (EBPSO) algorithm to solve the placement
problem for a complete observable network. First, to increase the search capability
of the algorithm, adaptive exponentially decaying inertia weight was used. Secondly,
a sigmoid function was used to obtain a discrete binary search space to update the
positions. Finally, a fresh solution between search locations from single mutation
operator to obtain.

4.4 Integer Linear Programming (ILP)

In [17], the author used novel-based linear integer programming (ILP) to solve prob-
lem about placing of PMUs by considering costs of PMU installations and communi-
cation channels, and observability of system provides by considering zero-injection
buses. In [18], by placing minimum PMUs, the authors could achieve full system
observability using integer programming (IP). In [19], binary integer linear program-
ming(BILP) is used to define the requirement of minimum numbers of PMUs and
their placement in a grid to obtain full network observability in normal and for possi-
ble disturbances [n − 1] security mode. In [20], PMUs were placed based on binary
integer linear programming (BILP) to run for the system intact case and voltage
Optimal Placement of PMU: A Brief Review 533

stability-based cases. In [21] For islanding and intact condition, mixed integer linear
programming (MILP) is used to maximize the redundancy for OPP and minimize
the number of PMUs for system observability. In [22], mixed integer linear pro-
gramming is used for OPP problem. In [23], integer linear programming (ILP) with
pseudo-PMU is used to find optimal location of PMUs.

4.5 State Estimation

In [24], the author has used state estimator (SE) for finding the PMU placement in
power system to fulfill the observability of the power system and also used UPFC &
STATCOM to improve in the voltage and system stability for a multi-bus system.

4.6 Graph Theory

In [25], graph theory is used to solve the OPP problem by converting the distribution
feeder into a graph model where different terms are used to find the minimum number
of PMU installation locations for complete observation.

4.7 Artificial Neural Networks

In [26], artificial neural network analysis (ANNs) tool has been used to calculate the
power system sensitivity factor for OPP problem to ensure the complete topological
observability of the system. In [27], PMU and artificial neural network (PMU-ANN)
are used for voltage stability monitoring in real time which uses the feedforward
neural network (FFNN) algorithm.

4.8 Sine Cosine Algorithm

In [28], sine cosine algorithm (SCA) is used for optimizing the placement of the
PMUs for power system visibility, while minimizing the number of PMUs placed in
power grid.
534 S. K. Mishra and M. Cherukuri

4.9 Binary Bat Algorithm

In [29], binary bat algorithm (BBA) is used to give a solution for OPP problem to
obtain full observable system by considering single PMU outage condition, and the
author also correlated this algorithm with few other algorithms for time execution
and convergence curve, where this algorithm has a higher convergence curve and a
lower execution time than several algorithms.

4.10 Exhaustive Search

In [30], the author has used reduced exhaustive search (RES) to reduce the com-
putational work to reduce the observability of the system. It also sees the PMU
outages.
Table 1 gives the detailed summary of the optimization techniques used so far in
the literature, and different factors have been taken into consideration to compare
and analyze them (Fig. 2 and Table 2).

5 Future Scope

Subsequent research efforts need to address a number of new problems. The possible
areas of research are as follows:
(a) It may so happen that one may get multiple optimal locations of PMU, but the
optimal number of PMUs is equal. Thus, there is a scope of doing research to
determine the best optimal solution based on the available optimal solutions.
(b) In order to bring a more pragmatic view of OPP of PMUs, it is desirable to
consider the failure of communication channels, environmental conditions and
so on.

6 Conclusion

Different methods are used to achieve solutions for optimum PMU position. Finding
the best position of PMUs in the power network requires the right technique. It can
be inferred that GA and PSO provide optimal placement solutions but do not lend
themselves mathematically in terms of calculating contingencies, especially for large
power systems that take time for computational research. In terms of phasing and
modeling, the computational framework of ILPs becomes easier to understand for a
large power network.
Table 1 Summary of optimization techniques used in the literature
S. Technique Software used System Effect Effect of Reliability Formulation characteristics Year References
No. considered of communication consideration
ZIB network (Y/N) (Y/N)
(Y/N)
1 Simulated Customized MATLAB IEEE 13, No No No • OPP problem 2016 [6]
annealing simulated annealing 34, 37 • System observability
(SA) buses
Simulated annealing MATLAB (NRIPG), No Yes No • Complete observability in 2016 [7]
(SRIPG), India power grid
(ERIPG),
(WRIPG),
(NERIPG)
2 Genetic Genetic algorithm MATLAB IEEE 14, Yes No Yes • Observability of power system 2018 [8]
algorithm 30, 57 bus during abnormal conditions
(GA) Genetic algorithm MATLAB IEEE 30, Yes No No • OPP problems 2018 [9]
Optimal Placement of PMU: A Brief Review

57, 118
bus
Genetic algorithm MATLAB IEEE-14 No No No • Reverse power flow detection 2018 [10]
and 39 bus
Multi-objective PSAT toolbox IEEE 39 Yes No No • Optimal PMUs placement 2018 [11]
genetic algorithm bus
(MOGA)
3 Partial swarm Binary partial swarm MATLAB IEEE 14, Yes No No • Achieve monitoring task of 2018 [12]
optimization optimization (BPSO) 24, 30, 39 power system network
(PSO) bus • Multi-objective OPP
Binary partial swarm MATLAB IEEE 14 Yes No No • OPP problem 2017 [13]
optimization (BPSO) bus • System observability
Binary partial swarm MATLAB IEEE 57 Yes No No • System observability 2017 [14]
optimization (BPSO) bus • OPP problem
(continued)
535
Table 1 (continued)
536

S. Technique Software used System Effect Effect of Reliability Formulation characteristics Year References
No. considered of communication consideration
ZIB network (Y/N) (Y/N)
(Y/N)
Improved binary MATLAB IEEE-30 Yes No No • OPP problem 2015 [15]
particle swarm bus • System observability
optimization
(IBPSO)
Exponential binary MATLAB IEEE 14, Yes No No • OPP problem 2015 [16]
particle swarm 30, 39 bus • System observability
optimization
(EBPSO)
4 Integer linear Novel integer linear MATLAB IEEE 30 Yes NO NO • OPP problems 2018 [17]
programming program (NILP) bus
(ILP) Integer linear MATLAB/MAT IEEE 14 Yes Yes No • Contingencies-based optimal 2017 [18]
programming (ILP) Power bus, PMU placement,
England
39 bus
Binary integer linear MATLAB IEEE 14, Yes No No • System observability 2017 [19]
programming (BILP) 24, 39,
110 bus
Binary integer linear PSAT Toolbox IEEE Yes No No • Voltage stability 2017 [20]
programming (BILP) 14-bus and
246-bus
Northern
Regional
Mixed integer linear MATPOWER IEEE 14, YES No No • Islanding 2016 [21]
programming(MILP) 30,118 and • Maximum redundancy
New • Contingency
England
39 bus
(continued)
S. K. Mishra and M. Cherukuri
Table 1 (continued)
S. Technique Software used System Effect Effect of Reliability Formulation characteristics Year References
No. considered of communication consideration
ZIB network (Y/N) (Y/N)
(Y/N)
Mixed integer linear MATLAB OPTI IEEE 14, Yes No No • Full system observability, 2016 [22]
programming toolbox 23, 30, 57, • location problem
(MILP) 118, 300
bus
Integer linear MATLAB IEEE 24 No Yes No • OPP problem 2016 [23]
programming (ILP) bus • System observability
5 State State estimation MATLAB IEEE 14 No No No • Voltage stability and power 2016 [24]
estimator bus system stability and control
(SE)
6 Graph theory Graph theory MATPOWER IEEE7, 1, No No No • Full system observability 2017 [25]
(GT) 30 and
Optimal Placement of PMU: A Brief Review

New Eng l
and 39
7 Artificial Artificial neural MATLAB IEEE 14 No No No • Security monitoring 2017 [26]
neural network (ANN) Bus
network
(ANN)
8 Sine cosine Sine cosine MATLAB IEEE 9, Yes No No • OPP problem 2018 [28]
algorithm algorithm (SCA) 14, 30 bus
(SCA)
9 Bat algorithm Binary bat algorithm MATLAB IEEE 14, No No No • System observability 2017 [29]
(BA) 23, 30, 57, • Optimal placement
118 bus
10 Exhaustive Reduced exhaustive MATLAB IEEE test Yes No No • OPP problem 2016 [30]
search (ES) search (RES) 9, 14, 24, • PMU outage
30 and 39
bus
537
538 S. K. Mishra and M. Cherukuri

8
7
ILP, 7
6
5
PSO, 5 Optimisation
4 Techniques
GE, 4 (2015-2019)
3
GT, 3
2
SA, 2 ANN, 2 BA, 1
1
SE, 1 SCA, 1 ES, 1
0
0 2 4 6 8 10 12

Fig. 2 Optimization techniques with no. of papers published (2015–2019)

Table 2 Advantages and disadvantages of purpose techniques

S. No Methods name Advantage Disadvantage
1 Genetic algorithm (GA) It provides the best Pareto Long execution time
optimal solution rather
than single solution
2 Simulated annealing (SA) This method provides full It may take too much time
observability of the power to find solutions. That is
system along with why a lot of effort has
valuable dynamic data been made to reduce the
measurement from the time to calculate
power systems at the same
time
3 Particle swarm Easy to implementation, Higher computing timing
optimization (PSO) having more efficiency and this also increases
control parameters and less with increasing the size of
parameter for adjusting the solution
4 Artificial neural network Provides multiple Long complexity of
(ANN) solutions based on network structure
computational models
5 Integer linear Computationally fast Constraints are nonlinear
programming (ILP) method, adaptable

References

1. Sefid, M., Rihan, M.: Optimal PMU placement in a smart grid: an updated review. Int. J. Smart
Grid Clean Energy 8(1), 59–69 (2019)
2. Jamuna, K., Selva, Y.L., Gnanasekaran, G.: Placement of PMUs for power system observability
using biogeography based optimization. In: Proceedings - 2013 3rd International Conference
on Advances in Computing and Communications, ICACC 2013, pp. 355–358 (2013)
Optimal Placement of PMU: A Brief Review 539

3. Mohanta, D.K., Murthy, C., Sinha Roy, D.: A brief review of phasor measurement units as
sensors for smart grid. Electr. Power Components Syst. 44(4), 411–425 (2016)
4. Zhong, J.: Phasor measurement unit (PMU) placement optimisation in power transmission
network based on hybrid approach (2012)
5. Chatterjee, S., Saha Roy, B.K., Ghosh, P.K.: Optimal placement of PMU considering practi-
cal costs in wide area network. In: 2017 14th IEEE India Council International Conference,
INDICON 2017 (2018)
6. Chen, X., Chen, T., Tseng, K.J., Sun, Y., Amaratunga, G.: Customized optimal µPMU
placement method for distribution networks. In: Asia-Pacific Power and Energy Engineering
Conference, APPEEC, vol. 2016-December, pp. 135–140 (2016)
7. Mallikarjuna, B., Gopakumar, P., Reddy, M.J.B., Mohanta, D.K.: A case study on optimal
phasor measurement unit placement for emerging Indian national smart grid. In: International
Conference on Signal Processing, Communication, Power and Embedded System, SCOPES
2016 - Proceedings, pp. 1956–1960 (2017)
8. Chandana Sushma, G., Jyothsna, T.R.: Observability constrained GA approach for optimal
PMU placement considering zero injection modeling. In: 2018 International Conference on
Control, Power, Communication and Computing Technologies, ICCPCCT 2018, pp. 174–178
(2018)
9. Laouamer, M., Mohammedi, R.D., Kouzou, A., Tlemcani, A.: Optimal placement of PMUs in
Algerian network using genetic algorithm. In: 2018 15th International Multi-Conference on
Systems, Signals and Devices, SSD 2018, pp. 947–951 (2018)
10. Al Rammal, Z., Daher, N.A., Kanaan, H., Mougharbel, I., Saad, M.: Optimal PMU placement
for reverse power flow detection. In: 2018 4th International Conference on Renewable Energies
for Developing Countries, REDEC 2018 (2018)
11. Tarif, T., Ladjici, A.A., Chabane, Y.: Optimal PMU placement for small-signal stability assess-
ment using Genetic algorithm. In: Proceedings of 2018 3rd International Conference on
Electrical Sciences and Technologies in Maghreb, CISTEM 2018 (2019)
12. Abdelkader Azzeddine, L., Ridh Djamel, M., Abdellah, K., Mohamed Mounir, R.: Optimal
PMU placement in power system based on multi-objective particle swarm optimization. In:
2018 15th International Multi-Conference on Systems, Signals and Devices, SSD 2018, p. 941
(2018)
13. Kumari, S., Walde, P., Iqbal, A., Tyagi, A.: Optimal phasor measuring unit placement by binary
particle swarm. (2017). In: 2017 8th International Conference on Computing, Communication
and Networking Technologies (ICCCNT) https://doi.org/10.1109/icccnt.2017.8204051
14. Rahman, N.H.A., Awalin, L.J.: Strategic PMUs placement considering power flow measure-
ments using binary particle swarm optimization method. In: 2017 International Conference
on Engineering Technology and Technopreneurship, ICE2T 2017, vol. 2017-January, pp. 1–3
(2017)
15. Rahman, N.H.A., Zobaa, A.F., Theodoridis, M.: Improved BPSO for optimal PMU placement.
In: Proceedings of the Universities Power Engineering Conference, vol. 2015-November (2015)
16. Maji, T.K., Acharjee, P.: Multiple solutions of optimal PMU placement using exponential
binary PSO algorithm for smart grid applications. IEEE Trans. Ind. Appl. 53(3), 2550–2559
(2017)
17. Li, Y., Li, J., Wu, L.: A Novel Integer Linear Programming Based Optimal PMU. Placement
Model. North American Power Symposium(NAPS) (2018). https://doi.org/10.1109/naps.2018.
8600551
18. El Hosainy, M., Seddik, K., Elezabi, A.: Fault-tolerant PMU placement using algebraic connec-
tivity of graphs. In: 2017 IEEE PES Innovative Smart Grid Technologies Conference Europe,
ISGT-Europe 2017—Proceedings, vol. 2018-January, pp. 1–6 (2018)
19. Isakov, R., Davletbaev, A., Suleymanova, N.: Observability of electric power networks factoring
in (n-1) security. In: 2017 International Conference on Industrial Engineering, Applications
and Manufacturing, ICIEAM 2017—Proceedings (20170
20. Sahu, P., Verma, M.K.: Optimal placement of PMUs in power system network for voltage
stability estimation under contingencies. In: 2017 6th International Conference on Computer
540 S. K. Mishra and M. Cherukuri

Applications in Electrical Engineering—Recent Advances, CERA 2017, vol. 2018-January,

pp. 365–370 (2018)
21. Singh, S.P., Singh, S.P.: Islanding based optimal placement of phasor measurement unit using
MILP. In: 2016 IEEE 7th Power India InternationalConference (PIICON) (2016). https://doi.
org/10.1109/poweri.2016.8077289
22. Khokhlov, M.V., Obushevs, A., Oleinikova, I., Mutule, A.: Optimal PMU placement for topo-
logical observability of power system: Robust measurement design in the space of phasor
variables. In: IEEE PES Innovative Smart Grid Technologies Conference Europe (2017)
23. Shukla, D., Singh, S.P., Singh, S.P.: Pseudo PMU for quasi-static analysis of power system. In:
2016 IEEE Annual India Conference, INDICON 2016 (2017)
24. Kiran, R., Lakshmikantha, B.R., Parimala, R.V.: A novel software model for enhancement of
system performance and security through an optimal placement of PMU & FACTS. In: 2016
IEEE Students’ Conference on Electrical, Electronics and Computer Science, SCEECS (2016)
25. Ghosh, P.K. Tahabilder, A.: Optimal PMU placement for complete system observability and
fault observability using graph theory. In 2017 International Electrical Engineering Congress,
IEECON 2017 (2017)
26. Kiran, R., Lakshmikantha, B.R., Parimala, R.V.: Optimal placement of PMUs and ana-
lytics on PMU data using ANN technique. In: 2017 International Conference on Energy,
Communication, Data Analytics and Soft Computing, ICECDS 2017, pp. 922–926 (2018)
27. Shah, H., Verma, K.: PMU-ANN based approach for real time voltage stability monitoring. In:
2016 IEEE 6th International Conference on Power Systems, ICPS 2016 (2016)
28. Laouamer, M., Kouzou, A., Mohammedi, R.D., Tlemcani, A.: Optimal PMU placement in
power grid using sine cosine algorithm. In: Proceedings of the 2018 International Conference
on Applied Smart Systems, ICASS 2018 (20190
29. Tawfik, A.S., Abdallah, E.N., Youssef, K.H.: Optimal placement of phasor measurement units
using binary bat algorithm. In: 2017 19th International Middle-East Power System Conferences
MEPCON 2017—Proceedings, vol. 2018-February, no. December, pp. 559–564 (2018)
30. Mabaning, A.A.G., Orillaza, J.R.C.: Complete solution of optimal PMU placement using
reduced exhaustive search. In: IEEE Region 10 Annual International Conference, Proceed-
ings/TENCON, pp. 823–826 (2017)
Design and Analysis of a PV-Based
Cascaded H-Bridge Seven-Level Inverter

Parul Gaur, Yajvender Pal Verma, and Preeti Singh

Abstract With the advancement of technology, multilevel inverters are preferred

over the two-level inverters due to its numerous advantages such as minimized total
harmonic distortion, better sinusoidal output voltage and lesser electromagnetic inter-
ferences. High-power industrial applications also make use of multilevel inverters
for obtaining better output. Two-level inverters are completely obsolete nowadays.
Keeping in view this, the focus is on to design, simulate and analyse a seven-level
inverter in this research work. Also, renewable energy sources such as solar, biomass,
hydro and wind can be naturally replenished during a life cycle. Uses of these renew-
able energy sources are more beneficial. Here, in this research work, photovoltaic
source is being used as the input to the cascaded H-bridge seven-level inverter rather
than direct current source. The designed seven-level inverter results in lesser har-
monics as well as lesser total harmonic distortion and is suitable for induction motor
drive applications.

Keywords Multilevel inverter (MLI) · Cascaded H-bridge (CHB) · Renewable

energy sources (RESs) · Photovoltaic (PV) · Total harmonic distortion (THD) ·
Harmonics

1 Introduction

Nowadays, multilevel inverters (MLIs) are emerging as strong power electronics

devices in the high-power industrial world as well as for commercial purposes. MLIs
have completely replaced the conventional two-level inverters, which was initially
used in the industries [1]. Conventional two-level inverters result in lesser sinusoidal

P. Gaur (B) · Y. P. Verma · P. Singh

UIET, PU, Chandigarh, India
e-mail: [email protected]
Y. P. Verma
e-mail: [email protected]
P. Singh
e-mail: [email protected]
© Springer Nature Singapore Pte Ltd. 2020 541
G. Pradhan et al. (eds.), Advances in Electrical Control and Signal Systems,
Lecture Notes in Electrical Engineering 665,
https://doi.org/10.1007/978-981-15-5262-5_40
542 P. Gaur et al.

output voltage with more harmonics. The concept of MLI has been introduced since
1975 as an alternative in high-power industrial applications [2]. As the number of
levels increases, a chance of getting more sinusoidal output voltage in terms of
levels also increases. Therefore, MLIs are the area of interest for the researchers
working in the power electronics domain. MLIs offer numerous advantages over the
conventional two-level inverters [3, 4]. The main advantage is the reduction in the
harmonics and overall total harmonic distortion (THD). THD is a measurement of
harmonic distortion present in a signal and is termed as ratio of the sum of powers of
all harmonic components of a signal to the power of the fundamental frequency. As
per the IEEE 519 standard, THD of the output voltage of a signal should be less than
5% [5]. The main motive in the designing of a MLI is to obtain minimum THD. With
increase in the number of levels of a MLI, THD reduces but the overall complexity
also increases. Therefore, maintaining a balance between complexity and THD is a
challenge for the researcher.
Another important aspect in the designing of a MLI is the optimal choice of
switching angles of the used power electronics devices. Numerous algorithms such as
sinusoidal pulse width modulation, space vector pulse width modulation, elimination
theory and selective harmonic elimination theory are available in the literature for
finding out the switching angles of a MLI [6–8]. But the careful selection of the
switching angles is required so as to minimize the THD and to increase the robustness
of the system.
In the present scenario, the field of research is shifted towards the renewable energy
sources (RESs) such as solar, wind, biomass, hydro and biofuel [9, 10]. These RESs
are the better option of power as these sources can be naturally replenished with time.
Many researchers are working on to integrate these RESs with the MLIs. Flexibility
to integrate these RESs with the MLIs has made MLIs more reliable options for drives
and smart grid applications. Here, in this research work, a solar-based seven-level
inverter has designed and simulated, and output waveforms are analysed in terms of
harmonics. The designed inverter is suitable for drive applications.

2 Multilevel Inverter and Its Topologies

Neutral point clamped (NPC), flying capacitor (FC) and cascaded H-bridge (CHB)
are the existing topologies of MLIs. Out of these three topologies, CHB is the most
widely used topology of MLI due to its advantages like elimination of clamping
diodes and capacitor unbalancing problems as are in NPC and FC topologies, respec-
tively. Figure 1 depicts the general configuration of a single-phase CHB inverter. CHB
configuration makes use of H-bridge, and each H-bridge consists of four switching
devices. Selection of switching angles of these devices is a very crucial step as
the optimal switching angles lead to minimized harmonics and THD in the output
waveform of the MLI.
Design and Analysis of a PV-Based Cascaded H-Bridge … 543

Fig. 1 Single-phase CHB inverter

Mathematically, THD is defined as the summation of all harmonic components

of the voltage or current waveforms compared against the fundamental voltage or
current and is given as:

(V22 + V32 + V42 + · · · + Vn2 )
THD = × 100% (1)
V1
544 P. Gaur et al.

2.1 Seven-Level CHB Inverter

Investigators in this research work present a photovoltaic (PV)-based seven-level

CHB inverter. Figure 2 represents a three-phase seven-level CHB inverter. The mag-
nitude of the output voltage of a seven-level CHB inverter with non-equal PV source
is represented as per the below equation:
∞
 4Vdc
VO = {(Vdc1 cos(nα1 ) + Vdc2 cos(nα2 ) + Vdc3 cos(nα1 )) sin(nωt)}
n=1,3,5,...
nπ
(2)

V O represents the output voltage; V dc is the direct current (DC) voltage of seven-
level CHB inverter. α 1 , α 2 and α 3 are the switching angles for three bridges of
seven-level CHB inverter. Another important parameter is modulation index (M),
which represents the fundamental voltage V 1 and is depicted in Eq. (3).

V1
M= 4  (3)
π
Vdcn

The system considered is three phase; therefore, the triplen harmonics get can-
celled by themselves. Fifth and seventh harmonics of the output waveform can be
eliminated using proper selection of switching angles. Figure 3 depicts the staircase
output waveform of a seven-level CHB inverter.

Fig. 2 Three-phase seven-level CHB inverter

Design and Analysis of a PV-Based Cascaded H-Bridge … 545

Fig. 3 Output waveform of

a seven-level CHB inverter

3 Photovoltaic Systems

A PV system is used to obtain solar energy and then to convert it into the usable elec-
trical energy for residential purposes as well as for massive utility-scale generation
plants. A PV system consists of solar panels for converting solar light into electrical
energy using the photovoltaic effect, inverters for conversion of obtained DC into
the alternating current (AC) and other electrical and mechanical hardware. Figure 4
depicts a residential grid-tied solar PV system. The electrical energy obtained from
the PV panels is in the form of DC. As most of the power electronics devices and
electrical grid use the AC form of electrical energy, therefore, inverters are used to
convert DC electricity into AC form. Then, this AC electricity is being transferred
to household appliances or the electrical utility grid.
Here, in this research paper, PV source is being considered as the input for DC
supply to the seven-level CHB inverter rather than taking DC battery source directly.
This promotes the use of RES such as solar and results in less environmental emission
and pollution.

Sun
Household
Appliances

Inverter Network Grid

DC-AC Meter

Solar Panels

Fig. 4 A residential grid-tied solar PV system

546 P. Gaur et al.

4 Proposed Work

This research work mainly aims to develop a seven-level CHB inverter with mini-
mum THD powered by a PV panel. A pulse width modulation-selective harmonic
elimination (PWM-SHE) modulation strategy is developed to provide the switching
to the switching devices in CHB MLI. Figure 5 represents the block diagram of
the proposed PWM-SHE modulation strategy-based CHB inverter powered by PV
panel.

4.1 PWM-SHE Modulation Strategy for THD Minimization

The main challenge in the designing of a seven-level CHB inverter is to minimize the
fifth and seventh harmonics in the output waveform. Analysis of a CHB inverter is
done using the trigonometric Fourier series representation as in Eq. (2). This equation
is converted into the SHE equations, which are depicted in Eq. (4) as under:

7 Level CHB Inverter

3 Phase
PV
Induction
Array
Motor

Personal Computer with MATLAB Simulink Models

PWM-SHE Modulation Strategy
FPGA Spartan 6 Board

Fig. 5 Block diagram of the proposed system

Design and Analysis of a PV-Based Cascaded H-Bridge … 547

Table 1 Obtained switching

Modulation index (M) α 1 (°) α 2 (°) α 3 (°)
angles of seven-level CHB
inverter 1.00 11.6817 31.1783 58.5774
0.90 17.5104 43.0523 64.1395
0.85 22.7654 49.3798 64.5562
0.80 29.2355 54.4383 64.4844
0.75 34.8935 54.4622 68.5500
0.70 38.3413 53.9297 73.9648
0.65 39.3876 55.5215 78.8979
0.60 39.4298 58.5839 83.1042

cos α1 + cos α2 + cos α3 = 3M

cos 5α1 + cos 5α2 + cos 5α3 = 0
cos 7α1 + cos 7α2 + cos 7α3 = 0 (4)

As explained earlier, α 1 , α 2 and α 3 are the switching angles for three bridges of
a seven-level CHB inverter and M is the modulation index. These three switching
angles must fulfil the criteria as:


0 < α1 < α2 < α3 < (5)
2
Now, the main aim is to minimize fifth and seventh harmonics in the output of
seven-level CHB inverter as depicted in Eq. (6). For this, objective function is defined
as in Eq. (7):

V 5 = cos 5α1 + cos 5α2 + cos 5α3

V 7 = cos 7α1 + cos 7α2 + cos 7α3 (6)

F = (abs(V 5) + abs(V 7)) (7)

The obtained three switching angles at different modulation indices are repre-
sented in Table 1. Optimal angles are 34.8935°, 54.4622° and 68.5500° that are
obtained at modulation index 0.75. These optimal angles are used for firing of the
switching devices of H-bridge of seven-level inverter.

4.2 Simulink Models and Results

Figure 6 depicts the Simulink model for PV-based cascaded H-bridge seven-level
inverter for one phase. Input source is PV rather than DC source. This is beneficial
548 P. Gaur et al.

Fig. 6 Simulink model for PV-based cascaded H-bridge seven-level inverter for phase A

in terms of utilization of more and more renewable sources. The obtained wave-
forms for phase A, B and C after the simulation are represented in Fig. 7. Table 2
depicts the THD obtained at different values of modulation index. The proposed CHB
seven-level inverter with PV source results in minimum harmonics and also suitable
for reactive power compensation devices such as static synchronous compensator
(STATCOM).
The THD obtained in the proposed seven-level CHB inverter can be compared
with the already existing and designed lower level CHB inverter [11]. Table 3 depicts
the comparison of the proposed system with the already existing lower level CHB
inverter.
Design and Analysis of a PV-Based Cascaded H-Bridge … 549

150

100

50
Amplitude (volt)

-50

-100

-150
0 2 4 6 8 10 12
4
Time (sec) x 10

(a) Phase A Output Voltage of a CHB 7 Level Inverter

150

100

50
Amplitude (volt)

-50

-100

-150
0 2 4 6 8 10 12
4
Time (sec) x 10

(b) Phase B Output Voltage of a CHB 7 Level Inverter

200

150

100
Amplitude (volt)

50

-50

-100

-150

-200
0 2 4 6 8 10 12
4
Time (sec) x 10
(c) Phase C Output Voltage of a CHB 7 Level Inverter

Fig. 7 Output voltage of CHB seven-level inverter

550 P. Gaur et al.

Table 2 Obtained THD in

Modulation index (M) THD (%)
PV-based CHB seven-level
inverter 0.7 9.27
0.75 9.21
0.8 9.33
0.85 9.54
0.9 9.57
0.95 9.35
1.0 9.40

Table 3 THD (%)

S. No. Levels in cascaded H-bridge inverter THD (%)
Comparison of the proposed
inverter with lower level 1 Three-level inverter 32.49
inverter 2 Five-level inverter 28.51
3 Proposed seven-level inverter 9.21

5 Conclusion and Future Scope

Multilevel inverters are very promising alternatives for high-power applications.

This paper presented a novel approach for designing of a renewable energy source,
i.e. PV-based cascaded H-bridge seven-level inverter. The simulation results depict
that the proposed system can eliminate the harmonics and results in lesser total
harmonic distortion at different values of modulation index. Further, SHE-PWM-
based strategy results in better performance in terms of lesser losses and harmonics.
Therefore, the designed system is well suited for renewable energy applications and
for reactive power compensation devices such as STATCOM. As a future scope, the
same work can be extended to higher levels of inverter suitable for renewable and
power electronics drive applications.

References

1. Rodriguez, J., Lai, J.S., Peng, F.Z.: Multilevel inverters: A survey of topologies, controls and
applications. IEEE Trans. Ind Electron. 49, 724–738 (2002)
2. Nabae, A., Takahashi, I., Akagi, H.: A new neutral point clamped PWM inverter. IEEE Trans
Industry Applications 17(5), 518–523 (1981)
3. Bernet, S.: Recent developments of high power converters for industry and traction applications.
IEEE Trans. Power Electron. 15(6), 1102–1117 (2010)
4. Kala, Peeyush, Arora, Sudha: A comprehensive study of classical and hybrid multilevel inverter
topologies for renewable energy applications. Renew. Sustain. Energy Rev. 76, 905–931 (2017)
5. Duffey, C.K., Stratford, R.P.: Update of harmonic standard IEEE-519-IEEE recommended
practices and requirements for harmonic control in electric power systems. IEEE Trans. Ind
Appl (1989)
Design and Analysis of a PV-Based Cascaded H-Bridge … 551

6. Gaur, P., Singh, P.: Various control strategies for medium voltage high power multilevel
converters: a review. In: Proceedings of RAECS, UIET, Panjab University, Chandigarh (2014)
7. Susheela, N., Kumar, P.S.: Performance evaluation of carrier based PWM techniques for hybrid
multilevel inverters with reduced number of components. Energy Procedia 117, 635–642 (2017)
8. Ye, M., Kang, L., Xiao, Y., Song, P., Li, S.: Modified hybrid modulation strategy with power
balance control for H-bridge hybrid cascaded seven level inverter. IET Power Electron. 11(6),
(2018)
9. Kalashania, M.B., Nazarpour, D.: New symmetric and hybrid multilevel inverter topology
employed in solar energy systems. Trans. Electr. Electron. Mater. 1–7, (2018)
10. Miranbeigi, M. Iman-Eini, H.: Hybrid modulation technique for grid-connected cascaded
photovoltaic systems. IEEE Trans. Ind. Electron. 63(12), (2016)
11. Singh, G., Garg, V.K.: THD Analysis of Cascaded H-bridge Multi-Level Inverter. In: 4th IEEE
International Conference on Signal Processing, Computing and Control, 21–23 Sep 2017,
Solan, India
Accurate Maximum Power Point
Tracking with New Modified Water
Cycle-Based Ridge Extreme Learning
Machine for Photovoltaic-Based
Microgrids

Sangram Keshari Pattnayak, Snehamoy Dhar, Durgesh Prasad Bagarty,

Meera Viswavandya, and Niranjan Nayak

Abstract This paper is proposing a new modified water cycle (WC) optimized
extreme learning machine (ELM) for maximum power point tracking (MPPT) of
a photovoltaic (PV)-based microgrid. The PV-based distributed generation (DG)
is considered with two-stage integration (DC-DC/DC-AC). The MPPT-based PV-
independent controller (DC-DC duty cycle for two-stage conversion) is primarily
estimating the control reference. The MPPT error (eMPP ) is escalated under PV
side (e.g. irradiation variation, partial shading) as well as grid side (e.g. symmet-
ric/asymmetric faults, load variation) contingencies. Thus, robust error reduced fast
MPPT is targeted in this paper against improved grid stability. The fastness is obtained
by Moore–Penrose inverse (MPI) calculation-based ELM, and the robustness is tar-
geted by reducing the randomness of initial weights through modified WC tech-
nique optimization. The proposed WC algorithm is incorporated with a sinusoidal
chaotic map, by virtue of which the nonlinearities during operational contingencies
are tackled properly. The effectiveness of the proposed MPPT is established in terms
of improved eMPP profile, and better stability (dynamic oscillation) towards grid
response. The performance evaluation is estimated in MATLAB script environment.

S. K. Pattnayak (B)
Indus School of Engineering, Bhubaneswar, India
e-mail: [email protected]
S. Dhar · N. Nayak
Siksha ‘O’ Anusandhan (Deemed to be University), Bhubaneswar, Odisha 751030, India
e-mail: [email protected]
N. Nayak
e-mail: [email protected]
D. P. Bagarty · M. Viswavandya
College of Engineering and Technology, Bhubaneswar, India
e-mail: [email protected]
M. Viswavandya
e-mail: [email protected]

© Springer Nature Singapore Pte Ltd. 2020 553

G. Pradhan et al. (eds.), Advances in Electrical Control and Signal Systems,
Lecture Notes in Electrical Engineering 665,
https://doi.org/10.1007/978-981-15-5262-5_41
554 S. K. Pattnayak et al.

Keywords Photovoltaic · Maximum power point tracking · Water cycle

algorithm · Extreme learning machine

1 Introduction

The requirement of distributed generations (DGs) to active distribution net-

work/microgrid is facing expeditious market growth from the last two decades. To
solve the expanding load requirement and to address the environmental (carbon)
contamination policy, renewable energy-oriented DGs are encouraging choice. More
than three hundred days of full solar irradiation is quite promising for photovoltaic
(PV)-based DGs in a nation like India [1]. To provide an economical PV solution the
maximum power (MP) should be extracted from PV panels, by effective maximum
power point tracking (MPPT) algorithms.
MPPT accuracy is directly influential for DG-to-grid (at point of common cou-
pling: PCC) stability margin. The erroneous MPPT will provide inaccurate control
references to primary controllers (PCs) [2]. The control hierarchy against PV-DG
integration to microgrid is dependent on PCs response (in terms of control reference)
to independent DG controllers (IDGCs). IDGCs are feedback path (closed-loop oper-
ation) to obtain proper duty cycle calculation/PWM reference generation for single
stage: DC-AC [3]/double stage: DC-DC+DC-AC [4] PV-based DG integration. Thus
accurate MP tracking is a very important problem towards grid stability (dynamic
oscillation damping).
MPPT schemes can be generally categorized for two sections: (a) linear and (b)
nonlinear techniques. Linear algorithms like P&O [4] Hill climbing [5] and incre-
mental conductance [6] are well established against simplicity (less computational),
and cost-effective solutions under irradiation change. But during local PV uncertain-
ties (e.g. partial shading, PV internal faults, etc.) where erratic profile of solar data
is inevitable, these algorithms are erroneous to extract MP. On the other hand, non-
linear MPPT techniques, e.g. fuzzy logic (FL) [7], artificial neural network (ANN),
[8] and evolutionary techniques, particle swarm optimization (PSO) [9], are effec-
tive to extract MP under such local uncertainties (especially partial shading). The
FL technique incorporates fuzzification, rule base decision data, inference action,
and defuzzification to track MP. This makes it tiresome against computation. Earlier
information about the PV systems’ behaviour is needed for the choice of accurate
fuzzy rule design and membership functions estimation. The FL-based MPPT design
mainly depends on designer’s prospective. Thus, for PV-based DG solution towards
rural microgrids is uneconomical. Recorded historical data (by data acquisition sys-
tem: DAS)-dependent ANN methods require large amount of training samples to
evolve according to the considered PV system. Conventional ANN-based MPPT
schemes are iterative and computationally complex to be implemented. These tech-
niques are system dependent, and without proper training for a specific system, the
outputs are inaccurate (increased MPPT error: eMPP ). PSO is evidenced potentially
to track exact MPP under shading condition, but its accuracy reduces when solar
Accurate Maximum Power Point Tracking … 555

data randomness is increasing (simultaneous partial shading, irradiation change for

certain time period). PSO is established as a complicated scheme which comprises a
large number of parameters which have to be decided on designers’ knowledge. The
increased number of parameters (random particles) provides enhanced convergence
to accurate MP, but simultaneous computational burden escalates too. To cope with
well-established conventional MPPTs’ drawbacks, this paper proposes a simplified
but accurate MPPT scheme method by considering extreme learning machine (ELM)
[10–12]. The conventional iterative learning (computational burden) is abolished in
terms of non-iterative Moore–Penrose pseudoinverse (MPPI) calculation. To increase
the efficacy of ELM-based MPPT learning, random hidden layer feature mapping-
based ELM (RELM) is introduced in this paper. The randomness in initial weights is
optimized with a new less computational modified water cycle (WC) algorithm. The
conventional WC algorithm [13] is a nature-inspired (water cycle process: streams,
rivers joining to the sea), meta-heuristic estimation similar to swarm intelligence.
Slow convergence with inaccurate optimality under multi-dimensional problems is
major drawback of WC. To overcome the drawbacks of conventional WC, mutation
operation is incorporate to make the action modified in terms of character diversity
of raindrops.
After a brief introduction in Sect. 1 the considered PV-based DG integrated micro-
grid architecture is presented in Sect. 2. The PV operation is obtained as single-stage
conversion for better reflection of MPPT error (eMPP ) effects on grid dynamics (with-
out voltage stability by DC-DC). The proposed modified WC optimized RELM-
MPPT is described in Sect. 3, where eMPP calculation is emphasized in terms of dif-
ferent standard formulations (e.g. root mean square error/RMSE). The performance
of the proposed MPPT scheme is evaluated under different operational contingen-
cies in Sect. 4. This evaluation study is obtained in MATLAB script environment.
The reduction of eMPP profile from conventional approaches is highlighted here along
with improvement in grid stability margin (in terms of dynamic oscillation damping).
Finally, conclusion and future scope of the work are presented in Sect. 5.

2 Considered PV-Based Microgrid

For the PV application as DG, microgrid integration requires three-phase AC power

and thus, a 6-pulse IGBT-based voltage source converter (VSC) is considered here,
Fig. 1. The considered single-stage conversion-based PV-DG is incorporated with
local loading (AC) with parallel capacitor (for voltage stability) at PCC. The MPPT
scheme is implemented in terms of measured/historical (DAS-based) solar irradia-
tion and PV panel temperature profile. The VSC injected power at PCC (P1 + j Q1 )
is directly influenced by feedback path (PWM generation). The feedback path or
IDGC is designed based on PCs’ reference (MPPT-based estimation) and local mea-
surements (PCC voltage: V 1,abc and I 1,abc ). Thus, eMPP profile is directly influencing
the grid stability. The grid design parameters are as mentioned in Table 1.
556 S. K. Pattnayak et al.

Fig. 1 The considered PV-based DG integration to microgrid operation

Table 1 Numerical data of

Parameters Values
the simulation system
VSC to PCC interfacing resistance (Ri ) 15 m
VSC to PCC interfacing inductance (L i ) 0.63 mH
Angular frequency (ω) 314.15 rad/s
PCC to grid interfacing resistance (Rg ) 10 m
PCC to grid interfacing inductance (L g ) 0.5 mH
Filter capacitor (C f ) 0.64 mF
DC link capacitor (C pv ) 0.5 mF
Conductance of local load (g2 ) 1.54 p u
Susceptance of the local load (b2 ) 1.03 p u

The PV system model is further described with proper mathematical descriptions

here. The MP data generation from local DG is possible to achieve while under-
standing the PV modelling. The single diode equivalent model is considered for the
present focus.
Accurate Maximum Power Point Tracking … 557

2.1 PV System Modelling

The solar cell is the elemental unit of any PV systems (array or module). A p-
n junction diode exposed to sunlight is able to transform light intensity (photon
energy) into the form of electrical energy (DC) by photoconduction effect. PV single
diode equivalent circuit is considered here for solar irradiation (W/m2 ) data to PV
parameter (MP power: PPV , voltage: V PV ) calculation [14, 15]. This PV model is
simulated in MATLAB/script environment for present focus where PV array (number
of series, parallel-connected modules) is depicted as:
   V +I R  
1 PV PV s −1
IPV = Npp × Iph − Npp × Irst e VT a Nse

 
Nse /Npp × VPV + IPV × R S
− (1)
Rsh

where I PV → PV array’s output current, N se → number of series modules, and N pp

→ number of parallel modules, VT = kT /q → thermal voltage. A real solar module
ELDORA-40 (10 kW PV system: 46 parallel × 6 series × 40 W) is developed here,
and other coefficient and parameters are described in Table 2.

3 Proposed Modified WC-Based RELM

The linear MPPT schemes (P&O, Hill climbing, etc.) are computationally effective
but under PV operational uncertainties like partial shading these schemes are erro-
neous (eMPP increases drastically). The nonlinear MPPT techniques (ANN, FL, etc.)
are computationally complex but effective under uncertainties. The conventional
ANN-based MPPTs are computationally complex majorly due to their architecture
(number of input and hidden layers) and iterative training process for large DAS-
based irradiation (W/m2 ). To provide a solution with fast computation, robust eMPP
reduction and less complexity RELM (non-iterative learning by MPPI)-based MPPT
are proposed in this paper. The initial random input weights are major reason for eMPP
profile deviation under irradiation data volatility. Thus to make the operation robust,

Table 2 Major descriptions

Parameters Values
of ELDORA-40 PV at
Standard Test Condition: STC MP power (PPV ) 40 W
(=1 kW/m2 , 25°C) Voltage at MP (V PV ) 17.4 V
Current at MP (I PV ) 2.3 A
Short circuit current (I sc ) 2.57 A
Open-circuit voltage (V oc ) 21.9 V
558 S. K. Pattnayak et al.

Fig. 2 Generalized ELM

structure

a modified (by mutation algorithm) WC algorithm-based optimized input weight

formulation is adopted for the proposed MPPT scheme.
ELM is generalized architecture for feed-forward ANN, with single hidden layer
configuration. Thus ELM consists of three layers (i.e. input: IP, hidden: Hd and out-
put: OP layer) architecture without dependency of Hd layer tuning or IP data sample
number. When the OP weights are estimated with proper mathematical background
(non-iterative MPPI), the IP and bias weights are selected randomly for conven-
tional ELM. The fast computation is possible in ELM by MPPI training as compared
with conventional support vector machine (SVM) [16], Least-square SVM [17]. The
architecture of generalized ELM is presented in Fig. 2.
In proposed configuration, ELM is having two IP nodes: Solar irradiation (GPV in
W/m2 ) and PV panel temperature (T PV in °C). The OP nodes are decided to be PV
system output voltage (V PV ), and power (PPV ). The data dimensions are considered to
be M number for IP nodes, N number for OP nodes, and Hd layer nodes are considered
as S. The IP node vector is defined as: [X IP ]1×M
K
= x1,IP K
, x2,IP
K
, . . . , x M,IP
K
; and the
OP node vector (target) is determined as: [TOP ]1×N = t1,OP , t2,OP , . . . , t NK,OP . Here
K K K

K is the instant of data sample out of Tot number of total data samples. Hd layer
output is obtained as: [Hd]1×S
K
= hd1K , hd2K , . . . , hd KM , and the output calculation
is expressed as:
 M
 
hd Kj = sinh φmj X mK + αm (2)
m=1

where j =1,2, …, S; [φ] M×S → IP weights (random selection between 0 and 1),
[α]1×S → bias weight at IP layer. The activation function is considered with sinh
function as in Eq. (2).
The OP vector for Kth instant (estimated) is assumed to be: [E OP ] K =
e1,OP , e2,OP
K K
, . . . , e NK ,OP , and the OP weights are: [γ ] S×N . Thus during training the
targeted MPPT is opted from least-square approach as:
K
[TOP ]1×N = [Hd]1×S
K
× [γ ] S×N
K
(3)

Now as the training is targeted to be achieved without iterative calculation, [Hd]

is not a square matrix, the unknown output layer weights (γ ) are calculated with
Accurate Maximum Power Point Tracking … 559

MPPI technique as:

K
[γ ] S×N
K
= (HdT × Hd)−1 HdT S×1
× [E OP ]1×N
K
(4)

This MMPI technique is indirectly minimizing the norm of loss function:

Jloss = Hd − E OP  (5)

Now to reduce the J loss function further (as random weights are assumed in IP
layer which has direct influence on γ estimation) a ridge regression theory (RRT )
is incorporated here. The singular matrix (HdT × Hd)−1 in Eq. (4) is combined
with a constant term (regularization coefficient: τ ). Thus, the proposed ridge ELM
(RELM)-based MPPT is now reducing the improved cost function:

1
Jloss = Hd − E OP 2 + γ 2 (6)
τ
Due to the modified loss function (Eq. 6) minimization, the OP weights are esti-
mated better with less randomness effect. The minimization is obtained by MPPI
theory as:

  K
I −1 T
[γ ] S×N
K
= Hd × Hd +
T
Hd × [E OP ]1×N
K
(7)
τ
S×1

here I → identity matrix, τ → 2C . The coefficient C is obtained between 6 and 10

for proposed consideration. Now the target T OP is estimated as:
K
[TOP ]1×N = [Hd]1×S
K

⎡ K ⎤K
 
I −1 T
× ⎣[γ ] S×N
K
= Hd × Hd +
T
Hd K ⎦
× [E OP ]1×N (8)
τ
S×1 S×N

As the IP and bias weights are random for initial selection, the eMPP is still not
possible to be reduced robustly. The eMPP profile escalates under solar data volatil-
ity (irradiation change, partial shading, etc.). Thus, the modified WC algorithm is
introduced to optimize the operation by the selection of proper IP weights.

3.1 Modified WC-Based IP Weight Optimization

The water flow is the basic idea behind this nature-inspired algorithm where evap-
oration, transpiration, condensation, precipitation and runoff take place to reach
560 S. K. Pattnayak et al.

a raindrop to sea (optimal point). Raindrop population is created initially. These

are equivalent to IP weights ([φ] M×S ). To minimize the eMPP, the cost function for
proposed optimization problem is derived as:
 K 
f (φ) = min TOP − E OP
K
/N where N = M. (9)

The raindrop population is defined as: [Rd] B×C = rd1K , rd2K , . . . , rdCK . The num-
bers of rivers are: Rv streaming towards sea (S E ). The number of streams (Str) can
be obtained as:
  
 CGk 
Str = round     × Str , CGk = f (φ)k − f (φ) SE +1 (10)
k=1 to S E CGk 

With each generation (k), the Str changes their flow and approaches towards
optimal point S E . The flow position is estimated modified by introducing chaos
signal (Chao) as:
 
i
kstr (t + 1) = kstr
i
(t) + Chao(t) × K SE (t) − K str (t) (11)
(i=1 to S E )

To make the flow modified towards optimal solution (S E ), the evaporation process
is influenced by chaotic mapping as:
 
 
 
k SE − kRv  < Chao(t)
i
 (i=1 to Rv )
 
 
 j 
or k SE − K Str  < Chao(t) or Chao(t) < 0.1 (12)
 ( j=1 to Strt )

Finally, when the process reaches to total generation and the optimal solution is
obtained for Eq. (9). The effectiveness of the proposed MPPT scheme is evidenced
in terms of eMPP . The standard formulation for this error profile is estimated in terms
of mean absolute percentage error (MAPE), root-mean-square error (RMSE) as:

N

1 |TOP − E OP |
MAPE = × 100
N k=1 TOP


1 N
RMSE =  |TOP − E OP |2 . (13)
N k=1
Accurate Maximum Power Point Tracking … 561

4 Result and Discussion

The effectiveness of proposed MPPT scheme is evidenced for DAS data volatility
as well as grid (PCC) parameter response. The DAS data is selected for 5 m data
interval, and the efficacy measures are various eMPP responses: RMSE, MAPE. The
MP power is estimated with proposed RELM-WC (modified WC optimized RELM).
The comparison is evidenced with existing techniques: conventional ELM, RELM
as shown in Fig. 3a. The data volatility profile is considered for the summer season.
The proposed RELM-WC depicts RMSE (in p u) 0.00058–0.00087 and MAPE (%)
0.0267–0.0532, respectively (Fig. 3b, c). The RMSEs (in p u) for conventional RELM
and ELM are estimated as 0.0037–0.0043 and 0.0067–0.0072, respectively. The
MAPEs (%) are evidenced to be in a range of 0.0475–0.0699 and 0.1076–0.2657
for conventional RELM and ELM respectively. The proposed RELM-WC based
MPPT shows superiority over conventional schemes in terms of robust reduction
in eMPP profile. The MPPT scheme is designed to be integrated with PV-based DG
integration towards AC utility (PCC: Fig. 1). Thus, controller hierarchy (i.e. PC,
IDGC, etc.) will be affected with MPPT error. The grid stability margin is in risk
under erroneous MP estimated operation. The PCC response (active power: P1 and
operational frequency: f ) is recorded for conventional scheme: ELM and proposed
scheme: RELM-WC. The proposed RELM-WC is evidenced as reference estimator
for PC, and the PC coordination to IDGC is evidenced in terms of PCC response as
in Fig. 4a–c.
The worst grid operation with simultaneous contingency is subjected to the con-
sidered microgrid. Partial shading condition is subjected at time t = 4.53–6.43 s
to PV irradiation. Again a three-phase fault is introduced at t = 6.1–6.2 s at bus
1. The PC’s reference is accurately calculated with proposed MPPT scheme, and
the dynamic oscillation is evidenced in 14–18 cycles. But the conventional ELM-
based erroneous reference shows decayed grid stability margin: 40–56 cycles. The
proposed RELM-WC-based MPPT is well suited for PV-DG integration towards
microgrid application. The performance of the proposed MPPT is obtained supe-
rior as compared with conventional techniques, under different grid contingencies
as shown in Fig. 5. The performance index is considered to be number of cycles (at
PCC response) for irradiation change, partial shading and fault at PCC.

5 Conclusion

In this paper, a new maximum power point tracking (MPPT) technique is presented
for photovoltaic (PV)-based microgrid/active distribution network, towards robust
MPPT error (eMPP ) reduction and resultant improved grid stability. The conventional
MPPTs (e.g. linear: P&O, Hill Climbing, etc., nonlinear: ANN, FL, etc.) are men-
tioned with their drawbacks against accurate MP tracking under various operational
562 S. K. Pattnayak et al.

(a)

(b)

(c)

Fig. 3 Performance validation of proposed RELM-WC MPPT under DAS data variation: a PV
system power output, eMPP profile under: b RMSE and c MAPE (%) error
Accurate Maximum Power Point Tracking … 563

Fig. 4 Performance
evaluation of proposed
MPPT-based PC’s reference
calculation against PCC
response: a operating
frequency in Hz, b PCC
active power, c PCC voltage

(a)

(b)

(c)

uncertainties (irradiation change, partial shading, etc.). To overcome existing draw-

backs and to obtain a fast, accurate MP tracker new ridge extreme learning machine
(RELM) is proposed with robust eMPP profile reduction. The robust error reduction
is achieved by omitting the randomness of proposed RELM’s IP weights by means
of new modified water cycle (WC)-based optimal weight selection. The modified
operation is presented with chaotic mapping of raindrops towards stream flow to sea
564 S. K. Pattnayak et al.

140

120

100
No. of Cycles

80
Irradiation Change
60
Partial Shading
40
Faults
20

0
Conv. RELM Prop.
ELM RELM-
WC

Fig. 5 Comparative study of proposed MPPT under different operational contingency: performance
index as grid stability limit (in number of cycles)

(optimal solution). The effectiveness of the proposed MPPT is validated in terms

of accuracy (eMPP decrement profile by standard error formulation: RMSE, MAPE).
The grid stability improvement (primary controller’s/PC’s reference to independent
DG controller/IDGC: PWM-based phase-locked loop/PLL) is evidenced in terms of
dynamic oscillation damping at point of common coupling (considered microgrid:
Fig. 1). The evaluation study is implemented in MATLAB script environment. The
local energy management (LEM) system design with proposed MPPT error provision
is considered as future scope of this paper.

References

1. Rathore, P.K., Chauhan, D.S., Singh, R.P.: Decentralized solar rooftop photovoltaic in India:
on the path of sustainable energy security. Renew. Energy 131, 297–307 (2019)
2. Shuai, Z., Fang, J., Ning, F., John Shen, Z.: Hierarchical structure and bus voltage control of
DC microgrid. Renew. Sustain. Energy Rev. 82, 3670–3682 (2018)
3. Jain, S., Agarwal, V.: Comparison of the performance of maximum power point tracking
schemes applied to single-stage grid-connected photovoltaic systems. IET Electr. Power Appl.
1(5), 753–762 (2007)
4. Kobayashi, K., Takano, I., Sawada, Y.: A study on a two stage maximum power point tracking
control of a photovoltaic system under partially shaded insolation conditions. In: 2003 IEEE
Power Engineering Society General Meeting (IEEE Cat. No. 03CH37491), vol. 4, pp. 2612–
2617 (2003)
5. Xiao, W., Dunford, W.G.: A modified adaptive hill climbing MPPT method for photovoltaic
power systems. In: 2004 IEEE 35th Annual Power Electronics Specialists Conference (IEEE
Cat. No. 04CH37551), vol. 3, pp. 1957–1963 (2004)
Accurate Maximum Power Point Tracking … 565

6. Tey, K.S., Mekhilef, S.: Modified incremental conductance algorithm for photovoltaic system
under partial shading conditions and load variation. IEEE Trans. Ind. Electron. 61(10), 5384–
5392 (2014)
7. Guenounou, O., Dahhou, B., Chabour, F.: Adaptive fuzzy controller based MPPT for
photovoltaic systems. Energy Convers. Manage. 78, 843–850 (2014)
8. Muthuramalingam, M., Manoharan, P.S.: Comparative analysis of distributed MPPT controllers
for partially shaded stand alone photovoltaic systems. Energy Convers. Manag. 86, 286–299
(2014)
9. Ishaque, K., Salam, Z., Amjad, M., Mekhilef, S.: An improved particle swarm optimization
(PSO)—based MPPT for PV with reduced steady-state oscillation. IEEE Trans. Power Electron.
27(8), 3627–3638 (2012)
10. Du, Y., Yan, K., Ren, Z., Xiao, W.: Designing localized MPPT for PV systems using fuzzy-
weighted extreme learning machine. Energies 11(10), 2615 (2018)
11. Satapathy, P., Dhar, S., Dash, P.K.: A firefly optimized fast extreme learning machine based
maximum power point tracking for stability analysis of microgrid with two stage photovoltaic
generation system. J. Renew. Sustain. Energy 8(2), 025501 (2016)
12. Behera, M.K., Majumder, I., Nayak, N.: Solar photovoltaic power forecasting using optimized
modified extreme learning machine technique. Eng. Sci. Technol. Int. J. 21(3), 428–438 (2018)
13. Huntington, T.G.: Evidence for intensification of the global water cycle: review and synthesis.
J. Hydrol. 319(1–4), 83–95 (2006)
14. Dhar, S., Sridhar, R., Mathew, G.: Implementation of PV cell based standalone solar power sys-
tem employing incremental conductance MPPT algorithm. In: 2013 International Conference
on Circuits, Power and Computing Technologies (ICCPCT), pp. 356–361 (2013)
15. Sridhar, R., Dhar, S., Dash, S.S.: Performance analysis of a stand alone PV system with reduced
switch cascaded multilevel inverter. Int. J. Power Energy Convers. 6(2), 107–127 (2015)
16. Smola, A.J., Schölkopf, B.: A tutorial on support vector regression. Stat. Comput. 14(3), 199–
222 (2004)
17. Suykens, J.A.K., Vandewalle, J.: Least squares support vector machine classifiers. Neural
Process. Lett. 9(3), 293–300 (1999)
Sensitivity Analysis of PI and Youla
Controllers for a PV-Fed Boost Converter

Nivedita Pati, Babita Panda, and Bhagabat Panda

Abstract The abundant use of energy from photovoltaic (PV) panels introduces
many challenges in the whole power system scenario in the form of grid integration.
A typical grid-connected photovoltaic (PV) system consists of a power converter to
connect these sources to the grid. Most readily available converter topologies are
designed to work under a constant source at its input, and hence they exhibit peculiar
characteristic when connected to a variable current source, the PV panels. This paper
thus intends to explore the dynamics and stability of the intermediate phase that is
controlled and regulated output of a PV-fed boost converter under an ohmic load
condition. The simulation platform is used for designing the PV circuit model and
gain model of converter and also to demonstrate the supremacy of one controller
over other. The classical PI controller is chosen for comparison with the controller
designed using Youla parameterization. This technique yields a simple and defined
solution to the problem through an arbitrary, proper and stable function named as Q
parameter. The controller performance is compared based on the variation in output
response due to variation in solar irradiance and load.

Keywords PV circuit model · State-space averaging · Youla controller ·

Sensitivity analysis

N. Pati (B) · B. Panda

School of Electrical Engineering, KIIT University, Bhubaneswar 751024, India
e-mail: [email protected]
B. Panda
e-mail: [email protected]
N. Pati
School of Electrical Engineering, KIIT University, Bhubaneswar 751024, India
B. Panda
Parala Maharaja Engineering College, Berhampur 761003, India
e-mail: [email protected]

© Springer Nature Singapore Pte Ltd. 2020 567

G. Pradhan et al. (eds.), Advances in Electrical Control and Signal Systems,
Lecture Notes in Electrical Engineering 665,
https://doi.org/10.1007/978-981-15-5262-5_42
568 N. Pati et al.

1 Introduction

A typical photovoltaic (PV) system consists of a solar converter, a power converter

to regulate the variable DC and a power conditioning device whose output can be
provided to various AC and DC loads with proper control circuitry. As photovoltaic
arrays exhibit nonlinear characteristics, various converter configurations are required
to process the electricity from PV output with a very high efficiency. In this paper, a
boost converter topology is considered as a case study as is mostly used to increase
the low-voltage output of PV. The mathematical model of the PV device is useful in
the study of the dynamic analysis of converters. The model is prepared based on the
circuit equations of a single-diode model.
Power converter working can be well imitated by the state-space model using the
state equations derived from the switching states of the converter. The application
of the converter output to more sophisticated appliances requires the regulated and
robust output using various current-mode and voltage-mode controllers [1].
The whole system under study is divided into three subsections. The first section
deals with the mathematical modeling of the PV panel and state-space modeling of
converter and proper interfacing of PV with converter. The second subsection deals
with closed-loop control using PI and Youla parameterization. The final section has
the simulation results and discussion. The control of the converters is achieved by
varying the duty cycle of the switch using the PWM technique. PI controllers are
mostly used simple and classical controller, but it limits the range of plants that they
can control satisfactorily. Youla parameterization process is based on plant model
inversion and takes into consideration the internal stability of the system through a
stable transfer function, Q(s) [2]. Internal stability for any practical feedback system
is a basic need as it guarantees the bounded condition of all signals in a closed-loop
system provided that the external disturbances are also bounded. In order to analyze
a system’s internally stable property, all the transfer functions related to input, output
and disturbance must be verified.

2 Mathematical Modeling

2.1 PV Array Circuit Model

The purpose of mathematical model is to simulate the true PV system which converts
the irradiance into a variable DC voltage and current. Several literatures reveal various
configurations of a PV model varying in accuracy and complexity. The single-diode
model shown in Fig. 1 is one of the most simple and easily accessible physical models
used to represent the electric characteristics of a single PV cell [3]. Applying KCL
and solving, the PV can be modeled as a current source using (1).
Sensitivity Analysis of PI and Youla Controllers … 569

Fig. 1 Circuit model of PV panel

   
V + Rs I V + Rs I
I = Ipv − I0 exp −1 − (1)
Vt a Rp

where I pv and I 0 are the photovoltaic and saturation currents of the array and V t is
the thermal voltage, respectively. The parameter dependency and improvement of
the model [3] are done using a set of equations described from (2) to (4)

 G
Ipv = Ipv,n + K I T (2)
Gn
Isc,n
I0,n =  (3)
Voc,n
exp aVt,n
−1
Isc,n + K I T
I0 =  (4)
Voc,n +K V T
exp aVt
−1

where Ipv,n , Isc,n , I0,n , Voc,n are the values at nominal temperature and irradiance.
The array is modeled through various subsystems of I pv , I 0 , I m a dependent current
source and the resistors in Simulink. Figure 1 is the source model which is simulated
using the data sheet values of KC200GT [3], and Fig. 2 shows the I–V characteristic
of the PV array for varying irradiance and temperature.
570 N. Pati et al.

Fig. 2 I–V characteristic with varying irradiance

Fig. 3 PV-fed boost converter

2.2 Power Converter Modeling

Boost converter is a non-minimum phase system exhibiting nonlinear and time-

variant characteristics. The switching operation of the converter results in the circuit
components being connected together in periodically changing configurations. For
ease in the analysis and controller design, the nonlinear system is characterized as
linear model within certain range and time period using state-space averaging (SSA)
Sensitivity Analysis of PI and Youla Controllers … 571

technique [4]. The converter is modeled using the gain model configuration [5] which
is achieved through the equations of the converter during the on and off states as (5–8)
 dil
    1
0 0 il
dt = + L Vg (5)
dvc
dt
0 − RC
1
vc 0
 
il
V0 = 0 1 (6)
vc
     1
dil
0 − L1 il
dt = 1 + L Vg (7)
dvc
dt C
− RC
1
vc 0
 
il
V0 = 0 1 (8)
vc

Parameter estimation of the converter elements is performed taking into account

30% of current ripple and 5% voltage ripple factor [6]. Using the values of L, C and
R, the transfer function (TF) of the converter for duty ratio as the input is calculated
from the averaged model [7]. The TF is given by (9).

V0 (s)
= C[s I − A]−1 .[(A1 − A2 )X + (B1 − B2 )Vin ] + (C1 − C2 )X
d(s)
A = A1 d + A2 d
    1
0 0 0 − L1
A= d + (1 − d), B1 = L = B2
0 − RC
1 1
C
− 1
RC
0
V0 (s) −1.44 ∗ 107 (s − 3333)
= 2 (9)
d(s) s + 5556s + 1.852 ∗ 107

The gain model of the converter is interfaced with the PV panel to obtain the
required output voltage as shown in Fig. 3. The source considered is PV for imple-
menting the source variations and analysis of robustness of controller [8]. The output
of PV panel is 29.14 V and gives a converter output voltage of 255 V with a ripple
of 13% (Fig. 4).

3 Controller Design

A successful application of the control law design needs to consider the reference
tracking and maintain the robustness toward load and irradiance variations. Vari-
ous non-linear controllers satisfy the robustness criteria as compared to its linear
counterparts [9]. The control performance will be affected by major issues like the
time-variant characteristics and the varying output of PV. A comparison is made
based on simulated results between two different controllers.
572 N. Pati et al.

Fig. 4 Output voltage of PV-fed boost converter

3.1 PI Controller

Linear Controllers are simple to design and easy to implement but has got the dis-
ability of not being robust. The objective behind design of controller is to ensure
regulated output with minimum overshoot and settling time in the presence of vary-
ing irradiance and temperature. This section deals with the design of a simple PI
controller by the classical Ziegler–Nichols first approach method.
The tuning of the controller parameters is made based on critical gain (K cr ) and
critical time period (T cr ) which is obtained from the point of sustained oscillations.
The critical gain and critical time period are 3.86e−4 and 6090, respectively. The
controller is of first order having a transfer function as:

0.202
C(s) = 1.74 ∗ 10−4 +
s
The converter output voltage which is our desired output is achieved through a
closed-loop structure where desired value is compared with a set point value and the
result is fed as pulse to the switch [10]. Figure 3 shows the closed-loop Simulink
model where the error produced is passed through the PI controller and is com-
pared with a ramp signal to generate the required duty cycle using the pulse width
modulation (PWM) technique.

3.2 Q-Controller (Youla Parameterization)

Controller design through Q-parameterization is a process of making a closed-loop
internally stable by selecting a stable and proper transfer function Q(s) [2]. For a
nominal plant G(s), the input–output TF can be of the form (10) where T 0 (jw) will
be 1 only at those frequencies where Q(jw) inverts the model.
Sensitivity Analysis of PI and Youla Controllers … 573

T0 (s) = G(s)Q(s) (10)

The process is easily applicable to stable case [9], whereas in systems having
a non-minimum phase zero a stable approximation of the plant is considered and
inverted. The plant TF, G(s) is factored into stable and unstable parts, and the inver-
sion is carried out with the stable part. Youla parameterization allows designers to
choose the closed-loop parameters of undamped natural frequency ωcl and damping
factor ξ cl [10]. Using these values, a TF Q(s) is generated in the structure of a poly-
nomial. In order to make Q(s) stable and proper, F Q (s) is multiplied with the inverted
plant model. F Q (s) acts as a filter of second order whose parameters are selected by
considering ξ and ωcl as 0.7 and 31,918.5 rad/s based on time response specifications.
The controller is then designed using the affine parameterization process using Q(s)
and G(s). The sequence of equations is from (11) to (15)

Q(s) = FQ (s)[G]−1 (11)

P(s)
Q(s) = (12)
E(s)
1
FQ (s) =
α2 + α1 s + 1
s2
α2 = 1/ωcl2 , α1 = 2ξcl /ωcl (13)

1
FQ (s) = (14)
9.82 ∗ 10−10 s 2 + 4.39 ∗ 10−5 s + 1
Q(s)
C(s) =
1 − Q(s)G(s)
s 2 + 5556s + 1.85e7
C(s) = (15)
47.13s 2 + 16,506,989.28s

The converter control is done by PWM method in which a pulse is fed to the
switch to give the desired result. The controller is connected in the forward path, and
the overall system is unity feedback system.

3.3 Sensitivity Analysis

The system’s robustness for a particular controller is validated by applying distur-

bances and nonlinearities in the model.
The gang of four transfer functions is known as sensitivity functions and is defined
as:
S(s): Nominal sensitivity function
574 N. Pati et al.

Fig. 5 Closed-loop Simulink model

T(s): Nominal complementary sensitivity

Si(s): Nominal input disturbance sensitivity
Su(s): Nominal control sensitivity
The sensitivity transfer function with respect to PI and Youla can be calculated
using Fig. 5 [2].

4 Simulated Results

The closed-loop structure in Fig. 3 is used to obtain the results. As converter is a

variable structure system, the control is achieved by duty ratio control. The results
for various cases are given in Figs. 6, 7, 8, 9, 10, 11, 12, 13 and 14.

Fig. 6 Converter output voltage for PI and Q-controller

Sensitivity Analysis of PI and Youla Controllers … 575

Fig. 7 Output voltage for change in irradiance (PI controller)

Fig. 8 Output voltage for change in irradiance (Youla controller)

The closed-loop controlled output of the converter using both the controllers
is shown in Fig. 6. Figures 7, 8, 9 and 10 show the output for change in solar
irradiance from 1000 to 800 to 1200 W/m2 at 0.1–0.2 s time interval. A 20% change
in load is carried out after 0.1 s which shows large variation in output using PI
controller. The control sensitivity function plot validates the result as for PI controller
the value is −68.5 dB whereas for Youla controller it is −33.5 dB. The sensitivity
plots give information for all the four nominal sensitivity functions which determines
the dominance of one controller over the other. Youla parameterization helps in
constructing a robust controller less sensitive to parameter variations.
576 N. Pati et al.

Fig. 9 Output voltage for change in load (PI controller)

Fig. 10 Output voltage for change in load (Youla controller)

5 Conclusion

The converter output is to be fed as input to inverter for a grid-connected PV system.

Therefore, output of the power converter needs to be regulated properly. This paper
had tried to compare two different controllers based on their ability to track the
reference and regulate the output under changing conditions. The Q-controller using
Youla parameterization shows exact tracking of set point with a minimum voltage
ripple of 2% as compared to PI whose output voltage ripple is around 5%. As PV
is the source, change in input due to change in irradiance is studied which reveals
no change in output using Q-controller and very slight changes for PI. Load change
variations are up to 0.15 s for a time interval of 0.05 s and settle down for Q, whereas
Sensitivity Analysis of PI and Youla Controllers … 577

Fig. 11 Closed-loop sensitivity analysis (S, T) PI

Fig. 12 Closed-loop sensitivity analysis (S, T), Youla

for PI controller the variations are large and are not settled at the reference value.
Further improvements can be incorporated to reduce the variations in output by
selection of proper Q.
578 N. Pati et al.

Fig. 13 Closed-loop sensitivity analysis (Si, Su) PI

Fig. 14 Closed-loop sensitivity analysis (Si, Su), Youla

References

1. Villalva, M.G., Gazoli, J.R., Filho, E.R.: Modelling and circuit based simulation of photovoltaic
arrays. In: 2009 Brazilian Power Electronics Conference, pp. 1244–1254 (2009)
2. Goodwin, G.C., Graebe, S.F., Salgado, M.E.: Control System Design. Prentice-Hall, Upper
Saddle River, NJ (2001)
3. Villalva, M.G., Gazoli, J.R., Filho, E.R.: Comprehensive approach to modeling and simulation
of photovoltaic arrays. IEEE Trans. Power Electron. 24(5) (2009)
4. Mahdavi, J., Emadi, A., Toliyat, H.A.: Application of state space averaging method to sliding
mode control of PWM DC/DC converters. In: Proceedings Conference Record IEEE IAS
Annual Meeting, vol. 2, pp. 820–827 (1997)
5. Su, J.-H., Chen, J.-J., Wu, D.-S.: Learning feedback controller design of switching converters
via MATLAB/SIMULINK. IEEE Trans. Educ. 45(4), 307–315 (2002)
Sensitivity Analysis of PI and Youla Controllers … 579

6. Mohan, N.: Power Electronics: Converters, Applications and Designs, 3rd edn. Wiley (2007)
7. Pati, N., Swain, N.: Application of H∞ controller to boost converter using model order
reduction. In: IEEE INDICON, pp. 1–6 (2015)
8. Ding, K., Bian, X.G., Liu, H.H., Peng, T.: A Matab-Simulink based PV module model and its
application under conditions of non uniform irradiance. IEEE Trans. Energy Convers. 27(4),
864–872 (2012)
9. Chan, C.-Y.: A nonlinear control for DC–DC power converters. IEEE Trans. Power Electron.
22(1) (2007)
10. Xiao, W., Dunford, W.G., Palmer, P.R., Capel, A.: Regulation of photovoltaic voltage. IEEE
Trans. Ind. Electron. 54(3), 1365–1374 (2007)
11. Pati, N.: A comparitive study of Youla and PID control algorithms for regulation of output
voltage of Buck Converter. In: IEEE 2nd International Conference on Electrical Energy Systems
(ICEES), pp. 267–271 (2014)
Improvement of Voltage Profile by Series
Active Filter in Grid-Connected PV
System

Rudranarayan Senapati, Sthita Prajna Mishra, Priyansha Sharma,

Rajendra Narayan Senapati, and Byomakesh Dash

Abstract Voltage fluctuation and harmonics in voltage has been the most significant
issue since the inception of renewable energy system into the grid network. Besides
this, the fluctuating pattern of load also makes worse the voltage profile. Under
such circumstances, the use of custom power device in the form of Series Active
Filter becomes the answer to the problem. The main objective of this work is to
improve the voltage profile through Series Active Filter implementing Sinusoidal
current control strategy. The strategy is a time-domain control strategy based on
instantaneous pq-theory. The control strategy has been elaborated here in detail and
has been implemented using MATLAB 2016A. The results have been given and
described in detail explaining the efficacy of the above control strategy. Since the
implemented strategy is a simple and effective control strategy, it has tremendous
potential for application in the Distributed Generation oriented system.

Keywords Power quality · Series active filter · Sinusoidal current control strategy

R. Senapati · P. Sharma
School of Electrical Engineering, KIIT Deemed to be University, Bhubaneswar, Odisha 751024,
India
e-mail: [email protected]
P. Sharma
e-mail: [email protected]
S. P. Mishra (B)
Department of EEE, GMR Institute of Technology, Rajam, Andhra Pradesh 532127, India
e-mail: [email protected]
R. N. Senapati
Bhubaneswar, Odisha, India
e-mail: [email protected]
B. Dash
Department of Electrical Engineering, Siksha ‘O’ Anusandhan (Deemed to be University),
Bhubaneswar, Odisha 751030, India
e-mail: [email protected]

© Springer Nature Singapore Pte Ltd. 2020 581

G. Pradhan et al. (eds.), Advances in Electrical Control and Signal Systems,
Lecture Notes in Electrical Engineering 665,
https://doi.org/10.1007/978-981-15-5262-5_43
582 R. Senapati et al.

1 Introduction

A usual assumption for most utilities is that the central generation utility produces
sinusoidal voltage. In the transmission system, voltage variation is less and may pos-
sibly be kept within the specified limit. But in distribution systems due to unbalanced
loading, at a large number of locations, the voltage distortions are significant. At sev-
eral load points, the current waveform rarely seems to be a sine wave. This anomaly
gives rise to the concept of harmonics, for the description of distortion in waveform
leading to the deterioration of the quality of electrical power with a decrease in the
efficiency of the system.
The following are the primary causes of voltage related issues:
1. Microprocessor-based and Microcontroller-based faster islanding and isolation.
2. High efficiency, adjustable-speed motor drives raising the level of harmonic
overpower systems.
3. Deregulation of utilities with reduced awareness of harmonic control and lower
reliability.
4. Highly interconnected network, where the failure of any component jeopardizes
the system stability.
5. Introduction of Distributed Generation (DGs) into the power with enhanced
harmonic levels.
The menace that runs through all the above reasons for rising stress over the
Power Quality (PQ) is due to the continuous drive from the manufacturer side for
the increase in productivity through faster, more productive and efficient machinery
for all utility customers who encourage the effort to make their customers more
profitable. The installed machineries and equipment suffer the most from common
power disruptions as well as they are the source of additional PQ issues. During
the entire process of automation, the competent operation of machineries and their
control moreover depends on the quality of power.

2 Power Quality

The phrase Power Quality (PQ) has grown into the most creative jargon due to electric
utilities and consumers as day by day all are more bothered about the quality of power.
From end-user point of view, PQ may be defined as any power issues illustrated with
voltage, current or frequency anomalies, bug the end user appliances. However,
there is almost no single accepted perception to appraise PQ, but it can ultimately be
measured by the capacity and act of the consumer appliances [1]. It is for sure that
the system is compact and reliable with the power electronics devices but leads to the
introduction of harmonics in the system. Use of power filters acts as a safeguard for
appliances overstating with destitute/low PQ and prevents the propagation resulting
instabilities in the power systems. The proposed sinusoidal current control strategies
Improvement of Voltage Profile by Series Active Filter … 583

have been realized to mitigate harmonics and various issues due to unbalanced or
unstable systems owed to non-linear loading. As based on converters, power filters
are connected in series or shunt to the distribution system, therefore, there are three
types of active filters, i.e. shunt active power filter (ShAPF), series active power filter
(SAF) and hybrid power filter (HPF).

3 Series Active Filter (SAF)

The SAF is connected in series with the power supply acts as a voltage booster com-
pensating voltage as a constant voltage source (CVS) [2, 3]. A single-phase SAF
has been designed to operate bidirectional without using dc capacitor with simpler
implementation in terms of hardware control. The SAF has two components, i.e.
PWM voltage control and active filter controller (AFC). SAF voltages are incorpo-
rated by the converter with a dc capacitor. The set voltage is calculated by the AFC
with load voltage and load current as the input signal. AFC will be processing the
signal for calculation of real-time instant compensating voltage to go through con-
tinuously into the PWM converter. The SAF operates in a closed-loop sensing the
voltage and computes the same for the compensating/series voltage reference value
Vc∗ for the PWM controller. A voltage source inverter (VSI) has been used in SAF for
its higher efficiency and compact size. A capacitor as an energy-storing component
connected at dc end of the converters for SAF acts as a compensator. In supplement,
the swapping of average energy continues to be zero between the power filter and
the system.

4 Sinusoidal Current Control Strategy (SCCS)

The design of the controller for SAF for non-sinusoidal and irregular supply volt-
ages and compensation of load voltage is carried by SAF assuring for a single opti-
mal compensation only. According to the choice of preferences, the design of the
controller for SAF is made as a major cause of inferring SCCS which is based on
instantaneous power theory depending on the transformation from abc to αβ0-frame.
But the reason for not adopting control in abc-frame is that, in 3-ϕ system, the three
phases are mutually dependent on each other, so independent control of the quantities
is difficult. Therefore, 3-ϕ quantities are converted into 2-ϕ mutually independent
quantities, for easier possible control in pq-domain, which is a stationary reference
frame as proposed by H. Akagi. The purpose of choosing this control strategy is its
simplicity in implementation. For a system with multiple renewable energy systems
integrated, implementing a robust control becomes cumbersome, as the control of
renewables requires lots of complexities [4].
The 3-ϕ quantities are used to evaluate the instantaneous powers in time domain,
both instantaneous active as well as reactive power can be estimated. By using low
584 R. Senapati et al.

pass filter, the harmonic power is extracted and used to generate the compensating
current with the known zero sequence power. The neutral point clamped capacitor
voltage is used to evaluate the zero sequence power. The above-mentioned method
is termed as SCCS due to sinusoidal nature of the compensating current.
Advantages of pq-theory over all other compensating theory is its validness for
both steady and transient state. Instantaneous power can be defined on αβ0-frame i.e.
in 3-ϕ form. So, 3-ϕ system can be considered as a single unit but not the addition of
three individual 1-ϕ circuits. The abc-frame to αβ0-frame transformation is known
as the Clarke transformation. For appropriate expression, zero sequence power, p0
is expressed in terms of αβ0-frame of reference, as the instantaneous watt power, p
and instantaneous VAR power, q are known from the instantaneous power theory.
Mathematically,
⎡ ⎤ ⎡ ⎤⎡ ⎤
p0 v0 0 0 i0
⎣ p ⎦ = ⎣ 0 vα vβ ⎦⎣ i α ⎦ (1)
q 0 vβ −vα iβ

Here, the 3-ϕ instantaneous active power is defined by both instantaneous active
power with the instantaneous zero sequence power. In the case of 3-ϕ 3-wire (3P3W)
system, instantaneous zero sequence power does not exist so for this type of system,
P3−φ can be treated as p only which may be noted as

P3−φ = vα i α + vβ i β (2)

p= p̄ + p̃ (3)
 
Average Value Oscillating Component
of the Active Power of the Active Power

From (3), it can be observed that instantaneous active power is divided into two
parts i.e. p̄ and p̃, where p̄ is referred to the average value or dc value of active power
implying total energy transfer in the system and p̃ defines the oscillating component
of active sequence power and the instantaneous imaginary power as:

1
q = vβ i α − vα i β = √ (vab i c + vbc i a + vca i b ) (4)
3

where, i a , i b , i c and va , vb , vc are the instantaneous current and voltage in abc-

frame, whereas i α , i β , i 0 and vα , vβ , v0 are the instantaneous current and voltage
in αβ0-frame. As the converters used nowadays are basically acts as non-linear load,
the energy flow between the systems has a boundary condition. Comparing to the
response of the converter and the generation of harmonic components and reactive
power with the conventional approaches the analysis of different types of power is
not sufficient using average or rms value as variables. So in a non-linear circuit,
Improvement of Voltage Profile by Series Active Filter … 585

time-domain analysis is carried out for analysis of energy flow. A 3-ϕ sinusoidal
voltage with positive and zero sequence voltages are considered for the realization
of zero sequence power. Symmetrical component in frequency domain is applicable
for steady-state operation. Hence, it can be converted into time domain for analysis
of both steady and transient state. For voltage, (5) is used as given
√ √ ⎫
va = √2V+ sin ωt + θv+ + 2V0√ sin ωt + θv0 ⎬
vb = √ 2V+ sin ωt − 2π + θv+ + √ 2V0 sin ωt + θv0 (5)
3 ⎭
vc = 2V+ sin ωt + 2π 3
+ θv+ + 2V0 sin ωt + θv0

whereas (6) used for current is as follows:

√ √ ⎫
i a = √2I+ sin ωt + θi+ + 2I0 √ sin ωt + θi0 ⎬
i b = √ 2I+ sin ωt − 2π + θi + √ 2I 0 sin ωt + θi (6)
3 + 0
⎭
i c = 2I+ sin ωt + 2π 3
+ θi+ + 2I0 sin ωt + θi0

In order to obtain zero sequence components, the above equation is required to be

converted into αβ0-frame by using Clarke transformation. For voltage, (7) is used
as follows:
√ ⎫
vα = + 3V+ sin ωt + θv+ ⎪ ⎪
√ ⎬
vβ = − 3V+ sin ωt + θv− (7)
√ ⎪
⎪
⎭
v0 = + 6V0 sin ωt + θv0

whereas (8) used for current is as follows:

√ ⎫
i α = + 3I+ sin ωt + θi+ ⎪
⎪
√ ⎬
i β = − 3I+ sin ωt + θi− (8)
√ ⎪
⎪
⎭
i 0 = + 6I0 sin ωt + θi0

From (7) and (8), the instantaneous zero sequence power can be obtained as

p = 3V0 I0 cos θv0 − θi0 − 3V0 I0 cos 2ωt + θv0 + θi0

⇒ p = p̄0 + p̃0 (9)

The instantaneous zero sequence power is divided into two parts consisting of
average power and oscillating component of power with double the line frequency.
Here p̄0 is unidirectional energy flow as conventional active power and p̃0 represents
the oscillating component with zero average value. The zero sequence power p̄0
cannot be obtained alone without the oscillating component p̃0 . Hence the total
zero sequence component always associated with both average as well as oscillating
component [6 and 7].
586 R. Senapati et al.

In the case of pq-theory, the instantaneous power so obtained is in time domain

without consideration of the rms value of voltage and current. It comprises of con-
ventional frequency-domain concept for which the theory is not contradictory rather
is a complementary theory in frequency domain.
The instantaneous zero sequence power component in the fundamental voltage
and current or in harmonics do not add any impact on instantaneous real power
and imaginary power. The total instantaneous active power is always the same as
the addition of instantaneous true power and instantaneous zero sequence power
includes both average and oscillating components. The instantaneous reactive power
reveals the energy exchanged between the system even in the harmonic and unbalance
condition. Imaginary power shows the energy exchanged between the phases, not
the energy transferred from one phase to another phase. The active and reactive
current components are derived from the instantaneous abc voltages and currents are
expressed as:
       
iα 1 vα vβ p 1 vα vβ 0
= 2 2 v −v
+ 2 2 v −v
(10)
iβ vα + vβ β α 0 vα + vβ β α q
   
Active Part Reactive Part

With the use of Inverse Clarke Transformation abc real and imaginary current
may be obtained as follows:
⎡ ⎤  ⎡ ⎤
i a( p) 1 √0  
v i + vβ i β vα
⎣ i b( p) ⎦ = 2 ⎢ 3 ⎥ α α
⎣ − 2 2√ ⎦
1
(11)
3 vα2 + vβ2 vβ
i c( p) − 21 − 23

and
⎡ ⎤ ⎡ ⎤
i a(q)   vbc
⎣ i b(q) ⎦ = v i + v i + v i
ca b ⎣
vca ⎦
ab c bc a
(12)
2
vab + vbc
2
+ vca
2
i c(q) vab

Also i a( p) , i b( p) , i c( p) and i a(q) , i b(q) , i c(q) are the real and imaginary current com-
ponents that generate real and imaginary power, respectively. The line voltage does
not contain any zero sequence component as: vab + vbc + vca = 0. As the line voltage
is free from zero sequence components, hence (12) and (13) can be re-written as:
⎡ ⎤ ⎡ ⎤ ⎡ ⎤ ⎡ ⎤
ia i0 i a( p) i a(q)
⎣ i b ⎦ = ⎣ i 0 ⎦ + ⎣ i b( p) ⎦ + ⎣ i b(q) ⎦ (13)
ic i0 i c( p) i c(q)
Improvement of Voltage Profile by Series Active Filter … 587

Fig. 1 Basic block diagram of 3-P-3-W SAPF

4.1 SCCC for SAF

The SAF compensates the current distortions resulted due to non-linear loads with an
appointment of high impedance path (by the generation of equal frequency voltage
as that of the current harmonics to be waived) enforcing the current of high frequency
to pass through the passive filter in parallel. The proposed work is carried out with
the simple control strategy based on instantaneous pq—theory over SAF, applied
on a 3P3W system. The basic block diagram of a 3P3W SAF for compensation of
voltage is illustrated in Fig. 1 and the control circuitry of SAPF in Fig. 2.
The SAF is for voltage compensation. The input to the control block is for cal-
culation of the instantaneous power with the phase voltages at the PCC and the line
currents of the non-linear load to be compensated. Assuming no zero sequence cur-
rent, the connection between source voltage, load voltage and active filter voltage
can be presented by
⎡ ⎤ ⎡ ⎤ ⎡ ⎤
vsa va vCa
⎣ vsb ⎦ = ⎣ vb ⎦ − ⎣ vCb ⎦ (14)
vsc vc vCc

The primary SAF voltages are incorporated by three 1-φ converters with a familiar
dc capacitor. The reference voltage for these converters is estimated by the AFC as
shown in Fig. 1, which has load voltages and currents (equal to the source currents)
588 R. Senapati et al.

Fig. 2 Control circuitry of SAPF

as input signals. Here the voltages are calculated by the dual pq-theory as given in
(15):
⎡ ⎤ ⎡ ⎤⎡ ⎤
p0 i0 0 0 v0
⎣ p ⎦ = ⎣ 0 i α i β ⎦⎣ vα ⎦ (15)
q 0 −i β i α vβ

From (15), the oscillating real power p̃ and the oscillating imaginary power q̃,
where the zero sequence powers p̄0 and p̃0 are assumed to be zero due to zero
sequence current. With these oscillating powers, the instantaneous voltages to be
injected by the SAF for load harmonic voltage compensation by using:
    
∗
vCα 1 i α −i β p̃
∗ = (16)
vCβ i α2 + i β2 i β i α q̃

A certain amount of  p̄ should be added to p̃ with an objective to compensate the

∗ ∗
losses. The reference voltage vCa and vCb can be transformed to the abc—reference
by
⎡ ∗
⎤  ⎡ ⎤ 
vCa 1 √ 0 ∗
⎣ v∗ ⎦ = 2 ⎣ −1/2 v
⎦ Cα (17)
Cb
∗ 3 √3/2 ∗
vCβ
vCc −1/2 − 3/2
Improvement of Voltage Profile by Series Active Filter … 589

The SAF generates the voltages mentioned in (17) for the harmonic voltage com-
pensation in the load producing oscillating active and reactive power at the load
end. This approach confirms the voltage from the source side has purely sinusoidal
∗
waveforms. The voltage vCabc needs to be obtained to compensate for the harmonic
component of load, producing the oscillating real and reactive power. The source
and load voltages thus obtained are purely sinusoidal in nature. The SAF integrated
into the system is implemented to take care of PQ issues and to inject grid as well
as load desired power. It consists of a PV system whose voltage is stepped up by a
boost converter.

5 Simulation Analysis

The simulation was carried using MATLAB/SIMULINK 2016a as per the control
strategy explained in Fig. 2 to verify the effectiveness of the proposed SCCS through
SAF on a 3P3W system for non-linear load in different environments. The specifica-
tions of loads are as follows: Load Resistance −60 , Load Inductance −0.15 mH,
Grid Voltage −230 V (RMS), DC-Link Capacitance–2000 μF, PV Voltage–104 V.
PV system is connected close to the load. The series active filter is meant to maintain
the load voltage so it is connected on the load side. A fixed RL load has been taken.
Figure 3a represents the performance of source voltage considering the SAF. This
SAF is the resulting source voltage characteristic in phase, and Fig. 3b presents
the performance of the source current of a 3ϕ3 W system. It is observed that the
SCCS in SAF maintains sinusoidal behaviour of source voltage while load voltage
harmonics are compensated. Because of non-linear load (i.e. RL rectifier circuit), the
load voltage was found to be distorted and unbalanced in absence of compensation,
but applying SCCS, the characteristics show the load voltage becomes balanced and
smooth as shown in Fig. 3a, b. The circuit breakers (CB) were given a time delay of
0.1 s to see the actual performance before the inception of the SAF into the circuit.
The non-linear behaviour of the load giving rise to distortions can be seen during
the first 0.1 s. After this, due to the effect of SAF on the system, the balanced and
distortion-free voltage can be observed. These waveforms disclose the compensation
of SAF for the disturbances at the source end. Figure 4a, b show the load voltage

(a) Source Voltages (b) Source Currents

Fig. 3 a Source voltages, b source currents

590 R. Senapati et al.

(a)Voltage (b)Current

Fig. 4 Load parameters a voltage, b current

and current waveforms respectively. The load voltage was found to have 231 V RMS
value and load current is found to be having an RMS value of 3.53 RMS value. SCCS
confirms the load voltage and current waveforms sinusoidal. Further, the strategy is
applied to variable grid voltage condition to see the performance of series active
filter compensation. Figure 5 shows the grid voltage at different instances. There is a
voltage sag between 0.5 and 1 s and voltage swell in between 1.5 and 2.0 s. But the
SAF is able to maintain the voltage level fixed at 311 V RMS. The injected voltage
by the series inverter is shown in Fig. 5. The load voltage is fixed due to the voltage
injected by the SAF as observed in Fig. 6. The injected voltage adds up to grid voltage

Fig. 5 Grid injected voltage

Fig. 6 Load voltage

Improvement of Voltage Profile by Series Active Filter … 591

during 0.5 to 1.0 s and nullifies from 1.5 s to 2.0 s. Here it is found that the response of
the SAF is quite fast and it tries to maintain the desired load voltage by injecting the
compensating voltage. The series injected voltage at the series transformer terminals
is presented in Fig. 7. There was an injection of voltage during voltage sag from 0.5 s
to 1 s. From 1.5 s to 2 s there was an interruption, taken care of by SAF by injection
of voltage during the same period. The harmonic analysis of load voltage is obtained
as seen in Fig. 8. The THD in load voltage is found to be 0.74%. The RMS value
of load voltage is 311.5 V. The Presence of higher order harmonics is suppressed
by SAF which can be seen from the THD analysis. The grid current is shown in
Fig. 9 and its harmonic analysis is given in Fig. 10. The THD in grid current was
found to be 22.35%. The RMS value of grid current was 8.623 A. Presence of 5th
order harmonics has been suppressed by SAF significantly. Its magnitude in respect
of fundamental is below 20%. The SAF based on SCCS assures against voltage sag
and swell. Besides this, it offers harmonic isolation to load voltage, as evident from
the results. The THD in load voltage is found to be 0.74% which is quite satisfactory.

Fig. 7 Series inverter injected voltage

Fig. 8 Harmonic analysis of load voltage

592 R. Senapati et al.

Fig. 9 Grid current

Fig. 10 THD of grid current

6 Conclusion

The conceptual study of SAF, using SCCS has been executed for 3-φ-3-W system
with passive and non-linear load. Nevertheless, the voltage and current in three Phases
experiences the disruption in the transient state, the said control strategy retreat by
drawing constant current by the load during steady-state condition. The load voltage
at the load terminal was obtained sinusoidal and balanced irrespective of sag, swell.

References

1. Senapati, R., Senapati, R.N., Behera, P., Moharana, M.K.: Performance analysis of unified
power quality conditioner in a grid connected PV system. In: International Conference on Signal
Processing, Communication, Power and Embedded System, pp. 416–420. IEEE (2016)
2. EL-Sadek, M.Z., Ibrahim, E.A., Kamel, R.M.: Series active filter for voltage harmonic sup-
pression and 50HZ voltage boosting. In: Power Systems Conference, MEPCON 2006. Eleventh
International Middle East, vol. 1, pp. 59–64. IEEE (2006)
3. Aredes, M., Watanabe, E.H.: New control algorithms for series and shunt three-phase four-wire
active power filters. IEEE Trans. Power Deliv. 10(3), 1649–1656 (1995)
Improvement of Voltage Profile by Series Active Filter … 593

4. Senapati, R.: Study and analysis of performance of 3-phase shunt active filter in Grid-tied PV-fuel
cell system employing sinusoidal current control strategy. Int. J. Renew. Energy Res. (IJRER)
8(1), 67–81 (2018)
5. Peng, F.Z., Akagi, H., Nabae, A.: A new approach to harmonic compensation in power systems-a
combined system of shunt passive, series active filters. IEEE Trans. Ind. Appl. 26(6), 983–990
(1990)
6. Senapati, R., Senapati, R.N., Moharana, M.K.: Sinusoidal current control strategy for UPQC in
grid connected PV-fuel cell microgrid. Int. J. Eng. Technol. 9(4), pp. 2800–2813 (2017)
Performance Improvement of Solar
Photovoltaic Array by Modified Bridge
Link Array Configuration

Chidurala Saiprakash, Alivarani Mohapatra, and Byamakesh Nayak

Abstract In this paper, a new tie-connected array configuration is proposed by

modifying the conventional bridge link (BL) array configuration. The BL array con-
figuration is popular due to the minimum number of tie connections, which reduces
the cable losses and yields the maximum power in uniform irradiance condition as
compared to the other conventional tie connected configurations. But in partial shad-
ing condition, BL array configuration is more prone to mismatch losses and gives
less power in comparison with total-cross-tied (TCT) array configurations. In this
paper, the BL array configuration is modified by considering the reduced number
of tie connections with improved performance. The performance of the modified
BL array configuration is compared with the BL array configuration in terms of the
maximum output power, shading loss, and mismatch loss. From the obtained results,
it is observed that the modified BL array configuration outperforms conventional BL
array configuration in all the cases of shading patterns which are considered in this
paper.

Keywords Solar photovoltaic · PV array configurations · Partial shading · PV

characteristics · Series–parallel (SP) · Bridge link (BL) · Honeycomb (HC) · Total
cross-tie (TCT)

C. Saiprakash (B) · A. Mohapatra · B. Nayak

School of Electrical Engineering, KIIT Deemed to Be University, Bhubaneshwar, Odisha 751024,
India
e-mail: [email protected]
A. Mohapatra
e-mail: [email protected]
B. Nayak
e-mail: [email protected]

© Springer Nature Singapore Pte Ltd. 2020 595

G. Pradhan et al. (eds.), Advances in Electrical Control and Signal Systems,
Lecture Notes in Electrical Engineering 665,
https://doi.org/10.1007/978-981-15-5262-5_44
596 C. Saiprakash et al.

1 Introduction

In the past few years, the usage of renewable energy sources is increasing gradually
due to the extinction of non-renewable energy sources like fossil fuels. These non-
renewable energy sources worsen the environment, and also, it is a burden to maintain.
The most abundant energy source is sunlight [1]. Sunlight is converted into electrical
energy by solar energy captured devices known as a photovoltaic (PV) cell. The
cluster of PV cell forms a PV module, and these PV modules connected to form
a PV array configuration. Keeping aside its advantages few cons such as partial
shading which causes the mismatch losses in the PV characteristics. During non-
uniform irradiance condition, it is very difficult to extract maximum power from the
PV array. Therefore, the efficiency of the PV system is get reduced [2, 3].
In partial shading condition, one or some of the modules in an array receives
variance irradiance levels, which will impact on the remaining modules. As a result,
the power extraction from the array will drastically reduce [4]. Partial shading effect
on the PV system can be reduced by considering the proper array configurations.
The existing accessible array configurations are series–parallel (SP) [5], bridge link
(BL) [6], total cross-tied (TCT) [7], and honeycomb (HC) array configuration [8].
In partial shading condition, the power extraction from the conventional SP con-
figuration is less due to the effect of mismatch losses. Under the partial shading
condition, alternative tie connected configurations are instigated to get better perfor-
mance. Under these tie configurations, BL configuration has minimum number of
tie connections than the other tie connected configuration, which will give the lower
cable losses and yields the maximum power in the uniform irradiance condition. But
BL configuration provides poor performance in the partial shading condition [9].
To improve the performance of the BL array configuration, in this paper modi-
fied BL array configuration is proposed with minimum number of tie connections
in comparison with conventional BL array configuration. The performance of the
proposed configuration is compared with the conventional BL array configuration
under different shading patterns, and its superiority has been discussed. The study
has been carried out in MATLAB–Simulink environment and from the obtained PV
characteristics; the performance of the modified BL array configuration is analyzed.
Compared to conventional BL, the proposed design gives more power with reduced
mismatch power loss.

2 PV Array Configurations

In the solar PV system to get the desired output power, appropriate array configura-
tions have to be preferred. The most widely connected configuration is SP configu-
ration. In this configuration, modules are allied in series as a string and then combine
these in parallel to configure the array. SP configuration is shown in Fig. 1a [10].
Coming to the partial shading condition, SP configuration will affect more and gives
Performance Improvement of Solar Photovoltaic Array … 597

Fig. 1 Solar array configurations of a SP b TCT c BL d HC e Modified BL

an inferior performance. So, alternate configurations come into pictures; these are
total cross-tie (TCT), bridge link (BL), and honeycomb (HC) [11].
For TCT configuration, first of all modules are connected in parallel and the
parallel string is connected in series. It seems like a parallel–series configuration. This
arrangement is shown in Fig. 1b. The presence of large number of tie connections TCT
exhibits more cable losses. Therefore, to reduce cable losses, in BL configuration
the tie connections are reduced to half [12].
Figure 1c represents the BL array configuration. In this configuration, modules
are configured in a 2 × 2 bridge rectifier manner. Even this configuration has less
number of tie connection compared to conventional tie configuration, it gives the
poor performance under the partial shading condition [13].
To improve the performance of the BL configuration, further it is modified into
HC configuration by changing the variable bridge sizes to 1 × 2 and 3 × 2, and this
configuration is shown in Fig. 1d [14]. But for HC configurations, the tie connections
are increased than the BL configuration.
In this paper, a modified bridge link array configuration has been proposed by the
reduced tie connections with the improved performance. In this configuration, the
bridge unit is modified into 3 × 2 instead of 2 × 2 in the bridge link configuration.
This modified configuration is shown in Fig. 1e. And the performance comparison
of BL and modified BL under various shading patterns is analyzed.

3 Analysis of Different Shading Patterns

3.1 Shading Patterns

The performance of the presented configurations is analyzed under partial shading

conditions for various shading patterns. All considered shading conditions are carried
out at 25 °C. For these configurations, 36 modules are considered and these are
arranged in 6 × 6 symmetrical array configurations.
In normal irradiance condition (STC), all the modules are considered at irradiance
levels of 1000 W/m2 and temperature at 25 °C.
598 C. Saiprakash et al.

For pattern-1 shading condition, column-wise shading pattern is considered. For

this, first column of an array is shaded, in that first half of the column modules receive
the irradiance of 500 W/m2 , another half of the modules in the first column get the
irradiance of 700 W/m2 , and remaining modules in an array receive the uniform
irradiance of 1000 W/m2 .
For pattern-2 shading condition, column-wise shading pattern is taken. For this,
the first two columns of an array are shaded; in that first column modules receive the
700 W/m2 of irradiance and the second column modules receive the 500 W/m2 of
irradiance.
For pattern-3 shading condition, row-wise shading pattern is considered. For this,
two rows from the bottom of the array receive the irradiance of 700 W/m2 .
For pattern-4 shading condition, row-wise shading pattern is considered. For this,
the first row from the bottom of the array receives the irradiance of 700 W/m2 , and
second row from the bottom gets the irradiance of 500 W/m2 .
For pattern-5 shading condition, diagonal shading pattern is considered. For this,
one of the diagonal modules in the array receives the irradiance of 700 W/m2 .
All these shading patterns are shown in Fig. 2.

1000 W/m2 700 W/m2 500 W/m2

Fig. 2 Various shading patterns at different irradiance levels
Performance Improvement of Solar Photovoltaic Array … 599

3.2 Simulation Model

The presented 6 × 6 array size of modified BL and BL configurations are modeled

in the MATLAB–Simulink environment. These simulated array configurations are
shown in Figs. 3 and 4. For this simulation model, PV module consisting of 54 PV cell
of KC200GT PV module is considered [15]. The parameter values of the considered
PV module are represented in Table 1.

Fig. 3 Simulation model of the bridge link array configuration

Fig. 4 Simulation model of the modified bridge link configuration

600 C. Saiprakash et al.

Table 1 Parameter values of

Module parameters Values
the PV module (KC200GT)
[15] Max. power (Pmpp ) 200 W
Current at max power (I MPP ) 7.61 A
Diode ideality factor (a) 1.3
Voltage at max power (V MPP ) 26.3 V
Series resistance (Rse ) 0.221 
Short-circuit current (I sc) 8.21 A
Shunt resistance (Rsh ) 415.405 
Open-circuit voltage (V oc ) 32.9 V
Number of cells connected in series in module 54
(N s )
V OC /K V (temperature coefficient) −0.123 V/K
I SC /K I (temperature coefficient) 0.0032 A/K

4 Results and Discussion

The PV characteristics of the modified bridge link configuration in comparison

with BL configuration is studied in the MATLAB. From the simulation results, the
obtained P-V curves for the presented configurations are shown in Fig. 5, and the
resultant maximum power values are given in Table 2.
From the PV characteristics, it is observed that the changes in the shading patterns
are affecting the performance output of the PV system. In the row-wise shading
pattern and diagonal shading pattern, the performance output is getting affected
more than the column-wise shading pattern. The modified BL array configuration,
yield maximum power values of 7164.98 W, 6631.65 W, 6177.95 W, 5530.91 W,
4690.86 W, 6461.62 W under shading patterns of uniform and pattern-1 to pattern-5
in that order. From the observed results, it is seen that the modified BL configuration
giving the highest maximum power than the BL array configuration under all the
considered shading patterns.

4.1 Shading Loss

Shading loss in PV array is defined as the difference of maximum power available

at STC to the sum of individual modules possible maximum power under partial
shading condition [16]. Shading loss is calculated in Eq. (1).

PSL = Pmax(STC) − Pmax(illuminated modules) (1)

In Eq. (1), PSL , Pmax(STC) , and Pmax(illuminated module) represent the shading loss,
maximum power available at STC, and sum of possible maximum power in individual
Performance Improvement of Solar Photovoltaic Array … 601

8000 7165 BL 7000 BL

Modified BL Modified BL

7000
6630
7160
6000 6620

Power (W)
6000
Power (W)

7155

5000 6610

5000 154 156 158 160

4000 154 156 158 160

4000
3000 3000
2000 2000
1000 1000
0 0
0 20 40 60 80 100 120 140 160 180 0 20 40 60 80 100 120 140 160 180
Voltage (V)
Voltage (V)
Uniform irradiance Shading pattern-1
7000 6178 BL
7000 BL

Modified BL Modified BL
5530

6000 6176 6000 5525

5520

Power (W)
5000 6174
5000 5515
Power (W)

6172
4000 155 156 157 158
4000
165 166 167

3000 3000
2000 2000
1000 1000
0 0
0 20 40 60 80 100 120 140 160 180 0 20 40 60 80 100 120 140 160 180 200
Voltage (V) Voltage (V)

pattern-2 Shading pattern-3

6000 BL
8000
BL
6600

7000
Modified BL Modified BL
4700
6400

5000 6200
Power (W)

4680
6000
Power (W)

6000

5800
4000 138 139 140 141 142
5000 5600

120 130 140 150 160

3000 4000
3000
2000
2000
1000 1000
0 0
0 20 40 60 80 100 120 140 160 180 0 20 40 60 80 100 120 140 160 180 200
Voltage (V) Voltage (V)
Shading pattern-4 Shading pattern-5

Fig. 5 PV characteristics of BL and modified BL configuration under different shading patterns

Table 2 Maximum power (W) output of array configuration under various shading patterns
Configuration Maximum available power(W) under different shading patterns
Uniform Pattern-1 Pattern-2 Pattern-3 Pattern-4 Pattern-5
Bridge link 7163.21 6620.40 6173.18 5526.42 4682.36 5924.67
Modified BL 7164.98 6631.65 6177.95 5530.91 4690.86 6461.62

modules under partial shading condition. From the calculated values, it is observed
that shading loss is the same in the given array configurations for each shading
pattern. The calculated values of shading loss are given in Table 3.
602 C. Saiprakash et al.

Table 3 Shading loss (W) values of array configuration under various shading patterns
Configuration Shading loss (W) under different shading patterns
Pattern-1 Pattern-2 Pattern-3 Pattern-4 Pattern-5
Bridge link 455.72 946.46 699.02 946.46 332.10
Modified BL 455.72 946.46 699.02 946.46 332.10

Table 4 Mismatch loss (W) values of array configuration under various shading patterns
Configuration Mismatch loss (W) under different shading patterns
Pattern-1 Pattern-2 Pattern-3 Pattern-4 Pattern-5
Bridge link 88.86 45.34 939.54 1536.16 908.31
Modified BL 77.61 40.57 935.05 1527.66 371.36

4.2 Mismatch Loss

Mismatch loss in PV array is defined as the difference of the sum of possible maxi-
mum power in individual modules under partial shading condition to the array global
maximum power point under partial shading condition [17]. The mismatch loss is
calculated in Eq. (2).

PML = Pmax(illuminated modules) − PGMPP (2)

In Eq. (2), PML , Pmax(illuminated module) , and PGMPP represent the mismatch loss, sum
of possible maximum power in individual modules under partial shading condition,
and the available global maximum power point under partial shading condition. From
the calculated values, it is observed that the modified BL array configuration having
the less mismatch loss in all considered shading conditions. The calculated values
of mismatch loss are given in Table 4.

5 Conclusion

In this paper, the performance evaluation of the modified bridge link array configura-
tion is compared with the conventional bridge link array configuration under different
shading patterns. For the performance analysis, the simulation model of the 6 × 6
BL configuration and modified BL array configurations under various shading con-
ditions have been implemented in the MATLAB–Simulink environment. From the
obtained results, it is observed that the performance of the modified BL array config-
uration is enhanced in terms of maximum available output power as compared with
Performance Improvement of Solar Photovoltaic Array … 603

conventional BL configuration under all considered shading patterns. The mismatch

loss is found less for the proposed modified BL array configuration in comparison
with the conventional BL array configuration which proves its superiority.

References

1. Ellabban, O., Abu-Rub, H., Blaabjerg, F.: Renewable energy resources: current status, future
prospects and their enabling technology. Renew. Sustain. Energy Rev. 39, 748–764 (2014).
https://doi.org/10.1016/j.rser.2014.07.113
2. Mohapatra, A., Nayak, B., Das, P., Mohanty, K.B.: A review on MPPT techniques of PV system
under partial shading condition. Renew. Sustain. Energy Rev. 80, 854–867 (2017). https://doi.
org/10.1016/j.rser.2017.05.083
3. Mohapatra, A., Nayak, B., Mohanty, K.B.: Performance improvement in MPPT of SPV system
using NN controller under fast changing environmental condition. In: 2016 IEEE 6th Interna-
tional Conference on Power Systems (ICPS), pp. 0–4 (2016). https://doi.org/10.1109/ICPES.
2016.7584159
4. Kumari, P.A., Geethanjali, P.: Parameter estimation for photovoltaic system under normal and
partial shading conditions: a survey. Renew. Sustain. Energy Rev. 84, 1–11 (2018). https://doi.
org/10.1016/j.rser.2017.10.051
5. Bingöl, O., Özkaya, B.: Analysis and comparison of different PV array configurations under
partial shading conditions. Sol. Energy (2018). https://doi.org/10.1016/j.solener.2017.12.004
6. Jha, V.: Triar US experimental verification of different photovoltaic array configurations under
partial shading condition. In: 2017 IEEE International WIE Conference on Electrical and
Computer Engineering (WIECON-ECE), 18–19 December (2017). https://doi.org/10.1109/
WIECON-ECE.2017.8468929
7. Wongwuttanasatian, T.: Effect of partial shading patterns and degrees of shading on Total Cross-
Tied (TCT) photovoltaic array configuration. Energy Procedia 153, 35–41 (2018). https://doi.
org/10.1016/j.egypro.2018.10.028
8. Mishra, N., Yadav, A.S., Pachauri, R., et al.: Performance enhancement of PV system using
proposed array topologies under various shadow patterns. Sol. Energy 157, 641–656 (2017).
https://doi.org/10.1016/j.solener.2017.08.021
9. Malathy, S., Ramaprabha, R.: Reconfiguration strategies to extract maximum power from
photovoltaic array under partially shaded conditions. Renew. Sustain. Energy Rev. 1–13 (2017).
https://doi.org/10.1016/j.rser.2017.06.100
10. Pendem, S.R., Mikkili, S.: Modeling, simulation and performance analysis of solar PV array
configurations (Series, Series–Parallel and Honey-Comb) to extract maximum power under
partial shading conditions. Energy Rep. 4, 274–287 (2018). https://doi.org/10.1016/j.egyr.2018.
03.003
11. Nayak, B., Mohapatra, A., Das, P.: Optimal hybrid array configuration scheme to reduce
mismatch losses of photovoltaic system (2016)
12. Kho, C.T.K., Ahmed, J., Kashem, S., Then, Y.L.: A comprehensive review on PV configurations
to maximize power under partial shading. In: IEEE Region 10 Annual International Conference,
Proceedings/TENCON (2017)
13. Ramaprabha, R., Mathur, B.L.: A comprehensive review and analysis of solar photovoltaic array
configurations under partial shaded conditions (2012). https://doi.org/10.1155/2012/120214
14. Bana, S., Saini, R.P.: Experimental investigation on power output of different photovoltaic
array configurations under uniform and partial shading scenarios. Energy 127 (2017). https://
doi.org/10.1016/j.energy.2017.03.139
15. Mohapatra, A., Nayak, B., Saiprakash, C.: Adaptive perturb & observe MPPT for PV system
with experimental validation. In: 2019 IEEE International Conference on Sustainable Energy
Technologies and Systems, pp. 257–261 (2019). https://doi.org/10.1109/icsets.2019.8744819
604 C. Saiprakash et al.

16. Vijayalekshmy, S., Bindu, G.R., Iyer, S.R.: A novel Zig-Zag scheme for power enhancement
of partially shaded solar arrays. Sol. Energy 135, 92–102 (2016). https://doi.org/10.1016/j.
solener.2016.05.045
17. Das, S.K., Verma, D., Nema, S., Nema, R.K.: Shading mitigation techniques: state-of-the-art
in photovoltaic applications. Renew. Sustain. Energy Rev. 78, 369–390 (2017). https://doi.org/
10.1016/j.rser.2017.04.093
A Simple Decision Tree-Based Efficient
Fault Classifier for PV-Based AC
Microgrid

Kanche Anjaiah and Rajesh Kumar Patnaik

Abstract Fault detection in the microgrid is necessary for control and smooth oper-
ation, as it enables the system for fast fault isolation and recovery. This paper presents
an efficient fault detection algorithm for different types of faults which occurred in
the distribution network of the PV-based AC microgrid. Different types of faults
are created in the PV-based AC microgrid with single and double feeders. By using
MATLAB/SIMULINK tools, the extracted fault currents are passed through change
detection filter (CDFT), to extract the feature sets which sets consists an entropy,
root mean square, mean and maximum peak, etc. After the analysis of target feature
sets-based CDFT, total six features are distinctively classified. Eventually all the
distinctively classified feature sets are proceed to a simple decision tree-based fault
classifier, for the identification of the type of fault current. The proposed technique
has been scrutinized for number of faults presented in the PV-based AC micro-
grid. The proposed technique for fault analysis has been demonstrated as a better-
quality technique as compared with the prevailing fault detection techniques based
on wavelet-based deep neural networks and support vector machines. The proposed
signal processing algorithm avoids the essential of additional processing technique
for the detection of various faults in the AC microgrid.

Keywords CDFT · PCC · PV · VSC

1 Introduction

As witnessed in the last few decades, a number of factors such as power demand,
depletion of conventional energy sources and highly developed technology called up
for the execution of distributed energy resources (DERs) [1]. The distributed energy
resources are renewable energy resources. From these sources, solar PV plays a vital
role in the power generation and this energy is necessary to meet energy demand;
nowadays, people are concentrated on DERs-based low-voltage distribution network,

K. Anjaiah (B) · R. K. Patnaik

GMRIT, Rajaam, India
e-mail: [email protected]

© Springer Nature Singapore Pte Ltd. 2020 605

G. Pradhan et al. (eds.), Advances in Electrical Control and Signal Systems,
Lecture Notes in Electrical Engineering 665,
https://doi.org/10.1007/978-981-15-5262-5_45
606 K. Anjaiah and R. K. Patnaik

i.e., microgrids, which are coupled with the transmission network at the point of
common coupling (PCC). According to the IEEE standard 1547.4, the microgrids
must have the capability and reliability to operate in both the modes (i.e., online
mode and islanded mode) [2].
Due to the presence of power electronic converters in the power system, it fails to
fault location and control, especially in the microgrid applications [3]. Most possible
faults in the microgrid are P-G, P-P, P-P-G, P-P-P, and P-P-P-G (i.e., P means pole, G
means ground and—means to). Fault between lines termed as the pole to pole faults
and fault between lines to ground is termed as the pole to ground [4]. According to
the literature survey, few methods for type of fault detection and location are discrete
Fourier transform (DFT) [5], discrete wavelet transform (DWT) [6], traveling wave
or injection-based algorithm [7], support vector machine and k-nearest neighbors
algorithm [8, 9], wavelet-based deep neural network algorithm [10]. These methods
are unable to find the exact type of fault; traveling wave or injection-based algorithm
suffers from the reflected wave detection and discrimination [7], and it is limited
for pole to ground faults only. Wavelet-based deep neural network algorithm is not
applicable for the noise in the current signals [10]. Rest of the algorithms required
synchronized data and over communication links.
To bridge the research gap in the fault detection for online solar PV-based micro-
grid, in this paper a new algorithm, i.e., change detection filter (CDFT), is proposed
for the detection of the type of fault in the microgrid. The major benefit of the pro-
posed technique is: It does not call for any extra signal processing algorithm, com-
putational complexity takes away, reduces the economical cost, additional accuracy,
and it makes the faster execution.
Primarily, the voltage and current samples near the target load feeders in the
planned model (i.e., PV-based) have been extracted, and these current samples are
computed to produce the preliminary feature set. These feature sets consist of energy,
entropy, mean, standard deviation, root mean square, kurtosis, maximum peak, and
variance [12] with the help of CDFT, for all cases as described in Sect. 3. After
produced, feature sets have been alienated with the distinct values of each feature from
afore-mentioned features. After analyzing different faults with the above-mentioned
feature set for all the cases, the target feature set has been fashioned which contains
entropy (Z 1 ), root mean square (Z 2 ), mean (Z 3 ), kurtosis (Z 4 ), variance (Z 5 ), and
maximum peak (Z 6 ). These feature sets subjected to a decision classifier as described
in Sect. 6.

2 Analysis of Change Detection Filter (CDFT)

Change detection filter (CDFT) will identify the minor changes in the signal, and
this technique working principle is based on reference signal; this signal every time
compares with every cycle in the signal. If the signal is mismatched with reference
signal, it will show error otherwise zero; this phenomena is explained below.
A Simple Decision Tree-Based Efficient Fault Classifier … 607

Fig. 1 Sine wave with 12 samples

Figure 1 has a reference cycle, which has been divided with twelve samples per
cycle samples are indicated with ‘T ’

T (1)−T (13) = 0
T (2)−T (14) = 0
.. ..
. .
T (11)−T (23) = 0
T (12)−T (24) = 0
T (13)−T (25) = 0
.. ..
. .
T (17)−T (29) = 0
T (18)−T (30) = 0

The above all expressions, i.e., from T 1 to T 24, indicated as zero; due to amplitude
is indistinguishable, at this situation CDFT shows on zero line. The outcome from
T 25 to T 30 indicates that amplitude is nonzero, because they are distinguishable.
From Fig. 1, it can observe all these cases. CDFT is easy to spot out the fault location
for plenty cycles in a signal.

3 Simulation-Based Solar PV Microgrid

The solar PV farm produces voltage, and currents are passed through DC-DC con-
verter; it will make the oscillated DC to flat DC, and this unchanged DC is fed to the
voltage source converter (VSC). It converts the DC-to-AC voltage, and then it will
boost up the voltage with transformer (T /F1); the outcome voltage from transformer
is 25 kV, and it connected to the grid. A feeder between grid and utility is 14 km
distance, and the feeder between grid and load is 8 km. Three-phase faults are formed
in these feeders in MATLAB/SIMULINK for analyzing the various faults in the AC
microgrid. Afore-mentioned process is indicated in Fig. 2, common faults in the AC
608 K. Anjaiah and R. K. Patnaik

DC Grid
V L 8KM Feeder
S Load
DC C
T/F1
Solar Farm 3-phase fault

Utility 14KM Feeder

120KV
2500MVA T/F2 3-phase fault

Fig. 2 Block diagram for PV-based with faults

microgrid distribution networks are [11].

1. Pole to ground (P-G)
2. Pole to pole (P-P)
3. Pole to pole to ground (P-P-G)
4. Pole to pole to pole (P-P-P)
5. Pole to pole to pole to ground (P-P-P-G)

3.1 Fault at 8 km Feeder

In this case, faults are created in the only one feeder, i.e., 8 km feeder, in the solar
PV-based microgrid for different irradiations. After creation of faults at 8 km feeder
in the MATLAB/SIMULINK, it will be simulated to get voltage and currents; at
the grid, these signals are passed through proposed algorithm (i.e., change detection
filter). With the help of target feature sets, distinctive faults are recognized (i.e.,
shown in Fig. 3) with distinctive values of target feature sets as shown in Table 1.
These distinctively classified events are passed through decision classifier.
Figure 3 shows the 400 irradiance of the PV-based microgrid shows very high
magnitude from all the graphs that means the fault current is more when low irradi-
ance. Similarly, it also shows that with the 1000 irradiance P-G fault shows very low
values to identify that fault is accurate; it is crucial a signal processing algorithm.
With the proposed algorithm (CDFT), it is very easy to identify the fault with the
help of target feature sets [12].

3.1.1 Target Feature Sets

i. Entropy (Z1 ):
A Simple Decision Tree-Based Efficient Fault Classifier … 609

8.00E+02

6.00E+02 400

4.00E+02 600
800
2.00E+02
1000
0.00E+00
Z1 Z2 Z3 Z4 Z5 Z6
A

5.00E+03

4.00E+03
400
3.00E+03
600
2.00E+03
800
1.00E+03 1000
0.00E+00
Z1 Z2 Z3 Z4 Z5 Z6
B
6.00E+02
5.00E+02
4.00E+02 400

3.00E+02 600

2.00E+02 800

1.00E+02 1000
0.00E+00
Z1 Z2 Z3 Z4 Z5 Z6
C

2.00E+03

1.50E+03
400

1.00E+03 600
800
5.00E+02
1000

0.00E+00
Z1 Z2 Z3 Z4 Z5 Z6
D

Fig. 3 Different types of faults with different irradiations (A: P-G, B: P-P, C: P-P-G: P-P-P, E:
P-P-P-G)
610 K. Anjaiah and R. K. Patnaik

1.00E+03
8.00E+02
400
6.00E+02
600
4.00E+02
800
2.00E+02
1000
0.00E+00
-2.00E+02 Z1 Z2 Z3 Z4 Z5 Z6
E

Fig. 3 (continued)

Table 1 Target features extracted for solar PV-based microgrid when fault is at 8 km feeder
Target features
F IRR Z1 Z2 Z3 Z4 Z5 Z6
P 1000 3.29 7.70 1.4*10−18 1.56*103 59.3 1.32*102
G 800 3.31 7.69 3.33*10−5 1.57*102 59.2 1.32*102
600 3.32 7.69 3.30*10−5 1.57*102 59.1 1.32*102
400 3.32 7.69 3.30*10−5 1.57*102 59.1 1.32*102
P 1000 3.28 1.11 1.4*10−18 8.86*102 1.34 2.73*101
P 800 3.26 1.05 3.33*10−5 9.77*102 1.24 2.66*101
600 3.26 1.00 3.33*10−5 1.04*103 1.17 2.68*101
400 3.25 0.985 3.33*10−5 1.06*103 1.15 2.70*101
P 1000 3.67 4.38 1.4*10−18 1.11*102 19.2 5.47*101
P 800 3.63 4.21 3.33*10−5 1.30*102 17.7 5.50*101
G
600 3.65 4.20 3.33*10−5 1.30*102 17.6 5.52*101
400 3.65 4.19 3.33*10−5 1.30*102 17.6 5.52*101
P 1000 3.74 1.29 2.9*10−19 5.83*102 1.74 2.96*101
P 800 3.59 1.47 3.3*10−5 4.03*102 2.22 3.07*101
P
600 3.46 1.76 3.33*10−5 2.30*102 3.15 3.31*101
400 3.28 1.24 3.30*10−5 6.65*102 1.62 3.01*101
P 1000 3.68 2.77 −3.3*10−5 2.11*102 7.73 6.301
P 800 3.53 2.86 −3.3*10−5 1.88*102 8.20 6.29*101
P
G 600 3.4 3.01 −6.7*10−5 1.56*102 9.14 6.29*101
400 3.22 2.75 −6.7*10−5 2.21*102 7.60 6.11*101

For a signal Z(p) with samples p = 1, 2, 3, 4, 5, …, k, its entropy is defined as

It is the average amount of random variable of the signal information from
the disorder signal or source signal
A Simple Decision Tree-Based Efficient Fault Classifier … 611


k
Entropy = Z 1 = C p logb C p (1)
p=1

ii. Root mean square (RMS) (Z2 ):

RMS is equal to DC current, and it can be defined as square root of the mean
squares of the signal

 k
1 
RMS = Z 2 =  (C 2 + C22 + · · · + C 2p ) (2)
k p=1 1

iii. Mean (Z3 ):

The ratio of the total of the samples to the total number of samples

k 
 
Cp
Mean = Z 3 = (3)
p=1
k

where k = total number of samples

iv. Kurtosis (Z4 ):
It is the quantifying of the signal whether the data are sharp or smooth relative
to normal distribution. It is also defined as the combined weight of the samples
related to the rest of the signal.

1
n
Kurtosis = Z 4 = ((C p − Z 3 )4 /Z S4 (4)
k p=1

where Z s is the standard deviation

v. Variance (Z5 ):
It is defined as the ratio of dissimilarity between the squared sums of the direct
samples and distribution mean (µ) to the number of terms in the distribution (D)


n
Variance = Z 5 = (C p − µ)2 /D (5)
p=1

vi. Maximum Peak Amplitude (Z6 ):

The signal having peak amplitude at a point or output of the signal is called
maximum peak amplitude.

Target feature sets are distinctively classified with the rule base model. Depending
on these features, classification of faults is done in Fig. 4.
From Fig. 4, it is clear that the P-P-P-G fault is more effective and it is clear
to identify; but P-G fault is very low, it is not that much easy to identify without
proper algorithm. Hence, the proposed algorithm will detect the minor changes in
612 K. Anjaiah and R. K. Patnaik

2500

2000
P-P-P-G
1500
P-P-P
1000 P-P-G
P-P
500
P-G
0
Z1 Z2 Z3 Z4 Z5 Z6
-500

Fig. 4 Analysis of different faults with target features (Z 1 –Z 6 )

the current waveform that will be illustrated with the help of the following decision
tree as discussed in Sect. 4. From Table 1, it will provide clear information for all
types of faults with its target features.

3.2 Fault at Both the Feeders

In this case, faults are created at both the feeders, i.e., 8 and 14 km feeders, in the
PV-based microgrid for different irradiations. After creation of faults at 8 km and
14 km feeders in the MATLAB/SIMULINK, it will be simulated to get voltage and
currents at the grid, and these signals are subjected to proposed algorithm (i.e., change
detection filter).
With the help of target feature sets as afore-mentioned, the distinctive faults are
recognized (i.e., shown in Fig. 5) with distinctive values of target feature sets as
shown in Table 2. These different faults are subjected to decision classifier. Different
types of faults are shown in Fig. 5 for PV-based microgrid when the faults are both
the feeders.

4 Simple Decision Tree-Based Fault Classifier

Figure 5a, c, e differs from Fig. 5b, d, and the main variation in Fig. 5a, c, e maintains
the constant in the magnitude of current for all irradiances. But rests of the faults
are continuously oscillating. The 400 irradiance of the PV-based microgrid shows
very high magnitude due to its low irradiance. Similarly with the 1000 irradiance,
P-G fault shows very low values to identify that fault is accurate. It is necessary a
signal processing algorithm. With the proposed algorithm (CDFT), it is very easy
to identify with the help of target feature sets. From all the faults, it is very clear
that irradiance gradually decreases and then fault current gradually increases. From
A Simple Decision Tree-Based Efficient Fault Classifier … 613

1.20E+03
1.00E+03
8.00E+02 400
6.00E+02 600
4.00E+02 800
2.00E+02 1000

0.00E+00
Z1 Z2 Z3 Z4 Z5 Z6
-2.00E+02 a
3.00E+03
2.50E+03
400
2.00E+03
1.50E+03 600
1.00E+03 800
5.00E+02
1000
0.00E+00
Z1 Z2 Z3 Z4 Z5 Z6
b
6.00E+02
5.00E+02
400
4.00E+02
3.00E+02 600
2.00E+02 800
1.00E+02
1000
0.00E+00
Z1 Z2 Z3 Z4 Z5 Z6
c
1.50E+03

400
1.00E+03
600
5.00E+02 800
1000
0.00E+00
Z1 Z2 Z3 Z4 Z5 Z6
d
5.00E+02
4.00E+02
400
3.00E+02
600
2.00E+02
800
1.00E+02
1000
0.00E+00
-1.00E+02 Z1 Z2 Z3 Z4 Z5 Z6
e

Fig. 6 Different types of faults with different irradiations (a P-G, b P-P, c P-P-G, d P-P-P, e P-P-P-G)
for both the feeders
614 K. Anjaiah and R. K. Patnaik

Table 2 Target features extracted for PV-based microgrid when fault is at both the feeders
Target feature sets
Fault Z1 Z2 Z3 Z4 Z5 Z6
P-G 3.29 7.7 1.4E–18 1.56E2 59.3 132
3.31 7.69 3.33E–5 1.57E2 59.2 132
3.32 7.69 3.3E–5 1.57E2 59.1 132
3.32 7.69 3.3E–5 1.57E2 59.1 132
PP 3.28 1.11 1.4E–18 8.86E2 1.34 27.3
3.26 1.05 3.33E–5 9.77E2 1.24 26.6
3.26 1.00 3.33E–5 1.04E3 1.17 26.8
3.25 0.98 3.33E–5 1.06E3 1.15 27.0
PP-G 3.67 4.38 1.4E–18 1.11E2 19.2 54.7
3.63 4.21 3.33E–5 1.30E2 17.7 55.0
3.65 4.20 3.33E–5 1.30E2 17.6 55.2
3.65 4.19 3.33E–5 1.30E2 176 55.2
PPP 3.74 1.29 2.9E–19 5.83E2 1.74 29.6
3.59 1.47 3.3E–5 4.03E2 2.22 30.7
3.46 1.76 3.33E–5 2.30E2 3.15 33.1
3.28 1.24 3.3E–5 6.65E2 1.62 30.1
PPP-G 3.68 2.77 −3.3E–5 2.11E2 7.73 63.0
3.53 2.86 −3.3E–5 1.88E2 8.20 62.9
3.4 3.01 −6.7E–5 1.56E2 9.14 62.9
3.2 2.75 −6.7E–5 2.21E2 7.60 61.1

Z 1 —in Fig. 6, it shows that except P-G fault all types of faults had more fault current
amplitude, because of faults are created in two feeders.
It is easy to classify and identify the types of faults under all the fault conditions
with the help of CDFT which was shown in Fig. 8, and it seems decision tree iden-
tification of fault created in both the feeders of the PV-based microgrid system. The
types of faults classification with features are shown in Table 2; according to Table 2
values, the generated results are shown in Fig. 7. The proposed algorithm accuracy
is discussed in Sect. 5.

5 Performance Evaluation

To analyze the recital of the proposed technique (CDFT), it is necessary to compute

the effectiveness of the technique. It defines the proposed technique ability to fruit-
fully categorize the correct faults for the number of target types of faults with unlike
irradiances.
A Simple Decision Tree-Based Efficient Fault Classifier … 615

Fig. 5 A simple decision tree-based fault classifier with CDFT

2000

P-P-P-G
1500
P-P-P
P-P-G
1000 P-P
P-G
500

0
Z1 Z2 Z3 Z4 Z5 Z6
-500

Fig. 7 Analysis of different faults with target features (Z 1 –Z 6 )

616 K. Anjaiah and R. K. Patnaik

Z1, Z2, Z3, Z4, Z5 AND Z6

If 4.22<Z1<4.25 && 1.53E1<Z1<1.54E1 && 1.48E2<Z4<1.48E2 && 2.36E2<Z5<2.36E2 LG

&& 2 49E2<Z6<2 49E2

If 1.05<Z2<1.98 && 1.45E-18<Z3<5.67E-4 && 1.25<Z5<3.92 LL

If 3.17<Z1<3.41&& 9.30<Z2<9.46 && 5.33E-4<Z3<5.67E-4

&&1.19E2<Z4<1.25E2 && LLG

If 3.73<Z1<3.78 && 2.60E2<Z4<3.91E2 && 3.48E1<Z6<3.71E1 LLL

If 4.35<Z1<4.40 && 4.62<Z2<4.71 && -5<Z3<-5E-4 &&

1.14E2<Z4<1.18E2 && 2.14E1<Z5<1.14E2 && LLLG

Fig. 8 A simple decision tree-based fault classifier with CDFT for both feeders

Efficiency can be defined as the number of truthful classified data sets divided into
the randomly selected data sets from the total data set (raw set). The mathematical
equation for efficiency is

Number of accurate classified data set

Efficiency(%) = × 100 (6)
Aimlessly selected data from overall set

5.1 Flowchart for Calculation of Efficiency

Step-1: Let us assume that N is the number of faults to be tested, D is number of

correctly classified, and initially, it is zero (D = 0).
Step-2: Generate an unsystematic number (Rn) for which unsystematic number
the initial prediction is generated (according to the following rule),
If Rn lies between 1 and 4, then the fault is P-G; then S set to be at 1.
If Rn lies between 5 and 8, then the fault is P-P; then S set to be 2.
If Rn lies between 9 and 12, then the fault is P-P-G; then S set to be 3.
If Rn lies between 13 and 16, then the event is P-P-P; then S set to be 4.
A Simple Decision Tree-Based Efficient Fault Classifier … 617

Table 3 Comparison
S. No. Techniques Efficiency (%)
between CDFT and existing
methods 1. Support vector machine [8] 93.3
2. Wavelet-based deep neural 99.3
networks [10]
3. Proposed method for feeder 1 99.89
4. Proposed method for both feeders 99.5

6 Simple Decision Tree-Based Fault Classifier

Step-3: all these data sets are subjected to simple decision tree for final forecasting
of proposed algorithm
Step-4: if decision tree conforms proposed algorithm according to the data sets,
it will be classified the fault currents.
Step-5: Finally, efficiency of the algorithm will be calculated with the help Eq. (6),
and it will generate correctly classified data sets.
In order to prove the accuracy of the discussed algorithm, it is necessary to compare
the efficiency of the proposed algorithm with existing methods [12] which are listed
in Table 3.
From the above deliberations (in Table 3), it is clear that the proposed method is
exactly classified the faults in the solar PV-based microgrid with the help of perfor-
mance indices. Hence, the time for computation is very less. Computational difficulty
of the proposed algorithm (CDFT) is reduced without affecting the efficiency. Draw-
back of this method is not able to give the information about its fault location; with
the help of advanced fuzzy-based algorithms, it will be shown in the extension of
this paper.

7 Conclusion

In this paper, a type of fault detection in solar PV-based microgrid has been made
with the help of change detection filter (CDFT) algorithm. With target features, total
five faults (P-G, P-P, P-P-G, P-P-P, P-P-P-G) are exactly classified. The proposed
technique (CDFT) illustrates considerable benefits such as improved classification
efficiency, and it does not necessitate further processing and less computational
complexity. The proposed algorithm is used for solar PV-based AC microgrid for
recognition and classification of the type of current faults. The proposed method
has the ability to identify the type of fault in the DC microgrid, that work will be
illustrated in the extension of this paper.
618 K. Anjaiah and R. K. Patnaik

References

1. Beheshtaein, S., et al.: Review on microgrids protection. IET Gener. Transm. Distrib. 13(6)
743–759 (2019)
2. IEEE guide for design, operation, and integration of distributed resource Island systems with
electric power systems. IEEE Std. 1547.4-2011, 1–54 (2011)
3. Castellan, S., et al.: A review of power electronics equipment for all-electric ship MVDC power
systems. Int. J. Electr. Power Energy Syst 96, 306–323 (2018)
4. Gu, J.-C., et al.: Design of a DC series arc fault detector for photovoltaic system protection.
IEEE Trans. Ind. Appl. 55(3), 2464–2471 (2019)
5. Xiao, J., et al.: Sizing of energy storage and diesel generators in an isolated microgrid using
discrete fourier transform (DFT). IEEE Trans. Sustain. Energy 5(3), 907–916 (2014)
6. Kar, S., Samantaray, S.R.: High impedance fault detection in microgrid using maximal over-
lapping discrete wavelet transform and decision tree. In: 2016 International Conference on
Electrical Power and Energy Systems (ICEPES). IEEE (2016)
7. Soleimanisardoo, A., Karegar, H.K., Zeineldin, H.H.: Differential frequency protection scheme
based on off-nominal frequency injections for inverter-based islanded microgrids. IEEE Trans.
Smart Grid 10(2) 2107–2114 (2018)
8. Jiang, J., et al.: Series arc detection and complex load recognition based on principal component
analysis and support vector machine. IEEE Access 7, 47221–47229 (2019)
9. Swetapadma, A., Yadav, A.: A novel single-ended fault location scheme for parallel transmis-
sion lines using k-nearest neighbor algorithm. Comput. Electr. Eng. 69, 41–53 (2018)
10. James, J.Q., et al.: Intelligent fault detection scheme for microgrids with wavelet-based deep
neural networks. IEEE Trans. Smart Grid 10(2), 1694–1703 (2017)
11. Hosseini, S.A., et al.: An overview of microgrid protection methods and the factors involved.
Renew. Sustain. Energy Rev. 64, 174–186 (2016)
12. Borges, F.A.S., Fernandes, R.A.S., Silva, I.N., Silva, C.B.S.: Feature extraction and power
quality disturbances classification using smart meters signals. IEEE Trans. Ind. Inform. 12(2),
824–833 (2016)
Application of Salp Swarm Optimization
for PI Controller to Mitigate Transients
in a Three-Phase Soft Starter-Based
Induction Motor

Monalisa Mohanty, Sanjay Kumar Sahu, Manas Ranjan Nayak,

Anshuman Satpathy, and Subhashree Choudhury

Abstract Induction motor (IM) has wide applications in both domestic and indus-
trial appliances. However, the major drawback of IM is the occurrence of heavy
inrush current at starting mostly due to switching on of a high inertial load. So, it is
necessary to control this inrush current which may interrupt the smooth functioning
of the system. A soft starter is a solid-state device that limits the sudden surge of
current and gives a gradual build-up to full speed. Proportional integral (PI)-based
soft starter is a classical and linear one, so it fails to response to changing speed of the
motor. So in this paper, a new heuristic salp swarm optimization (SSWO) technique is
proposed which dynamically tunes the PI control parameters to respond correctly to
the nonlinear nature of IM. SSWO is a nature-inspired meta-heuristic algorithm that
imitates the swarming behaviour of salps when navigating and foraging in ocean. The
SSWO-tuned PI-controlled IM model along with a thyristor-based soft starter (SS)
is designed using MATLAB/Simulink software and simulated to obtain minimum
overshoot, rated current during the starting of the motor, a controlled voltage supply
and a steady speed. Further, the distortion of the system with proposed SSWO-based
PI and conventional PI controller is discussed through the total harmonic distortion
(THD) calculation.

Keywords Induction motor (IM) · Soft starter (SS) · Slap swarm optimization
(SSWO) · Proportional integral (PI) controller · Transient stability · Total
harmonic distortion (THD)

1 Introduction

Induction motor (IM) is a synchronous motor that draws heavy input current instan-
taneously for motor starting and develops high torque [1]. The large surge current,

M. Mohanty · S. K. Sahu · M. R. Nayak · A. Satpathy · S. Choudhury (B)

Department of Electrical and Electronics Engineering, Institute of Technical Education and
Research, Siksha ‘O’ Anusandhan (Deemed to be University), Bhubaneswar, Odisha 751030,
India
e-mail: [email protected]

© Springer Nature Singapore Pte Ltd. 2020 619

G. Pradhan et al. (eds.), Advances in Electrical Control and Signal Systems,
Lecture Notes in Electrical Engineering 665,
https://doi.org/10.1007/978-981-15-5262-5_46
620 M. Mohanty et al.

known as inrush current, has many disadvantages such as it creates the dip in power
grid voltage, excessive heating of the windings, etc. The rapid acceleration leads to
mechanical stress on motor shaft, electrodynamic stress on power cable, slip of belt
in belt and pulley application, this ultimately leading to wear and tear of the system
[2]. This gradually reduces the efficiency and lifespan of the motor.
Soft starter (SS) is an electrical solid-state device made up of thyristor or silicon
controlled rectifier (SCR), whose main function is to limit the voltage [3]. It can
also be called as thyristor-based alternating current–voltage regulator (TACVR).
There are different conventional methods to start up an IM, such as direct online
(DOL) starter, auto-transformer, primary resistance and star–delta starter. These are
basically categorized into three classes: (i) mechanical, (ii) electro-mechanical and
(iii) electrical or solid-state devices [4]. The solid-state devices have advantages over
other two classes such as [5]: (i) adjusting SCR or thyristor firing angle can control
the motor terminal voltage, (ii) the inrush current remains under the rated value, (iii)
there is a steady increase in the speed, and (iv) the mechanical torque (Tm) and
electrical torque (Te) are reduced.
Proportional integral (PI) control method is one of the classical and conventional
controllers which is widely used. Its significant feature is (i) easy implementation and
(ii) simple in structure. There are two ways of tuning in PI controller: the open-loop
method, i.e. without a feedback, and the closed-loop method, i.e. with a feedback. In
closed-loop control method, nature-inspired algorithm is used to tune the PI param-
eters, i.e. kp and ki, in order to respond optimally to the nonlinearity of the system.
There are many meta-heuristics algorithms that are developed in the literature by
many authors. The natural behaviour of organism is studied such as particle swarm
optimization (PSO) [6]. In this paper, a PID controller along with PSO algorithm is
applied to a plant of higher order to study the response of PSO-tuned PID controller
over the conventional Ziegler–Nichols (ZN) tuning method, ant colony optimiza-
tion (ACO) [7]. An ACO-tuned PI controller is designed and simulated to control
the current during the starting on an induction motor with an objective to achieve
good dynamic response, and again it is experimentally verified using a less costly
microcontroller. A cuckoo search (CS) algorithm-tuned PI controller is simulated in
comparison with ZN-tuned PI controller, and it is observed that CS provides faster
rise time for starting current of IM and has a few parameters to be tuned [8]. Whereas
same work carried using bat algorithm (BA), it gives faster convergence as compared
to CS, CS has faster rise, but BA has minimum overshoot [9], salp swarm optimiza-
tion (SSWO) technique [10], etc. However, SSWO has the following advantages over
other algorithm, such as: (i) a zero overshoot is obtained, (ii) settling time is very
less, and (iii) the SSWO-tuned PI controller gradually attends the reference speed
with a relatively zero overshoot.
To study the effectiveness of the proposed controller in tuning the PI controller
parameters effectively, the system is modelled using MATLAB/Simulink. All the
characteristic curves of the IM are found out, and a detailed comparison between the
proposed technique and classical technique is shown. Further to test the distortion to
be within IEEE constraints, THD analysis is made and a comprehensive comparison
has been studied.
Application of Salp Swarm Optimization for PI Controller … 621

The organization of this paper is as follows. Section 2 gives the detailed modelling
of IM and SS. Section 3 elaborates the detailed control structure of the conventional PI
controller and SSWO-tuned PI controller. Section 4 presents the MATLAB/Simulink
model and the result analysis, and finally in Sect. 5, the conclusion from the entire
study has been drawn.

2 Modelling of System Parameters

2.1 Modelling of IM

The two most frequently used dynamic models, for modelling of IM, depend upon:
(i) space vector theory derivation and (ii) direct-quadrature (d-q)-axis derivation [11].
Unlike space vector theory, the d-q-axis derivation does not use complex numbers
or variables, which makes the equations complicated. Both the models are valid for
transient and performance analysis of IM. The modelling of IM using synchronous
and stationary reference frame is basically used. IM can be represented using space
vector which comprises of the following equations [12].
(i) The equations of voltage are as follows:

Vds = Rs i ds + ρλds − ωλqs (1)

Vqs = Rs i qs + ρλqs + ωλds (2)

Vdr = Rr i dr + ρλdr − (ω − ωr ) (3)

Vqr = Rr i qr + ρλqr + (ω − ωr )λdr (4)

(ii) The flux linkage equations are given as

λds = L ls i ds + L m (i ds + i dr ) (5)

 
λqs = L ls i qs + L m i qs + i qr (6)

λdr = L lr i dr + L m (i ds + i dr ) (7)

 
λqr = L lr i qr + L m i qs + i qr (8)
622 M. Mohanty et al.

(iii) The motion equation is stated below

J
ρωr = Te − Tl (9)
p

(iv) The electromagnetic torque is given by:

3PLm  
Te = i qr i ds − i ds i qr Nm (10)
4
where d is direct axis; q is quadrature axis; S is stator variable; R is rotor variable;
Vds , Vqs are stator voltages of d-axes and q-axes; Vdr Vqr are rotor voltages of d-axes
and q-axes; i ds , i qs are d-axis and q-axis stator currents; i dr , i qr are rotor currents of
d-axes and q-axes; Rs , Rr are winding resistances of stator and rotor; ω is rotational
speed; ωr is angular speed of rotor (electrical); ρ is operator used for derivative =
d/dt; Tl is load torque; Te is electromagnetic output torque; J is moment of inertia;
P are total poles; L m is magnetizing inductance; L s , L r = are stator and rotor self-
inductances; L ls , L lr = are stator and rotor leakage inductances; λds , λqs are d-axis
and q-axis stator flux leakages; and λdr , λqr are d-axis and q-axis rotor flux leakages.
Equations 1, 2, 3 and 4 are used to draw the circuit representation of d-axes and
q-axes. Figure 1 shows the d-q-axis equivalent circuit of the IM in the synchronous

Fig. 1 d-axes and q-axes representation of IM

Application of Salp Swarm Optimization for PI Controller … 623

Fig. 2 Circuit for

connection of soft starter
with the motor

rotating reference frame.

2.2 Soft Starter

Soft starter is an electrical solid-state device, i.e. used for limiting the electrical
quantities to the rated value for smooth starting of an IM. Inside the soft starter,
thyristors are connected in two anti-parallel manners for each phase. The conventional
method for starting an IM is: (i) DOL starter, (ii) auto-transformer, (iii) primary
resistance, (iv) star–delta starter, etc. Thyristors are used due to more efficiency and
rating of power. The connection of soft starter with the motor is shown in Fig. 2. The
main objective of this paper was designing of a soft starter model for controlling the
torque developed and current at the starting of IM drive. After the required condition
is achieved, the SS is disconnected from the IM [13].

3 Control Structure

3.1 PI Controller

The PI controller is a linear controller which is very simple in structure and has
easy implementation. ‘P’ stands for proportional controller having a constant ‘K p ’
which helps in enhancing the rise time. ‘I’ stands for integral controller having a
constant ‘K i ’ which helps in reducing error. Both work unanimously to control a
624 M. Mohanty et al.

signal perfectly. The combination of PI controller is taken as a feedback to the IM,

to stabilize the high oscillations that occur during the starting of motor. In IM, it
is necessary to have a gradual increase in motor speed, with terminal voltage and
input current remaining under the rated value. The negative response of the integral
controller proves to be beneficial here [14]. However, the major drawback is it cannot
respond to nonlinearity in the system. So, the parameters of the PI controller need to
be tuned. The PI controller mathematical representation is given in Eq. (11).

t
u(t) = K p × er(t) + K i ∫ er(t)dt (11)
0

3.2 Salp Swarm Optimization (SSWO)

Salp swarm optimization (SSWO) was developed by Mirjalili et al. in 2017. It is

a nature-inspired algorithm. The behaviour of the salp swarm has been mimicked
here. Salp is the member of the Salpidae family [10]. They have a very interesting
swarming behaviour, while navigating and searching for food. Mathematically, it is
of two groups: (i) leader or the head salp and (ii) followers of the leader. In deep
water bodies, the salps form a swarm called as salp chain. The leader or the head
salp guides the swarm [15]. Their main target is to search the food source called
‘F’ in the search space. The leader’s location is amended with reference to the food
source, and accordingly the follower salp position is updated.
In SSWO, the food source location is updated regularly. The salp chain has the
potential to chase the food source in the best possible way, even if the population
degrades. Another advantage is the leader salp position is frequently updated with
reference to the food source. There is only one controlling parameter. The number
of iteration gradually decreases as the space for search is explored and exploitation
of the search space begins [16]. Local solution to the problem is avoided by this
algorithm which is advantageous over existing optimal algorithm. Figure 3 shows
the flow chart of salp swarm optimization (SSWO).

Advantages of SSWO
SSWO has the following advantages over other algorithm [17], such as: (i) a zero
overshoot is obtained, (ii) settling time is very less, and (iii) the SSWO-based PI
controller gradually attends the reference speed with a relatively zero overshoot.
SSWO Technique Application for Tuning PI Parameters
Here, the problem formulated is a minimization function and SSWO technique is
used to obtain the best optimum parameter values of the PI controller such that the
high inrush current of the IM can be controlled optimally.
Here, integral absolute error (IAE) method of error calculation is adopted. Let I 
be the rated current of IM (rms value) and I be the IM actual current. Let er = (I  − I )
be the error.
Application of Salp Swarm Optimization for PI Controller … 625

Fig. 3 Flow chart of salp swarm optimization (SSWO)

The objective function to be minimized can be formed as:

Minimize,
t
J (θ ) = ∫(er(t))2 (12)
0

Such that, (θinitial ≤ θ ≤ θfinal )

Here,

  t
θ = K p , K i J (θ ) = ∫(er(t))2 (13)
0

In the current work, a PI controller is designed using a heuristic algorithm which

is nature-inspired for optimization, i.e. salp swarm optimization algorithm. For opti-
mization, the objective function is described by Eqs. 12 and 13. Equation 12 is
626 M. Mohanty et al.

continuously computed according to the algorithm, and the computed result ranks
the parameters which gives the minimal value.

4 Simulation Result and Analysis

Figure 4 comprises MATLAB/Simulink model of a three-phase induction machine

with a soft starter which consists of a set of anti-parallel thyristors for each phase.
In technical terms, a soft starter is a power electronic device which reduces the
high starting current and torque of machine by allowing more and more voltages
through semiconductors by controlling the firing angle every half cycle. In this paper,
the various characteristics of the induction machine have been plotted with soft
starter with the conventional PI controller and proposed SSWO-tuned PI controller.
The results obtained illustrate that the SSWO-tuned PI controller is better than the
classical PI controller in terms of system stability, dynamic response and control
parameters.
Figures 5 and 6 show the induction machine’s mechanical torque and electro-
magnetic torque, respectively. It is observed that the system shows a considerable
amount of ripple and harmonics for both mechanical and electromagnetic torques
with the conventional PI controller. However, the SSWO-tuned PI controller shows
its effectiveness by reducing ripple and harmonics from the system. Figures 7 and 8
depict the speed and angular speed, respectively. It is seen that the SSWO-tuned PI
controller possesses more speed and better angular speed than that of the classical
PI controller.

Fig. 4 MATLAB/Simulink model of the system

Application of Salp Swarm Optimization for PI Controller … 627

Fig. 5 Mechanical torque

Fig. 6 Electromagnetic
torque

Fig. 7 Speed

Fig. 8 Angular speed, wm

628 M. Mohanty et al.

Figures 9, 10, 11 and 12 show the IM’s three-phase voltage with conventional
PI controller, voltage with SSWO-tuned PI controller, current with conventional PI
controller and current with SSWO-tuned PI controller, respectively. It can be seen
that with the conventional PI controller the system fails to reduce initial high current
and the harmonics from the system. However, the SSWO-tuned PI controller proves
its robustness by reducing the initial system current and harmonics from voltage.

Fig. 9 Grid voltage with conventional PI controller

Fig. 10 Grid voltage with SSWO-tuned PI controller

Fig. 11 Grid current with conventional PI controller

Application of Salp Swarm Optimization for PI Controller … 629

Fig. 12 Grid current with SSWO-tuned PI controller

Figures 13, 14, 15 and 16 show the THD values of voltage with conventional
PI controller, voltage with SSWO-tuned PI controller, current with conventional PI
controller and current with SSWO-tuned PI controller, respectively. It can be seen
from the above figures that the conventional PI controller fails to eliminate system
THD greatly as grid voltage with conventional PI is 9.73%, and current is 12.80%
which shows there are nonlinearities in the system, whereas the proposed SSWO-
tuned PI controller proves its superiority on the conventional PI controller by reducing
the system THD to a greater extent, i.e. 0.47 and 0.91% of grid voltage and current,
respectively. Table 1 summarizes the THD values for grid voltage and current for
proposed SSWO-tuned PI controller and conventional PI controller.

Fig. 13 Grid voltage with PI

Fig. 14 Grid voltage with

SSWO-tuned PI
630 M. Mohanty et al.

Fig. 15 Grid current with PI

Fig. 16 Grid current with

SSWO-tuned PI

Table 1 THD values of grid

Grid parameters Controller
voltage and current
SSWO-PI PI
THD values THD values
PI 0.47% 9.73%
THD values 0.91% 12.80%

5 Conclusion

This paper gives a detailed discussion of advantages of SSWO technique over the
conventional PI controller in mitigating the overshoot problem of IM. This intelli-
gence controller along with the soft starter was capable in reducing the IM starting
problem. It also reduces the sudden torque building of an IM. The model is simu-
lated in MATLAB/Simulink software. The results obtained give the justification of
enhancement in system steady state, limitation of heavy inrush current at starting,
increase in system dynamics by using the proposed SSWO technique for tuning the
PI control parameters dynamically. This technique can be implemented in industrial
plants or for stability of any higher-order system, where cost is a major constraint.
For justifying the stability, THD analysis has been done and the values agree well
with the IEEE limits. SSWO algorithm can also be used to tune a PID controller,
and its result can be compared with the technique proposed in this paper; besides,
Application of Salp Swarm Optimization for PI Controller … 631

the linear controller integration with other hybrid algorithms can be designed to get
better results in three-phase synchronous machine.

References

1. Larabee, J., Pellegrino, B., Flick, B.: Induction motor starting methods and issues. In: IEEE
PCIC-2005, pp. 217–222 (2005)
2. McElveen, R.F., Toney, M.K.: Starting high-inertia loads. IEEE Trans. Ind. Appl. 37(1), 137–
144 (2001)
3. Nwachukwu, C.O., Ajumuka, M.I., Izuegbunam, F.I.: Three phase soft starter controller scheme
for asynchronous machine response to transients using MatLab/Simulink. In: 2017 IEEE 3rd
International Conference on Electro-Technology for National Development (NIGERCON).
IEEE (2017)
4. Nithin, K., Mathew, B., Rafeek, M.: An improved method for starting of induction motor
with reduced transient torque pulsations. Int. J. Adv. Res. Electr. Electron. Instrum. Eng. 2(1),
462–470 (2013)
5. Pandey, S., Bahadure, S., Kanakgiri, K., Singh, N.M.: Two-phase soft start control of three-
phase induction motor. In: 2016 IEEE 6th International Conference on Power Systems (ICPS),
New Delhi, pp. 1–6 (2016)
6. Mahapatra, S., et al.: Induction motor control using PSO-ANFIS. Procedia Comput. Sci. 48,
753–768 (2015)
7. Sundareswaran, K., Nayak, P.S.: Ant colony based feedback controller design for soft-starter
fed induction motor drive. Appl. Soft Comput. 12(5) (2012)
8. Yildiz, A.R.: Cuckoo search algorithm for the selection of optimal machine parameters in
milling operations. Int. J. Adv. Manuf. Technol. (2012)
9. Bora, T.C., Coelho, L.D.S., Lebensztajn, L.: Bat-inspired optimization approach for the
brushless DC wheel motor problem. IEEE Trans. Magn. 48(2), 947–950 (2012)
10. Mirjalili, S., et al.: Salp swarm algorithm: a bio-inspired optimizer for engineering design
problems. Adv. Eng. Software 114, 163–191 (2017)
11. Itoh, J.-I., et al.: Design of flux-axis angular speed estimation using induced voltage in speed
sensor-less field oriented control for induction motor. In: 2019 10th International Conference
on Power Electronics and ECCE Asia (ICPE 2019-ECCE Asia). IEEE (2019)
12. Ratnani, A., Punit, L., Thosar, A.G.: Mathematical modelling of an 3 phase induction motor
using MATLAB/Simulink. Int. J. Mod. Eng. Res. (IJMER) 4(6), 62–67 (2014)
13. Rakhmawati, R., Murdianto, F.D., Alim, M.W.: Soft starting & performance evaluation of PI
speed controller for brushless DC motor using three phase six step inverter. In: 2018 Inter-
national Seminar on Application for Technology of Information and Communication. IEEE
(2018)
14. Ahmed, A.M., et al.: Brushless DC motor speed control using both PI controller and fuzzy PI
controller. Int. J. Comput. Appl. 109(10), 29–35 (2015)
15. Faris, H., et al.: An efficient binary salp swarm algorithm with crossover scheme for feature
selection problems. Knowl. Based Syst. 154, 43–67 (2018)
16. Sayed, G.I., Khoriba, G., Haggag, M.H.: A novel chaotic salp swarm algorithm for global
optimization and feature selection. Appl. Intell. 48(10), 3462–3481 (2018)
17. Faris, H., et al.: Salp swarm algorithm: theory, literature review, and application in extreme
learning machines. Nature-Inspired Optimizers, pp. 185–199. Springer, Cham (2020)
An Improved MPPT Technique
for Increasing Efficiency of PV Module

Arjyadhara Pradhan, Babita Panda, Bhagabat Panda, and Aradhana Khillo

Abstract The electrical output of the Photovoltaic panels depends on several

factors like irradiance, temperature, soiling, shading, humidity, etc. Even the total
power produced is not completely transferred to the load. The factors affecting the
panel performance cannot be avoided as these are natural phenomenon, hence extract-
ing maximum power under all these conditions can improve the efficiency of the Pho-
tovoltaic system. Maximum power point tracking is an electronic tracking system
that extracts maximum power from the Photovoltaic panel by changing the electrical
operating point. By using the maximum power point tracker (MPPT) the life of the
Photovoltaic system is increased. In this paper, a new optimization technique for
maximum power point tracker (MPPT) is designed named Fish Search Optimization
(FSO). By using six samples of maximum power (function of PV voltage and cur-
rent) as fishes the best duty cycle is calculated for tracking. Even a boost converter is
designed for load matching operation. The efficiency of the system using this novel
technique is found to be 97.8%.

Keywords Boost · Unregulated · MPPT · Photovoltaic · Converter

A. Pradhan · B. Panda (B)

School of Electrical Engineering, KIIT University, Bhubaneswar, India
e-mail: [email protected]
A. Pradhan
e-mail: [email protected]
B. Panda · A. Khillo
Department of Electrical Engineering, Parala Maharaja Engineering College, Berhampur, India
e-mail: [email protected]
A. Khillo
e-mail: [email protected]

© Springer Nature Singapore Pte Ltd. 2020 633

G. Pradhan et al. (eds.), Advances in Electrical Control and Signal Systems,
Lecture Notes in Electrical Engineering 665,
https://doi.org/10.1007/978-981-15-5262-5_47
634 A. Pradhan et al.

1 Introduction

Renewable energy sources are one of the promising areas for reducing the burden of
consumption of fossil fuels, which creates pollution. Among the various renewable
sources, Photovoltaic technology is most widely used and simple technology that
directly converts solar energy to electrical energy. Maximum power point tracking is
generally used for maximizing power extraction from a Photovoltaic system or wind
system under various weather conditions. Studies show that throughout the day the
solar irradiance level does not remain the same rather it keeps on changing. Even
during rainy months, the effect of the cloud causes changes in the amount of sunlight
received by the panel. There are various MPPT methods of tracking and many more
novel methods are developed by various researchers to make the efficient utilization
of solar output power [1]. The various MPPT methods are classified based on cost,
convergence speed, hardware implementation, etc. P and O method is called hill-
climbing method. It is a very simple and reliable method that is based on oscillations
of output power. In this case below, the MP point the voltage increases whereas above
the MP point the voltage decreases. The incremental conductance method is based
on incremental changes in the module current and voltage to determine the effect
of voltage. It performs better under changing weather conditions. This method is
more complicated in comparison to other methods of MPP tracking [2–4]. Extreme
seeking control method is an online method which gives a real-time optimization
using feedback control [5, 6]. In current sweep method, a sweep waveform is used
for PV module current. Here the IV curve can be drawn and updated at constant
time intervals. Abhraham et al. developed a new MPPT technique called successive
approximation register method where the tracking time is very fast, and hence energy
saving can be done [7]. In open-circuit voltage method or the constant voltage method,
the operating point is fixed at MP point by regulating module voltage and matching
it to that of the fixed (open-circuit) voltage.
It is an empirical method as the power delivery to the load is interrupted to match
with the Voc [8–10]. Other than these methods, there are various methods of tracking
like Fuzzy logic, ANN method, modified P and O and modified Inc Cond, etc. [11].
In case of traditional P and O method, if the step size is small tracking time is long but
accuracy can be achieved at low irradiance condition. In rapidly changing weather
condition the step size has to be increased where the accuracy of the system cannot
be obtained. In case of using soft computing techniques like Fuzzy logic control, the
step size can be accurately fixed to track the MP point, and hence efficiency of the
system can be improved [12]. Figure 1 shows the block diagram representation of
the Photovoltaic standalone system.
In the Block diagram, PV system is connected to the DC-DC Converter here in
this case Boost converter is used and to a dc load. A PWM signal is used to trigger
the switch of the converter. The signal to the converter circuit is obtained by using
the Pulse width Generator. Hence, MPPT block is used to track maximum power and
fix the operating point considering the load matching effect. Thus using the MPPT
technique the duty cycle is adjusted.
An Improved MPPT Technique for Increasing Efficiency of PV Module 635

Fig. 1 Block diagram

representation of
photovoltaic standalone
system

2 Proposed Method

In the proposed method a new algorithm for MPPT has been described with the help
of a flowchart. The optimization technique used is called Fish Search Optimization
(FSO). The overall system consists of input source as a Photovoltaic system con-
nected with a boost converter and a resistive load. Here the signal to the converter
switch, i.e. duty cycle is obtained from the MPPT block. The new technique used
basically is a modified version of P and O technique giving better results in com-
parison to other methods. Maximum power point tracking is based on the concept
of operating the PV system at maximum power point irrespective of any weather
and load conditions. Thus in the proposed technique, the maximum power is taken
as fishes (function of PV voltage and current) and the main objective is to find out
the prey, i.e. the knee point. Instead of using fixed step size as in traditional P and
O method here we use six sets of values for duty cycle (fitness value). Initially, the
fishes are set with a fixed fitness value. Then the present value of power and previous
power are evaluated and checked for the best power (prey). The iterations proceeds
for six cycles and hence the prey (maximum power) with best fitness value (duty
cycle) is fixed. Hence this duty cycle is given to the gate signal of converter switch
to operate at the maximum powerpoint. This method is simple, provides high-speed
response, less costly, highly efficient, requires less computational time and even user
doesn’t require sophisticated mathematical knowledge (Fig. 2).
In the traditional Perturb and Observance method initially, module voltage and
module current are recorded. Then Power is calculated and change in power is com-
pared. If it is positive then the present voltage is compared with the past voltage. If the
result is satisfied the voltage is incremented, i.e. moves forward to reach MPP point
or else the voltage is decreased. Here a constant and small perturbation is applied
step by step until maximum power is reached (Fig. 3).
636 A. Pradhan et al.

Fig. 2 Flow chart of fish search optimization (FSO) technique

Fig. 3 Flow chart of Perturb and Observance method

An Improved MPPT Technique for Increasing Efficiency of PV Module 637

Table 1 Solar cell

S. No. Parameters Values
specifications
1 Short-circuit current, (I sc ) 7.34 A
2 Open-circuit voltage (V oc ) 0.6 V
3 Quality factor 1.5 V
4 Series resistance 0
5 Output voltage 43 V

3 DC-DC Converter

In this work, boost converter is used to step-up voltage from 42 V Even though
CUK and Sepic converter have more advantages as their input current is continuous,
but the number of passive components in these converters is more, hence voltage
and current are high with high switched power stress. Hence boost converter with
less number of passive components is less complex and is more efficient. They have
better loop stability and respond more to dynamic actions. The following are the
design parameters considered for design of boost converter [13–18]. Inductor value:
6.3 mH, Load resistance: 164 O, capacitance: 4400 microfarad, duty ratio 0.74 and.
switching frequency (Fs): 25 kHz. Voltage ripple factor 5% and Ripple current is
considered 20% of the output current and as per IEC standard. Here IGBT is used for
switching purposes. For the design purpose, the values of internal resistance taken
are 1000 m and the snubber resistance is taken as 105 .

4 Result Analysis

The entire work is carried out in MATLAB/Simulink platform to study and compare
the result of the proposed method over the existing method. Here the blocks of Sim
power systems are used. In this work, we use 12 panels (in series connection), in
each, there are six solar cells (in series connection) were utilized, totally 72 solar
cells were used. The total output voltage of PV module is 43.2 V. A boost converter
is designed using IGBT switch with Internal Resistance and Snubber Resistance.
The value of RC filter used in the input side is taken as R = 0.005, C = 100 µF. The
values of inductor and capacitor are L = 7.3 mH, C = 4400 µF. The specification of
solar cell used is given in Table 1 (Figs. 4, 5, 6, 7 and 8).

5 Simulation Results

The results of the proposed system carried out in MATLAB/Simulink are given in
the below figures.
638 A. Pradhan et al.

Fig. 4 Overall model of PV system connected with boost converter and resistive load using proposed
MPPT technique

Fig. 5 Subsystem of PV module

Fig. 6 Series connection of six number of solar cells

An Improved MPPT Technique for Increasing Efficiency of PV Module 639

Fig. 7 Simulation of boost converter

Fig. 8 Shows simulation model of FSO technique and duty cycle generation

Figure 9 gives the value of maximum power obtained by using the novel MPPT
technique (FSO) and is found to be 244.5 W. For the input power of 250 W, 244.5
W output power is obtained.
Figure 10 shows the convergence time of the two methods used. The new method
takes less time to reach the MP point than the traditional P and O method.
Figure 11 shows the convergence time of the two methods used. The traditional
P and O method takes more time to reach the MP point.
Figure 12 shows the duty cycle generated by using Pulse width modulation gen-
erator. This PWM signal is used to trigger the switch of the converter. The signal to
the converter circuit is obtained by using Pulse width generator. The MPPT block
is used to track maximum power and fix the operating point considering the load
matching effect.
Figure 13 gives the value of input voltage to the converter, i.e. 43 V. The boost
640 A. Pradhan et al.

Fig. 9 Maximum power using FSO method

Fig. 10 Convergence time by using novel method

converter is designed using IGBT switch with Internal Resistance and Snubber
Resistance.
Figure 14 gives the output voltage of boost converter using Perturb and Observance
methods (traditional), i.e. 162 V.
Figure 15 shows the output voltage of the converter using novel technique and is
found to be 185 V.
Table 2 gives the value of efficiency considering output power and input power
for different values of irradiance using both the methods.
An Improved MPPT Technique for Increasing Efficiency of PV Module 641

Fig. 11 Shows the convergence time by using traditional P and O method

Fig. 12 Duty cycle generated by using PWM generator

6 Conclusion

The proposed FSO technique for MPPT tracking based on the selection of best fitness
value was implemented using MATLAB/Simulink software. A total of 72 number of
solar cells used in the design provide a voltage of 43 V. The designed system consists
of a Photovoltaic panel connected with a boost converter and a resistive load. The
642 A. Pradhan et al.

Fig. 13 Input voltage to the boost converter

Fig. 14 shows the output voltage of the converter using Perturb and Observance method

DC-DC converter used is highly efficient as it uses only one switch to perform load
matching between source and load. The proposed technique was evaluated for its
performance at various irradiance levels. From the simulation results, it was clear
that the efficiency of this technique is more and remains almost constant for different
levels of irradiance, whereas using traditional Perturb and Observance method the
efficiency is less and even varies with changing irradiance. The system efficiency is
found to be 97.8%.
An Improved MPPT Technique for Increasing Efficiency of PV Module 643

Fig. 15 Shows the output voltage of the converter using fish search optimization

Table 2 Efficiency of the proposed method

S. No. Solar Existing Proposed Existing Proposed
irradiance method (Pout ) method (Pout ) method method
(W/m2 ) efficiency (%) efficiency (%)
1 400 230 238.75 92.0 95.5
2 500 230.5 239.5 92.2 95.8
3 650 233 240.75 93.2 96.3
4 800 234.1 242.1 93.6 96.8
5 900 235 243 94.0 97.2
6 1000 235.75 244.5 94.3 97.8

Acknowledgements I express my heartfelt thanks to the School of Electrical Engineering, KIIT

University, Bhubaneswar for providing laboratories with all software facilities to carry out the test.

References

1. Esram, T., Chapman, P.L.: Comparison of photovoltaic array maximum power point tracking
techniques. IEEE Trans. Energy Convers. 22, 2 (2007)
2. Bianconi, E., Calvente, J., Giral, R., Mamarelis, E., Petrone, G., Ramos-Paja, C.A., et al.:
Perturb and observe MPPT algorithm with a current controller based on the sliding mode. Int.
644 A. Pradhan et al.

J. Electr. Power Energy Syst. 44(1), 346–356 (2013)

3. Hua, C., Lin, J., Shen, C.: Implementation of a DSP-controlled photovoltaic system with peak
power tracking. IEEE Trans. Ind. Electron. 45(1), 99–107 (1998)
4. Abdelsalam, A.K., Massoud, A.M., Ahmed, S., Enjeti, P.N.: High-performance adaptive perturb
and observe MPPT technique for photovoltaic based microgrids. IEEE Trans. Power Electron.
4, 26 (2011)
5. Panda, B., Pati, N., Panda, B.: Hardware implementation of PV Z source inverter using max-
imum boost control method. In: Proceedings of the Third International Conference on Smart
Computing and Informatics, vol. 2, pp. 685–695. Springer (2018). Kalinga Institute of Industrial
Technology Deemed to be University
6. Bazzi, A.M., Krein, P.T.: Concerning maximum power point tracking for photo voltaic opti-
mization using ripple-based extremum seeking control. IEEE Trans. Power Electron. 26(6),
1611–1612 (2011)
7. Panda, B., Panda, B., Hota, P.K., Bhuyan, S.K.: A comparative analysis of maximum power
point techniques for photovoltaic system. In: IEEE Power, Communication and Information
Technology Conference (PCITC), Bhubaneswar, pp. 732–737 (2015)
8. Enslin, J.H.R., Wolf, M.S., Snyman, D.B., Swiegers, W.: Integrated photo voltaic maximum
power point tracking converter. IEEE Trans. Ind. Electron. 44, 769–773 (1997)
9. Panda, B., Panda, B., Hota, P.K.: A comparative analysis of PWM methods of ZSI used for
PV system. In: Proceedings of the 5th International Conference on Frontiers in Intelligent
Computing: Theory and Applications FICTA 2016, vol. 1, pp. 281–291. KIIT University,
Bhubaneswar (2016)
10. Selvan, S., Nair, P., Umayal, U.: A review on photo voltaic MPPT algorithms. Int. J. Electr.
Comput. Eng. (IJECE) 6(2)
11. Panda, B., Sarkar, A., Panda, B., Hota, P.K.: A comparative study of PI and fuzzy controllers
for solar powered DC-DC boost converter. In: International Conference on Computational
Intelligence and Networks (CINE), Bhubaneshwar, pp. 47–51 (2015)
12. Pradhan, A., Ali, S.M., Parashar, S.K.S., Banerjee, G.: Effect of panel orientation on the
performance of PV system. J. Adv. Mater. Eng. 3(1), 1–6 (2016)
13. Panda, B., Hota, P.K., Panda, B.: Fault analysis of grid connected photovoltaic system.
American J. Electr. Power Energy Syst. 5(4), 35–44 (2016). https://doi.org/10.11648/j.epes.
20160504.12.arjyadhara. Pradhan, A., Panda, B.: Performance analysis of photovoltaic mod-
ule at changing environmental condition using MATLAB/Simulink. Int. J. Appl. Eng. Res.
12(13), 3677–3683 (2017)
14. Pradhan, A., Panda, B.: Experimental analysis of factors affecting the power output of the PV
module. Int. J. Electr. Comput. Eng. 7, 141–149 (2017)
15. Panda, B., Hota, P.K., Panda, B.: A simple current control strategy for single stage grid
connected three-phase PV inverter. American J. Electr. Electron. Eng. 4(4), 102–109 (2016)
16. Pradhan, A., Panda, B.: Analysis of ten external factors affecting the performance of PV system.
In: International Conference on Energy, Communication, Data Analytics and Soft Computing
(ICECDS), SKR Engineering College, Tamil Nadu, IEEE Madras Section 1–2 August 2017
17. Pradhan, A., Panda, B.: Design of DC-DC converter for load matching in case of PV system.
In: International Conference on Energy, Communication, Data Analytics and Soft Computing
(ICECDS), SKR Engineering College, Tamil Nadu, IEEE Madras Section 1–2 August 2017
18. Pradhan, A., Panda, B., Ghosh, S., Panda, B.: Efficiency improvement of photovoltaic sys-
tem using maximum power point tracking. In: International Conference on Electrical, Elec-
tronics, Computers, Communication, Mechanical and Computing (EECCMC), Priyadarshini
Engineering College, Tamil Nadu, IEEE Madras Section 28–29 January 2018
A Review of Advancements in DC Grid
Technology

Jagadish Kumar Patra, Ramaprasad Panda, Smruti Pragnya Dash,

and Akanksha Rani

Abstract The intensifying requirement for energy and high depletion rate of fossil
fuels has led to the extensive use of renewable sources. DC grid system provides the
best platform for utilization of the available renewable sources in both commercial
and domestic applications. In comparison with AC grid, the number of interfacing
converters required in case of DC grid is less. This paper presents us an analysis of
upgradation in the field of DC grid till date. Various dominant factors that have to
be acknowledged in architecture of a DC grid have been put forward. For example,
grid voltage level, grid protection and battery integration. As we know in AC grid
there is not any one fixed specific voltage level that can be followed to form a DC
grid. Therefore, different proposed voltage levels have been analyzed by explaining
their advantages and disadvantages. Most importantly, here, we have discussed about
battery integration, since in DC grid, it assures system reliability. Discussions have
also been made on the areas related to DC grid protection.

Keywords AC · DC · Grid · Converter

1 Introduction

A considerable change came along with power electronics innovations which is

favorable toward DC. Importance of DC increased constantly, since DC voltage is
required for the operation of the components of power electronics. With the nonstop
rise in use of renewable energy sources and DC appliances, increases interest in using
DC electrical system [1–3]. Nowadays almost all of the electronic appliances used
work on DC. This renovation will bring multiple beneficial outcomes along with
it. DC grid plays an important role since the power generated from the renewable,
storage devices and the power consumed by most of the domestic and commercial
loads is in the form of DC.

J. K. Patra (B) · R. Panda · S. P. Dash · A. Rani

Department of Electrical and Electronics Engineering, Silicon Institute of Technology, Silicon
Hills, Bhubaneswar, Odisha 751024, India
e-mail: [email protected]

© Springer Nature Singapore Pte Ltd. 2020 645

G. Pradhan et al. (eds.), Advances in Electrical Control and Signal Systems,
Lecture Notes in Electrical Engineering 665,
https://doi.org/10.1007/978-981-15-5262-5_48
646 J. K. Patra et al.

Generation from renewable

energy sources
Storage
devices
Generating DC--DC
station

DC--AC AC--DC

AC
Grid

AC--DC
AC--DC Either
AC or
DC--DC DC
DC--AC
Lightening Heating
Other
Loads loads
Loads

Motor
Loads

Fig. 1 Structure of AC microgrid

When we are considering distributed power sources, DC grid is dominant over AC

in various ways. Contracted energy losses, uplifted power density and smooth union
with renewable energy sources and availability of storage systems are some of the
advantages of DC grid system. Also, DC grid system is comparatively more efficient
as we do not have to worry about the losses due to reactive power component which
is one of the major concerns in AC grid system [4–8].
Huge energy loss that takes place due to power electronics convertors (especially
rectifiers) can be eliminated by the application of DC grid system. A considerable
amount of loss, volume and weight is contributed by the AC line whose only objective
is to generate DC voltage that can be used by the devices. So the number of interfacing
converters required can be reduced if DC grid will put into action [9, 10]. As a
result, the overall system efficiency will increase since DC grid drastically reduce
the number of power converters, making a simpler and cost-effective system. The
appreciable number of reduction in the number of converters has been illustrated in
Figs. 1 and 2. In case of DC grid, synchronization of the phase and frequency of
voltage is not required, making the system simpler and easily controllable.

2 Voltage Level

While constructing a DC grid, the selection of DC voltage level is one of the major
points to be considered. There is no predefined particular standard for deciding
the voltage level. DC grid seems to be our future that is why different researches
are continuing in numerous parts of the world on the very same issue. In different
A Review of Advancements in DC Grid Technology 647

Generation from renewable

energy sources
Storage
devices
Generating
station

AC--DC DC--DC

DC
Grid

Either
AC or
DC--DC DC
DC--AC
Lightening Heating
Other
Loads loads
Loads

Motor
Loads

Fig. 2 Structure of DC grid

articles, for various applications, multiple levels of DC voltage have been introduced.
The selection of voltage levels for various applications has their own justification.
In real-time DC grid has been tested and established in different places, for which,
they considered ETSI EN 300 132-3-1 standard to set the level of DC voltage [11].
According to his standard voltage level should lie in between 260 and 400 V. This
condition must be satisfied when we are considering ICT and telecommunications
equipments. Two standard nominal voltages for feasible operation of DC grid are
354 V and 380 V.
Globally, the most acknowledged voltage level in various commercial applications
is 380 V. The author has highlighted various cases of commercial buildings deploying
DC microgrid in them as shown in [12]. The DC microgrid at Xiamen University
has accepted 380 V as its voltage level. Its installed capacity is 90 kw of load and
150 kw of solar panels as shown in Fig. 3, which is a stable system.
Fraunhofer IISB situated at Erlangen implemented one more DC grid with same
voltage level of 380 V [13, 14]. The total efficiency of the system is 91.7% which is
approximately 5% greater than AC grid. Also, relatively the distribution efficiency
has improved.
China telecom has chosen a voltage level of 275 V to supply quality power [15].
Control at this voltage is very stable, proved by both simulation and experimental
results. 400 V is considered as the voltage level in [16, 17] for forming DC bus. To
realize the setup as a robust system, authors of both the papers have used different
control mechanisms. For the buildings used for commercial purpose, 380 V is the
best DC voltage level as concluded from the above description about standard ETSI
EN 300 132-3-1.
648 J. K. Patra et al.

Utility AC/DC

Monitor
AC/DC AC/DC AC/DC

380V DC Bus

380V/24V Data EV Home

Light Center Charger AC Appliances Expansion

Fig. 3 Structure of 380V DC microgrid

While taking domestic applications into consideration, still the query remain the
same about choosing appreciate voltage level. Basically, we can say that none of
the standards have properly defined a particular voltage level to be accepted over
the world, although there are so many DC voltage levels are described in different
research papers. While considering a distribution setup, two voltage levels have
been introduced in the paper [18], 380 and 24 V. Choosing 380 V as DC voltage
level is completely justified as 380 V can easily be accessed and also specified in
EMerge Alliance. Talking about 24 V, as we know that below the voltage level of
30 V insulation is not a requirement, for residential utilities, 24 V can be considered
as a safe DC voltage level. From the above discussion, we can definitely draw the
conclusion that this architecture is more cost-effective than that of conventional AC
system.
For domicile application, another voltage level is also suitable, i.e., ±170 V as
specified in [19]. To serve a stable DC voltage of 170 V, this system used renewable
sources, converters and also a collection of supercapacitor. An example of 380 V
DC grid for domestic appliances has been discussed in [20]. According to the load
requirement, this system makes use of many converters to vary the voltage. The work
done in [21] deals with multiple voltage levels less than 72 V and gives us the best
voltage level for domestic purpose. The voltage levels considered are 12, 24, 48 and
72 V. Based on percentage loading, cable drop and branching, these four voltage
levels were analyzed. It was studied that 72 V DC bus remained preferable in all the
considerations except percentage loading where best results were obtained at 48 V.
A Review of Advancements in DC Grid Technology 649

3 Converters

In theory, various DC-DC converters have been demonstrated, as its efficiency will
decide the overall efficiency of DC system. Another reason behind discussing these
converters is for interfacing the devices which can store energy and to integrate
various distributed energy sources to the DC grid. Different techniques are there
like current-fed isolated DC-DC converters, dual active bridge converters, isolated
boost and full-bridge converters, buck-boost converters, LLC resonant converters,
etc. [22–37]. In the paper, the authors have chosen a DC-DC converter having high
efficiency and gain. As you can see in Fig. 4, it uses a switched capacitor with a
coupled inductor to get greater efficiency. This technique has used a diode and a
capacitor as a pair for recycling of the leakage inductance energy. For a supply of
30 V, an output of 400 V was obtained [38].
The above-mentioned paper demonstrates a converter, whose input side has half-
bridge resonant boost converter and output side has voltage quadruple topology as
shown in Fig. 5. Using this topology, an input and output voltage level of 42 and
380 V, respectively, can be achieved. Also, this design shows different praiseworthy

Fig. 4 Proposed converter Sw1 C1

in [38]
D2

C3

D1

D3

D4 VBUS
C2 C0
PV
Panel

Fig. 5 Proposed Converter

in [39]
M1 D1 C1 D3

L C3 C5 C7
Lr
D4
V0
C6
CP

D5
Cin C2 C8
M2 D
PV 2
C4
D6
650 J. K. Patra et al.

aspects like turn-on of ZCS for diodes and ZVS for switches, which reduces the
ripple content in output voltage and clamp the switch voltage without using snubber
circuit arrangements. To reduce the ripple content of output voltage and magnetic
size, it operates at a very high frequency, i.e., at 150 kHz from this converter [39].
In this paper, the focus is on the solution of two major necessities of DC microgrid.
Multilevel modular capacitor clamped converter is the best for microgrid application
providing bidirectional power flow and grounding isolation as demonstrated in Fig. 6
[40]. Another article explained an interleaved DC-DC forward converter having 2–3
winding transformers as shown in Fig. 7. It has high efficiency and also the size of
the output filter is less. This converter has less ripple content in output voltage and
current [41].
The authors interrogated a CCM boost converter with high gain, i.e., almost double
of conventional boost converter. The design included ZCS turn-off of diodes and ZVS
turn-on of switches, thus, reducing the switching losses. The converter shown in Fig. 8

Fig. 6 Proposed Converter RHV1 RLV1

in [40]
C3
SR11 SB11 SR13
C2

C1 C4
SR12 SB13
SB12 SR14

RHV2 RLV2

SR21 C2 SB21 C3 SR23

C1 C4
SR22 SB23

SB22 SR24

Fig. 7 Proposed Converter

in [41]
D3 D1 L

Vin NP NS D2 C RL

T Nd
A Review of Advancements in DC Grid Technology 651

Fig. 8 Proposed Converter

in [42]

S2
D1
C2
L1 L2
C1

Ro
D2
Vo
Vi S1
C3

gives an adoptable voltage of 400 V for an input of 80 V. But the major drawback is
the high voltage stress across its components [42].
The explanation in [43] gives idea about non-isolated step-up DC-DC converter
with RPWM switching method. In this soft switching design, i.e., ZVS turn-on of
switch was done along with a particular switching method that associated the use of
Lr and Cr. Thus, the economization of both turn-on and turn-off losses were done.
This exact switching design also reduces the losses involved with diode reverse
recovery. The major drawbacks of this design are limited operating frequency and
high current at turn-off.
The authors recommended a new three-level DC-DC converter where the input
voltage can be stabilized by making use of this converter design [44]. For bipolar
DC microgrid employment, the authors have introduced a combination of CUK and
SEPIC converter. The arrangement is of a single switch making the control uncom-
plicated. Fusion of these two converters fulfills the purpose of current redistribution
and also voltage balancing [45].
In [46], the assessment of a multi-cell dual active bridge converter is accomplished
as shown in Fig. 9. Application of the suggested converter increases reliability of the
system, i.e., it provides uninterrupted power to load by sharing the total load among
the remaining converters even if breakdown of one of its converter occurs. One part
of the cell in series and the other in parallel reduces voltage stress across switches
and high gain. Other crucial advantages related to design are load sharing among the
cells and an option for extension of power system capability.
The authors introduced a different DC-DC converter bearing unidirectional char-
acteristic for its usage in bipolar DC bus as shown in Fig. 10. It uses the LLC design
for its operation and has numerous advantages by operating in a limited portion of
the gain curve. The advantages of this particular design are easy control, ZCS and
high efficiency, voltage balancing. The converter has a voltage gain that is beyond
unity leading to large increased efficiency and decreased currents [47].
This article presented a design on how we can employ tapped inductor concept as
shown in Fig. 11, to obtain large gain with low duty cycle. To acquire and send data,
CANBUS protocol along with power line communications have also been integrated
652 J. K. Patra et al.

Cb L

S1 S3
S5 S7

C1 C2
V1 V2

S2 S4
S6 S8

Fig. 9 Proposed Converter in [46]

Fig. 10 Proposed Converter

in [47]
S1 S3

Cr

Vin

S2 S4

Fig. 11 Proposed Converter Enable/

in [48] Lin Lt1
Disable D1

LMOUT
SW1
Lt2
Cin VO
COUT
PV D2
SW2

with the converter. For better durability, the converter has been fabricated using
critical elements, thus, assures 25 years of operating guarantee [48]. In [49], for
A Review of Advancements in DC Grid Technology 653

microgrid applications, a novel quasi-Z-source converter galvanically isolated DC-

DC converter design has been proposed. For using PV, monotonous input current is
desired which is considered by this design. This topology has combined features taken
from inductor and transformer-based converters. For an input of 15 V and 125 W, the
output voltage of 400 V was obtained. This further confirms the improvised voltage
step-up the ability of this converter. In this paper [50], the authors provided a solution
to associate two DC buses, i.e., by implementing a three-phase bidirectional current-
fed push-pull DC-DC converter. With the motive of achieving soft switching for all
the devices and voltage clamping of primary side devices, a modulation technique
has been added on the secondary side of the topology.
Another converter design recommended in [51] allows multiple inputs. It consists
of boost legs and buck-boost in the input side, thus, facilitating easy integration
of renewable sources. This converter has advantages like reliability, flexibility and
is easily scalable. Idea conveyed in [52] provides an asymmetric series resonant
converter for offshore DC grid usage. Transformer size is reduced because of soft
switching property of series resonant converter. The proposed converter well satisfies
the bidirectional power flow which is a major necessity in offshore scenarios. A dual
active bridge converter is designed in [53], which is bidirectional in nature and acts
as interface between grid and ultra-capacitor. This topology gives the minimal ripple,
thus, reducing the size of filter and resulting in increase in power density. For input
variation from 25 V to 48 V, the output voltage of 200 V can be obtained.
One of the important applications of the DC grid is electric vehicle. In [54], for
interfacing electric vehicle with grid, the authors have proposed a bidirectional DC-
DC converter. This converter is shown in Fig. 12 exhibits a special feature, i.e., it
allows overlapping of input and output voltage limits. Higher efficiency is achieved
by reduction of switch stress by using half-bridge switches. For input voltage range
of 200 V to 400 V, this design can provide output in the range 300–400 V. Moreover,
for the same electric vehicle application, a CLLLC type resonant converter has been

Fig. 12 Proposed Converter

in [54]

S1 S3
L1
L2
C1
D
C2
C
S2 S4
Battery B
U
S
654 J. K. Patra et al.

introduced in [55]. By including additional inductor and capacitor than LLC type
resonant converter upgrades it with many more positive characteristics. The design
fabricated in article [56] is an enhanced full-bridge three-level converter. The most
important application of this converter is in assimilation of wind turbine with DC
grid.
With the addition of a passive filter with the converter, working of the converter
improves and the transformer voltage stress also decreases. In article [57], a dual input
DC-DC converter has been inspected which works great for DC grid applications.
This design allows bucking action for one of the sources and bucking as well as
boosting action for the other source. Therefore, either of the two sources continuously
supply to the load with diminished ripple current (should always be low) which is a
vital feature.
The article [58] presented a LCL DC-DC converter which can obtain high power
conversion up to megawatts. Use of transformer is eliminated by using LCL circuit as
shown in Fig. 13. In case of external DC faults, this converter also has the capability
of regulating current. At 100 MW power, this circuit provided 300 kV output for an
input voltage of 20 kV. A DC-DC converter without a transformer has been suggested
by the author in [59]. This design integrates a series–parallel resonant converter with
a boost converter as presented in Fig. 14 which provides soft switching of diodes and
switches. We get advantages like modularity and reduced voltage stress across the
switches with this topology. Output voltage of 500 V was obtained with 50 V input,
thus, higher voltage ratio is obtained. In [60], a resonant converter has been examined.
Integration of two asymmetrical pulse width modulated resonant topologies results
in the suggested converter. Using this converter, an efficiency of 93.1% was possible
to obtain.
In [61], the authors examined an isolated modular converter. This converter has
integrated capacitor voltage balancing control and soft switching providing multiple
benefits. This convertor has advantages of low cost, high efficiency and small size.

Fig. 13 Proposed Converter

in [58]

V1 C2 V2
C1 L1 L2
S1 S3 S5 S7
C
L1 L2
C
V1 C2 V2
C1 S4 S2 S8 S6
A Review of Advancements in DC Grid Technology 655

Fig. 14 Proposed Converter Lb

in [59] D3

S1 D5 C1
D1
Cr1 Lr
Cr2
Vin Vo
D4

S2 C2
D2

4 Protection

The various problems in functioning of DC microgrid are in the form of faults.

Therefore, the protection from these faults is required. In order to have uninterrupted
performance, the grid protection immediately separate the faulty part. Simplicity,
reliability, speed, performance and economics are a few of the multiple aspects
highlighted in [62–64]. Use of circuit breaker is proposed in [65]. It prevents over
current by simply disconnecting the current circuit. In case, the over current condition
continues, it elongates the arc length. Hence, by melting the contact metals, the circuit
gets opened and over current situation is prevented. Similar device is a fuse which
operates in almost similar way [66]. Because of cable capacitance and inductance,
connecting these types of protective equipment suffers very less setbacks.
Another method for protection has been discussed in [67], i.e., by using an electro
mechanical switch. The stated approach firstly deactivates the system and activates
the safe section of the system after the fault has been removed. But this method is
comparatively more tedious. In order to stop the current quickly, a complete solid-
state imitation of circuit breakers called active current limiters or interrupters can
be used. [68, 69] also discussed about similar type of circuits. Still due to factors
like high cost, weight and size these are not widely used in industries. [70] discusses
about the current advancements in the domain of semiconductors, thus, leading to
more powerful and affordable solid-state limiters.
DC systems are tough to control because of absence of zero crossing. Hence,
hybrid and oversized solid-state breakers are used for its handling [71, 72]. These
types of circuit breakers are faster in action but incur a considerable amount of loss.
Corzine [73] Introduces an effective Z-source breaker. This breaker has various
merits such as quick fault clearance and no dependence on sensing units. 400 V
DC microgrid is installed in various renewable energy sources and energy storage
devices as shown in [74]. Using suitable protection devices, the above system is
protected from various faults. Alarm signal are used to remove faults such as fault in
switch of inverter bank (fuel cell system) and DC pole to ground fault (solar system).
Stimulating the circuit breakers can remove DC pole to pole fault (solar system) or
656 J. K. Patra et al.

fault in multiple switches of inverter (fuel cell system). [75] establishes the assess-
ment of protection for radial DC systems. The advantages of DC circuit breakers
are perfectly utilized by radial DC systems which are excellent for small-scale DC
grids. Handling of hybrid CB on one side and DC/DC converter on the other side
of every linking cable is recommended in this paper. Security to radial DC grid and
inbuilt isolation from DC faults is achieved by using DC/DC converters. The above
procedure brags of numerous features compared to other protection schemes such
as minimized cost, easy control and powerful protection. [76] illustrates a different
type of robust differential protection scheme. Unfortunately, the illustrated method
comes with the major demerit of communication delay, which in turn is directly
proportional to DC cable length. Failure in immediate termination of fault can lead
to increased fault current and require CB of higher rating thus increasing its cost. As
stated earlier, DC/DDC converters are excellent for fault protection. [77] emphasizes
on simultaneously using LCL voltage source converter and mechanical DC circuit
breaker for protection. LCL voltage source converter has a fundamental characteris-
tic of fault current control so it can be operated for extended time. Lesser cost and
higher reliability make this system preferable for employment.
Modular multilevel converter is well suited for conditions where fast auto reclos-
ing is required and fuses cannot be used as discussed in [78]. The MMC removes the
short circuit fault in an appreciable way if it is in a completely controlled mode. The
reactive power is kept constant so as to remove the short circuit fault.

5 Conclusion

This paper gives an overall review on three considerable fields regarding DC micro-
grid, they are voltage level, converters and protection. Voltages that are realized
across the world have been analyzed for both commercial and domestic applica-
tions. DC-DC converters play a crucial role in the integration of renewable courses
with microgrid. Advantages and disadvantages of various converter methodologies
have been discussed here. At the end, for safe and reliable operation, the protection
is very much necessary. Different protection strategies have been analyzed in this
paper. Subsequent research must include converter methodologies with larger gain
and higher efficiency, fixation of DC bus voltage for domestic application and lastly
protection strategies with minimum fault clearance time.

References

1. Ott, L.: Power electronics for low-voltage DC grids in commercial buildings. VDE Congress
(2014)
2. Reed, G.F., Technologies, D.C.: Solutions to electric power system advancements. Power
Energy Mag. IEEE 10, 10–17 (2012)
A Review of Advancements in DC Grid Technology 657

3. Engelen, K., Leung Sun, E., Vermeyen, P., Pardon, I., D’hulst, R., Driesen, J., Belmans, R.:
The feasibility of smallscale residential DC distribution. In: The 32nd Annual Conference of
the IEEE Industrial Electronics Society, pp. 2618–2623 (2006, November)
4. Hatziargyriou, N., Asano, H., Iravani, R., Marnay, C.: Microgrids. IEEE Power Energ. Mag.
5(4), 78–94 (2007)
5. Kakigano, H., Miura, Y., Ise, T., Uchida, R.: DC Micro-grid for super high quality distribution—
system configuration and control of distributed generations and energy storage devices. In: IEEE
IPEMC Power Electron Motion Control Conference, pp. 1740–1745 (2004)
6. Guerrero, J.M., Vasquez, J.C., Teodorescu, R.: Hierarchical control of droop-controlled DC
and AC microgrids: a general approach towards standardization. IEEE Conf. Ind. Electron.
(IECON) 4305–4310 (2009)
7. Xu, L., Chen, D.: Control and operation of a DC microgrid with variable generation and energy
stage. IEEE Trans. Power Deliv. 26(4), 2513–2522 (2011)
8. Justo, J.J., Mwasilu, F., Lee, J., Jung, J.W.: AC-microgrids versus DC microgrids withy
distributed energy resources: a review. Renew. Sustain. Energy Rev. 24, 387–405 (2013)
9. Lago, J., Moia, J., Heldwein, M.L.: Evaluation of power converters to implement bipolar dc
active distribution networks—DC–DC converters. Proc. Energy Convers. Congr. Expo 985–990
(2011)
10. Landsman, E.: A unifying derivation of switching dc-dc converter topologies. In: Proceedings
of IEEE Power Electronics Specialists Conference, pp. 239–243 (1979)
11. ETSI EN 300 132-3-1—Environmental Engineering (EE); Power supply interface at the input
to telecommunications and datacom (ICT) equipment; Part 3: Operated by rectified current
source, alternating current source or direct current source up to 400 V; Sub-part 1: Direct
current source up to 400 V, European Standard EN 300 132-3-1 V2.1.1(2012-02)
12. Zhang, F., Meng, C., Yang, Y., Sun, C., Ji, C., Chen, Y., Wei, W., Qiu, H. and Yang, G.:
Advantages and challenges of DC microgrid for commercial building. In: IEEE ICDCM First
International Conference on DC Microgrids, pp. 355–358 (2015)
13. Wunder, B., Ott, L., Szpek, M.: Energy efficient DC-grids for commercial buildings. In: IEEE
INTELEC 36th International Telecommunications Energy Conference, pp. 1–8 (2014)
14. Weiss, R., Ott, L., Boeke, U.: Energy efficient low-voltage DC-grids for commercial buildings.
In: IEEE ICDCM First International Conference on DC Microgrids, pp. 154–158 (2015)
15. Xu, H.G., He, J.P., Qin, Y., Li, Y.H.: Energy management and control strategy for DC micro
grid in data center. In: 5th International Conference on Electricity Distribution, Shanghai, 5–6
Sept 2012
16. She, X., Lukic, S., Alex, Q.H.: DC zonal micro-grid architecture and control. In: 36th Annual
Conference on IEEE Industrial Electronics Society, pp. 2988–2993 (2010)
17. Eid, A.: Stand-alone DC micro-grids: analysis and control. In: The 2nd International Conference
on Engineering and Applied Science (2013)
18. Webb, V. J.: Design of a 380 V/24 V DC micro- grid for residential DC distribution (2013)
19. Kakigano, H., Miura, Y., Ise, T.: DC micro-grid for super high quality distribution-system
configuration and control of distributed generations and energy storage devices. In: IEEE
Power Electronics Specialists Conference, pp. 1–6 (2006)
20. Ryu, M.H., Kim, H.S., Baek, J.W., Kim, H.G. Jung, J.H.: Effective test bed of 380-V DC
distribution system using isolated power converters. IEEE Trans. Ind. Electron. 62 (7) (2015)
21. Chaidez, J.E.: DC house modeling and system design (2011)
22. Kim, H.-S., Ryu, M.-H., Baek, J.-W., Jung, J.-H.: High-efficiency isolated bidirectional AC-DC
converter for a DC distribution system. IEEE Trans. Power Electron. 28(4), 1642–1654 (2013)
23. Jung, J.-H., Kim, H.-S., Ryu, M.-H., Baek, J.-W.: Design methodology of bidirectional CLLC
resonant converter for high frequency isolation of DC distribution systems. IEEE Trans. Power
Electron. 28(4), 1741–1755 (2013)
24. Chen, W., Rong, P., Lu, Z.: Snubberless bidirectional DC-DC converter with new CLLC res-
onant tank featuring minimized switching loss. IEEE Trans. Ind. Electron 57(9), 3075–3086
(2010)
658 J. K. Patra et al.

25. Garcia, O., Flores, L.A., Oliver, J.A., Cobos, J.A., de la Pena, J.: Bidirectional DC-DC converter
for hybrid vehicle. In Proceedings of IEEE PESC, pp. 1881–1886 (2005)
26. Wu, T.-F., Chen, Y.-C., Yang, J.-G., Kuo, C.-L.: Isolated bi-directional full-bridge DC-DC
converter with a flyback snubber. IEEE Trans. Power Electron. 25(7), 1915–1922 (2010)
27. Zhu, L.: A novel soft-commutating isolated boost full- bridge ZVS-PWM DC-DC converter
for bidirectional high power applications. IEEE Trans. Power Electron. 21(2), 422–429 (2006)
28. Naayagi, R.T., Forsyth, A.J., Shuttleworth, R.: High-power bidirectional DC-DC converter for
aerospace applications. IEEE Trans. Power Electron. 27(11), 4366–4379 (2012)
29. Oggier, G.G., Garcia, G.O., Oliva, A.R.: Switching control strategy to minimize dual active
bridge converter losses. IEEE Trans. Power Electron. 24(7), 1826–1838 (2009)
30. Zhao, B., Yu, Q., Sun, W.: Extended-phase-shift control of isolated bidirectional DC-DC con-
verter for power distribution in microgrid. IEEE Trans. Power Electron. 27(11), 4667–4680
(2012)
31. Krismer, F., Kolar, J.W.: Accurate power loss model derivation of a high-current dual active
bridge converter for an automotive application. IEEE Trans. Ind. Electron. 57(3), 881–891
(2010)
32. Tan, N.M.L., Abe, T., Akagi, H.: Design and performance of a bidirectional isolated DC-DC
converter for a battery energy storage system. IEEE Trans. Power Electron. 27(3), 1237–1248
(2012)
33. Krismer, F., Kolar, J.W.: Efficiency-optimized high- current dual active bridge converter for
automotive applications. IEEE Trans. Ind. Electron 59(7), 2745–2760 (2012)
34. Xuewei, P., Rathore, A.K.: Novel bidirectional snubberless naturally commutated soft-
switching current- fed full-bridge isolated DC/DC converter for fuel cell vehicles. IEEE Trans.
Ind. Electron 61(5), 2307–2315 (2014)
35. Rathore, A.K., Prasanna, U.R.: Analysis, design, experimental results of novel snubberless
bi-directional naturally clamped ZCS/ZVS currentfed half-bridge DC/DC
36. Jeong, D.K., Ryu, M.H., Kim, H.G., Kim, H.J.: Optimized design of bi-directional dual active
bridge converter for low-voltage battery charger. J. Power Electron. 14(3), 468–477 (2014)
37. Zhao, B., Song, Q., Liu, W.: Experimental comparison of isolated bidirectional DC-DC convert-
ers based on all-Si and all-SiC power devices for next-generation power conversion application.
IEEE Trans. Ind. Electron. 61(3), 1389–1393 (2014)
38. Vineeth Kumar, P.K., Sreenivasan, M.K.: Hardware implementation of high efficient and
high voltage gain DC-DC converter for DC microgrid applications. In: IEEE International
Conference on Electrical, Computer and Communication Technologies, pp. 1–5 (2015)
39. Patil, D., Rathore, A.K., Srinivasan, D., Panda, S.K.: High-frequency soft-switching LCC
resonant current-fed DC/DC converter with high voltage gain for DC microgrid application.
In: 40th Annual Conference of the IEEE Industrial Electronics Society, pp. 4293–4299 (2014)
40. Sintupatsuk, P., Khomfoi, S., Paisuwanna, P.: A dc to dc multilevel modular capacitor clamped
converter with electrical grounding isolation and bidirectional power flow for a dc microgrid
application. In: 9th International Conference on Electrical Engineering/Electronics, Computer,
Telecommunications and Information Technology, pp. 1–4 (2012)
41. Habumugisha, D., Chowdhury, S., Chowdhury, S.P.: A DC-DC interleaved forward converter
to step—up DC voltage for DC microgrid applications. In: IEEE Power and Energy Society
General Meeting, pp. 1–5 (2013)
42. Park, S., Choi, S.: Soft-switched CCM boost converters with high voltage gain for high-power
applications. IEEE Trans. Power Electron. 25(5) (2010)
43. Park, Y., Jung, B., Choi, S.: Nonisolated ZVZCS resonant PWM DC–DC converter for high
step-up and high-power application. IEEE Trans. Power Electron. 27(8) (2012)
44. Zhang, X., Gong, C., Yao, Z.: Three-level DC converter for balancing DC 800-V voltage. IEEE
Trans. Power Electron. 30(7) (2015)
45. Ferrera, M.B., Litrán, S.P., Durán, E., Andújar, J.M.: A SEPIC-Cuk converter combina-
tion for bipolar DC microgrid applications. In: IEEE International Conference on Industrial
Technology, pp. 884–889 (2015)
A Review of Advancements in DC Grid Technology 659

46. Mamede, H.R., dos Santos, W.M., Coelho, R.F., Martins, D.C.: A multicell dual-active bridge
converter for increasing the reliability of power supply in a DC microgrid. IN: IEEE First
International Conference on DC Microgrids, pp. 274–279 (2015)
47. Zubieta, L.E., Lehn, P.W.: A high efficiency unidirectional DC/DC converter for integrat-
ing distributed resources into DC microgrids. In: IEEE First International Conference on DC
Microgrids, pp. 280–284 (2015)
48. Krzywinski, G.: Integrating storage and renewable energy sources into a DC microgrid using
high gain DC DC boost converters. In: IEEE First International Conference on DC Microgrids,
pp. 251–256 (2015)
49. Chub, A., Vinnikov, D., Jalakas, T.: Galvanically isolated quasi-Z-source DC-DC converters
with combined energy transfer for renewable energy sources integration. In: IEEE International
Conference on Industrial Technology, pp. 2896–2900 (2015)
50. Bal, S., Rathore, A.K., Srinivasan, D.: Naturally clamped snubberless soft-switching bidirec-
tional current-fed three-phase push-pull DC/DC converter for DC micro-grid application. In:
IEEE Applied Power Electronics Conference and Exposition, pp. 717–724 (2015)
51. Kwasinski, A., Krein, P.T.: Multiple-input dc-dc converters to enhance local availability in
grids using distributed generation resources. In: Twenty Second Annual IEEE Applied Power
Electronics Conference, pp. 1657–1663 (2007)
52. Meyer, C., De Doncker, R.W.: Design of a three-phase series resonant converter for offshore
DC grids. In: 42nd IAS Annual Meeting. Conference Record of the 2007 IEEE Industry
Applications Conference, pp. 216–223 (2007)
53. Zhou, H., Duong, T., Sing, S.T., Ashwin, M.., Khambadkone, A.M.: Interleaved bi-directional
dual active bridge DC-DC converter for interfacing ultra capacitor. In: IEEE International
Symposium on Industrial Electronics, Micro-Grid Application, pp. 2229–2234 (2010)
54. Khan, M.A., Husain, I., Sozer, Y.: A bi-directional DC-DC converter with overlapping input
and output voltage ranges and vehicle to grid energy transfer capability. In: IEEE International
Electric Vehicle Conference, pp. 1–7 (2012)
55. Zahid, Z.U., Dalala, Z. M., Chen, R., Chen, B., Lai, J.-S.: Design of bidirectional DC–DC
resonant converter for vehicle-to-grid (V2G) applications. IEEE Trans. Transp. Electrification
1(3) (2015)
56. Deng, F., Chen, Z.: Control of improved full-bridge three-level DC/DC converter for wind
turbines in a DC grid. IEEE Trans. Power Electron. 28(1) (2013)
57. Veerachary, M., Reddy, B.A., Kumar, E. V: Two-input DC-DC converter for DC-grid applica-
tions. In: International Conference on Green Computing, Communication and Conservation of
Energy, pp. 574–579 (2013)
58. Jovcic, D., Zhang, L.: LCL DC/DC converter for DC grids. IEEE Trans. Power Deliv. 28(4)
(2013)
59. Athab, H., Yazdani, A., Wu, B.: A transformerless DC–DC converter with large voltage ratio
for MV DC grids. IEEE Trans. Power Deliv. 29(4) (2014)
60. Lam, J., Jain, P. K: A new step-up DC/DC resonant converter with a capacitive output filter for
medium voltage (MV) DC grid in wind energy power systems. In: 40th Annual Conference of
the IEEE Industrial Electronics Society, pp. 1084–1089 (2014)
61. Xing, Z., Ruan, X., You, H., Yang, X., Yao, D., Yuan, C.: Soft-switching operation of isolated
modular DC/DC converters for application in HVDC grids. IEEE Trans. Power Electron. 31(4)
(2016)
62. Nigim, K.A., Lee, W.-J.: Micro grid integration opportunities and challenges. In: Power
Engineering Society General Meeting. IEEE, pp. 1–6 (2007)
63. Johnson, B.K., Lasseter, R.H., Alvarado, F.L., Adapa, R.: Expandable multiterminal DC
systems based on voltage droop. IEEE Trans. Power Deliv. 8(4), 1926–1932 (1993)
64. Nikkhajoei, H., Lasseter, R.H.: Microgrid protection. In: Power Engineering Society General
Meeting. IEEE, pp. 1–6 (2007)
65. Brozek, J.P.: DC overcurrent protection—where we stand. IEEE Trans. Ind. Appl. 29(5), 1029–
1032 (1993)
660 J. K. Patra et al.

66. Gregory, G.D.: Applying low voltage circuit breakers in direct-current systems. IEEE Trans.
Ind. Appl. 31(4), 650–657 (1995)
67. Ciezki, J.G., Ashton, R.W.: Selection and stability issues associated with a navy shipboard DC
zonal electric distribution system. IEEE Trans. Power Deliv. 15(2), 665–669 (2000)
68. Brice, C.W., Dougal, R.A., Hudgins, J.L.: Review of technologies for current-limiting low-
voltage circuit breakers. IEEE Trans. Ind. Appl. 32(5), 1005–1010 (1996)
69. Meyer, C., Kowal, M., De Doncker, R.W.: Circuit breaker concepts for future high-power
DC-applications. In: IEEE-IAS Conference Record, pp. 860–866 (2005)
70. Meyer, C., Kowal, M., De Doncker, R.W.: Solid-state circuit breakers and current limiters
for medium-voltage systems having distributed power systems. IEEE Trans. Ind. Appl. 19(5),
1333–1340 (2004)
71. Bolanowski, B., Wojcik, F.: A fast DC hybrid circuit breaker. In: IEE Colloquium on Electronic-
Aided Current-Limiting Circuit Breaker Developments and Applications, pp. 8/1 (1989)
72. Meyer, C., Kowal, M., De Doncker, R.W.: Circuit breaker concepts for future high-power
DC-applications. IN: IEEE Industry Applications Society Conference, pp. 860–866 (2005)
73. Corzine, K.: DC micro grid protection with the Z-source breaker. In: 39th Annual Conference
of the IEEE Industrial Electronics Society, pp. 2197–2204 (2013)
74. Lee, W.-S., Kang, S.-H.: Protection for distributed generations in the DC micro-grid. In: 2nd
IEEE PES International Conference and Exhibition on Innovative Smart Grid Technologies,
pp. 1–5 (2011)
75. Jovcic, D., Taherbaneh, M., Taisne, J.-P., Nguefeu, S.: Offshore DC grids as an interconnection
of radial systems: protection and control aspects. IEEE Trans. Smart Grid 6(2) (2015)
76. Descloux, J., Raison, B., Curis, J.B.: Protection strategy for undersea MTDC grids. In:
Proceedings of IEEE PowerTech, Grenoble, France, pp. 1–6 (2013)
77. Hajian, M., Zhang, L., Jovcic, D.: DC transmission grid with low-speed protection using
mechanical DC circuit breakers. IEEE Trans Power Deliv. 30(3)
78. Jäger, V.M. K., Rothstein, A., Steimel, A., Meyer, D., Bartelt, R., Heising, C.: Short circuit
protection in DC ship grids based on MMC with full-bridge modules. In: International Confer-
ence on Electrical Systems for Aircraft, Railway, Ship Propulsion and Road Vehicles, pp. 1–5
(2015)
A Novel Approach for RFID-Based
Smart EVM System

Dipanjan Sen, Savio Jay Sengupta, Abhishek Sharma, Wasim Reja,

Bijoy Goswami, and Subir Kumar Sarkar

Abstract A customized and highly secured EVM system has been designed by
using RFID (Sengupta et al in Eur J Sci Res 97(4):592–603 (2013) [1]) to achieve
a reliable high-end communication system for the purpose of election and voting.
In this work, an overview of the existing problems related to the electronic voting
machine has been introduced initially and then the proposed systems are designed
using RFID (Sengupta et al in Eur J Sci Res 97(4):592–603 (2013) [1]), (Bag and
Sarkar in Int J Radio Freq Ident Technol Appl 4(2):197–211 (2013) [2]) in detail.
The problems related to security breeching at the voting centers are very common. A
detailed study has been done considering the security breaching and access in case
of the complete voting system using EVM. Keeping in mind, the present scenario
of high-end securities for the corporate industries, banks, etc., we have designed a
secure and flexible RFID-based EVM system using low frequency RFID technology
(Bag and Sarkar in Int J Radio Freq Ident Technol Appl 4(2):197–211 (2013) [2])
and biometric sensors. The system has been practically implemented in laboratory,
and the experimental data shows that this system can be widely used by the election
commission at the voting centers to maintain the confidentiality.

D. Sen (B) · S. J. Sengupta · W. Reja · B. Goswami · S. K. Sarkar

Department of Electronics and Tele-Communication Engineering, Jadavpur University, Kolkata,
India
e-mail: [email protected]
S. J. Sengupta
e-mail: [email protected]
W. Reja
e-mail: [email protected]
B. Goswami
e-mail: [email protected]
S. K. Sarkar
e-mail: [email protected]
A. Sharma
Dunzo Digital Pvt. Ltd, Bangalore, India
e-mail: [email protected]

© Springer Nature Singapore Pte Ltd. 2020 661

G. Pradhan et al. (eds.), Advances in Electrical Control and Signal Systems,
Lecture Notes in Electrical Engineering 665,
https://doi.org/10.1007/978-981-15-5262-5_49
662 D. Sen et al.

Keywords RFID · LF range · Arduino · EVM system · High-end security ·

Biometric sensors

1 Introduction

Radio frequency identification or RFID is one of the most reliable wireless commu-
nication technologies that can be applied to various fields of wireless applications
[2–4]. RFID is a non-contact wireless technology that identifies objects that are
attached to RFID tags. RFID readers obtain information of objects and surroundings
through communication with tag antennas. RFID has spread its wings in different
sectors like business, retail, healthcare, smart home, smart universities, etc. [4, 5].
In our proposed idea, we would like to implement this technology in securing the
voting systems for an election and to provide a more secure and flexible system.
The term RFID describes a system that identifies an item by using magnetic field
or radio frequency. The most important components of this system are a tag and a
reader. The tag will be attached to the item which is needed to be tracked and the
reader which is nothing but a device will sense the tag, i.e., it will read the unique
IDs.
that were stored in the tag [5]. A computer system is also required that will
store the unique IDs of the tag in the database. Readers and tags can communicate
with each other in different frequency regions such as low frequency (in short LF,
125 kHz), high frequency (in short HF, 13.56 MHz), ultra-high frequency (in short
UHF, 860–960 MHz) and microwave frequency (2.4 GHz) [5].
In this article, a smart EVM system has been designed with the help of RFID
technology and biometric sensors. The voter ID card has been replaced by a RFID
tag (card), where the required information of an individual has been stored. The
modification of the existing EVM is executed by interfacing a biometric sensor
(fingerprint sensor) with the EVM. Hence, the proposed EVM system can provide
high-end security and also more flexibility [5, 6].
In Sect. 2, we have mentioned few problems related to the voting system as per
as the security is concerned. Section 3 describes the main proposed design. In this
section, all the necessary details are given with a flowchart. In Sect. 4, we have shown
the hardware design and all the necessary information needed for the proposed design.
Section 5 contains the conclusion part.

1.1 Need for the Proposal

In the past few decades, the voting system during the election in our country has
faced major issues related security [6]. Voting is a fundamental process that keeps
our system of government working. Because of the severe importance of voting in
any country, the procedure of voting is crucial. The whole process must be clean,
A Novel Approach for RFID Based Smart EVM System 663

transparent, and most importantly, safe and not rigged. The importance of having a
standard system for voting with intense security, which cannot be rigged or broken,
is an issue of concern in this country and that is what the aim of our work is to
ensure a safe and flexible system which can counteract fraud, forging of card, voting
for someone else and rigging the final result. Therefore, this article consists of a
highly secure [6] and flexible RFID-based system for designing an electronic voting
machine.

1.2 Proposed Structure

In this paper, we have proposed a smart EVM system to secure voting in the coun-
try [7]. Considering one voting center, all the eligible voters will appear with LF
(125 kHz) RFID tags, which is a smart voter card that can be used for identifying an
individual (voter). The following components are used in our proposed system:
1. Passive RFID (125 kHz) tags, which in our case are smartcards.
2. Low frequency RFID readers.
3. Arduino Uno processor.
4. A high-end computer and biometric (Adafruit) sensors.
The figure below shows the used RFID tag:
See Fig. 1.
The figure below shows the low frequency RFID reader:
See Fig. 2.
The flowchart of our proposed system is given below:

Fig. 1 Used LF RFID card tags

664 D. Sen et al.

Fig. 2 Used parallax RFID

reader

Fig. 3 Structure of our

proposed system

From Fig. 3, at first, an individual has to place the card near the reader before
he/she enters the voting room. The necessary information will be verified with the
existing database of the election commission. The new ID cards allocated to the
existing users should be designed by building a link with the aadhar details, and
hence, it consumes less amount of time. The proposed system provides two-level of
security by introducing biometric sensor interface with the EVM. After the verifying
the information of a voter ID card (RFID tag) through a reader, the voter is allowed
to place his/her vote. In this case, the role of a presiding officer is passive as he/she is
not involved in activating the EVM. The EVM will communicate with the biometric
sensor (fingerprint sensor) at the time of voting. At the moment before casting the
vote, the voter needs to place his/her finger on the fingerprint sensor for the purpose
of verification and the identity of the voter will be checked and verified with the
existing information of the user in the central database [7, 8]. Then, the EVM will
be active for few seconds for voting. In this proposed work, the EVM will operate
automatically based on biometric sensor interface and will be inactive after the casting
of an individual vote. Therefore, few security issues related to the manual operation
of the EVM by the presiding officer can be reduced with efficient design of the system
architecture.
A Novel Approach for RFID Based Smart EVM System 665

1.3 Hardware Design and Necessary Information

The proposed structure consists of RFID (125 kHz) card tag, LF RFID reader, fin-
gerprint sensor (Adafruit), Arduino Uno, LCD shield, jumper wires, button, bread
board and a computer. The following picture shows the proposed structure (Fig. 4).
The system works in a very simple way. Each voter has been provided a tag that
has a unique ID and other necessary information required to verify the identity. At

Fig. 4 Proposed system a security layer at the entrance of the voting room, b security system to
activate the EVM
666 D. Sen et al.

the very beginning, the ID cards (tags) are registered, i.e., each card ID is registered
with the card holder’s name, D.O.B and address is stored in the database with other
biometric information and signature. The readers are attached at the doors of voting
rooms.
For our testing purpose, five RFID tags (voter ID cards) have been used for our
system and for the same the database is provided in Table 1.
Now, when a voter brings the RFID tag (voter ID card) in front of the reader, then
the tag ID will be read by the reader and it will check the current read ID with the
stored IDs in the database. After the verification of the necessary data, the voter will
be allowed to enter the voting room. Figure 5 shows the verification of voter ID at
the entrance point.
Now, a biometric sensor, processor interface attached with the EVM system will
verify the fingerprint of the respective voter before activating the EVM for a particular
time period. The EVM system will only get active when the stored data (fingerprint)
in the central database matches with the current input data. In all other cases, the
EVM will be inactive (Table 2).

Table 1 Registered database I

Id Name Sex Date of birth Address
030001222B Amit Dutta M 10/01/1965 20/C Sahapur colony, New Alipore
Sahapur, Kolkata: 700053
03004022F0 Bijoy Tudu M 05/02/1970 19/A Allenby Rd, Near Northern Park,
Bhowanipore, Kolkata: 700025
030001BA61 Himi Ray F 04/05/1980 Scheme III-M Phool Bagan CIT Road,
Kankurgachi, Kolkata: 700054
030001AE81 Sonu Khan M 14/11/1975 19/22 A, Anandpur, Near Ruby general
hospital, Kolkata: 700107
03002F2CE6 Gargi Das F 12/05/1990 25/6/1C Bin Nagar, Kolkata: 700086

Fig. 5 Gate mechanism when tag matches

A Novel Approach for RFID Based Smart EVM System 667

Table 2 Registered database

Id Name Fingerprint
II
030001222B Amit Dutta

03004022F0 Bijoy Tudu

030001BA61 Himi Ray

030001AE81 Sonu Khan

03002F2CE6 Gargi Das

2 Paper Preparation

After verifying the data with the existing information in the central database, the
access will be given to the eligible voter. The biometric sensor, processor inter-
face connected with the EVM system performs the decision-making criteria and it
activates the EVM for few seconds for the purpose of voting [9] (Fig. 6).

2.1 Algorithmic Approach (Step by Step)

Step 1: The tag IDs are registered in the central database.

Step 2: The RFID smart card is brought in close proximity of the reader.
Step 3: The unique ID of the card is detected by the reader.
Step 4: The current read ID is now compared with the stored tag IDs.
Step 5: If the input ID matches with the stored ID, then the voter is allowed to enter
the voting room.
668 D. Sen et al.

Fig. 6 Granting the access

Step 6: If the input fingerprint matches with the stored fingerprint, then the EVM
system is activated.

2.2 Conclusion

The paper provides a solution to the issues related to voting in this country by
introducing the design and implementation of an RFID-based smart EVM system.
This system will secure the complete voting procedure and from the experimental
outcome, it has been observed that it can provide a layer-wise security and data
verification. One of the advantages of this system is low cost, reliability and high-
end security. In our laboratory, the system that we have implemented is an offline
one and also in LF range. In this article, a novel approach has been incorporated to
secure the voting method and to preserve the democratic rights of an individual. In
the future, this
system can be implemented by adding more advance, secure and flexible features
to prevent forging, rigging, etc.

Acknowledgements Authors of this work would like to thank the financial support obtained from
the DST-WB project. Ref. No: 287(Sanc.)/ST/P/S&T/6G-43/2017.

References

1. Bag, J., Rajanna., K.M, Sarkar, S.K.: Design and FPGA implementation of a Zig-bee enabled
processor for RFID reader suitable for power efficient home/office automation. Eur. J. Sci. Res.
97(4), 592–603 (2013)
2. Bag, J., Sarkar, S.K.: Development and VLSI implementation of a data security scheme for
RFID system using programmable cellular automata. Int. J. Radio Freq. Ident. Technol. Appl.
4(2), 197–211 (2013)
A Novel Approach for RFID Based Smart EVM System 669

3. Bag, J., Roy, S., Kantha, B., Sarkar, S.K.: Advanced multiple step security (AMSS) scheme using
PCA for RFID system and its FPGA implementation. Int. J. RFID Technol. Appl. Inderscience
4(4), 325–341 (2015)
4. Siio, I., Hayasaka, T.: Putting information to things: RFID tags and their application. IPSJ Mag.
40(8), 846–850 (1999)
5. Wang, S.-W., et al.: RFID application in hospitals: a case study on a demonstration RFID project
in a Taiwan hospital. HICSS2006, vol. 8, 4–7 Jan 2006
6. Ranasinghe, D.C., Engels, D.W., Cole, P.H.: Security and privacy solutions for low-cost RFID
system. In: Proceedings of IEEE ISSNIP, Melbourne, Australia, pp. 337–345, 5–8 Dec 2005
7. Sarma, S.E., Weis, S.A., Engels, D.W.: RFID systems ans security and privacy implications.
Technical report MITAUTOID-WH-04. Auto Center, MIT, US (2002)
8. Zhang, J., Tian, G.Y., Marindra, A.M.J., Sunny, A.I., Zhao, A.B.: A review of passive RFID tag
antenna-based sensors and systems for structural health monitoring applications. Sensors 17(2),
265 (2017)
9. Juels, A.: RFID security and privacy: a research survey. IEEE J. Sel. Areas Commun. 24, 381–394
(2006)
Design of a Low Power and Low Phase
Noise VCO Using Active Resistor
and DTMOS

Shasanka Sekhar Rout, Satabdi Acharya, and Kabiraj Sethi

Abstract This brief represents a low power and low phase noise voltage controlled
oscillator (VCO) using dynamic threshold MOSFETs (DTMOS) and active resistors.
One pair of nMOS and one pair of pMOS transistors are associated in between
CMOS cross-coupled pair, which serves as active resistors in series with the cross-
coupled pair. DTMOSs are used instead of normal MOSFETS, in which the body
is dynamically connected to the gate so that the body to source voltage will be
adjusted with the change in gate voltage. This is responsible for threshold voltage
reduction. With a decrease in the threshold voltage, the output voltage swing of
VCO increases and the phase noise decreases accordingly. In this work in addition to
conventional DTMOS, a capacitive divider circuit is used, which solves the issue of
forward biasing of pn-junction at body to source of MOSFETs. The proposed VCO
architecture is implemented in 180 nm CMOS technology and simulated in cadence
tool. It demonstrates a tuning range of 1.78 GHz to 1.81 GHz and lower phase noise
of −118 dBc/Hz at carrier offset of 1 MHz and −48 dBc/Hz at carrier offset of 1 kHz
with a power consumption of 0.15 mW. So, this work will be a proper building block
for the design of a customized receiver.

Keywords Active resistor · Carrier offset · DTMOS · Phase noise · VCO

1 Introduction

Low-frequency flicker (1/f) noise in MOS transistors is higher than that found in
other semiconductor counterparts and the corner frequency shows a range of some
hundred kHz to several MHz. The 1/f noise effect becomes more serious with scaling.
In CMOS oscillators, since the 1/f noise gets upconverted to the close-in phase noise
close to the carrier frequency [1], the performance of phase noise at lower offsets

S. S. Rout (B)
GIET University, Gunupur 765022, India
e-mail: [email protected]
S. Acharya · K. Sethi
Veer Surendra Sai University of Technology (VSSUT), Burla, Odisha 768018, India
© Springer Nature Singapore Pte Ltd. 2020 671
G. Pradhan et al. (eds.), Advances in Electrical Control and Signal Systems,
Lecture Notes in Electrical Engineering 665,
https://doi.org/10.1007/978-981-15-5262-5_50
672 S. S. Rout et al.

from the carrier frequency is conquered by the flicker noise [1]. This up-conversion
of 1/f noise to close-in 1/f3 phase noise is a major problem in oscillators [1]. The
up-gradation of 1/f3 noise from 1/f phase noise can be minimized considering the
explanations and methods used in previous works.
In MOSFETS, the noise spectrum depends on the square of the transconductance.
Ideally, an oscillator deals with a huge output swing. The large signal swing may
lead to higher 1/f noise in oscillators as larger signal swing implies a larger transcon-
ductance variation. Also when cross-coupled pairs enter linear region, the channel
resistance comes into effect, which lowers the Q-factor. It leads to poor phase noise
presentation where the phase noise spectrum depends inversely on the quality fac-
tor of LC-tank. So source degeneration resistors [2] and drain resistors [3, 4] can
be used, which limit the transconductance and also disinfect the tank from smaller
channel resistance as these resistors are in series with the channel resistances. So it
ultimately suppresses the excess 1/f noise current. But these methods show poor 1/f2
performance due to thermal noise of resistors and along with this drain resistance
with parasitic capacitances at drain creates phase-delay [3], so in this work active
resistors [5] are used which does not sacrifice 1/f2 phase noise.
In [3] and [4], four resistors are connected with the drain of transistors, which
lowers the phase noise, but it creates a phase shift in impulse sensitive function (ISF)
with parasitic capacitances at the drain of transistors. And also these resistors in series
with small conducting resistances of cross-coupled transistors degrade Q-factor of
LC-tank. In [5], two active resistors (consisting of pMOS transistors) are connected
in between cross-coupled pairs. These active resistors exhibit small resistance when
cross-coupled pairs reside in a saturation region and high resistance when cross-
coupled pairs reside in linear region. This operation of change in resistances in
two different regions prevents small conducting resistances of cross-coupled pair in
linear region to degrade Q-factor. But with the extra MOSFETS, power consumption
is a little bit more. In [6], for reduction of power consumption, normal MOSFETs
are replaced by DTMOSs [7], which work on dynamic threshold voltage method.
Application of this method reduces the threshold voltage which results in low power
consumption [8].
Also, the performance of phase noise stands proportionally on amplitude of the
output variation [1], which can be controlled by changing threshold voltage. The
reducing of threshold voltage enhances the variation of output voltage primary to
lower phase noise further. In [6], the conventional DTMOS VCO is described without
the concept of active resistors which does not consider the linear region operation
of cross-coupled pairs. On the other hand DTMOS based VCO supports a limited
output variation where the body to source pn-junction will not be forward-biased
[6, 9].
In this paper, a design of VCO is presented where cross-coupled pairs of VCO are
consisting of DTMOSs and normal pMOS and nMOS transistors as active resistors.
Here a capacitive divider circuit is introduced between the body and gate of cross-
coupled transistors to control the threshold voltage dynamically without the limitation
of body-source forward biasing. The active resistors are used to avoid linear region
Design of a Low Power and Low Phase Noise VCO Using Active … 673

operation. As a result, it gives lower 1/f3 phase noise. This work is implemented
in 180 nm CMOS technology and achieves low 1/f3 close-in phase noise devoid of
much affecting the 1/f2 phase noise [10–12].
The paper is planned as follows. The proposed design with active resistor and
DTMOS concept are analyzed with a brief explanation in Sect. 2. Different CMOS
VCOs results with the corresponding post-simulation outcomes are reported in
Sect. 3. Finally, Sect. 4 concludes it.

2 Circuit Design and Analysis

The schematic of the proposed circuit design is presented in Fig. 1. Here pair of nMOS
and pMOS transistors MN3,4 , and MP3,4 are used as active resistors connected between
the cross-coupled pairs MP1,2 and MN1,2 . To increase the tuning range of VCO, a
capacitor bank with MOS switch topology is used in addition to MOS varactors
CVAR1 and CVAR2 and the tank circuit consisting of LTANK , CTANK, and RTANK . The
varactors are controlled by the control voltage, Vcntr . Two pairs of C-R are used for
better biasing of varactors with a biasing voltage vb . Four capacitor divider circuits
of C1 − C2 are used between the body and gate of cross-coupled transistors to control

VDD

C1 MP1 MP2 C1

C2 C2
vd1 vd2
r1 r2
MN3 MN4
Vcntr
Vn1
CVAR1 CVAR2
C C
R Vb R

vout+ LTANK vout-

v1 v2
RTANK

CTANK
MP3 Vn2 MP4
r3 r4
vd3 vd4
C1 C1

C2 C2
MN1 MN2

Fig. 1 Proposed VCO design schematic

674 S. S. Rout et al.

the threshold voltage dynamically. Vn1 and Vn2 are gate voltages of MN3,4 , and MP3,4
respectively. Buffers are also connected at the output nodes to confirm better output.
MP3,4 and MN3,4 transistors always operate as resistors as these transistors are
applied with a bulky overdrive voltage, so that the state of triode region (|vgd | >
|Vth |) is always satisfied, where vgd and Vth are gate-drain and threshold voltage,
respectively. The conducting resistances r is given in Eq. (1).

L
r=    (1)
K W vgd  − |Vth |

where K is the process coefficient, L is the length of channel and W is the width of
transistors.
After VCO reaches a steady-state, v1 and v2 swing around VDD, meanwhile, vd3
and vd4 follow v1 and v2 separately. When MN1,2 operates in the saturation region, v1
and v2 are about equivalent to the balanced point (VDD ). The transconductance gm of
MN1,2 will not be deteriorated by voltage drop on MP3,4 as the denominator of Eq. (1)
is high at the starting point. So higher gm and quicker switch in saturation state are
achieved, which are requirements for improved phase noise. When v1 (v2 ) is rising
and vd4 (vd3 ) is falling, MN1 (MN2 ) is progressively pressed into deep triode state.
On the other side, MP3 (MP4 ) is enforced to depart deep triode state, as over-driven
voltage of MP3 (MP4 ) begins to decrease. As a result, r3 (r4 ) increases significantly
and this isolates the tank from small resistance of cross-coupled nMOS transistors
effectively. Thus it prevents the degradation of Q-factor by disconnecting the tank
from channel resistances of cross-coupled transistors, so phase noise behaviour is
improved. The same explanation is valid for cross-coupled pMOS transistors (MP1,2 )
and NMOS transistors acting as active resistors (MN3,4 ).
Since active resistors are realized by extra transistors, the flicker noise of these
transistors can appear as 1/f3 phase noise and they also consume a little bit more
power. Hence to reduce the power consumption, dynamic threshold MOSFETS with
lower threshold voltage are used instead of normal transistors. This paper uses a
capacitive divider circuit for employment of DTMOS concept that deals well with
the problem of body-source forward biasing with higher voltage swing which is
a problem in conventional DTMOSs. This capacitor divider circuit comprises two
capacitors C1 and C2. The threshold voltage is given in Eq. (2).
  
 
Vth = Vth0 + γ 2F + Vs − C1 VG  − 2 F (2)
 C1 + C2 

where Vth0 is the zero body bias threshold voltage, γ is the body effect coefficient,
F is a substrate work function and VG and VS are the gate and source voltage
respectively [8]. For ease, the parasitic capacitances are ignored owing to the large
capacitance value of C1 and C2 .
The threshold voltage changes with the change in gate voltage of transistors and
the change in threshold voltage lead to modification in drain current. Thus the output
Design of a Low Power and Low Phase Noise VCO Using Active … 675

voltage swing increases as well as the phase noise. The capacitor divider circuits of
C1 − C2 control the threshold voltage dynamically. The net transconductance of the
cross-coupled transconductor is calculated as follows:

gm gmb C1
Gm = − − (3)
2 2(C1 + C2 )

So from Eq. (3), it is observed that the output voltage swing increases as the
net transconductance increases. From Eq. (2), it is found that the capacitance ratio
C1 /(C1 + C2 ) changes the threshold voltage, which leads to a change in the dc current.
As a result, the swing amplitude gets changed, which is proportionally associated
to the dc current in the current-limited regime but restricted by the supply voltage
in the voltage-limited regime. Further, it can be studied from Eq. (2) that when the
capacitance proportion improves; the output variation enhances in the current-limited
regime, but at the same time reduces owing to forward biasing of pn-junction and
hence the Q-factor degradation in the voltage-limited regime [8]. Also, the output
variation and Q-factor affect the phase noise performance. To preserve the advantages
of high output swing and high net transconductance and thus fast turn-on with lower
power supply and lower power dissipation, the capacitance ratio in Eq. (2) is to be
optimized. Also, this method of using DTMOS preserves the Q-factor of the LC-tank
and dc current. Only the start-up time is affected by the net transconductance given
by Eq. (3). The net transconductance is −gm /2 for the static body-biasing, −(gm
+ gmb )/2 for conventional DTMOS whereas Eq. (3) is given for the DTMOS with
capacitive divider. So this topology encounters moderate start-up time. Therefore,
this topology increases the swing amplitude while maintaining the start-up time, as
compared to the static body-biasing. Also, it restricts the chance of forward biasing
of pn-junction of body and source.

3 Post-simulation Result

The proposed VCO is implemented with 180 nm CMOS technology in cadence tool.
Here, Transient, Periodic Steady State (PSS) and Pnoise analysis are made to find
out the power dissipation, oscillation frequency, phase noise, tuning range, etc. The
device parameters used for the proposed VCO design post-simulation are reported
in Table 1. The tank elements are accepted to get 2.4 GHz frequency. This VCO
consumes 0.15 mW from the supply voltage of 1.12 V. Figure 2 shows the tuning
range with control voltage deviation in the range 0–1.8 V. The frequency tuning range
is achieved from 1.78 to 1.81 GHz (300 MHz) which can be increased by increasing
the branches in the capacitor bank.
The phase noise versus relative frequency plot is drawn in Fig. 3, which results in
phase noise of −118 dBc/Hz at 1 MHz offset frequency and −48 dBc/Hz at 1 kHz
offset frequency. The proposed VCO’s layout is exposed in Fig. 4, where it results
in the silicon area of 0.658 × 0.803 mm2 .
676 S. S. Rout et al.

Table 1 Device parameters

Device parameters Proper value
used for proposed VCO
design (W/L)MP1–MP4 148 µm/0.18 µm
(W/L)MN1–MN4 74 µm/0.18 µm
R 136.6 
RTANK 110.4 
C, CTANK 400 fF
C1 2 pF
C2 1 pF
LTANK 4.6 nH

Fig. 2 Proposed design tuning range

The performance of the proposed work is compared with the performance of

VCOs available in the literature [4–6], which is shown in Table 2. It is found that this
work performs better in terms of power dissipation and phase noise when compared
with [4, 5].
This implemented design results in an improved phase noise of −118 dBc/Hz in
comparison to −113.4 dBc/Hz of [4] and −110 dBc/Hz of [5]. It is due to the active
resistors, which prevent the channel resistance of cross-coupled pairs from degrading
Q-factor of LC-tank. And also due to the use of DTMOS, Vth can be obtained to be
lower, which outcomes in superior voltage variation and therefore lower phase noise.
This design results in the power consumption of 0.15 mW, which is much lower than
0.72 and 0.42 mW of [4] and [5] respectively. The low power consumption is resulted
due to the lowering of Vth . The proposed VCO gives a fundamental frequency of
1.78 GHz which is higher than 1.14 GHz of [6]. Thus, this work results in low phase
noise and low power consumption with a moderate frequency range.
Design of a Low Power and Low Phase Noise VCO Using Active … 677

Fig. 3 Phase noise plot of the proposed VCO

Fig. 4 Layout diagram of proposed VCO

678 S. S. Rout et al.

Table 2 Parameter evaluation of various VCOs

Refs. Process Frequency Supply Power Phase noise Phase noise
(GHz) voltage (v) dissipation in dBc/Hz in dBc/Hz
(mW) (1 K offset) (1 M
offset)
[4] 65 nm 3.3 1.2 0.72 −47 −110
[5] 65 nm 5.71 0.6 0.42 −43.3 −113.4
[6] 0.18 µm 1.14 0.34 0.103 – −121.2
This work 0.18 µm 1.78 1.12 0.15 −48 −118

4 Conclusion

The proposed work using a dynamic threshold MOSFET (DTMOS) and active resis-
tors are designed and implemented with 180 nm RF CMOS technology. The proposed
technique achieves a low power and optimizes the swing amplitude, so it reduces the
phase noise according to the capacitance ratio as well as it prevents forward-biased
body to source pn-junction. Therefore, this proposed VCO can be used in mixer
design of the receiver front end.

References

1. Hajimiri, A., Lee, T. H.: Phase noise in CMOS differential LC oscillators. In Symposium on
VLSI Circuits, pp. 48–51. Honolulu (1998). https://doi.org/10.1109/vlsic.1998.687999
2. Yun, S.-J., Cha, C.-Y., Choi, H.-C., Lee, S.-G.: RF CMOS LC-oscillator with source damping
resistors. IEEE Microw. Wirel. Compon. Lett. 16(9), 511–513 (2006). https://doi.org/10.1109/
LMWC.2006.880700
3. Pepe, F., Bonfanti, A., Levantino, S., Samori, C., Lacaita, A. L.: Suppression of flicker noise
up-conversion in a 65 nm CMOS VCO in the 3.0-to-3.6 GHz band. IEEE J. Solid-State Circuits.
48(10), 2375–2389 (2013). https://doi.org/10.1109/jssc.2013.2273181
4. Levantino, S., Zanuso, M., Samori, C., Lacaita, A.: Suppression of flicker noise upconversion
in a 65 nm CMOS VCO in the 3.0-to-3.6 GHz band. In IEEE International Solid-State Circuits
Conference, pp. 50–51, San Francisco (2010). https://doi.org/10.1109/isscc.2010.5434054
5. Sun, J., Boon, C.C., Zhu, X., Yi, X., Devrishi, K., Meng, F.: A low-power low-phase-noise
VCO with self-adjusted active resistor. IEEE Microw. Wirel. Compon. Lett. 26(3), 201–203
(2016). https://doi.org/10.1109/LMWC.2016.2521167
6. Jang, S.-L., Lee, C.-F.: A low voltage and power LC VCO implemented with dynamic threshold
voltage MOSFETS. IEEE Microw. Wirel, Compon. Lett. 17(5), 376–378 (2007). https://doi.
org/10.1109/LMWC.2007.895720
7. Assaderaghi, F., Parke, S., Sinitsky, D., Bokor, J., Ko, P.K., Hu, C.: A dynamic threshold voltage
MOSFET (DTMOS) for ultra-low voltage operation. IEEE Electron Device Lett. 15(12), 510–
512 (1994). https://doi.org/10.1109/55.338420
8. Rout, S.S., Acharya, S., Sethi, K.: A low phase noise gm -boosted DTMOS VCO design in
180 nm CMOS technology. Karbala Inter. J. of Modern Science. 4(2), 228–236 (2018). https://
doi.org/10.1016/j.kijoms.2018.03.001
Design of a Low Power and Low Phase Noise VCO Using Active … 679

9. Levantino, S., Samori, C., Zanchi, A., Lacaita, A. L.: AM-to-PM conversion in varactor-tuned
oscillators. IEEE Tran. Circ. Syst. II: Analog Digital Signal Process. 49(7), 509–513 (2002).
https://doi.org/10.1109/tcsii.2002.804051
10. Lee, J.-H., Kim, S.-Y., Cho, I., Hwang, S.: 1/f noise characteristics of sub-100 nm MOS
transistors. J. Semicond. Tech. Sci. 6(1), 37–41 (2006)
11. Pepe, F., Bonfanti, A., Levantino, S., Samori, C., Lacaita, A., L.: Analysis and minimization of
flicker noise up-conversion in voltage-biased oscillators. IEEE Trans. Microw. Theory Tech.
61(6), 2382–2394 (2013). https://doi.org/10.1109/tmtt.2013.2259257
12. Bonfanti, A., Pepe, F., Samori, C., Lacaita, A. L.: Flicker noise up-conversion due to harmonic
distortion in van der Pol CMOS oscillators. IEEE Tran. Circ. Syst. I: Regul. Pap. 59(7), 1418–
1430 (2012). https://doi.org/10.1109/tcsi.2011.2177132
A Stand-Alone Sine-Wave Inverter
with Reduced Switching Loss

Nanda Kishor Panda, Ramprasad Panda, and Jagadish Kumar Patra

Abstract Sine pulse width modulation is one technology used mostly in power
inverters nowadays to reduce bulky filter requirements and give a pure sinusoidal
wave. This paper brings forward a novel stand-alone sine-wave inverter utilizing
sine PWM technology in a full-bridge inverter with a modified topology having
two additional buck switches connected at the output of a conventional H-bridge
topology. This inverter totally deals with six switches from which the two additional
switches are the only ones operating at high frequency while the other switches
operate at the low (line) frequency. This improves the voltage control and improves
the overall efficiency by reducing switching loss. The above topology aims to reduce
the switching losses by half as compared to a standard H-bridge hence increasing
the efficiency as well as increasing the reliability of the high switching switches
as they operate alternatively for only half cycle. The entire idea was simulated and
verification of the same was done in laboratory utilizing a prototype model.

Keywords Sine PWM · H-bridge · Buck switch · Reliability · Efficiency ·

Switching loss

1 Introduction

An inverter is required to transfer power from a DC source to an AC load and has a

wide variety of applications such as adjustable-speed AC motor drives, uninterrupt-
ible power supplies, power controlling devices, and flexible AC transmission system.
A full bride inverter bears four switches as shown in Fig. 1. They are managed by
phase-shifted square wave drives or utilizing PWM techniques.

N. K. Panda
Department of Electrical Engineering, University of Technology, Eindhoven PO Box-513, The
Netherlands
R. Panda · J. K. Patra (B)
Department of Electrical and Electronics Engineering, Silicon Institute of Technology, Silicon
Hills, Bhubaneswar, Odisha 751024, India
e-mail: [email protected]

© Springer Nature Singapore Pte Ltd. 2020 681

G. Pradhan et al. (eds.), Advances in Electrical Control and Signal Systems,
Lecture Notes in Electrical Engineering 665,
https://doi.org/10.1007/978-981-15-5262-5_51
682 N. K. Panda et al.

Fig. 1 Full-bridge inverter

Switching harmonic is easier to filter in this case as the switching devices are
operated at far superior frequency when compared with the output frequency. On
the other hand, square wave inverters are steered by the power frequency itself thus
leading to lower switching losses while leading to greater harmonic injection which
in turn warrants the need for bulky filters and additional losses.
There are many PWM techniques such as unipolar, bipolar, phase displacement
control, and modified sine [1]. Unipolar switching tends to be more complex in its
implementation but its output waveform has a upper hand with respect to quality
when compared with waveform of bipolar switching. The other techniques [2] tend
to provide better voltage regulation at the cost of waveform quality. These PWM
techniques are applied to generate various waveforms such as sinusoidal, trapezoidal,
staircase, and stepped for modulation. Of these, the sine PWM is the frequently used
one even though it endures a drawback of low fundamental voltage.
Reduction in switching losses is done by implementation of soft switching meth-
ods as integration of PWM mostly leads to high switching losses. [3, 4] Higher current
and voltage stresses on main switches accompany the soft switching techniques by
introduction of auxiliary switches with larger ratings, thus reducing the power output
limit.
One of the most efficient PWM techniques is hybrid switching [5, 6] where two
switches are steered by high frequency PWM signals while the others follow line
frequency square wave signals. Devices operating at low switching frequency can
now be used instead of devices with greater switching frequency; thus, leading to
reduction in conduction losses. The problem of unequal distribution of switching
loss still persists in hybrid PWM technique when operating with larger loads or at
greater switching frequencies. Due to this, the reliability of the system reduces as the
switches operating at higher frequencies dissipate more heat when set side by side
with the other switches.
In [7], the author proposes a particular method for HPWM full-bridge inverter by
which the problem of unequal distribution of switching losses among the switches
can be eradicated.
A Stand-Alone Sine-Wave Inverter with Reduced Switching Loss 683

In this paper, as a substitute of giving sine PWM signals to the switches, two
additional switches are connected to which the sine PWM gate signals are given,
each for half the time period. The inverter switches are operated at line frequency.
As only two switches are operated at high frequency while the remaining switches
are operated at low frequency, overall switching losses are slashed down consider-
ably thus improving efficiency. This setup allows simpler control of frequency and
modulation index.
Inverters are classified on the basis of their operation as voltage sources and
current source inverters. Inverters can also be classified on the basis of the network
as bridge, series, and parallel inverters. Bridge inverters require the use of switches
which are connected in a network and operated using a control signal in a particular
sequence to obtain the required output waveform. There exist various kinds of bridge
inverters topologies, foremost of them being have wave and full wave inverters. While
half wave inverters utilize just two switches instead of the four switches commonly
resent in full wave inverters, and they lag behind in efficiency and waveform quality.
A full-bridge inverter topology consists of four switching devices as shown in Fig. 1.
They are steered by making use of phase-shifted square wave drives or by PWM
techniques.
Full-bridge inverters commonly employ PWM switching techniques to obtain
pure sine-wave outputs. This is due to their various advantages, namely
Easier control of output voltage
Minimization of lower order harmonics
Lower filter requirements
Lower power consumption
PWM inverters are classified on the basis of their levels, i.e., two-level and three-
level as depicted in Fig. 2. A triangular carrier wave when compared with reference
sine-wave provides us with PWM signal in the easiest possible method. Comparator
takes the carrier and reference wave as its input and provides output as two-level
PWM signal. Three different voltage levels that is zero, low, and high can be incor-
porated to obtain three-level PWM possessing greater closeness to true sine wave.
The output obtained, i.e., three-level PWM signal to correspond to a sine wave, and
the comparison state should also be three-level. A triangular wave is used after mak-
ing its amplitude half and summing it with a sine signal at a time. One half of the
bridge circuit can be steered utilizing the output PWM signal, while the other half
of the bridge network is controlled by square wave of same frequency and phase as
the sine signal. Voltage polarity across the load is monitored by this other half. Sim-
ilarly, higher levels of PWM can be achieved using the previous levels for improved
performance.
It can be noted in Fig. 3, to create a dead time between the switching pulses
for +ve and −ve half cycle, the modulating signal is compared with a small bias
voltage(0.1 V) instead of the ground potential. The proposed topology adds two buck
switches (MOSFET) in back-to-back connection across the output of the H-bridge
inverter. The two switches work consecutively in the two positive and negative half
cycles. As shown in Figs. 3 and 4, they are connected in series with two diodes
to block the residue currents in closed conditions. In the first cycle, the switch M5
684 N. K. Panda et al.

Fig. 2 PWM levels

Fig. 3 Generation of sine PWM

A Stand-Alone Sine-Wave Inverter with Reduced Switching Loss 685

Fig. 4 Proposed topology

superimposes the positive wave form with sine PWM whereas the switch M6 does
that in the negative half cycle.

2 Simulation Results

The proposed topology was simulated using PSIM. The carrier signals generated
were (i) sinusoidal (50 Hz, 8 V) and (ii) unipolar triangular (20 kHz, 10 V). The
modulation index was fixed at 0.8(mi = 0.8).The circuit was simulated and output
waveform of all the gate signals along with the waveform of voltage across the load
with and without filers which were generated and are shown in Fig. 5 (Fig. 6).

3 Calculation and Discussion

A prototype was prepared in the laboratory to verify the utility of the suggested
inverter. The inverter was given 50 V DC input and operated at 80watts. The gate
signals for S1, S4, S2, and S3 were generated by using IC LM741 whereas the
gate signal for S5 and S6 was generated from the same synchronized carrier signals
using IC LM311. All the gate signals were properly conditioned and isolated using
MOS driver (IR2110) before feeding it to the gate of the MOSFETs. The MOSFETs
used were IRF540. The modulation index for the sine PWM was fixed at 0.8. The
final output was filtered using a low-pass LC filter with corner frequency 1 kHz
and 3db disturbances. The various waveforms at different point were recorded. The
components of the low-pass filter were designed as per the following equations:

Eout ∗ Toff
L= = 1.458 mH (1)
0.5 ∗ Iout
δ ∗ Iout
C= = 1674 µF (2)
8 ∗ f ∗ δVout
686 N. K. Panda et al.

Fig. 5 Gate signal and Inverter output without filter

Fig. 6 Inverter output with filter

A Stand-Alone Sine-Wave Inverter with Reduced Switching Loss 687

Fig. 7 Gate signals

Fig. 8 Output of inverter

with filter

The losses [8] incurred in the proposed topology were compared with a standard
full-bridge inverter as shown in Table 1. The characteristics are considered for IR2110
MOSFET [9] operating at 25 °C and tested with the hardware setup shown in Fig. 9
(Figs. 7 and 8).
 
Switching loss = (0.5) ∗ UDD ∗ IDoff ∗ tr + t f ∗ f sw (3)

U DD Input voltage
I Doff Output current
tr Rise time
tf Fall time
f sw Switching frequency
688 N. K. Panda et al.

Table 1 Switching loss

Parameters Value
tr 44 ns
tf 43 ns
U DD 300 V
I Doff 1A
f sw1 50 Hz
f sw2 1 kHz
Switching loss for circuit shown in Fig. 1 0.0522 W
Switching loss for proposed topology 0.02871 W

Fig. 9 Hardware setup

We can see a saving of 45% in switching losses between the proposed topology
and a standard full-bridge inverter with minimum THD as verified by the hardware.
This showcases the effectiveness of the proposed topology as a viable alternative and
the cause for further development in this design.

4 Conclusion

By modifying a standard H-bridge topology, this paper puts forward a novel sine-
wave inverter topology which provides us improved efficiency. The output THD
was found out to be well under control. Easy methodology was used for switching
of the involved devices. The overall efficiency was boosted by making reduction
A Stand-Alone Sine-Wave Inverter with Reduced Switching Loss 689

in switching losses as only two of the switches are operating at high frequency.
The proposed topology can be utilized for both low power and also medium power
applications in the field of photovoltaic, backup power in UPS, etc. As only two
switches use sine PWM signals for harmonic compensation in active filters, using
this topology will be a better choice.

References

1. Bitar, A.S.J., Crowley, I.F.: PWM techniques : a pure sine wave inverter, p. 91 (2011)
2. Panda, R., Tripathi, R.K.: A symmetrical hybrid sine PWM switching technique for full bridge
inverters. In: Proceedings of India International Conference on Power Electronics, IICPE 2006,
pp. 345–348 (2006)
3. De Uberhdia, U.F.: Soft switched single phase inverter with special PWM technique (1998)
4. He, X.H.X., Kong, J.K.J., Qian, Z.Q.Z., Williams, B.W.: Analysis and design of composite soft
switching single phase\ninverters with PWM control. In: Proceedings of IEEE 1999 International
Conference on Power Electronics and Drive Systems PEDS’99 (Cat. No.99TH8475), vol. 2
(1999, July)
5. Yousefzadeh, V., Maksimović, D., Li, Q.: A zero voltage switching single-phase inverter using
hybrid pulse-width modulation technique. In: PESC Record—IEEE Annual Power Electronics
Specialists Conference, vol. 2, pp. 1274–1279 (2004)
6. Mao, H., Abu-Qahouq, J., Luo, S., Batarseh, I.: Zero-voltage-switching half-bridge DC-DC
converter with modified PWM control method. IEEE Trans. Power Electron. 19(4), 947–958
(2004)
7. Ai, T.H., Chen, J.F., Liang, T.J.: A random switching method for HPWM full-bridge inverter.
IEEE Trans. Ind. Electron. 49(3), 595–597 (2002)
8. Graovac, D., Pürschel, M., Andreas, K.: MOSFET power losses calculation using the data-sheet
parameters. Infineon Technol. AG, 1, 1–23 (2006)
9. Siliconix, V.: IRF540, SiHF540, no. V
Mitigation of Power Quality Issues Using
DSTATCOM Through the Theory
of Direct-Detection Light Wave Receivers
Using Optical Amplifiers

S. M. Sanjay, M. Saritha, and M. H. Sidram

Abstract In power systems, usually it is difficult to maintain the quality of electrical

power within the acceptable limits. Mainly the use of power electronic devices that
act as the nonlinear loads is responsible for the degradation of the quality of power.
Power equipment has rapidly expanded with a variety of sensitive loads that require
ideal sinusoidal supply voltage for their reliable operation. So, it became necessary
to include different types of compensation with the intention to meet IEEE standards.
In this paper, DSTATCOM is used as a compensating device to mitigate the power
quality issues in the radial distribution system. Theory of Direct-Detection Light wave
Receivers using Optical Amplifiers is used for the generation of reference current
and for installation of DSTATCOM unit that result in power loss reduction, minimum
load balancing index and maximum voltage profile. This work also concentrates on
the impact of phase noise. In order to pre-identify the most candidate bus for the
optimal placement of DSTATCOM Newton–Raphson technique is used, which is an
iterative approach for the load flow analysis. The proposed methodology is verified
on IEEE-30 bus radial distribution system, by implementing in MATLAB simulation
platform.

Keywords DSTATCOM · IEEE-30 bus · Maximum voltage profile · Minimum

load balancing index · Optical amplifiers · Power loss reduction

1 Introduction

Quality of power defines the importance of electrical power to consumer equip-

ment without any disturbance or damage appliances. Power quality issues involve
many possible situations where the load current or voltage waveforms deviates from
the ideal sine wave provided standard frequency and amplitude for any three-phase
machines [1]. The main cause for variations in voltage profile is due to various power
quality issues such as sag/swell, transients, harmonics and flicker are restricted. In the

S. M. Sanjay (B) · M. Saritha · M. H. Sidram

Department of E&EE, JSS S&TU, Mysuru 570006, India
e-mail: [email protected]

© Springer Nature Singapore Pte Ltd. 2020 691

G. Pradhan et al. (eds.), Advances in Electrical Control and Signal Systems,
Lecture Notes in Electrical Engineering 665,
https://doi.org/10.1007/978-981-15-5262-5_52
692 S. M. Sanjay et al.

recent era, the load disturbances can be differentiated depending upon the cause as
well as issues that relate to the quality of voltage delivered and the requirement of the
consumer [2]. Generally, IEEE 1159 standard is used for the operation of the power
system and grounding of the equipment appropriately to overcome power quality
issues. Electrical equipment tends to react to disturbances easily. Therefore a rise in
demand to overcome the disturbances has been increased from the past many years.
Hence scope for mitigation of power quality issues has been increased deliberately
[3].
Custom power devices have been used to mitigate the power quality issues in the
distribution system to overcome the disturbances caused by voltage profiles such as
sag/swell, harmonics, and short interruptions [4]. Voltage dips are mainly avoided
by proposing specific custom power devices such as voltage source converter (Heart
of the DSTATCOM) connecting in parallel to the equipment [5]. In modern grids,
current is injected using devices such as STATCOM or DSTATCOM to compensate
reactive power [6].
DVR, DSTATCOM, and UPQC are the devices used alternatively to overcome
voltage and current based power quality issues. The major problems being faced
nowadays at the distribution side totally affects the quality of power issued to
the consumer. Hence, DSTATCOM is the best available technology that is used
to compensate the majority of the issues involved in power quality [7].
The optimal allocation of DSTATCOM, results in a decrease in losses during
the distribution of power in electrical machines also improves the voltage profile.
Various advantages that involve by allocating DSTATCOM, concurrently depicts a
notable solution on lowering power loss, compensation of voltage profile, electric-
ity aspect correction, balancing of load, electricity pleasant development, on-height
working fees discount, freeing the overloading of distribution traces, device balance
development, pollutant emission reduction and expanded typical strength efficiency
[8].
The main application of optical amplifier is to improve the sensitivity of the
receiver by using an additional preamplifier. Recently, demand has been created to
use optical pre-amplifiers in order to improve the performance and detection by the
sensitivity of light wave receivers. Additionally, this work proposes an evaluation
technique which is very simple to figure out the effect of noise caused by phase
on direct detection of light wave receiver using optical amplifiers [9]. The paper is
organized as follows. Section 2 discusses the proposed methodology. Results and
discussion have been described in Sect. 3. The conclusion is outlined in Sect. 4.

2 Proposed Methodology

The main objectives of the proposed work are to regulate voltage at the point where
load is connected sensitively, machine disturbances along with minimum load bal-
ancing index, and power loss reduction. This work specifically examines the behavior
of DSTATCOM which compensates the current drawn from a utility by comparing
Mitigation of Power Quality Issues Using DSTATCOM Through the … 693

Fig. 1 Block diagram of the

proposed methodology

the Q limits that are set initially, without affecting the voltage. The objectives of the
proposed work are achieved effectively by the implementation of optical amplifiers,
whose gain and spontaneous emission noise are controlled to enhance the overall
result of the system along with the optimal placement of DSTATCOM (Fig. 1).

2.1 Load Flow Analysis

In power systems, load flow analysis is used to deal with steady-state operation of a
network involving various node voltages and branch power flows in the system by
designing a specific transmission structure. In order to pre-identify the most candidate
bus for allocating DSTATCOM. The load flow evaluation is achieved for IEEE-30
bus radial distribution system with Newton–Raphson approach, which is an iterative
technique for the load flow analysis.

2.2 Concept of DSTATCOM

DSTATCOM is the voltage source converter used to mitigate the power quality issues
by injecting reactive power into the system where it is powered by small duration
instant electric power which is stored in a dc capacitor. The filters of DSTATCOM
usually the specifications that are pre-defined in the software [10]. The main principle
behind DSTATCOM is that, compensation of reference currents. The device first
generally injects three-phase currents after comparing it with reference currents by
canceling the disturbances caused by the load. Hence scope to overcome power
quality issues using injection of reference currents has been recommended [11].
With shunt connected DSTATCOM, voltage regulation can be done very precisely
694 S. M. Sanjay et al.

by compensating various disturbances. But the only criteria of this device to overcome
live performance of the compensator while mitigating power quality.
The operation of VSI is done with a suitable dc shunt capacitor device. The reaction
of the DSTATCOM is very massive even as compensating AC or DC equipment [12].

2.3 Optical Amplifiers

An optical amplifier is a tool that amplifies an Optical Signals at once to the electrical
signals without the need of a converter. The concept involving laser without an optical
hollow space or one wherein outputs from the hollow space is nothing but the optical
amplifier. They are used as optical repeaters in the lengthy distance fibrotic which
bring a whole lot of the arena’s telecommunication links.

2.4 Description of the Receiver

The input signal that is amplified with a flat gain G in a frequency filtered and
constrained by band and obtained by a photodetector. For the receiver of Fig. 2a, the
detector output is sampled as soon as every bit interval and relying on the outcomes.
The receiver makes a decision whether the active power is to be injected or not by
comparing the Q limits. The incoming signal is represented in Eq. (1).
√
ei (st) = 2P cos[2π f t + ∅(t)] (1)

Fig. 2 a Single-filter direct-detection lightwave receiver with an optical preamplifier. b Dual-filter

direct-detection lightwave receiver with an optical preamplifier
Mitigation of Power Quality Issues Using DSTATCOM Through the … 695

where f, P, ∅(t)is the optical frequency, optical electricity and the laser phase noise
respectively.
The amplifier output consists of a signal that is amplified with the addition of
big-band Gaussian noise n(t) also with spontaneous emission that is amplified as
given in Eq. (2).
√
e0 (t) = 2G P cos[2π f t + ∅t] + n (2)

The desired signal is passed by the polarization filter which blocks the light within
the orthogonal polarization. The bandpass filter with bandwidth B is a finite impulse
response filter output. The optical wave e f (t) that is filtered (amplified signal plus
noise) is detected by a photodetector of quantum efficiency ï(≤1), yielding an output
current represented by an Eq. (3).
ηq
i(t) = |E F (t)|2 + NS + NT (3)
hf

where q, E f , N S and N T are the electron charge, complex amplitude of ef (t), shot
noise and thermal noise respectively.
The fluctuations of the thermal noise and shot noise can be made negligible if
sufficient gain of an amplifier is considered. Receiver block diagram as shown in
Fig. 2a depicts the ideal case where amplified signal does not consist of phase noise,
i.e. (t) ≡ zero. The detection that is proved in Fig. 2a happens in very rare conditions.
Figure 2b represents the light wave receiver which detects the optical alerts which
are impaired with the help of phase noise for everyday direct detection. The signal
spectrum gets broadened by the phase noise where larger bandpass filter’s (BPF)
usage is depicted. General performance of electrical equipment is affected by the
big bandpass filter outputs due to large bandwidths because of extra spontaneous
emission noise. Hence, the post-detection filtering method is employed using the
low-bypass filter as illustrated in Fig. 2b to eliminate the effect of spontaneous
emission noise. The evaluation of the overall performance of the receiver shown
in Fig. 2b is very rare. One evaluation method has been provided for the special case
of negligibly small line widths.

3 Results and Discussions

Performance of the direct detection lightwave receiver can be concluded that it

depends mainly on the width of transmitter laser line and bandwidth of the opti-
cal BPF. If the optimum filter bandwidth is wider, then instantly performance of
the system will become poor due to excess spontaneous emission noise. Hence,
post-detection filter is used to overcome such problems. Standard IEEE-30 bus is
employed in this work, where load flow analysis is carried out by Newton–Raphson
696 S. M. Sanjay et al.

method to pre-identify the most candidate bus and results are tabulated as shown in
Table 1. Load flow results of base case have been tabulated in Table 2.
From the above load flow evaluation with the help of Newton–Raphson method
obtained the graph for voltage profile as shown in Fig. 3. It can be observed that the
voltage dip is maximum in 26th and 30th bus of IEEE-30 bus system (Table 1).
Installation of DSTATCOM unit is perceived by using Theory of Direct-Detection
Light wave Receivers with the usage of optical amplifiers. The proposed work is
implemented in MATLAB simulation with version R2018a. After allocating the
DSTATCOM unit in the 26th and 30th buses the voltage profile of all of the buses
in the IEEE-30 bus system is improved to 1 p.u. as shown in Table 3. Table 4 offers
the records about the most desirable placement of DSTATCOM in IEEE-30 bus
system. Comparison of Voltage Profile of IEEE-30 Bus System with and without
DSTATCOM is shown in Fig. 4.
According to the “Theory of Direct-Detection Light Wave Receivers Using Opti-
cal Amplifiers” the power loss graphs for distinctive instances are obtained as shown
below. Figure 5 suggests the load balancing index of the equipment. Figures 6, 7
and 8 explain approximately the power losses with respect to the voltage profile by
considering the gain of 25.99 dB, 26.14 dB and 35.35 dB respectively. Power losses
with respect to the voltage profile by means of considering the spontaneous emis-
sion noise N1, N2, N3 and ratio of N2 to N1 are shown in Figs. 9, 10, 11 and 12,
respectively.
From the above result, it is also concluded that the power loss is minimum for
maximum gain and also for the spontaneous emission noise ratio of N2 to N1, in

Table 1 Bus voltages before the installation of DSTACOM

Bus No. Bus voltage Angle (degree) Bus No. Bus voltage Angle (degree)
(p.u) (p.u)
1 1.0600 0.0000 16 1.0304 −15.6251
2 1.0430 −5.3543 17 1.0188 −15.8687
3 1.0196 −7.5308 18 1.0114 −16.6067
4 1.0104 −9.2840 19 1.0066 −16.7658
5 1.0100 −14.1738 20 1.0095 −16.5502
6 1.0096 −11.0581 21 1.0082 −16.2178
7 1.0020 −12.8649 22 1.0120 −15.9811
8 1.0100 −11.8193 23 1.0085 −16.2294
9 1.0392 −14.0644 24 0.9991 −16.3007
10 1.0215 −15.6706 25 1.0032 −16.0720
11 1.0820 −14.0644 26 0.9852 −16.5038
12 1.0496 −15.1245 27 1.0145 −15.6559
13 1.0710 −15.1245 28 1.0078 −11.7163
14 1.0320 −16.0018 29 0.9944 −16.9077
15 1.0251 −16.0084 30 0.9828 −17.8067
Mitigation of Power Quality Issues Using DSTATCOM Through the … 697

1.1

1.08

1.06
Voltage in p.u

1.04

1.02

0.98

0.96

0.94

0.92
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30
Bus No.

Fig. 3 Voltage profile of IEEE-30 bus system without DSTATCOM

Table 2 Load flow results for base case

P-generation 0.4 MW
Q-generation 1.511 Mvar
P-load 2.834 MW
Q-load 1.262 Mvar
Total power loss 2.679 MW

Table 3 Bus voltages after the installation of DSTATCOM

Bus No. Bus Voltage Angle (degree) Bus No. Bus Voltage Angle (degree)
(p.u) (p.u)
1 1.0600 0.0000 16 1.0315 −15.9675
2 1.0430 −5.3525 17 1.0202 −15.8422
3 1.0200 −7.5318 18 1.0128 −16.5781
4 1.0108 −9.2848 19 1.0080 −16.7371
5 1.0100 −14.1692 20 1.0110 −16.5223
6 1.0101 −11.0627 21 1.0102 −16.1952
7 1.0023 −12.8655 22 1.0147 −15.9655
8 1.0100 −11.8168 23 1.0106 −16.2084
9 1.0402 −14.0508 24 1.0033 −16.3066
10 1.0231 −15.6455 25 1.0125 −16.1605
11 1.0820 −14.0508 26 1.0000 −16.7794
12 1.0503 −15.0935 27 1.0240 −15.7112
13 1.0710 −15.0935 28 1.0091 −11.7434
14 1.0330 −15.9691 29 1.0076 −17.0359
15 1.0263 −15.9809 30 1.0000 −18.0250
698 S. M. Sanjay et al.

Table 4 Optimal placement

Bus No. Vsh p.u Thsh degree Qsh p.u
of DSTATCOM in IEEE-30
bus system 26 1.0014 −16.7872 −0.0137
30 1.0020 −18.0321 −0.0202

1.1

1.08

1.06

1.04
Voltage in p.u

1.02

0.98

0.96

0.94

0.92
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30
Bus No.

Fig. 4 Comparison of voltage profile of IEEE-30 bus system with and without DSTATCOM

Fig. 5 Load balancing index

Mitigation of Power Quality Issues Using DSTATCOM Through the … 699

Fig. 6 Power loss with gain of 25.996

Fig. 7 Power loss with gain of 26.143

which N1 and N2 are the spontaneous emission noise and N3 is the spontaneous
emission noise distinction between N1 and N2.
Implementation of optical amplifiers also helps us to improve the overall efficiency
of the system by considering the higher gains, which results in minimum power loss as
shown below, instead of considering only DSTATCOM to improve the voltage profile
and decrease the power loss of the system. Hence, it is concluded that by controlling
700 S. M. Sanjay et al.

Fig. 8 Power loss with gain of 35.352

Fig. 9 Power loss with spontaneous emission noise N1

the spontaneous emission noise and gain of the optical fiber, the efficiency of the
overall system can be improved.
Mitigation of Power Quality Issues Using DSTATCOM Through the … 701

Fig. 10 Power loss with spontaneous emission noise N2

Fig. 11 Power loss with spontaneous emission noise N3

4 Conclusion

In this paper, a Theory of Direct-Detection Light Wave Receivers using Optical

Amplifiers Technique has been employed for the installation of DSTATCOM unit
702 S. M. Sanjay et al.

Fig. 12 Power loss with spontaneous emission noise ratio of N2 to N1

with the objective of power loss reduction, minimum load balancing index and max-
imum voltage profile. The method is executed using MATLAB simulation with ver-
sion R2018a. The developed technique has been implemented in IEEE-30 bus system
and results are compared for different gain and spontaneous emission noises. The
proposed technique has resulted in better performance in terms of maximum gain
when compared to other minimum gains and spontaneous emission noise ratio of N2
to N1 when compared to spontaneous emission noise N1, N2 and N3.
From this work, it can be observed that quality of power improved through DSTAT-
COM with Theory of Direct-Detection Light Wave Receivers with the usage of opti-
cal amplifiers for the generation of reference current and for its installation. The
simulation work can be carried out using different software like PSCAD, ETAP,
EMPT-RV, LabVIEW and PSpice.

References

1. Axente, I., Ganesh, J.N., Basu, M., Conlon, M.F., Gaughan, K.: A 12-kVA DSP-controlled
laboratory prototype upqc capable of mitigating unbalance in source voltage and load current.
IEEE Trans. Power Electron. 25(6) (2010, June)
2. Kolhatkar, Y.Y., Das, S.P.: Experimental investigation of a single-phase UPQC with minimum
VA loading. IEEE Trans. Power Deliv. 22(1) (2007)
3. Kwan, K.H., Chu, Y.C., So, P.L.: Model-based H∞ control of a unified power quality
conditioner. IEEE Trans. Industr. Electron. 56(7) (2009)
4. Basu, M., Das, S.P., Dubey, G.K.: Comparative evaluation of two models of UPQC for suitable
interface to enhance power quality. Electr. Power Syst. Res. 77, 821–830 (2007)
5. Kumar, G.S., Vardhana, P.H., Kumar, B.K., Mishra, M.K.: Minimization of VA loading of
unified power quality conditioner (UPQC). In: IEEE Powering 2009 Lisbon, Portugal, pp. 18–
20 (2009)
6. Vardhana, P.H., Kumar, B.K., Kumar, M.L.: A robust controller for DSTATCOM. In: IEEE
Powering 2009 Lisbon, Portugal, pp. 18–20 (2009)
Mitigation of Power Quality Issues Using DSTATCOM Through the … 703

7. Srinivas, M.: Combined LMS-LMF based control algorithm of DSTATCOM for power quality
enhancement in distribution system. IEEE Trans. Power Electron. 63(7) (2016)
8. Yuvaraj, T.: Multi-objective simultaneous DG and DSTATCOM allocation in radial distribution
networks using cuckoo searching algorithm. Alexandria Eng. J. 57(4) (2018)
9. Tonguz, O.K.: Theory of direct-detection lightwave receivers using optical amplifiers. J.
Lightwave Technol. 9(2) (1991)
10. Shuai, Z., Luo, A., Shen, Z.J., Zhu, W., Lv, Z., Wu, C.: A dynamic hybrid var compensator
and a two-level collaborative optimization compensation method. IEEE Trans. Power Electron.
24(9) (2009)
11. Shukla, A., Ghosh, A., Joshi, A.: State feedback control of multilevel inverters for DSTATCOM
applications. IEEE Trans. Power Deliv. 22(4) (2007)
12. Singh, B., Solanki, J.: A comparison of control algorithms for DSTATCOM. IEEE Trans. Ind.
Electron. 56(7) (2009)
13. Kikuchi, K., Okoshi, T., Nagamatsu, M., Henmi, N.: Degradation of bit-error-rate in coherent
optical communications due to spectral spread of the transmitter and the local oscillator. J.
Lightwave Technol. LT-2(6), 1024–1033 (1984)
14. Way, W. I., Zah, C. E., Lee, T. P.: Applications of traveling-wave laser amplifiers in subcarrier
multiplexed light wave systems. IEEE Trans. Microwave Theory Technol. 38 (1990)
15. Ross, M.: Laser Receivers: Devices, Techniques, Systems. Wiley, New York (1966)
16. Shuai, Z., Luo, A., Shen, Z. J., Zhu, W., Lv, Z., Wu, C.: A dynamic hybrid var compensator
and a two-level collaborative optimization compensation method. IEEE Trans. Power Electron.
24(9) (2009)
17. Murty, V.V., Kumar, A.: Optimal placement of DG in radial distribution systems based on new
voltage stability index under load growth. Int. J. Electr. Power Energy Syst. 69, 246–256 (2015)
18. Yuvaraj, T., Ravi, K., Devabalaji, K.R.: DSTATCOM allocation in distribution networks
considering load variations using Bat algorithm. Ain Shams Eng. J. 8, 391–403 (2017)
A Brief Review on Synchro Phasor
Technology and Phasor Measurement
Unit

Monalisa Mohanty, Ravi Kant, Asit Kumar, Debasis Sahu,

and Subhashree Choudhury

Abstract In power system monitoring, supervisory control and data acquisition

(SCADA) system has played an important role by computing the voltage and current
signal for a local area. However, when quicker response is needed, phasor measure-
ment unit (PMU) serves as an effective alternative owing to its faster reporting rate. In
this paper, a brief review on PMU and its contribution to synchro phasor technology
(SPT) has been comprehensively summarized.

Keywords Synchro phasor technology (SPT) · Supervisory control and data

acquisition system (SCADA) · Optimal placement of PMU (OPP) and phasor data
concentrator (PDC) · Power system (PS) · Phasor measurement unit (PMU)

1 Introduction

Nowadays, phasor measurement unit finds its application as a measurement system

by all most all power grid around the world. Phasor measurement unit device is
used to compute the amplitude and phase of electrical quantity like voltage signal
and current signal of the power network taking a universal time signal as reference
for synchronization. The geostationary positioning satellite (GPS) satellite provides
the time synchronization to the data collected from the remote terminal unit (RTU)
and thus, helps in real-time monitoring. PMUs are able to produce sampled phasor
data from the input analog signals. This whole concept is about synchronizing the
data with a common reference, basically time and the measurement is termed as
synchro phasor. PMUs can produce 30–60 measurements per seconds in comparison

M. Mohanty · R. Kant · A. Kumar · S. Choudhury (B)

Department of Electrical and Electronics Engineering, Institute of Technical Education and
Research, Siksha ‘O’ Anusandhan (Deemed to be University), Bhubaneswar, Odisha 751030,
India
e-mail: [email protected]
D. Sahu
Department of Electrical and Electronics, CIT, Centurion University of Technology and
Management, Bhubaneswar 752050, India

© Springer Nature Singapore Pte Ltd. 2020 705

G. Pradhan et al. (eds.), Advances in Electrical Control and Signal Systems,
Lecture Notes in Electrical Engineering 665,
https://doi.org/10.1007/978-981-15-5262-5_53
706 M. Mohanty et al.

to SCADA. So, it can be used for real-time monitoring and helps it developing better
controlling and protection strategy. It can also measure the frequency of power grid
and can also be integrated with a protective relay. This paper briefs about PMUs, its
application and some of the ongoing projects in India [1].
The entire paper organization is as follows. The details about synchro phasor
technology have been enumerated in Sect. 2. PMU and its performance are clearly
described in Sect. 3. The phasor data concentrator (PDC) is detailed in Sect. 4.
Section 5 discusses about PMU and various algorithm for optimal placement of PMU
(OPP). Sections 6 and 7 illustrate the application of SPT and concept of hybrid state
estimation, respectively. Section 8 highlights the installation of PMUs for enhance-
ment of present scenario of Indian Power System. Finally, in Sect. 9, the conclusion
from the entire study has been drawn.

2 Synchro Phasor Technology (SPT)

In power system, analog signals obtained from current transformers (CTs) and poten-
tial transformers (PTs) are fed to SCADA system that communicate the data and send
to load dispatch centre’s energy management system (EMS). In, EMS different Algo-
rithms are present which will compute these data and give the status of operating
condition of power system and necessary control action needed, if it has any issue.
The EMS gets the data from the SCADA in every 4–6 s. The duration of getting
data may vary within less than 4 s or at max at 10 s. This system is applicable only
when small area is concerned but when wide area monitoring (WAM) is considered,
which may be geographically apart having different location, data may be lost in time
queue. In this condition, it is necessary that all data are time stamped centrally or
data are synchronous. This has the advantage of not only monitoring data per cycle
considering the voltage and magnitude but also the phase angle and magnitude can
be kept tracked. Power network stress is marked by separation in phase angle that
helps in real-time monitoring and post analysis especially in case of contingencies
and blackout. Such benefits can be achieved with the help of PMUs and the new
technology called synchro phasor technology [2].
Synchro phasor means synchronized phasor. The concept was first put forth by
Charles Proteus Steinmetz in 1893 [3]. This concept was later used for real-time
phasor measurement with a synchronized time reference provided by GPS satellite
resulted in the invention of PMU in 1988 [4].
Figure 1 gives a structural idea about the synchro phasor technology and how the
synchronized data are used in a wide area monitoring system for providing protection
and control and for post-mortem analysis in case of any outage or disturbances.
Synchro Phasor is defined as “time-synchronized numbers that represent both the
magnitude and phase angle of the sine waves of voltages and currents which takes in
reference an absolute point of time provided by Global Positioning Satellite”. Phasor
and amplitude representation of sinusoidal voltage signal is shown in Fig. 2, V 1 and
V 2 represents the magnitude of the phasor and an angle θ1 and θ2 are basically
A Brief Review on Synchro Phasor Technology and Phasor … 707

Fig. 1 Synchro phasor technology structure [8]

Fig. 2 Phasor and amplitude representation of sinusoidal voltage signal of two buses [3]
708 M. Mohanty et al.

difference in angle between signal’s crest and time reference at time t = 0. This time
reference, t = 0, is given by GPS satellite which provides time synchronizations to
this measurement [5].

3 Phasor Measurement Unit

3.1 Definition of PMU

IEEE standard defines phasor measurement unit as an electronic telemetering appara-

tus that takes voltage signals and current signals obtained from current transformers
and potential transformers as input and generates current phasor and voltage phasor,
respectively, which are time synchronized or time stamped. PMU device also gener-
ates frequency and rate of change of frequency (ROCOF). The synchronizing time
signals are obtained from global positioning satellite through a GPS clock [6].

3.2 Historical Development of PMU

In 1965, a blackout occurred in North-Eastern Power Grid of North America which

led to the necessity of development of digital computer-based relay. Later, symmetri-
cal component concept was formulated to meet the computational challenge, which
led to designing of symmetrical component distance relay. The first prototype of
Phasor Measurement Unit (PMU), which was given by Virginia Tech in 1988 com-
bining the concept of Phase and Symmetrical Component and Global Positioning
Satellite was developed in 1980 [7]. Later, this institute in collaboration with many
manufacturers has developed PMU on commercial basis.

3.3 Architectural Structure of PMU

As per IEEE, PMU is a telemetering device that consists of measurement unit, syn-
chronization unit and data unit [8]. The analog voltage signal and analog current
signal reach PMU from potential transformers and current transformers. This forms
the measurement layer. This is further passed through an anti-aliasing filter following
the Nyquist-Shannon theorem [9]. To prevent aliasing error, IEEEC37.118 has done
it compulsory to use an anti-aliasing filter. Since Butterworth filter (BPF) gives a
flat response in pass band in comparison to Bessel filter, Chelbyshev filter, elliptical
filter, it is highly preferred [10]. Figure 4 shows the response of various anti-aliasing
filters. The signals are then passed through an analog-to-digital (A/D) converter,
which converts analog signals with continuous time and amplitude to discrete signal
A Brief Review on Synchro Phasor Technology and Phasor … 709

with a predefined sampling speed which depends on the converter’s sampling rate.
The anti-aliasing filter, phasor measurement unit processor and A/D converter form
the measurement unit. After the data are obtained from the measurement unit, it is
transmitted with the help of a modulator and demodulator (MODEM).
The synchronization unit consists of phase-locked oscillators and GPS. The data
are time stamped by taking reference of time signals obtained from GPS satellite. The
phase-locked oscillator locks the pulse signal from the satellite with the sampling
clock. The architectural structure of PMU is shown in Fig. 3. Different algorithms
used for phasor estimation are (i) Zero-crossing demodulation, (ii) Sliding discrete
Fourier transform (SDFT), (iii) Phasor measurement angle changing, (iv) Discrete
Fourier transform (DFT), (v) Kalman filters and (vi) Least error square.
Discrete Fourier transform, Zero-crossing and sliding discrete Fourier transform
are broadly used for phasor estimation techniques. The algorithms used for phase
estimation techniques can be tested using simulation software like MATLAB and
Lab View. Figure 4 shows the response of using various anti-aliasing filters.

Fig. 3 The architectural structure of PMU [6]

Fig. 4 Response of various

anti-aliasing filters [10]
710 M. Mohanty et al.

Fig. 5 Block diagram of internal structure of phase estimator

Table 1 System frequency of

Power network frequency Reporting rate (Fs− frames
power system and
(Hz) per second)
corresponding reporting rate
50 10–50 frames per second
60 10–60 frames per second

Figure 5 represents the internal structure of PMU. The voltage and current sig-
nals are sampled at a particular frequency, which are fed to discrete Fourier trans-
form block (DFT). DFT block extracts the samples for the fundamental frequency
using different algorithms for estimation. Electrical quantities, i.e., voltage pha-
sor and current phasor that are three-phased, frequency, circuit breaker status and
ROCOF are derived that depicts the state of the system. The quantities estimated by
phasor measurement unit arrive at reporting rate, which varies for different countries
power system standards. Table 1 depicts the system frequency of power system and
corresponding reporting rate.

3.4 Class of PMU

PMU has two performance classes, which are (a) Performance lass (P-Class) (b)
Measurement class (M-Class). Table 2 illustrates difference between both the classes
A Brief Review on Synchro Phasor Technology and Phasor … 711

Table 2 Difference between various classes of PMUs

Performance class PMU Measurement class PMU
Quick response applications are met by using It is implemented for application that does not
this class PMU require fast reporting speed and can allow
being affected by aliased signals
It does not requires explicit anti-aliasing It requires explicit anti-aliasing filtering
It needs shorter window length It needs longer window length

[11, 12].

3.5 Comparison of SCADA and Synchro Phasor Technology

A comparative study of SCADA and synchro phasor technology is given in Table 3.

4 Phasor Data Concentrator (PDC)

It is the data collection unit of synchro phasor technology. It collects the time syn-
chronized voltage and current data from all PMUs optimally placed in an area. These
PDCs then communicate to the super PDC which provide information to control room
for monitoring, control and protection for a wide area monitoring system. Figure 6
gives the idea of PMU in integration with PDC work in a wide area monitoring
system [13].

5 Optimal Placement of PMU (OPP)

Phasor measurement unit device is more advantageous than SCADA, as the data are
recorded in fraction of second. This gives the benefit in state estimation of bus in
a power system, which helps in wide area monitoring, control and operation. But

Table 3 Comparative study of synchro phasor technology and SCADA technology

SCADA technology Synchro phasor technology
Data are not time synchronized universally Data are time synchronized universally
Phasor angles are calculated from the data Estimation of phasor angles can be done
collected at EMS locally (at the sub-station level)
Rate of reporting is once in 4–6 s Rate of reporting is in 10/12/15/20/30/60
frames per second
712 M. Mohanty et al.

Fig. 6 Working of PMU along with PDC in a wide area monitoring [14]

since it is a costly device, it is essential to place PMU judiciously so that the system
becomes maximum observable with the help of minimum number of PMUs [14].
This illustrates the optimal placement of PMUs. Several techniques are developed and
categorized in the literature mostly into two main groups (i) Mathematical techniques
and (ii) Heuristic approach.
In mathematical techniques, integer programming is the most often used algo-
rithm. It has integer design variables where the objective function and the constraints
are linear, non-linear or quadratic [15]. In heuristics approach, different algorithms,
like genetic algorithm [16], simulated annealing [15], Tabu search algorithm [16],
particle swarm optimization algorithm [17], artificial neural network (ANN) [18],
flower pollination algorithm [19], differential evolution [20], immune genetic algo-
rithm [16], spanning-tree search [21], greedy algorithm [21], recursive security algo-
rithm [22], teaching–learning base optimization technique [23], improved binary par-
ticle swarm optimization [24], Best-first search algorithm [25] and modified binary
cuckoo optimization algorithm [26], etc., are used to find the local minima which
is very difficult in conventional mathematical algorithms. The above algorithms are
tested in different IEEE bus systems.

6 Application of Synchro Phasor Technology (SPT)

Table 4 gives the classification of some of the offline and online applications of
synchro phasor technology.
A Brief Review on Synchro Phasor Technology and Phasor … 713

Table 4 Synchro phasor technology applications

Offline application Online application
Validation of power system model Monitoring of a wide area of the power network and
visualization
State estimation
Post-disturbance analysis Oscillation monitoring of power system
Supervising of frequency stability
Monitoring of voltage stability

6.1 Validation of Power System Model

Modelling of the power system in terms of mathematical equations is indispensable.

Models can be designed from component levels to further move to be at system
level, and it is necessary that the model to be an accurate one. The SCADA data
obtained once in 4–6 s cannot capture all system dynamics and will not help in
designing correct system models. Phasor measurement unit data can be obtained at a
count of 10–60 F/s and can capture system dynamics and thus helping in designing
correct system model. It is advisable to take into consideration PMU models while
performing system model validation using PMU data [27, 28].

6.2 Post-disturbance Analysis

After the occurrence of blackout, the scenario is recreated using the time synchro-
nized data available to study the occurrence of problem. It becomes very important
when multiple power system components are included in the disturbances. Syn-
chro phasor technology provides wide area time synchronized phasor data and thus,
reducing the time required for completing post-disturbance analysis.

6.3 Monitoring of a Wide Area of the Power Network

and Visualization

The time synchronized data collected by PMU from the RTU which are geograph-
ically far apart helps in wide area monitoring. PMUs determine the magnitude and
phase angles of voltage phasor and current phasor and output data at reporting rate
of (10–60 F/s) making monitoring of power system dynamics possible. [29].
714 M. Mohanty et al.

6.4 State Estimation

It estimates the state of the power system from the telemetered data collected from
the network.These data come from various devices present at various nodes point.
However, it may not be present at all the node point of a network to the control room.
The EMS converts these data into state variables, i.e., the voltages and the phase
angles. In static state estimation, the voltage, current, reactive power flow and real
power flow measurements are available. The phase angles can be calculated from
these data. In this Estimation, data are calculated at a time interval of (0–5) s. This
method is actively used in SCADA technology. In dynamic state, estimation data are
calculated at a very short time interval say 0.01 s and all these data are synchronized
with the help of a GPS clock. These data are collected from the measurement devices,
i.e., PMU which are basically present at locations that are geographically very far
from each other of a particular region. A group of PMUs communicate with the PDC
of that region and the PDC of different region submit the data to the main control
centre for further action of monitoring and control. This forms a vital part of synchro
phasor technology. State estimation is a very important tool for online monitoring
and disturbance analysis. The phase angle difference clearly indicates that the system
stress and transient present. State estimation is used to check bad data handling,
error detection and elimination of redundant data. The EMS operation depends on
the working of operating states. The measurement equation is given by Eq. (1):

M = H (x) + E (1)

where M = measurement vector formed by voltage magnitude, real and reactive

power flows and power injections; H (x) = the nonlinear function relating the error-
free measurements to the system states; x = state vector; e = noise in measurements;
m = number of measurements and n = number of state variables.
In conventional state estimator, weighted least square estimator is commonly used.
It is used to reduce the error in the measurements. Since, it is nonlinear, iterative and
time consuming. So, linear state estimator is used to overcome this drawback. It uses
current and voltage data as state variables. As, a result, the problem of state estimation
is linear non-iterative. PMU-based state estimation has advantage of linearity and is
non-iterative. So speed of computation is very fast. It also helps in case of online
implementation [30].

6.5 Oscillation Monitoring of Power System

Owing to enhanced reporting rate of PMUs, early detection of poorly damped nodes is
possible. Local and inter-area oscillations can be damped by triggering proper control
actions at the right time. Figure 7 gives the detailed events involved in oscillation
monitoring [31].
A Brief Review on Synchro Phasor Technology and Phasor … 715

Fig. 7 Block diagram of oscillation monitoring system

6.6 Supervising of Frequency Stability

The mismatch between the generation and the load is indicated by the deviation
in power system frequency. PMUs along with voltage and current phasor can also
measure frequency, which allows designing of suitable control action to prevent an
instability situation [32].

6.7 Voltage Stability Monitoring

The PMU can measure the voltage phase, which gives accurate indication of system
stress. Figure 8 shows the common approaches for online static analysis of voltage
stability [33].
716 M. Mohanty et al.

Fig. 8 Block diagram of voltage stability monitoring

7 Hybrid State Estimation

Though PMU has more promising advantages in comparison to SCADA technology,

it is costly to implement. In literature many authors have proposed a unique hybrid
state estimation algorithm merging PMU technology and SCADA technology. In
this method, the complete power system is observable by SCADA. Consider that
at any one bus of power system PMU is placed. So, the bus and other bus adjacent
to this node point are observable by PMU, while others are unobservable by PMU.
In between two SCADA measurements, the system states are interpolated using
interpolation matrix (H ). In between two SCADA samples, if PMU changes sig-
nificantly, the interpolation matrix is compensated with a sensitivity change matrix
(H ) [34]. To compute H , the method classifies measurement set into four sub-
set, i.e., PMU measurements, SCADA measurements of PMU boundary buses with
significant change, adjacent to selected boundary cases and remaining SCADA mea-
surements and apply a modified least square method to each subset of measurements
with different weights. This proposed scheme has been tested on different IEEE
benchmark test systems. Figure 9 gives hybrid state estimations structural outline
[35].

8 Installation of PMU for Improving the Performance

of Indian Power System Scenario

The Indian Power Grid is categorized into five regions, i.e., Eastern, Western, South-
ern, Northern and North-Eastern [36]. After the world wide implementation of PMU
in the national grid of various countries, in 2010, The Power System Operation
Corporation (POSOCO) took an initiative to go for synchro phasor technology in
Indian Power System Operation as their first Pilot Project with an objective to include
A Brief Review on Synchro Phasor Technology and Phasor … 717

Fig. 9 Hybrid state estimations structural outline

the wide area monitoring system and its application in electrical grid. The synchro-
nized data are collected from PMU installed at nine locations in the Northern Western
region and are communicated through optic fibre to reach the PDC located at Northern
Region Load Dispatch Centre (NRLDC), as shown in Fig. 10.
Since the commissioning of this project, POSOCO is utilizing synchro phasor
data for improving the electrical grid performance in many ways. Some of these are
as mentioned below:
• Real-time monitoring is achieved
• d f /dt setting of the frequency relay is done.
• Post-disturbance analysis of fault can be done to determine exact location of fault
and time required to clear a fault.
• Power system oscillations detection and investigation are done.
• Angular separation value observed through the PMU also helps in validating the
network model in the SCADA as well as offline simulation packages.
• Authentication and determining the need of special protection schemes.
Initially, four PMUs were installed strategically considering large geographical
distance, load angle separation and fibre optic communication link with load centre. A
hybrid approach was undertaken taking into consideration of the data from SCADA.
As of 14 PMUs are implemented, nine are installed in Northern region, two in
Western region and three in Southern region [37]. These PMUs communicate to their
718 M. Mohanty et al.

Fig. 10 PMU and PDC location in North-Western region [37]

respective PDC present at the respective Regional Load Despatch Centres (RLDCs).
For Northern region, PDC is installed at New Delhi. Similarly, for Western region, the
PDC is located at Mumbai and for Southern region, it is at Bengaluru. The existing
WAMS at Western region and Southern region are demonstration projects, while it
is was a pilot project in the Northern region. Figure 11 depicts the locations having
phasor measurement units in India.

9 Conclusion

This paper brings about a brief review on synchro phasor technology and PMU and
its application in the present power system scenario in India. Researches are going
on for dynamic state estimation, Hybrid state estimation and optimal placement of
PMU, as PMUs are expensive to implement. The optimization problem designed
so far works only on a small area with an objective for making the system more
observable with less number of PMUs, and this can also take into consideration
other constraints like the communication issues, the state of the environment, etc.
Mathematical algorithm in combination with heuristic algorithm, can be used as
hybrid algorithm to obtain better results. Emphasis is given on PMU testing due to
different phasor estimation algorithm used in PMU by different vendors, which may
lead to error. The PMU integrated with power grid can add value to the smart grid.
A Brief Review on Synchro Phasor Technology and Phasor … 719

Fig. 11 Current locations showing existing phasor measurement units in India [37]

Micro PMUs are nowadays proposed for its application in distribution network. It
will be helpful in both offline and online monitoring of the distributed generation
(DG)-integrated power system. Many works are also being carried out by using PMUs
data for analysis of power system stability by developing different algorithms. These
data can help designing of adaptive protection system, control systems, after event
disturbance analysis, etc. To prevent cyber breach and maintain good communication
between the power network and the customer, a complex architectural designing is
needed for PMU. Mitigation of harmonics using PMU can be done and provide better
response to electrical vehicle.These electrical vehicle can serve as an alternative
during contingencies in distribution side. So, it would not be wrong to say PMU is a
renaissance which not only helps in post-disturbance analysis, but also helps in fault
elimination and prevention of blackout problems.

References

1. Bonanomi, P.: Phase angle measurements with synchronized clocks-principle and applications.
IEEE Trans. Power Apparatus Syst. 12, 5036–5043 (1981)
2. Phadke, A.G., Thorp, J.S.: Synchronized phasor measurements and their applications, vol. 1.
Springer, New York (2008)
720 M. Mohanty et al.

3. Adamiak, M., Kasztenny, B., Premerlani, W.: Synchrophasors: definition, measurement, and
application. In: Proceedings of the 59th Annual Georgia Tech Protective Relaying, Atlanta,
GA, pp. 27–29 (2005)
4. Phadke, A.G.: Synchronized Phasor measurements-a historical overview. In: IEEE/PES
Transmission and Distribution Conference and Exhibition, vol. 1. IEEE (2002)
5. Schweitzer, E.O., et al.: Advanced real-time synchrophasor applications. In: Proceedings of
the 35th Annual Western Protective Relay Conference, Spokane, WA (2008)
6. Phadke, A.G., Tianshu, B.I.: Phasor measurement units, WAMS, and their applications in
protection and control of fs. J. Modern Power Syst. Clean Energy 6(4), 619–629 (2018)
7. Phadke, A.G., Thorp. J.S.: History and applications of phasor measurements. In: 2006 IEEE
PES Power Systems Conference and Exposition. IEEE (2006)
8. Usman, M.U., Faruque, M.O.: Applications of synchrophasor technologies in power systems.
J. Modern Power Syst. Clean Energy 7(2), 211–226
9. Nanda, P., Panigrahi, C.K., Dasgupta, A.: Application of compressive sampling in computer
based monitoring of power systems. Adv. Comput. Eng. (2014)
10. Nanda, P., Panigrahi, C.K., Dasgupta, A.: Phasor estimation and modelling techniques of
PMU—a review. Energy Procedia 109, 64–77 (2017)
11. Castello, P., et al.: A fast and accurate PMU Algorithm for P + M class measurement of
synchrophasor and frequency. IEEE Trans. Instrum. Meas. 63(12), 2837–2845
12. Roscoe, A.J., Abdulhadi, I.F., Burt, G.M.: P and M class phasor measurement unit algorithm
s using adaptive cascaded filters. IEEE Trans. Power Deliv. 28(3), 1447–1459 (2013)
13. Kanabar, M., Adamiak, M.G., Rodrigues, J.: Optimizing wide area measurement system archi-
tectures with advancements in phasor data’s concentrators (PDCs). In: 2013 IEEE Power &
Energy Society General Meeting. IEEE (2013)
14. Sefid, M., Rihan, M.: Optimal PMU placement in a smart grid: an updated review. Int. J. Smart
Grid Clean Energy 8(1), 59–69 (2019)
15. Gou, Bei: Generalized integer linear programming formulation for optimal PMU placement.
IEEE Trans. Power Syst. 23(3), 1099–1104 (2008)
16. Aminifar, F., et al.: Optimal placement of phasor measurement units using immunity genetic
algorithm. IEEE Trans. Power Deliv. 24(3), 1014–1020 (2009)
17. Maji, T.K., Acharjee, P.: Multiple solutions of optimal PMU placement using exponential
binary PSO algorithm for smart grid applications. IEEE Trans. Ind. Appl. 53(3), 2550–2559
(2017)
18. Gupta, A.K., Verma, K.: PMU-ANN based real time monitoring of power system elec-
tromechanical oscillations. In: 2016 IEEE 1st International Conference on Power Electronics,
Intelligent Control and Energy Systems (ICPEICES). IEEE (2016)
19. Hassanin, K.M., Abdelsalam, A.A., Abdelaziz, A.Y.: Optimal PMUs placement for full
observability of electrical power systems using flower pollination algorithm. In: 2017 IEEE
International Conference on Smart Energy Grid Engineering (SEGE). IEEE (2017)
20. Peng, C., Sun, H., Guo, J.: Multi-objective optimal PMU placement using a non-dominated
sorting differential evolution algorithm. Int. J. Electr. Power Energy Syst. 32(8), 886–892
(2010)
21. Manousakis, N.M., Korres, G.N., Georgilakis, P.S.: Taxonomy of PMU placement methodolo-
gies. IEEE Trans. Power Syst. 27(2), 1070–1077 (2012)
22. Farsadi, M., Golahmadi, H., Shojaei, H.: Phasor measurement unit (PMU) allocation in
power system with different algorithms. In: 2009 International Conference on Electrical and
Electronics Engineering-ELECO 2009. IEEE (2009)
23. Raj, A., Venkaiah, C.: Optimal PMU placement by teaching-learning based optimization
algorithm. In: 2015 39th National Systems Conference (NSC). IEEE (2015)
24. Rather, Z.H., Liu, C., Chen, Z., Thøgersen, P.: Optimal PMU placement by improved particle
swarm optimization. In: 2013 IEEE Innovative Smart Grid Technologies-Asia (ISGT Asia).
IEEE (2013)
25. Venkatesh, T., Jain, T.: Optimal PMU placement using best first search algorithm with pruning.
In: 2014 Eighteenth National Power Systems Conference (NPSC). IEEE (2014)
A Brief Review on Synchro Phasor Technology and Phasor … 721

26. Dalali, M., Karegar, H.K.: Optimal PMU placement for full observability of the power network
with maximum redundancy using modified binary cuckoo optimisation Algorithm. IET Gener.
Transm. Distrib. 10(11), 2817–2824 (2016)
27. Zhang, D.X., et al.: Study on load model validation and its adjustment approach based on
calibration of simulations against disturbances. Power Syst. Technol. 31(4), 24–31 (2007)
28. Huang, Z„ et al.: Generator dynamic model validation and parameter calibration using phasor
measurements at the point of connection. IEEE Trans. Power Syst. 28(2), 1939–1949 (2013)
29. Soonee, S.K., et al.: The view from the wide side: wide-area monitoring systems in India. IEEE
Power Energy Mag. 13(5), 49–59 (2015)
30. Vanfretti, L., et al.: A phasor-data’s-based State estimator incorporating phase bias correction.
IEEE Trans. Power Syst. 26(1), 111–119 (2010)
31. Annakage, U.D., et al.: Application of phasor measurement units for monitoring power system
dynamic performance (2017)
32. Uhlen, K., et al.: Monitoring amplitude, frequency and damping of power system oscil-
lations with PMU measurements. In: 2008 IEEE Power and Energy Society General
Meeting-Conversion and Delivery of Electrical Energy in the 21st Century. IEEE (2008)
33. Gong, Y., Schulz, N., Guzman, A.: Synchrophasor-based real-time voltage stability index. In:
2006 IEEE PES Power Systems Conference and Exposition. IEEE (2006)
34. Göl, M., Abur, A.: A hybrid state estimator for systems with limited number of PMUs. IEEE
Trans. Power Syst. 30(3), 1511–1517 (2014)
35. Das, K., et al.: real-time hybrid state estimation incorporating SCADA and PMU measurements.
In: 2012 3rd IEEE PES Innovative Smart Grid Technologies Europe (ISGT Europe). IEEE
(2012)
36. Soonee, S.K., et al.: Experiences of synchrophasors initiative in India. rue 2, 109 (2014)
37. Sharma, R.B., Dhole, G.M.: Synchrophasor measurement network and its applications in
Indian grid. In: 2016 International Conference on Emerging Trends in Electrical Electronics
& Sustainable Energy Systems (ICETEESES). IEEE (2016)
Evaluation of the Standard Procedure
to Assess the Sphericity and Size of Pellets
Using Image Processing Techniques

Anuja Nanda, Akshaya Kumar Patra, Santisudha Panigrahi,

Alok Kumar Mishra, and Lalit Mohan Satapathy

Abstract Through Pelletization iron ore fines produced spheres of typically 9––
16 mm diameter which can be used as the raw material to the blast furnace. In the
process of pelletizing the pellets may be produced of different sizes. Since the perme-
ability of the blast furnace depends on the shape and size of the pellet it is desirable to
produce nearly spherical pellets with a diameter range of 9–16 mm. There are many
physical methods such as using a scale or caliper and solid displacement method
to measure the size of the pellet. The shape and size of the pellet can be changed
during measurement since they are wet. However, without physically handling them
it is very difficult to measure the shape and size. This paper describes digital image
processing techniques that are used to extract sphericity and size information from
the 2D image of iron ore pellets. The techniques are simple but very useful for indus-
trial applications. Online image processing method is also presented for calculating
the sphericity and the size of a single/multiple pellets by software and its toolboxes.
The technique can suitably derive the size distribution of a number of pellets. The
technique is validated by conducting a number of experiments.

Keywords Pelletization · Sphericity · Image processing · Size distribution

A. Nanda (B) · A. K. Patra · A. K. Mishra · L. M. Satapathy

Department of Electrical and Electronics Engineering, Institute of Technical Education and
Research, Siksha ‘O’ Anusandhan (Deemed to be University), Bhubaneswar, Odisha 751030,
India
e-mail: [email protected]
A. K. Patra
e-mail: [email protected]
A. K. Mishra
e-mail: [email protected]
L. M. Satapathy
e-mail: [email protected]
S. Panigrahi
Department of Computer Science and Engineering, Institute of Technical Education and Research,
Siksha ‘O’ Anusandhan (Deemed to be University), Bhubaneswar, Odisha 751030, India
e-mail: [email protected]

© Springer Nature Singapore Pte Ltd. 2020 723

G. Pradhan et al. (eds.), Advances in Electrical Control and Signal Systems,
Lecture Notes in Electrical Engineering 665,
https://doi.org/10.1007/978-981-15-5262-5_54
724 A. Nanda et al.

1 Introduction

India has adequate reserves of iron ore. While catering to the demands of all the
DRI/steel plants, there is an excessive generation of fines, during mining and from
crushing units, which are underutilized. In order to increase the availability of iron
ore in meeting the ever-increasing demand of steel, i.e. 100 Million tonnes by 2020
as indicated by industry experts and steel Ministry, Pelletization of iron ore fines is
the only route that is going to dominate the Indian steel industry. Preparation of iron
ore fines/concentrate, as a feed to blast furnace, involves the production of spherical
pellets of typically 9-16 mm diameter through pelletization. The process combines
agglomeration and thermal treatment to convert the raw ore into pellets suitable for
use in a blast furnace. A typical flow diagram is shown in Fig. 1. A disc-type pelletizer
is a cylindrical tray, rotating at a certain rpm and of a suitable diameter inclined at
an angle of around 45° to the horizontal. Fine chromes ore continuously fed into the
rotary tray by a conveyer belt and water is sprayed over fine chromes. Fine iron ore
particles are form lumps due to the capillary attraction of water and form granulated
arrows. Due to the rotating effect of the tray, the granulated materials grow to the
desired pellet size and are discharged as green pellets by overflowing from the rim
of the pelletizer. The flow rate of water sprayed determines the pellet size. However
the structure of pellet tends to differ due to the varying moisture content of ores and

Fig. 1 Typical flow diagram of pelletization

Evaluation of the Standard Procedure to Assess the Sphericity … 725

the properties of the iron ore fines/concentrates, it has been difficult to ensure desired
size control solely by the visual inspection of human operators. From this point of
view, there has been a strong demand for the development of a system for measuring
the average size of pellets in a continuous and non-contact manner and thereby
automatically controlling the pelletizer. The use of pellets of varying shape and size
has attracted much interest. The permeability of the blast furnace is dependant on the
size and shape of the pellets. There is a requirement of generating pellets of uniform
size and the shape should be spherical. The productivity of pelletization is based on
the generation of pellets of the size range of 9–16 mm. There is a need to analyze the
shape and size. There is no straightforward method except for human inspection.
In the process of pelletization, the pellets are formed of different sizes. This may be
accepted if it is within certain desired limits. Factors affecting the pellet sphericity
in the pelletization process are (i) particle size of the starting material (ii) balling
speed (iii) residence time (iv) moisture content of the granulated mass. Any change
is such parameters may affect the quality and quantity of the desired pellets. The size
may vary and the shape may not be spherical. A spherical pellet can withstand more
stress which is desirable.
Therefore, it is essential to measure the size and provide a quantitative value for
the shape of the pellets. The size can be measured by many methods such as using
a sieve or caliper and solid displacement method, etc. However, it is very difficult
to measure the size and shape of green pellets without physically handling them.
Since they are wet and they change their size and shape during measurement. Some
researchers have worked using imaging techniques for pellet size analysis. Some of
the reported works are briefly discussed below.
Shi et al. [1], have employed an edge detection technique to obtain a pellet image
boundary robustly to calculate the pellet sphericity. In this paper, dual active con-
tour model is used to relieve this problem by combining with a local shape model to
improve the parameterization. In the method presented by Harayama et al. [2], online
measurement of size of the pellet by image processing technique has been imple-
mented. The moment of the power spectrum of pellet image is proposed as a new
index of pellet size. Experimental results indicate that this new index corresponds to
the real size of the pellet. Advantages of this method are (i) the average size of pellets
accumulated in a pile is measured, (ii) the index has good correspondence to human
senses, and (iii) the index has covered a wide measurement range. Andersson et al.
[3] have employed a 3D laser camera system in a laboratory environment to capture
a 3D surface of data of pellets and steel balls. Validation of the 3D data against a
spherical model has been performed and demonstrates that pellets are not spherical
and have a physical structure that a spherical model cannot capture. This method
used to capture the 3D surface data, segment it, determine segmented regions’ sizes,
and evaluate the spherical model. Thurley [4] has attempted online 3D surface mea-
surement of iron ore green pellets using laser ray and image processing techniques.
Andersson et al. [5], proposed a method to make a distinction between entirely visi-
ble and partially visible pellets. This is a significant distinction as the size of partially
visible pellets cannot be correctly estimated with existing size measures and would
bias any size estimate. Size estimates of the surface of a pellet pile show that the
726 A. Nanda et al.

overlapped particle error is overcome only by estimating the surface size distribution
with entirely visible pellets [6–7].

2 Proposed Technique for Size Measurement

The image of pellet is taken and the background noise is removed. The area in terms
of pixel covered by a pellet is determined. The pixel area in terms of square centimeter
is calibrated. The camera used captures 640 × 480 pixels and the area covered by
these pixels is calibrated experimentally. The area depends on the camera height. If
the camera height is more the area covered with these 640 × 480 pixels is more.
Once the area of a pellet is found out in terms of sq cm, the radius or diameter can
be found out. This assumes that the pellet is spherical. The two-dimensional image
gives the circular image and by rotating the pellet, the diameter of the other side is
found out and the average diameter of these entire rotated images would give the
size of the pellet [8].

2.1 Algorithm for Offline Image Processing for Shape

Analysis

Offline image processing work is carried out to see the feasibility and effectiveness
of the algorithm before it applies to online image processing [9–10]. First the image
file is read. Then the image is thresholded. The background noise is removed. The
boundary is extracted by using region prop. Algorithm used for determining spher-
ical object: To determine which objects are round, estimate each object’s area and
perimeter. Use these results to form a simple metric indicating the roundness of an
object: This metric is equal to one only, for a circle and it is less than one for any
other shape. The discrimination process can be controlled by setting an appropriate
threshold. In this example, a threshold of 0.94 is used, so that only the pills will be
classified as round. Regionprops is used to obtain estimates of the area for all of the
objects. Notice that the label matrix returned by bw boundaries can be reused by
regionprops. stats = regionprops (L,‘Area’,‘Centroid’); Simulation results of image
of different shapes are shown in Fig. 2.
From the above simulation results, it can be clearly seen that the sphericity metric
of a rounded object is about 0.9 or more. Other types of objects show the metric value
less. This experiment is conducted to evaluate the correctness of sphericity measuring
algorithm. A single pellet is subjected to the same algorithm and the sphericity is
measured at different angles by rotating the same pellets. The results are shown in
Fig. 3.
The same algorithm is subjected to a group of pellets subject to the condition that
they are not touching each other. The results are shown in Fig. 4.
Evaluation of the Standard Procedure to Assess the Sphericity … 727

Fig. 2 Simulation result of shape recognition: a original image. b Simulation result of different
shapes

Fig. 3 Simulation result of

single pellets

2.2 Algorithm for Size Distribution of a Group of Pellets

It is important to find the size distribution of a group of pellets as the productivity

of pelletization depends on the desired size distribution. The present study is to
determine the size distribution of a group of pellets. This is nothing but finding out
the number of pellets of a set of particular sizes. The area in terms of no. of the pixel
is found out by using the algorithm for size analysis as discussed above. To find
the area in square centimeter following procedure is followed. Camera to be fixed
at a particular height so that the camera captures a particular fixed area. The image
captured by the camera is 640 × 480 pixels. By using this value and scale of the area
viewed we can compute the area of the pellet in square cm. Also, the radius of the
728 A. Nanda et al.

Fig. 4 Simulation result of group of nontouching pellets: a original image. b Sphericity result of
pellets

Fig. 5 Simulation result of shape and size distribution of non-overlapped pellets: a image of group
of pellets. b Image sphericity. c Size distribution histogram

pellet can be found out. Histogram of the radius with respect to reference values is
taken to find out the size distribution of the pellet. The simulation result is shown in
Fig. 5.

2.3 Algorithm for Finding the Size of Group of Touching

Pellets

The algorithm discussed in the previous section assumes that the pellets are subjected
to camera view and are placed separated from each other. However, the algorithm will
be more robust if it can estimate the size of the pellets are even touching each other.
Evaluation of the Standard Procedure to Assess the Sphericity … 729

Sometimes it is required that the size to be estimated from a group of pellets moving
on a conveyer belt. It is not easy to separate each and every pellet and calculate
the size of individual pellets to find the size distribution. However, even if one can
estimate and say that pellets are either big or medium-sized or small, it is very useful
for pellet industry. In the pelletization disc sometimes the pellet grows in undesired
size and sometimes the growth is below the desired size. Such situations can be
identified online if such an algorithm is derived. This is the main motivation behind
this work.

2.3.1 Principle of Size Estimation

The image is a two-dimensional vector and if we check the row vectors and column
vectors of the edges of an image containing bigger sized pellets we find they are
more separated and we may say they are low frequency. If we check the same in an
image of small-sized pellets we find the edges are more closely placed and hence
high frequency.
To see the frequency of row and column matrix of the image
Big pellets → low frequency
Small pellet → High frequency
The process of the above algorithm is as follows. First, the image is taken. Then the
image is thresholded. The background noise is removed. The boundary is extracted
by using region prop. Then the edge of the image is extracted by using a Sobel filter
and image morphology.
Then by taking the Fourier transform (FFT) of the rows and column of the edge
extracted image, the frequency is determined. The results are shown in Figs. 6 and
7.

3 Measurement of Pellets by Using 2D-Digital Online

Image Processing Techniques

After doing exhaustive simulation study on various algorithms, an In this study,

an attempt is made to measure the shape and size of the pellets using 2D-digital
image processing techniques. An online camera-based image processing system is
established using Matlab software and its toolboxes and is shown in Fig. 8. A sample
result of a set of pellets after image processing is shown in Fig. 9.
730 A. Nanda et al.

(a) (b)

Magnitude in cm.
Magnitude in cm

Frequency in Hz Frequency in Hz
(c) (d)

Fig. 6 Simulation result of group of touching pellets (very_ big2_ pellet.jpg): a pellet image.
b Edge detection. c Plot of centre row of edge detected image. d FFT of centre row of edge detected
image

3.1 Experimental Method

The principle of size measurement used here is as follows. The image of the pellet is
captured and the background noise is removed by using different image processing
techniques. The area in terms of the pixel covered by a pellet is determined. The
pixel area in terms of square centimeter is calibrated. The camera used in this exper-
iment captures 640 × 480 pixels and the area covered by these pixels are calibrated
experimentally. The area depends on the camera height. If the camera height is more,
the area covered with these 640 × 480 pixels is more. Once the area of a pellet is
found out in terms of sq cm, the radius or diameter can be found out. This assumes
that the pellet is spherical and 2-D image capture is circular. The two-dimensional
image gives the circular image and by rotating the pellet, the diameter of the other
side is found out and the average diameter of these entire rotated images would give
the estimate of the size of the pellet. The size distribution of the pellets can also be
found out by finding the histogram of the radius with respect to the reference values
of the pellet.
Evaluation of the Standard Procedure to Assess the Sphericity … 731

(a) (b)
Magnitude in cm

Magnitude in cm

Frequency in Hz Frequency in Hz

(c) (d)

Fig. 7 Simulation result of group of touching pellets (small-pellet.jpg). a Pellet image. b Edge
detection. c Plot of centre row of edge detected image. d FFT of centre row of edge detected image

Fig. 8 Experimental setup

for online imaging for shape
and size analysis of pellets

The shape of the pellet should be ideally spherical and hence the shape of the
pellet is quantified as its sphericity. The sphericity metric is found out by measuring
the pixel perimeter and the pixel area of the pellet. The metric used for sphericity
measure is sphericity = 4π(area/perimeter2 ). Figure 10 shows the image processing
results of a typical set of pellets and its spericity metric.
732 A. Nanda et al.

Fig. 9 Measurement of pellets sphericity

Fig. 10 Results of size distribution algorithm: a online original image. b Image processed image.
c Size distribution histogram

3.2 Result and Discussions

A number of simulation as well as online experiments are conducted to validate

the algorithm proposed. The algorithm is coded using Matlab software with image
acquisition and image processing toolboxes. A web camera is used to capture the
image. The resolution of the camera is 640 × 480.
The present technique is capable of finding out sphericity of pellets accurately.
A relatively spherical pellet shows its sphericity in the range of 0.90 to 0.97. The
sphericity and size analysis of a typical pellet is shown in Table 1. From Table 1, it is
seen that the average radius of the pellet is about 1.1 c.m and the spehericity metric
is about 0.90. This is a good pellet.
In another experiment, a group of pellets has been subjected to an online size
distribution algorithm. The results of the size distribution algorithms are shown in
Fig. 10.
In Fig. 10, the original image captured by the camera is shown in Fig. 10a, the
processed image is shown in Fig. 10b and the size distribution histogram is shown in
Evaluation of the Standard Procedure to Assess the Sphericity … 733

Table 1 Size analysis and sphericity of a typical pellet

S. Area of Total Capture Capture Height Pellet Radius Sphericity
No. the captured d area d area of area (cm)
pellet area in length breath camera (cm2 )
in no pixels (cm) (cm) (cm)
of pixel
1 10,190 307,200 13 9 16 3.8810 1.1115 0.91
2 10,521 307,200 13 9 16 4.0070 1.1294 0.91
3 10,536 307,200 13 9 16 4.0127 1.1300 0.90
4 10,549 307,200 13 9 16 4.0177 1.1309 0.89
5 10,100 307,200 13 9 16 3.8467 1.1065 0.91
6 10,377 307,200 13 9 16 3.9522 1.1216 0.91
7 10,473 307,200 13 9 16 3.9887 1.1268 0.89
8 10,645 307,200 13 9 16 4.0542 1.1360 0.90
9 10,749 307,200 13 9 16 4.0939 1.1415 0.91
10 10,599 307,200 13 9 16 4.0367 1.1335 0.90

Fig. 10c. From the results in Fig. 10c, it is observed that there are three small pellets
and four largest pellets in the lot. Figure 10b shows the sphericity of each pellet. The
measured sphericity metric of each pellet is shown near the pellet. The value is more
if it is more spherical.
In another experiment, the pellet size and sphericity of five different pellets are
measured and tabulated in Table 2. The nomenclature of the pellet specification
‘Pellet-n-h’ shows the nth pellet is measured at hth height. It looks that the pellet
size is slightly varied at a different height. This variation is due to an illumination
problem. However, the algorithm clearly distinguishes different sized pellets. The
measurement error at 12 cm height is evident from the data shown in Table 2.

4 Conclusions

Techniques for image processing have been successful in recent years due to low
cost of devices for image acquisition and manipulation and an improvement in the
technical characteristics (speed, resolution, processing time, etc.). Shape recognition
and size analysis based on the method of determining rounding object by classifying
object based on their boundaries has been discussed here. A metric based on the
perimeter and the area is used to determine the sphericity of the pellets. The size
analysis tool is very useful for pelletization industry as well as a research organization.
This method can easily replace the sieving method of size analysis. The algorithm and
method stated are very simple and can be extended for very high-end applications.
The proposed algorithm works when the pellets are not touching each other. The size
734 A. Nanda et al.

Table 2 Sphericity and pellet size measurement at different heights

Specification Length Breath Height Diameter Sphericity
in cm in cm in cm in mm
Pellet_1_10 8 6 10 20.952 0.91
Pellet_1_12 9 8.8 12 23.288 0.91
Pellet_1_14 10.7 7.7 14 20.132 0.90
Pellet_1_16 12.5 9.3 16 21.048 0.90
Pellet_1_18 13.5 10 18 20.316 0.89
Pellet_1_20 15 11 20 20.316 0.90
Pellet_2_10 8 6 10 17.7 0.90
Pellet_2_12 9 8.8 12 21.526 0.89
Pellet_2_14 10.7 7.7 14 17.708 0.91
Pellet_2_16 12.5 9.3 16 18.077 0.89
Pellet_2_18 13.5 10 18 18.108 0.91
Pellet_2_20 15 11 20 18.434 0.89
Pellet_3_10 8 6 10 16.160 0.90
Pellet_3_12 9 8.8 12 18.482 0.90
Pellet_3_14 10.7 8 14 16.692 0.90
Pellet_4_10 8 6 10 16.086 0.91
Pellet_4_12 9 8.8 12 15.755 0.91
Pellet_4_14 10.7 7.7 14 15.202 0.90
Pellet_5_10 8 6 10 14.016 0.91
Pellet_5_12 9 8.8 12 14.952 0.90

distribution algorithm for a set of pellets that are touching with each other is in the
process of development.
A real-time implementation of shape and size analysis algorithm has been devel-
oped by using the image acquisition toolbox of Matlab. A detailed study has been
made to study the image processing techniques for pellet shape and size analysis.
A number of algorithms are proposed which can be used in mineral processing
industries and research Institutions.

References

1. Shi, D.M., Heng, P.W.S., Chen, F.: A pellet sphericity measure system based on dual active
contour models. Proceedings of the 2003 International Conference on Machine Learning and
Cybernetics (IEEE Cat. No.03EX693), Xi’an, vol. 5, pp. 2781–2784 (2003)
2. Harayama, M., Uesugi, M.: On-line measurement of average pellet size with spatial frequency
analysis. Proceedings of the 1992 International Conference on Industrial Electronics, Control,
Instrumentation, and Automation, San Diego, CA, USA, vol. 3, pp. 1613–1618 (1992)
Evaluation of the Standard Procedure to Assess the Sphericity … 735

3. Andersson, T., Thurley, M.J., Marklund, O.: Pellet Size Estimation Using Spherical Fitting.
2007 IEEE Instrumentation & Measurement Technology Conference IMTC 2007, Warsaw,
pp. 1–5 (2007)
4. Thurley, M.J.: On-line 3D surface measurement of iron ore green pellets. 2006 International
Conference on Computational Inteligence for Modelling Control and Automation and Inter-
national Conference on Intelligent Agents Web Technologies and International Commerce
(CIMCA’06), Sydney, NSW, pp. 229–229 (2006)
5. Andersson, T., Thurley, M.J., Marklund, O.: Pellet Size Estimation Using Spherical Fitting.
2007 IEEE Instrumentation & Measurement Technology Conference IMTC 2007, Warsaw,
pp. 1–5 (2007)
6. Allen, T.: Particle Size Measurement, 5th edn, pp. 128–137. Chapman & Hall, London (1997)
7. Baert, L., Fanara, D., Remon, J.P., Massart, D.: Correlation of extrusion forces, raw materials
and sphere characteristics. J. Pharm. Pharmacol. 44(8), 676–678 (1992, August)
8. Baert, L., Remon, J.P., Elbers, J., Van Bommel, E.: Comparison between a gravity feed extruder
and a twin screw extruder. Int. J. Pharm. 99(1), 7–12 (1993)
9. Chen, S., Yue, Z.Q., Tham, L.G.: Digital image-based numerical modeling method for
prediction of inhomogeneous rock failure. Int. J. Rock. Mech. Min. 41(6), 939–957 (2004)
10. Sadr-Kazemi, N., Cilliers, J.J.: An image processing algorithm for measurement of flotation
froth bubble size and shape distributions. Miner. Eng. 10(10), 1075–1083 (1997)
Enhanced and Energy-Efficient Program
Scheduling for Heterogeneous
Multi-Core Processors System

Lavanya Dhanesh, S. Deepa, P. Elangovan, and S. Prabhu

Abstract Scheduling is essential for the proper functioning of multi-core proces-

sors for parallel processing. A real-time embedded system has been extensively used
for different fields such as control, scheduling, and monitoring. They will perform
multiple tasks under schedule time constraints. Single-core processor system can run
only one process at a time. Single-core processor cannot satisfy the applications of
Real-Time applications. This system consumes more power which is not acceptable
when scheduling through the Multi-core processor. To avoid these issues introduced
Heterogeneous Multi-core Processors (HMP) which schedules the tasks much better
when compared to homogenous multi-core processors. The main proposal of the
study is to provide a solution to computational starving in real-time field. The starv-
ing mainly occurs due to the time spent for the scheduling of the real-time tasks in
a multiprocessor system. This paper proposes an optimized multi-task scheduling
algorithm that schedules the multiple tasks on different cores of a multi-core pro-
cessor in an efficient way. This proposed algorithm increases the overall efficiency
and it automatically allocates a suitable core processor for reducing time. The Pro-
posed system is evaluated to priority, pipeline, preemption, and cyclic task scheduling
which minimizes power consumption, response time, and avoid overload.

Keywords Heterogeneous · Parallel processing · Task scheduling · Multi-core

processor · Round-Robin · And first come first serve bases · Shortest job first ·
Worst-Case execution time · Relative deadline · Interrupt latency · Load
balancing · Power consumption

L. Dhanesh · S. Deepa
Department of EEE, Panimalar Institute of Technology, Chennai 600123, India
P. Elangovan (B)
Department of EEE, Sreenivasa Institute of Technology and Management Studies, Chittoor,
Andhra Pradesh 517127, India
e-mail: [email protected]
S. Prabhu
Department of EEE, Sree Vidyanikethan Engineering College, Tirupathi, Andhra Pradesh 517102,
India

© Springer Nature Singapore Pte Ltd. 2020 737

G. Pradhan et al. (eds.), Advances in Electrical Control and Signal Systems,
Lecture Notes in Electrical Engineering 665,
https://doi.org/10.1007/978-981-15-5262-5_55
738 L. Dhanesh et al.

1 Introduction

Scheduling of tasks is necessary for each system. A scheduler becomes indispensable

part in OS. Without scheduler, tasks cannot execute the order in which an OS requires.
The task to be scheduled in a single-core processor is easy to choose. Some algorithms
are the First Come First Server, Priority-based, Shortest Job First, Round-Robin, etc.
The multi-core processor has processed with many elements on a single-core chip.
Rapid work should be carried out by a real-time system and it should serve the
function within a specified time interval. Real-time systems’ key drawbacks are
following timing constraints and performing the tasks within timing limits. The
powerful processor is needed to meet the computational hungry applications due to
an increase in the problems of real-time applications. For many applications, multi-
core processors are the best solution. Multi-core processors are an immature stage and
address the issue of multi-core partitions with a view to all cores that are used equally.
In order to serve many applications, the more dominant processor is needed. The
growing difficulty of real-time applications cannot be met by a single-core processor.
Single-core processors consume more power that cannot be accepted. The multi-
core processor provides the solution to real-time applications that are difficult and
computationally ravenous. Even at low power consumption, multi-core processors
provide advanced computational power. Here introduced heterogeneous multi-core
processors provide an innovative development for various computing abilities. The
processor power consumption and heat dissipation are the main challenges found
today. Heterogeneous multi-core processors are effectively reducing the processor
energy consumption and consider the execution speed performance. The multi-core
processor reduces the frequency range of processor and reduces the temperature level
of the system. In this processor, instructions can run simultaneously processed with
individual cores and increase the amount of parallelism.
This paper proposes an Optimized Multi-task scheduling (OMTS) algorithm
which schedules the multiple tasks on different cores of a multi-core processor.
The proposed algorithm allocates the core to the tasks based on the nature of the task
to be done. This scheduling not only increases the efficiency of the task but it also
reduces the time of serving the task. This proposed system includes the parameters
Worst-Case Execution Time, Relative Deadline, Interrupt Latency, Load Balanc-
ing, Power Consumption, and Utilization Bounds. These parameters are compared
with the already existing heterogeneous task scheduling algorithms and proved that
the proposed OMTS algorithm is better in Worst-Case Execution Time, Relative
Deadline, Interrupt Latency, Load Balancing, Power Consumption, and Utilization
Bounds. The main goal of task scheduler is to optimize the performance so that it
can minimize the task executing time and maximize resource utilization.
The structure of this paper as follows: The previous heterogeneous multi-
core architecture and various multi-core scheduling styles have been analyzed in
Sect. 2.Sect. 3 proposes a latest heterogeneous multi-core scheduling system and
proposes a multi-core heterogeneous scheduling algorithm. Section 4 describes the
Enhanced and Energy-Efficient Program Scheduling … 739

results and discussions and the implementation part and previous author techniques
are compared with proposed techniques.

2 Related Work

Hsiang-Kuo Tang proposed a scheduling approach for heterogeneous Resources

where the hard and soft real-time periodic tasks are combined. In this presentation,
the focus is on scheduling occasional tasks alongside a hard time restrain inter-
mittent tasks with requirements of heterogeneous ongoing frameworks [1]. Even
they disclose a technique to redesign periodic task responsiveness without parcel-
ing intermittent undertaking time restrain this ensures by first scheduling occasional
assignments that are disconnected, at that point powerfully scheduling periodic tasks
in the rest of the asset will have a delay.
Zhe Wang et al. are proposing temperature-aware RT scheduling tasks. In this
paper, they clarified task allocation as a noteworthy way of reducing the peak tem-
perature in embedded systems running either on an arrangement of occasional hetero-
geneous undertakings with an indistinguishable period or intermittent heterogeneous
allocation with a singular period [2].
Pengliu Tan et al. exhibits hybrid multi-core architecture for the scheduling of the
real-time approach. They give a hybrid scheduling approach in this presentation for
the constant task of heterogeneous multi-core design. That strategy takes advantage
of the best level and a planning method at the base level [3]. For each scheduling
strategy, a sporadic server is submitted in the best level plan. Growing intermittent
database is used to plan the tasks dispatched in accordance with its scheduling policy.
An RM OS scheduler is adopted in the base level plan to supervise and prepare the
best kind of intermittent servers.
Sung IL Kim and Jong-Kook Kim showed that for performance and energy effi-
ciency improvement in an HMP environment, intelligent task scheduler is criti-
cal. The actual machine consists of two multi-core processors that are supported
by dynamic voltage/frequency scaling on-the-fly. They extracted several important
components from previously proposed algorithms and combined them to build bet-
ter performance algorithms. The results showed that some of the best combinations
reduced energy consumption and improved environmental completion rates. [4].
Derong Liu, and K. Li, X. Tang et al. proposed heterogeneous multi-core Systems
Pipeline-based Scheduling. The pipeline-based scheduling algorithm proposed in
this paper to improve the performance of heterogeneous multi-core. This algorithm
reduces communication and reduces Worst-Case Execution Time (WCET), relative
deadlines and power utilization, and inactive time for transmitting from parent-kid
task sets to a similar processor, and valuable processor working time during a balance
condition [5–7].
Park et al. proposed the group scheduling algorithm to outline a streamlined
multi-core architecture program. This schedule enhances support space control and
results in inertia reduction and reduction of Worst-Case Execution Time (WCET),
740 L. Dhanesh et al.

Relative Date and Power consumption, CPU utilization, deadlock-free feedback loop
while maintaining comparative overhead synchronization with current techniques
[8]. There is no scheduling strategy that assesses multi-core processor issues and
system conflicts.

3 Proposed Work

3.1 Overview

In distributed and parallel system have more than one processor’s process with paral-
lel programs. The total amount of processing time is needed to execute with overall
process assign with the processor is known as the workload of the processor [9].
The system of scattered computers is connected with tens or hundreds of computers
relatively increase the high-speed networks have many advantages over a system
also known as individual computers. A distributed system provides resource sharing
is one of the main advantages, which also provides better reliability performances
compared with traditional systems. The main issues of distributed and parallel sys-
tems are developing effective techniques for distributing the workload of multiple
processors [10]. Lavanya Dhanesh et al. proposed the Cyclic Priority Preemptive
Task scheduling algorithm which schedules the real-time tasks in the single-core
processor [11]. Lavanya Dhanesh et al. also proposed the power saving of the CPU
by improving the performance of the real-time system kernal using the PSCPPTS
algorithm [12–15].
A load matching is a process of improving the performance of distributed and
parallel systems redistributes a load between the processors. This paper analyzes
the performance of the different optimized multi-task algorithm which is based on
various parameters by considering the two typical load balance approaches namely
the static and dynamic. This approach provides a solution to complex and compu-
tational starved real-time applications. This paper proposes an optimized multi-task
scheduling algorithm that schedules multiple tasks on various cores of multi-core pro-
cessors. The optimized multi-task algorithms not only increase the efficiency of task
scheduler but automatically allocates the appropriate core processors which reduce
the time. This proposed system consists of various parameters are Relative Deadline,
Worst-Case Execution Time, Load Balancing, Interrupt Latency, Utilization Bounds,
and Power Consumption.

3.2 Task Scheduling

The scheduling is a procedure of Central Processing Unit or system factors that are
processing with threads and specify the order of data. Task scheduling is a process of
Enhanced and Energy-Efficient Program Scheduling … 741

assigning the specific time in which each task is completed with overall operation. In
real-time task scheduling system should be used to support the privacy assignment of
the schedule program provides temporal dividing applications. Real-time processors
classification or task scheduler depends on the time criticality.
Scheduling is an essential aspect of real-time, time-constrained systems. Task
scheduling involved resource allocation and task time to perform the performance.
In a real-time application, basic executable objects are scheduled for real-time tasks.
These tasks can be periodic or a-periodic, as well as real-time constraints that are hard
or soft. For scheduling purposes, a task set consists of different scheduling methods
in order to execute the request for the task that they will fulfill. Multiprocessor has
emerged as a more powerful computing resource for real-time applications for the
organization.

3.3 Heterogeneous Scheduling Parameters

Deadline: It is described as the time allotment from exhibiting an undertaking to

the time it is accomplished. A respectable scheduling algorithm reliably attempts to
keep the assignments performed within the due date imperative.
Worst-case Execution Time: This is the exact time it took for the projects to be
done.
Completion Time: This is the time it takes to spread the whole execution of the
work. It joins the cloud system’s execution time and deferral. A number of existing
algorithms for scheduling consider minimizing task completion time.
Performance: Quality shows all the items called efficiency is provided together
by the scheduling algorithm to give the customers great administrations according to
their needs. Optimized multi-task algorithms will consider the execution at the end of
the user as well as the end of the service provider. Since the past decagon, multi-core
processors have also been in effect, but it has gained more noteworthiness late due to
a portion of the innovative constraints such as vast throughput and long-haul battery
existence with gigantic vitality skills.
Load balancing: In addition, it is the procedure for dispersing the entire load in
a system across finished diverse nodes, associates so that no nodes and associations
are stacked immediately to improve efficiency. In order to increase the efficiency of
the system, the proposed scheduling algorithms try to keep the load balanced in a
network.
Power consumption: Thus the optimized multi-task algorithms give preference
to both the higher priority and lower priority tasks in a cyclic way. This reduces
the interrupt latency thus making the improvement in the performance of the CPU
thereby reducing the power consumption of the CPU as a whole.
742 L. Dhanesh et al.

3.4 Multi-Core Processor

Multi-core processors have changed the advanced world today which is possessed
by us and still the unlimited endeavors are being completed to produce quicker and
more astute chips. Since the most basic bearing to broaden the execution of processor
is multi-core, the makers are exceedingly revolved around this development. The
multi-core processor is processed from CPU in a real-time environment.
In which Sort the task is increasing the order of their core, Assign priority level
based on these tasks dynamically gets highest priority tasks, the first task assigns
remaining list of the global queue to the core to have least WCET and the remaining
tasks until wait for next tasks arrives at an instant time. If the new tasks arrive then
the presence of previous tasks and the new arrival the last task is waiting in the queue.
During all processor is busy from the average fraction of time.
A vital part of parallel processing in multi-core systems is to allocate tasks to
processors to accomplish the greatest performance. The objectives of task schedul-
ing algorithms are to maximize system throughput by assigning a task to a proper
processor, make the most of resource utilization, and drop execution time.
The objective of multiprocessor scheduling is to discover an effective solution
to lessen the overall execution time for a collection of subtasks that compete for
working out (Fig. 1).

3.5 Task Structure in Multi-Core Processor

Due to many advantages of multiprocessors compare with single-core processor are

facing the major issues in technology. Many software programs can run slowly with
multi-core processors when compared to single-core processors.

3.6 Optimized Multi-Tasking Algorithm

Optimized Multi-tasking Algorithm is a computer networking system for allocating

work across multiple cores network connections, central processing units, disk drive,
and other devices to optimize resource utilization, increase performance, decrease
response time and avoid overload. The multi-task scheduler is to increase the perfor-
mance so that it can minimize the task waiting time and maximize resource utilization
and finally reduce the power consumption.
Enhanced and Energy-Efficient Program Scheduling … 743

Fig. 1 Proposed multi-core processor system

4 Result and Discussion

This section clearly shows that the differences in all cases using single-core and
multi-core processors. The three graphs visually show the modification of enhanced
results mainly used for large environments.
Figure 2 shows the comparisons of the two techniques are single-core processors
744 L. Dhanesh et al.

performance (%)
100
single core

Accuracy
90 processor
80
single core Multi core
Multi core
processor processor
processor

Techniques

Fig. 2 Comparison between single-core and multi-core processors

and multi-core processors. The proposed techniques using a multi-core processor

provides efficient output than other existing techniques.
Table 1 shows that the comparison of the table with the existing algorithm and
proposed algorithm provides an excellent result that is very efficient. In Table 1,
various parameters like the processing speed, execution time, and power consumption
were compared. The present work provides efficient results and accuracy.
Figure 3 represents the accuracy level in performance analysis of multi-core pro-
cessors with several existing techniques are optimized multi-tasking algorithm, Sta-
ble matching algorithm, and Market Basket Analysis Algorithm. The proposed opti-
mized multi-tasking algorithm is producing an efficient output than other existing
techniques.

Table 1 Comparison of the parameters of three algorithms of multi-core processor

S. No. Techniques Execution time Processing speed Power
(%) (%) consumption (%)
1 Stable matching 60 60 80
algorithm
2 Optimized 35 90 40
multi-tasking
algorithm
3 Market basket 55 85 68
analysis algorithm

92 Market Basket Analysis Algorithm

Accuracy performance (%)

90 Stable matching algorithm

88
Optimized Multi tasking algorithm
86
84
82
80
Market Basket Analysis Stable matching Optimized Multi tasking
Algorithm algorithm algorithm
Multi-core processor techniques

Fig. 3 Comparison between multi-core processor algorithms

Enhanced and Energy-Efficient Program Scheduling … 745

Fig. 4 Comparison of execution time

Fig. 5 Comparison of processing speed

The above Figs. 4 and 5 show a comparative analysis of optimized multi-tasking

algorithm with attributes of execution time and processing speed on the multi-core
processor. This optimized multi-tasking algorithm compared with two existing tech-
niques named Stable matching algorithm, and Market Basket Analysis Algorithm.
The proposed algorithm is producing an effective result than existing techniques.

5 Conclusion

The performance of modern multi-core architectures obviously indicates the feasi-

bility for high performance of multiple tasks is processed within time. In this paper,
746 L. Dhanesh et al.

multi-core processor platform has been designed in the way of providing efficient
parallelism and suitable resource management of workload process with different
resources individuality. The solution of the proposed system consists of dynamic
priority lists for each task and cores. Here, the individual core will be assigned a task
is based on the first priority level in the core’s list.
The proposed Optimized Multi-task scheduling algorithm provides the following
achievements:
1. Increases the performance efficiently
2. Increases the Processing speed
3. Minimizes the task waiting time
4. Maximizes the resource utilization
5. Finally, it reduces power consumption.
Thus the proposed algorithm of Heterogeneous multi-core processors system
provides enhanced efficient output compared with other existing techniques.

References

1. Tang, H.K., Ramanathan, P., Compton, K.: Combining hard periodic and soft aperiodic
real-time task scheduling on heterogeneous compute resources. In: 2011 IEEE International
Conference on Parallel Processing
2. Wang, Z., Ranka, S., Mishra, P.: Temperature-aware task partitioning for real-time scheduling
in embedded systems. In: 2012 IEEE 25th International Conference on VLSI Design
3. Tan, P., Shu, J., Wu, Z.: A hybrid real-time scheduling approach on multi-core architectures.
J. Softw. 5(9) (2010)
4. Kim, S.I., Kim, J.K.: A method to construct task scheduling algorithms for heterogeneous
multi-core systems. IEEE Access 7, 142640–142651 (2019)
5. Liu, D., Jing, M., Wang, Y., Yu, Z., Zeng, X., Zhou, D.: Pipeline-based scheduling for
heterogeneous multicore systems. IEEE (2012)
6. Li, K., Tang, X., Veeravalli, B.: Scheduling precedence constrained stochastic tasks on
heterogeneous cluster systems. IEEE Trans. Comput. 64(1), 191–204 (2015)
7. Li, K., Tang, X., Yin, Q.: Energy-aware scheduling algorithm for task execution cycles with
normal distribution on heterogeneous computing systems. In: 41st International Conference
on Parallel Processing, pp. 40–47 (2012)
8. Park, J., Dally, W.J.: Buffer-space efficient and deadlock-free scheduling of stream applications
on multi-core architectures. In: Proceedings of the 22nd ACM Symposium on Parallelism in
Algorithms and Architectures, pp. 1–10 (2010)
9. Geng, X., Xu, G., Fu, X., Zhang, Y.: A task scheduling algorithm for multi-core-cluster systems.
J. Comput. 7(11) (2012)
10. Yun, Y., Shi, S.: Tasks scheduling algorithm for parallel system with multi-core processor.
Comput. Appl. 28(12), 280–283 (2008)
11. Dhanesh, L., Murugesan, P.: Smart scheduling of the real -time tasks using the CPPPS
algorithm. J. Comput. Theor. Nano Sci. 14(3), 1–8. ISSN 1546-1955
12. Dhanesh, L., Murugesan, P.: Power saving of the CPU by improving the performance of the real
-time system kernal using the PSCPPTS algorithm. Int. J. Appl. Eng. Res. 10(5), 12465–12473.
ISSN 0973-4562 (2015)
13. Dhanesh, L., Murugesan, P.: A novel approach in scheduling of the real- time tasks in hetero-
geneous multicore processor with fuzzy logic. Int. J. Power Electron. Drive Syst. (IJPEDS)
9(1), 80–88 (2018)
Enhanced and Energy-Efficient Program Scheduling … 747

14. Deepa, S., Anipriya, N., Subbulakshmy, R.: Design of controllers for continuous stirred tank
reactor. Int. J. Power Electron. Drive Syst. (IJPEDS) 5(4), 576–582 (2015)
15. Deepa, S., Praba, S., Deepalakshmi, V., Jayaprakash, L., Manimurugan, M.: A fuzzy Ga based
Statcom for power quality improvement. Int. J. Power Electron. Drives 8(1), 483–491 (2017)
A Prototype of Density-Based Intelligent
Traffic Light Control System Using
Image Processing Technique
and Arduino Microcontroller in Lab
VIEW Environment

Anita Mohanty, Subrat Kumar Mohanty, and Jitesh Kumar

Abstract Nowadays traffic congestion is a serious issue associated with transporta-

tion, the backbone of the economy of a city. Because of the rise in population and
number of vehicles on a road, traffic jam is very common all over the world. Conges-
tion not only rises pollution, stress, frustration but also wastes money, fuel. Another
serious cause of congestion is the delay in red light at a junction. In our work, we
proposed a technique to optimize the red light ON duration of traffic light controller
depending on traffic. Regulation of road traffic at each junction in a city is the main
aim of our work. Our system measures traffic density at different lanes at a junction
and accordingly changes the time delay of red light. This system controls the traffic
light by image processing using MATLAB. Cameras are installed for each lane to
capture the image, which is analyzed by Lab VIEW to detect congestion on a partic-
ular lane and according to congestion, the green light of each lane is controlled from
that lane using Arduino microcontroller.

Keywords Congestion · Image processing · Edge detection · Traffic light ·

Arduino microcontroller

1 Introduction

Nowadays traffic jam is a big issue in countries all over the world which have a
greater impact on the transportation system. In a daily basis, it is very irritating to
be in traffic. By this, there has been a decrease in the average velocity of vehicles.
There has been a loss in time, money, opportunities, productivity from workers, trade
opportunities, delaying in delivery goods. Rapid growth of a number of automobiles
and the constant rise of road users can’t be accompanied by existing infrastructure.
To solve these congestion created problems, we have to build new infrastructures

A. Mohanty (B) · J. Kumar

Silicon Institute of Technology, Silicon Hills, Bhubaneswar, Odisha, India
e-mail: [email protected]
S. K. Mohanty
College of Engineering Bhubaneswar, Bhubaneswar, Odisha, India
© Springer Nature Singapore Pte Ltd. 2020 749
G. Pradhan et al. (eds.), Advances in Electrical Control and Signal Systems,
Lecture Notes in Electrical Engineering 665,
https://doi.org/10.1007/978-981-15-5262-5_56
750 A. Mohanty et al.

and facilities like new roads, flyovers, bypass roads, intercity trains, and the creation
of rings for smooth and efficient traffic flow. Expansion of roads and lanes is not
possible in all situations, but incorporating intelligence into the road traffic systems
with advanced and reliable technology is surely possible. For Intelligent Transport
System (ITS), traffic data needs to be collected and distributed. Traffic information
may be collected from a variety of sensors like inductive loop [1], infrared light
sensors [2], magnetic sensors [3], piezoelectric sensors [4] and acoustic sensors [5].
By this traffic will be interrupted during installation. They have limited coverage.
Their cost may rise and reliability concerns may arise due to repair and failure. Hence
intelligence-based traffic light control system is a better solution. In conventional
traffic light systems, prior a fixed timing is set for the lights. But they are not desirable
as they don’t work for actual road conditions. Depending on the change in timer
value, the lights are getting ON and OFF automatically. Operation of traffic signals
is important for the control of vehicles in road when properly used. They will provide
orderly movement of traffic, co-ordination for uninterrupted flow, increase motorist
confidence by appointing correct path.
This can be achieved by using graphical programming language Lab VIEW, a view
based video camera system. It is useful for finding traffic parameters in wide area
as it is a real-time image processing technique [6–9]. Through image measurement,
road congestion can be described qualitatively as well as traffic status such as speed
of vehicles, number of vehicles can be provided quantitatively. These quantitative
measures give traffic parameters that contribute us complete information about traffic
flow [8].
This proposed method is an elementary, economical actual time traffic light control
structure able to differentiate the appearance and nonappearance of automobiles in
road images. If road is vacant, traffic light goes to RED and goes to go GREEN in
case of appearance of vehicles on the roadway.

2 Related Work

In [10], the authors proposed the sensor connected microprocessor-based traffic light
control system. Traffic light timing change creates a negative effect on traffic. Too
early applied modification in timing may proceed to jam in other lane and much
delay may create chaos, queuing in vehicles, road rage, accidents, etc. In [11], the
authors proposed an embedded-based system by using sensor networks. The red and
green lights timings at every intersection of roadway are sensibly determined on
whole vehicles in each lane. The users who need to get the latest position of traffic
in advance are provided with GSM cell phone interface. This helps the drivers to
choose another Route in case of a jam. In [12], An ECODE protocol is proposed
to find the segments of road affected by severe traffic jams. It handles multi-hop
communication and geo-cast techniques to collect and disperse data among vehicles.
A multiagent-based technique known as CARAVAN is proposed in [13] to avoid
traffic congestion using virtual agent recognition. An event-driven architecture is
A Prototype of Density-Based Intelligent Traffic Light Control … 751

proposed in [14] to find traffic jams in the background of an allocated traffic instruc-
tion system. In [15], fuzzy logic based advanced traffic light operational structure is
proposed to optimize the duration of lights. But this technique is having limitations
because of a shortage of learning function and experts proposed these function’s
confidence in many rules. In [16], the authors proposed an algorithm for measuring
the density of traffic using image processing to adjust the timing of traffic lights. But
the operation of traffic lights after congestion estimation by using the algorithm is
not proposed. In [17], a fuzzy logic based traffic light control system is proposed to
minimize the congestion at a junction. The density of vehicles is calculated by using
an image processing tool in Lab VIEW. Lack of confidence in forming the fuzzy
rules is the limitation of this technique. In [18] authors proposed a real-time vehicle
detection based on background differencing, morphological operations, and edge
detection. Traffic parameters such as counting the number of vehicles and speed of
vehicles are calculated using threshold techniques. Authors in [19] have depicted an
efficient video-based vehicle detection system based on Harris-Stephen corner detec-
tion algorithm. They implemented a traffic detection system on embedded computer
platforms by taking video images of roads. In [20], the authors proposed an automatic
algorithm based on artificial neutral network and fuzzy logic controller to control
traffic light time based on image processing by acquiring images of vehicles. The
efficiency of this system can be increased by proper alignment of cameras. The error
associated with this system can be decreased by increasing training data with new
data of vehicles. In paper [21], traffic density is estimated by measuring the total area
occupied by vehicles on the road. The traffic cycle is set depending on the total traffic
density of all the roads at the junction. A weight is calculated depending on the traffic
density for each road. This model is applicable for less number of interconnected
traffic junctions, but not applicable for large number of interconnected traffic junc-
tions. In paper [22], the authors proposed a traffic density estimation technique using
image processing based on the area occupied by the edges of vehicles. This method
estimates traffic congestion to control traffic signals but still requires development
to get higher accuracy.
This proposed system is cable to detect intensity of vehicles on all four lanes of
the junction, and use the following two forms of control:
1. Depending on the traffic density in each lane of the junction, on priority base,
it runs the sequence of traffic lights. The lane with the more vehicles is getting
the highest priority and the sequence continues till congestion detected in other
lanes. This will help to avoid the jam on roads with more traffic.
2. It also adjusts the time duration of the signal lights at the intersection depending
on the number of vehicles. Hence a road with less number of vehicles will have
green signal for a shorter duration, in the meantime a road with more number of
vehicles will have green signal for a longer duration.
752 A. Mohanty et al.

3 Proposed Method

Today’s existing system for road traffic control at different junction of a city is based
on “time” which is already assigned in the system for a junction. According to these
assigned time, the signals are working for each lane. But sometimes the situation
will occur when all the vehicles in a lane are still in waiting condition as the time
is not over and signal is still red. Nowadays as time is very valuable, so these types
of systems are very inefficient. The suggested method is minimizing the chances of
traffic congestion due to traffic lights by allowing the vehicles in the lane with higher
number of vehicles. By this, less number of vehicles will be there at a junction and
can reduce the waiting time of people at different junctions.
The main attraction of this developed system is to estimate the total vehicle queue
in the congested area using image processing techniques in MATLAB and LAB
VIEW environment. Its results are utilized in the control of traffic light system which
is implemented using Arduino.
The designed smart traffic light control system is having four directional roads as
shown in Fig. 1. The intersection of these roads is supplied with traffic three colored
lights: Red, Green, and Orange. At the road junction four still cameras placed. Each
still camera captures the image of the road in a clockwise direction starting from the
north. Then, the image processing is done using MATLAB to found the congestion
on the road by computing the vehicle count. Based on its value the traffic light
system is controlled easily by allocating the required time for a green signal for
each lane using Lab VIEW. Here because of suitability in computation, MATLAB is
preferred, whereas LAB VIEW is suitable for the acquisition of images, processing,
and illustrating traffic signals.
The methodology proposed in the simulation model is, at a moment only traffic
from one direction is allowed to go and in the rest three directions, the moving of
vehicles is blocked. To maintain a constant flow of traffic in all the directions we are
assuming the following conditions.
Condition 1: Shows green light against north direction and red lights for remaining
directions.

Fig. 1 Lane diagram

A Prototype of Density-Based Intelligent Traffic Light Control … 753

Condition 2: Shows green light to west direction and red lights for rest of
directions.
Condition 3: Displays green light towards south direction and red lights to rest of
the directions.
Condition 4: Displays green light to east direction and red lights to rest of
directions.
A specific time gap is assigned for all of the above conditions, as constant traf-
fic flow is assumed in every direction. This sequence of operation repeats itself
depending on the congestion on the lane to avoid waste of green light time for free
roads.

4 Model Descriptions

Figure 2 shows the block diagram of the suggested system. The cameras are installed
to capture images of the four lanes. The images are captured repeatedly even if the
current traffic light sequence is working. But the new sequence happens only after
the end of the ongoing sequence.
The algorithm to measure traffic density and to operate traffic lights accordingly
is expressed by the flowchart as depicted in Fig. 3. The density of traffic is measured
in each lane for adjustment of the timing of the traffic light.

Number of
vehicles in each lane
Image Processing
( Matlab)

LabVIEW Environment

Algorithm

Ardiuno
Traffic Lights

Fig. 2 The block representation of the suggested system

754 A. Mohanty et al.

Fig. 3 Flowchart of the

proposed system

The proposed model of the system has the following modules:

(a) Software
(b) Interface
(c) Hardware.

4.1 Software Module

This includes MATLAB 2015 version for image processing to detect traffic conges-
tion in a lane by estimating the intensity of vehicles. This processing is done in Lab
VIEW environment by using Lab VIEW 2012 software. The simulation model of
operation of traffic light is constructed by using Lab VIEW.
A Prototype of Density-Based Intelligent Traffic Light Control … 755

4.1.1 MATLAB for Estimation of Number of Vehicles

The process of estimating the intensity of vehicles in each lane includes the following
steps:
Step 1: Acquisition of Images:
Image of each lane is acquired through the camera placed on each lane as it is a
small prototype; the images of different roads stored in.jpg format are used for image
processing by using MATLAB. Image from all four lanes are acquired simultaneously
in clockwise sequence starting from the north and saved accordingly. The location
of the directory is specified in the program to retrieve the correct image.
Step 2: Processing of Images:
The captured images are sent for processing to obtain the count value in the same
clockwise sequence starting from the north. They are transformed into greyscale
image, then to a binary image containing only two colors, black and white. Thresh-
olding of this image is for radical reduction of information to simplify further pro-
cessing. This threshold image is accomplished for further image processing. The
steps involving image processing are shown in Fig. 4.

(a) Read Captured Image:

To read the captured image, imread() function is used. It retrieves the stored image
for processing. We have to specific the location of directory in order to retrieve the
correct image. The imshow() function is used to display the image while simulation.

(b) RGB to GRAY Conversion:

To convert RGB to GRAY we use, rgb2gray() function. It reduces a color image to a

grayscale image, i.e., image with different scales of gray, for reduction of processing
time, and hence it could be proved advantageous for real-time implementation as
there we have to take a decision in a short period of time.

Read RGB to GRAY to

Captured Image GRAY BINARY
Conversion Conversion

(a) Edge
Counting Complementing
Detection the Image

Fig. 4 Steps of image processing

756 A. Mohanty et al.

(c) GRAY to BINARY Conversion:

GRAY scale image is converted to BINARY image gray2binary() function is used.

This function reduces the image to further from different shades of gray to binary
color i.e., either black or white which can be easily read by the processor. The shade
of gray which has more intensity is converted to white and shade having less intensity
is converted to black.

(d) Complementing the Image:

The complementing the image means to convert the white part of binary image to
black and vice-versa. It is done so that the small patches present in the image could
be eliminated. The function used for this is imcomplement() function.

(e) Edge Detection:

This is used to detect the edge of the object and hence to count the number of objects.
The function used here is edge() function.

(f) Counting:

Ultimately, the number of objects detected is counted and stored in a variable from
where it is sent for processing.

4.1.2 Lab VIEW for Traffic Light Execution

In LAB VIEW, the programs developed are known as virtual instruments or VIs, with
extension.vi. These VIs are receiving data from the user or from the PC interfaces
with the process. They process them and display, store, or transmit them remotely.
A VI has the following components:
• Front Panel—works as the user interface.
• Block Diagram—has a graphical source code which defines the functionality of
the VI.
• Icon and Connector Pane—finds the interface to the VI so that a VI can be used
in another VI. A subVI is a VI within another VI. It is a subroutine in text-based
Lab VIEW programming languages.

4.2 Interface

The interfacing between the software running in a PC with the prototype is established
by using PC parallel port with USB cable.
A Prototype of Density-Based Intelligent Traffic Light Control … 757

4.3 Hardware Module

This module includes PC, Arduino Microcontroller, and LEDs. A general-purpose

PC is used for image processing in Lab VIEW environment. The prototype is designed
with microcontroller and LEDs.
Changing the duration of traffic light
The time gap operation of the traffic signal is operated by following the steps given
below:
(1) The clicked pictures by cameras are input to MATLAB for the processing of
pictures/images.
(2) After processing in MATLAB, the priority of each lane is finalized as per its
intensity of traffic.
(3) The highest prioritized lane is the lane with the highest relative intensity of
traffic.
(4) The green signal of the lane with highest priority is having the duration such
a way that the vehicles can cross the junction which ultimately reduces the
congestion. The rest of the lanes have their green signals as per the original
sequence.
(5) When all the lanes have shown their green signals; the sequence of operation of
traffic light completes its one cycle of traffic light monitoring and congestion
control.
(6) This process gets repeated for whole of day if allowed uninterrupted.

5 Experimental Results

The acquired images of each lane at a particular instant of time are stored in four
different.jpeg files and are shown in Fig. 5.
The block diagram in Lab VIEW to detect congestion is represented in the diagram
Fig. 6.
After counting the number of vehicles in each lane, the output is used to operate
traffic lights in Lab VIEW and also the data is used in Arduino microcontroller to
operate traffic lights present in the prototype. The Lab VIEW program for connection
to Arduino and traffic lights is shown in Fig. 7.
The Front Panel in Lab VIEW for control of traffic light is represented in Fig. 8.
It contains the following modules:
• Indicators to show the number of vehicles in each lane
• 3xLEDs indicators for each lane, belonging to the traffic lights.
The density of vehicles based on the proposed algorithm in each lane is calculated
and listed in Table 1.
758 A. Mohanty et al.

Route 1 Route 2

Route 3 Route 4

Fig. 5 The images captured in all four lanes at a junction

Fig. 6 The block diagram (Lab VIEW) of proposed system to detect congestion
A Prototype of Density-Based Intelligent Traffic Light Control … 759

Fig. 7 The block diagram (Lab VIEW) of proposed system to connect Arduino to operate traffic
light

Fig. 8 Lab VIEW front

panel for detecting
congestion and controlling
traffic lights

Table 1 Number of vehicle

Name of the lane Number of vehicles estimated
of vehicles detected in each
lane Route 1 136
Route 2 1
Route 3 202
Route 4 16
760 A. Mohanty et al.

Table 2 Duration of lights in

Name of the light in each lane Duration
each lane at a junction
Red 1000 ms
Green 1000 * number of vehicles of
the lane
Yellow 1000 ms

From Fig. 8 and Table 1, it is cleared that number of vehicles calculated in Route
3 is more. Hence the traffic light sequence starts from Route 3 as shown in Fig. 8.
The traffic light’s duration depends on the number of vehicles of that lane. As more
number of vehicles are detected in Route 3, the GREEN light duration for that lane
is more to minimize the traffic in that lane (Table 2).
When one of the lanes get large number of vehicles, that the corresponding lane
is accomplished first and the rest of the lanes follow as per the sequence of opera-
tion starting from the congested lane. This task performs continuously by advanced
Arduino microcontroller which controls the traffic lights. The lane comparison is
graphically represented by using LAB VIEW as shown in Fig. 8.
The implementation of this system requires the materials: an Arduino UNO, 12
LEDs (4 Red, 4 Yellow, and 4 Green), and 12 × 220 O resistors and to connect
jumpers wires and are shown in Fig. 9.
These materials are used to implement the prototype to fulfil the requirement of
a traffic light control system. The experimental set up is shown in Fig. 10.
After the implementation of the prototype and compilation, the Lab VIEW pro-
gram is uploaded to Arduino board and LEDs will glow. The condition of the traffic
lights is checked and voltages at each LED light are listed in Table 3.
The proposed method is compared with regular method on different aspects. The
different aspects and their comparisons are listed in Table 4. From this comparison,
it is cleared that the proposed method is better than regular method.

Fig. 9 a Arduino Uno board, b USB cable, c LED and Resistors

A Prototype of Density-Based Intelligent Traffic Light Control … 761

Fig. 10 Experimental set up

Table 3 The voltage levels for each LED for each lane for a single sequence of operation
Condition East West North South
Normal GREEN ON RED ON (Lane RED ON (Lane RED ON (Lane
sequence (Lane 1, 3.8 V) 2, 3.5 V) 2, 3.7 V) 2, 3.9 V)
YELLOW ON YELLOW ON YELLOW ON YELLOW ON
(Lane 1, 3.5 V) (Lane 2, 3.7 V) (Lane 1, 3.6 V) (Lane 1, 3.5 V)
RED ON (Lane GREEN ON RED ON (Lane RED ON (Lane
1, 3.8 V) (Lane 2, 3.9 V) 1, 3.5 V) 1, 3.5 V)

Table 4 Comparison of proposed method with regular methods

Aspects Regular method Proposed method
Traffic personnel prerequisite in peak hours Yes No
Congestion chances High Low
Chances of varying duration of traffic light No As per requirement
Waiting time in traffic More Less
CO2 emission More Less

6 Conclusion

The suggested method is created to minimize waiting time for drivers at a road junc-
tion due to traffic. By following this proposed method traffic signal duration will
adjust depending on the actual time requirements as a change in the road traffic is
762 A. Mohanty et al.

encountered. This intensifies the effectiveness in management of road traffic, which

ultimately rise the efficiency of control of traffic flow especially in cities where the
roads are very congested. This technique is proficient enough to judge the flood of
vehicles at traffic junction and manage the traffic lights depending on the traffic flow
towards each direction. The image processing technique is considered to be an effi-
cient method to analyze and operate road traffic lights. This technique is considered
as an easy method for raising the effective management of traffic and peak hours’
congestion control. Also can replace the work and involvement of human being in
control of traffic. It depicts the scenario in real-time. Hence it is efficient enough in
comparison to other methods based on traffic control using sensors and infrared light
techniques.

References

1. Ki, Y., Baik, D.: Model for accurate speed measurement using double-loop detectors. IEEE
Trans. Veh. Technol. 55(4), 1094–1101 (2006)
2. Deekshitha., Disha, D., Malavika, Soumya.: Traffic monitoring system using IR sensors. Int.
J. Adv. Res. Ideas Innovations Technol. 3(3), 1045–1057 (2017)
3. Haoui, A., Kavaler, R., Varaiya, P.: Wireless magnetic sensors for traffic surveillance. Transp.
Res. Part C: Emerg. Technol. 16(3), 294–306 (2008)
4. Zhang, J., Lu, Y., Lu, Z., Liu, C., Sun, G., Li, Z.: A new smart traffic monitoring method using
embedded cement-based piezoelectric sensors. Smart Mater. Struct. IOP Sci. 24(2), 1–8 (2015)
5. Barbagli, B., Manes, G., Facchini, R., Manes, A.: Acoustic sensor network for vehicle traffic
monitoring. In: International Conference on Advances in Vehicular Systems, Technologies and
Applications, VEHICULAR-2012, pp. 1–6 (2012)
6. Beymer, D., McLauchlan, P., Coifman, B., Malik, J.: A real-time computer vision system
for measuring traffic parameters. In: IEEE Conference on Computer Vision and Pattern
Recognition, p. 495–501 (1997)
7. Fathy, M., Siyal, M.Y.: An image detection technique based on morphological edge detection
and background differencing for real time traffic analysis. Pattern Recogn. Lett. 16, 1321–1330
(1995)
8. Ferrier, N.J., Rowe, S.M., Blake, A.: Real-time traffic monitoring. In: Proceedings of the Second
IEEE Workshop on Applications of Computer Vision, pp. 81–88 (1994)
9. Cucchiara, R., Piccardi, M., Mello, P.: Image analysis and rule-based reasoning for a traffic
monitoring system. IEEE Trans. Intell. Transp. Syst. 1(2), 119–130 (2000)
10. Sehgal, V.K., Dhope, S., Goel, P.: An embedded platform for intelligent traffic control. In:
UKSim Fourth European Modeling Symposium on Computer Modeling and Simulation, IEEE
Computer Society, pp. 541–545 (2010)
11. Chavan, S.S., Deshpande, R.S., Rana, J.G.: Design of intelligent traffic light controller using
embedded system. In: Second International Conference on Emerging Trends in Engineering
and Technology, ICETET-09, IEEE Computer Society, pp. 1086–1091 (2009)
12. Maram, Y.B., Azzedine, B.: Efficient traffic congestion detection protocol for next generation
VANETs. In: ICC. IEEE, pp. 3764–3768 (2013)
13. Fernando, T.S., Mercedes, V.V., Cristina, S.M.: A cooperative approach to traffic congestion
detection with complex event processing and VANET. IEEE Trans. Intell. Transp. Syst. 13(2),
914–928 (2012)
14. Prajakta, D., Seng, L.W., Aniruddha, D., Jugdutt, S.: CARAVAN: congestion avoidance and
route allocation using virtual agent negotiation. IEEE Trans. Intell. Transp. Syst. 14(3), 1197–
1207 (2013)
A Prototype of Density-Based Intelligent Traffic Light Control … 763

15. Azimirad, E., Pariz, N., Sistani, M.B.N.: A novel fuzzy model and control of single intersection
at urban traffic network. IEEE Syst. J. 4(1), 107–111 (2010)
16. Al okaishi, W., Atouf, I., Benrabh, M.: Real-time traffic light control system based on back-
ground updating and edge detection. In: International Conference on Wireless Technologies,
Embedded and Intelligent Systems, IEEE Publication (2019) 1–5
17. Patela, V.K., Patel, M.N.: Development of intelligent traffic control system by implementing
fuzzy-logic controller in lab view and measuring vehicle density by image processing tool in
Lab VIEW. Am. Sci. Res. J. Eng. Technol. Sci. (ASRJETS). Global Soc. Sci. Res. Researchers
26(4), 406–417 (2016)
18. Prutha, A.: Morphological image processing approach of vehicle detection for real-time traffic
analysis. Int. J. Eng. Res. Technol. (IJERT) 3(5) (2014)
19. Chintalacheruvu, N., MuthuKumar, V.: Video based vehicle detection and its application in
Intelligent transportation systems. Sci. Res. J. Transp. Technol. 2, 305–314 (2012)
20. Suhweil, Y., Al Yaman, M.: Smart controlling for traffic light time. In: IEEE Jordan Conference
on Applied Electrical Engineering and Computing Technologies (AEECT), pp. 1–5 (2017)
21. Hasan, M.M., Saha, G., Hoque, A., Majumder, M.B.: Smart traffic control system with applica-
tion of image processing techniques. In: International Conference on Informatics, Electronics
& Vision, pp. 1–4 (2014)
22. Uddin, M.S., Das, A.K., Taleb, M.A.: Real-time area based traffic density estimation by image
processing for traffic signal control system: Bangladesh perspective. In: International Confer-
ence on Electrical Engineering and Information & Communication Technology (ICEEICT),
pp. 1–5 (2015)
Modeling and Performance Analysis
of n-FinFETs: A Comparative Study

J. Jena , T. P. Dash, E. Mohapatra, S. Das, J. Nanda, and C. K. Maiti

Abstract A comparative analysis of full electrostatic performance and RF charac-

teristics of n-type FinFETs is carried out by using the different physical models.
Quantum confinement effects are shown to be significant in sub-100 nm FinFET
devices. In the simulation, the ATLAS device simulation tool is used in which vari-
ous quantum corrections are implemented in models such as (Drift-Diffusion) DD,
(Density-Gradient) DG, (Energy Balance) EB, and (Bohm Quantum Potential) BQP.
It has been shown that the FinFETs possess the key advantages for applications in RF
electronics, around ~15 GHz of transition frequency (ft) where the unilateral power
gain-bandwidth product is important for the devices.

J. Jena (B) · T. P. Dash

Department of Electrical and Electronics Engineering, Institute of Technical Education and
Research, Siksha ‘O’ Anusandhan (Deemed to be University), Bhubaneswar, Odisha 751030,
India
e-mail: [email protected]
T. P. Dash
e-mail: [email protected]
E. Mohapatra · C. K. Maiti
Department of Electronics and Communication Engineering, Institute of Technical Education and
Research, Siksha ‘O’ Anusandhan (Deemed to be University), Bhubaneswar, Odisha 751030,
India
e-mail: [email protected]
C. K. Maiti
e-mail: [email protected]
S. Das
Department of Electronics and Communication Engineering, Silicon Institution of Technology,
Silicon Hills, Bhubaneswar, Odisha 751024, India
e-mail: [email protected]
J. Nanda
Department of Physics, Institute of Technical Education and Research, Siksha ‘O’ Anusandhan
(Deemed to be University), Bhubaneswar, Odisha 751030, India
e-mail: [email protected]

© Springer Nature Singapore Pte Ltd. 2020 765

G. Pradhan et al. (eds.), Advances in Electrical Control and Signal Systems,
Lecture Notes in Electrical Engineering 665,
https://doi.org/10.1007/978-981-15-5262-5_57
766 J. Jena et al.

Keywords Bohm quantum potential · Energy balance model · Transition

frequency (ft) · FinFET

1 Introduction

Multigate FET technology is now known to be the best alternative that can extend
CMOS scaling to sub-10 nm technology nodes with minimum additional process-
ing costs [1, 2]. From the fabrication perspectives, the most likely candidate for
widespread adoption among the multigate devices is the FinFETs. Process steps
play an important role in determining device performance [3]. Though FinFET pro-
cess steps are reported by many research groups [4, 5], detailed optimized fabrication
process steps are not available in the public domain. As of now, the device’s physical
attributes are not yet generalized, in this work, we have done some studies on device
performance prediction by comparing several quantum transport models. We investi-
gate several transport models for efficient multi-dimensional simulation of quantum
confinement effects in advanced FinFETs to analyze both the dc and ac performance.
The purpose of models is described briefly as the DG model can predict transport
properties and also accurately reproduce the carrier concentration predicted by the
Schrodinger–Poisson model. The BQP model has some better flexibility over the
density-gradient model as better convergence and calibration to the Schrodinger–
Poisson model. However, the Schrodinger–Poisson model predicts the bound state
energies or wave functions. In the EB models, unlike the charge transport DD model
neglects non-local effects, through the use of a higher-order approximation of the
Boltzmann Transport Equation (BTE) are functions such as mobility and impact
ionization of the local carrier temperature rather than the local electric field [6].
We have also done the analyses of the mobility models of the electron and hole
as they are also responsible for RF performance [7]. At the high doping profile of
the channel, cause serious degradation because of Coulomb scattering in the carrier
mobility for high-frequency analysis. The fringing capacitances play an important
role in the unilateral power gain calculation at the high-frequency response. Because
of the capacitance effect, power gain performance degraded and transition occurs
between passband and stopband. The cut-off frequency is f t = f (gm, Cgs, Cgd),
whereas gm, Cgs, and Cgd are known as conductance, gate to source capacitance,
and gate to drain capacitance, respectively [8]. B. Chen et al., shown an RF analysis
of the FinFET by comparing three types of devices as a varying number of Fins with
wide drain extension Fins (wide drain FinFET), drain extension (hybrid FinFET),
and lower drain resistances and observed the maximum cut-off frequency of 53 GHz
at hybrid FinFET [9]. In the frequency response plots, we have shown the cut-off
frequency at 0 dB for different models.
Through the device simulations, one can perform the operations of DC, AC, and
mixed-mode simulations. In this work, TCAD simulations have been performed for
tri-gate FinFET with 30 nm gate length. Different device simulations have been per-
formed to understand the electrical response of the devices. To evaluate the device
Modeling and Performance Analysis of n-FinFETs … 767

performance we have analyzed ballistic and non-ballistic transport models consid-

ering the quantum mechanical effect. The EB model and DD model are compared
along with the necessary quantum corrections. The RF performance has been studied
and cut-off frequency (ft) has also been calculated.
The paper is organized as follows: Section II describes the device details and
simulation environment. A brief description of different transport models has been
presented in this section. Section III describes the device simulation results along
with the frequency response of the FinFETs. Section IV summarizes the conclusions.

2 Device Details and Simulation Methodology

The 3D n-FinFET device structure used in the simulation is shown in Fig. 1. The
design parameters of the simulated device are listed in Table 1.

Fig. 1 The 3D n-FinFET

device structure used in the
simulation

Table 1 Geometrical details

Geometrical parameters Value
of the device
Gate length 30 nm
Fin height 50 nm
Fin width 30 nm
S/D doping 1 × 1020 /cm3
Substrate doping 1 × 1015 /cm3
Gate oxide thickness 1 nm
768 J. Jena et al.

The net doping profile of the simulated FinFET is shown in Fig. 2. The 3D doping
profile (see Fig. 2a) shows the overall dopant distribution throughout the device.
The dopant concentration in the active region is a very important concern for device
performance. The 2D doping profile (XZ-plane) is shown in Fig. 2b which is obtained
by taking a cut-plane at the mid of the y-axis. It shows the doping profile across the
active area (source, drain, and channel) of the simulated device. Figure 2c shows
the quantitative doping value (1D profile) in n and p-type across the structure from
source to drain. The channel doping concentration is the substrate doping which is
1 × 1015 /cm3 , where the source and drain doping are maintained at 2 × 1018 /cm3 .
Such device considerations are verified from the electrical characterization of
the devices using suitable transport models and its RF performance for unilateral
power gain. The simulation results are studied form the characteristics plots of Id-
Vg, Id-Vd, and frequency response using different models. Advanced transport mod-
els that support predictive technology development in the microelectronics industry
with required calibration [10, 11] are used for the behavioral analysis of the device
structure shown in Fig. 1.
Different device simulations have been performed to understand the electrical
response of the semiconductor devices. The quantum effect comes into the picture
in device characteristics when the device scales down to nanometer range. In the
classical drift-diffusion transport model, the quantum effect is not taken into account.
The position of an electron is described by the probability density function. When
density-gradient is needed, one can also make use of the Poisson solution with the
Schrodinger equation (self-consistent) for correct calibration [12, 13].

2.1 Drift-Diffusion (DD) and Density-Gradient (DG) Models

The solution of the Boltzmann equation is particularly complex in its entirety. The
models of DD which can be derived from the Boltzmann equation by the method of
moments:
−
→ 
Jn (r, t) = |q|Dn ∇n(r, t) + |q|n(r, t)µn E (1)

−
→  p(r, t) + |q| p(r, t)µ p E
J p (r, t) = −|q|D p ∇ (2)

kT kT
Dn = µn , D p = µp (3)
|q| |q|

whereas J, D, µ, k, q, E, and T represent the current density, the diffusion coeffi-
cient, mobility, Boltzmann constant, charge of a carrier, electric field, and temper-
ature, respectively. The indices p and n in Eqs. (1)–(3) are associated with holes
and electrons, respectively. The electron and hole mobility is a combination of three
components which are the mobility due to intervalley phonon scattering, surface
Modeling and Performance Analysis of n-FinFETs … 769

Fig. 2 a 3D Net doping profile of a simulated device, b 2D net doping profile in the YZ plane,
and c 1D plot of the doping concentration of source-channel-drain (from left to right). Doping
concentration is shown in the logarithmic scale
770 J. Jena et al.

roughness scattering, and acoustic phonon. However, from advanced transport mod-
els, in a high field regime, the DD simulations must have currents consistent resulting
in the context of ultra-short channel transistors. In the DG model, the quantum cor-
rection potentials for holes and electrons have been added to the current relation of
the DD model and, therefore, it has been extended further [2]. It is derived from the
method of moments applied to the Wigner equation.

2.2 Bohm Quantum Potential Model (BQP)

The BQP is used to determine the quantum mechanical effects. It introduces a

position-dependent quantum potential. The advantage of the BQP model consists
in the possibility to correct model quantum effects without the explicit solution of
the Schrodinger equation [14].

2.3 Energy Balance Model (EB)

The EB model is used for the effects of ballistic carrier transport. It significantly
increases drain currents and is the cause of the decrease in the output impedance
of the Id-Vd curves. It also uses the BQP model to take into account the quantum
confinement of carriers in 3D which further decreases the drain currents [6].

3 Results and Discussion

As the drift-diffusion model considers the scattering effect, hence it is expected that
it would yield a lower current than that of the energy balance model. Drift-diffusion
model along with quantum mechanical effect also performs in the same way as in
the case of the EB model, it reduces the value of current due to quantum confinement
[6]. Figure 3 shows the electron concentration profile in a particular 2D (XZ) plane
of the device using the EB-BQP model. In this model, the maximum charge carriers
participate in the conduction than other models.
In Fig. 4, mobility profiles have been shown for electron and hole using the BQP
model with the graphical representation from source to drain. As electron mobility
is larger than hole mobility, the maximum electron and hole mobility is observed
about 480 and 220 cm2 /V-s at the channel region, respectively. Total current density
has been observed for different models in Fig. 5.
The Lombardi Mobility model (CVT) is used for the transverse field dependence
study of particular device simulation. The Bohm Quantum Potential (BQP) model
has been used as it has the advantages over the density-gradient model for the better
convergence to the Schrodinger–Poisson solution.
Modeling and Performance Analysis of n-FinFETs … 771

Fig. 3 Electron concentration profile in the XZ direction of the devices using the EB model

Hot carrier transport equations (HCTE) are used by the model to include the
electron or hole carrier temperature. Figure 5 shows the current density profile of
different transport models. It has been observed with the effect of BQP + HCTE and
BQP + CVT. From the density profile, it has been observed that the effect of BQP
+ HCTE is more than BQP + CVT. Figure 6 shows the transfer characteristics for
the FinFET considering all transportation models discussed so far. The drain current
becomes highest in the EB model as it follows the ballistic transport equations. The
reduction of current from only the EB model is due to the quantum confinement
of some of the carriers out of total carriers that take part in current conduction. To
compare with the non-ballistic transport model, we have also obtained the Id-Vg
characteristics for the drift-diffusion model along with its quantum correction (with
BQP) model.
The Id-Vd curves have been shown in Fig. 7 for different transport models. For
different transport models, the increase in drain current with a ballistic model is
because of reduced scattering and output impendence. Drain current is highest with
an energy balance model followed by drift-diffusion without considering quantum
effect. With consideration of quantum correction, the drain current decreases up to
40% using an energy balance model and 50% using the drift-diffusion model. While
the EB model and DD model are compared, the EB model shows 35% and 44%
increment of current compared to the DD model with and without consideration of
quantum effects, respectively.
772 J. Jena et al.

Fig. 4 a Electron and b hole mobility profile of the BQP model in the XZ direction from source
to drain
Modeling and Performance Analysis of n-FinFETs … 773

Fig. 5 Total current density profile of the different models in the XZ direction

Fig. 6 Transfer 1E-4 60

characteristics with different @ Vd=0.05V
models EB 50
Id (µA) Linear Scale

DD
Id (A) Log scale

1E-5 EB+BQP
DD+BQP 40

30
1E-6
20

1E-7 10

0
0.0 0.2 0.4 0.6 0.8 1.0
Gate Voltage(V)

3.1 RF Characterization

Like in planar MOSFETs, the fringing capacitance between the gate electrode and the
source-drain electrode is large, which is one of the drawbacks of multigate FinFETs.
774 J. Jena et al.

Fig. 7 Output
characteristics with different 300 EB Vg=1V
models DD
250 EB+BQP
DD+BQP
200

Id ( µA)
150

100
Gate length 30 nm
50 Fin Width 30 nm
S/D Doping 1x10^20 /cm3
0 S-S Doping 1x10^15 /cm3

0.0 0.2 0.4 0.6 0.8 1.0

Drain Voltage(V)

As the fringing capacitance consists of in the outer and in the inner with respect
to the effective gate area. The outer fringing gate mainly refers to the capacitance
between gate and source/drain extension, gate, and source/drain while the inner fring-
ing capacitance is the fringing capacitance inside the physical gate. Liu et al., have
reported the cut-off frequency of 14.8 GHz of the nonplanar nanowire MOSFETs
[15].
FinFET may not compensate its loss in the RF performance still for the better
advantage of electrostatic gate control of the device could compromise the processing
issues. The unilateral gain has been calculated for different models and is shown in
Fig. 8 from which we got a maximum transition frequency (f t ) of ~15 to 20 GHz.
From the frequency response curve, we observed CVT and CVT + BQP having the

Fig. 8 Comparisons of the 50

frequency response curve of BQP+HCTE
the device using different 40 CVT+BQP
Unilateral Power gain (dB)

models CVT
HCTE
30

20

10

-10 ft ~ 15 to 20 GHz
-20
8 9 10 11
10 10 10 10
Frequency (Hz)
Modeling and Performance Analysis of n-FinFETs … 775

same response as like HCTE and HCTE + BQP. The main reason is that there is no
effect of quantum correction model (BQP) on the cut-off frequency calculation.

4 Conclusions

In this work, TCAD simulation of tri-gate FinFETs has been studied with different
transport models and its RF performance has also been analyzed. The EB model
shows an effective result compared to other ballistic carrier transports which signif-
icantly increases the drain current. However, when the effect of quantum confine-
ment is taken into account, the drain current decreases significantly. The frequency
response has been plotted and the corner or cut-off frequency (f t ) has been cal-
culated to be around ~15 to 20 GHz. Hence, an appropriate 3D TCAD modeling
approach should be taken to compensate for the new FinFET-specific design and
process challenges for future technology node devices.

References

1. Dash, T.P., Dey, S., Das, S., Jena, J., Mohapatra, E., Maiti, C.K.: Performance Comparison of
strained-SiGe and bulk-Si channel FinFETs at 7 N technology node. J. Micromech. Microeng.
29, 104001 (2019)
2. Jena, J., Dash, T.P., Mohapatra, E., Dey, S., Das, S., Maiti, C.K.: Fin shape dependence of
electrostatics and variability in FinFETs. J. Electron. Mater. 48, 6742–6752 (2019)
3. Dash, T.P., Dey, S., Das, S., Jena, J., Mohapatra, E., Maiti, C.K.: Source/drain stressor design
for advanced devices at 7 nm technology nodes. Nanosci. Nanotech. Asia (2019)
4. Choi, Y.-K., Chang, L., Ranade, P., Lee, J.S., Ha, D., Balasubramanian, S., Agarwal, A., Ameen,
M., King, T.J., Bokor, J.: FinFET process refinements for improved mobility and gate work
function engineering. In: Proceedings of IEDM, pp. 259–262 (2002)
5. Huajie, Z., Yi, S., Qiuxia, X., Yongliang, L., Huaxiang, Y.: Fabrication of Bulk-Si FinFET
using CMOS compatible process. Microelectron. Eng. 94, 26–28 (2012)
6. SilvacoInc, Atlas User’s Manual (2016)
7. Doria, R.T., et al.: Junctionless multiple-gate transistors for analog applications. IEEE Trans.
Electron Devices 58(8), 2511–2519 (2011)
8. Subramanian, V., et al.: Planar bulk MOSFETs versus FinFETs: an analog/RF perspective.
IEEE Trans. Electron Devices 53(12), 3071–3079 (2006)
9. Chen, B., et al.: Analog and RF characteristics of power FinFET transistors With different
drain-extension designs. IEEE Trans. Electron Devices 65(10), 4225–4231 (2018)
10. Maiti, C.K.: Introducing Technology Computer-Aided Design (TCAD)—Fundamentals,
Simulations, and Applications. Pan Stanford Publishing Pte. Ltd., Singapore (2017)
11. Baumgartner, O., Stanojevic, Z., Schnass, K., Karner, M., Kosina, H.: VSP—a quantum-
electronic simulation framework. J. Comput. Electron. 12, 701–721 (2013)
12. Stanojevic, Z., Karner, M., Kosina, H.: Exploring the design space of non-planar channels:
shape, orientation and strain. In Proceedings of IEDM, pp. 332–335 (2013)
13. Stanojevic, Z., et al.: Physical modeling—a new paradigm in device simulation. In: Proceedings
of IEDM, pp. 5.1.1–5.1.4 (2015)
776 J. Jena et al.

14. Gerrer, L., Georgiev, V., Amoroso, S.M., Towie, E., Asenov, A.: Comparison of Si <100>
and <110> crystal orientation nanowire transistor reliability using Poisson-Schrödinger and
classical simulations. Microelectron. Reliab. 55, 1307–1312 (2015)
15. Liu, L., Han, Q., Makovejev, S., Trellenkamp, S., Raskin, J-P., Mantl, S., Zhao, Q-T.: Analog and
RF analysis of gate all around silicon nanowire MOSFETs, In: Joint International EUROSOI
Workshop and International Conference on Ultimate Integration on Silicon (EUROSOI-ULIS),
Athens, pp. 176–179 (2017)
DOA Estimation on Fractal-Based Array

P. Raiguru and R. K. Mishra

Abstract The paper presents a Fractal-based array geometry for the Direction of
Arrival estimation performance. In general, the Uniform Linear Array (ULA) with
large numbers of uniformly placed physical antennas gives accuracy in the estimation
of DOA. However, it suffers highly by the Mutual Coupling (MC) between the
physical elements. The steering vectors are affected which affects the DOA estimation
performance. So The Fractal Array (FA) geometry arrangement followed by the
Cantor set is used in this paper. It gives a reduced Mutual Coupling effect between
the antenna elements as compared to ULA. The performance of DOA estimation
is analyzed using model-based approach. But this paper analyzes the performance
using the subspace method. The resolution of closed sources is analyzed in both ideal
and non-ideal conditions using the subspace method.

Keywords Direction of arrival · Fractal array · Mutual coupling

1 Introduction

The Direction of Arrival (DOA) estimation problem has been widely studied in mul-
tiple areas with a very diverse application such as target tracking [1], communication
[2], remote sensing [3], biomedical [4], astronomy [5], etc. The performance of the
system is can be enhanced by improving many factors presented in the literature.
In particular, multiple antennas offer several advantages. Since the antenna arrays
generate the beams with a maximum towards the desire users and nulls with different
directions of interferences, they play an important role in improving the performance

P. Raiguru (B)
Department of Electronics and Communication Engineering, Institute of Technical Education and
Research, Siksha ‘O’ Anusandhan (Deemed to be University), Bhubaneswar, Odisha 751030,
India
e-mail: [email protected]
R. K. Mishra
Department of Electronics Science and Technology, Berhampur University, Berhampur, India
e-mail: [email protected]

© Springer Nature Singapore Pte Ltd. 2020 777

G. Pradhan et al. (eds.), Advances in Electrical Control and Signal Systems,
Lecture Notes in Electrical Engineering 665,
https://doi.org/10.1007/978-981-15-5262-5_58
778 P. Raiguru and R. K. Mishra

of both base stations and mobiles. However, practically the performance is affected
by the mutual coupling between the elements present in the array. In general K ele-
ment Uniform Linear Array (ULA) can estimate (K − 1) the number of signal’s
DOA in the absence of mutual coupling. But the performance is degraded as an array
manifold of ULA is highly suffers from the mutual coupling [6]. This limitation can
be solved by maintaining the element positions in such a way that will not affect the
performance of DOA estimation.
The fractal having fractional dimension is generated recursively. There are num-
bers of fractal arrays are available in literature in the field of fractal antenna engineer-
ing which emphasizes the theory and design of fractal arrays. These fractal arrays
are generally found from the generating sub-arrays. The Fractal-based array, i.e.,
Triadic Cantor Linear Array is a compact and multi-band antenna element design
technique [9, 10]. Its theory is applied to design antenna array, radiation pattern
computation and adaptive beam-forming scheme [11, 12]. In general, the subspace
methods are the high-resolution method such as MUSIC and ESPRIT [7, 8] which
used for DOA estimation as it gives an accurate estimation as compared to other
conventional methods. In this paper, we present the Fractal-based array arrangement
to analyze the performance of DOA estimation using the sub-spaces method. The
Mutual coupling effect of Fractal Array (FA) is analyzed. The resolution between
closed spaced signals is compared under both ideal and non-ideal conditions.

2 Signal Model with Mutual Coupling Matrix

The K element Cantor Linear Fractal Array receives the L numbers of narrowband
uncorrelated signal sources from directions θ. The received signal [12] in matrix
form is given by

x = CAs + n (1)

where C is the K × K mutual coupling matrix, s is a signal vector and n is an AWGN

noise vector and A is the array manifold given by

A = [a(θ 1 ), a(θ 2 ), . . . , a(θ k )] (2)

where a(θ ) is the steering vector. The mutual coupling matrix C can be obtained as
follows:

c|b1 −b2 | , if |b1 − b2 | ≤ B,
Cb1,b2 = (3)
0, otherwise,

where c0 is the self-coupling coefficient and its value is 1 for each existing antenna
element. Here c1 is the nearest mutual coupling coefficient and it can be expressed
DOA Estimation on Fractal-Based Array 779

as c1 = ce jπ/3 where c is the value varies in between 0.1 and 1 [6]. Other coupling
coefficients can be found from it with the equation cg = c1 e− j (g−1)/8 /g, 2 ≤ g ≤ B.
It is noted here that if the MC effect is not considered than C becomes the Identity
matrix. The Co-variance matrix of received signals [5] is given by

R = xH x (4)

The Eigen analysis of R yields the Eigenvectors and Eigenvalues. Eigenvalues are
sorted in ascending order and the Eigenvectors U corresponding to (K − L) numbers
of lowest Eigenvalues are found. The MUSIC spectrum can be estimated by

1
J (θ ) =   (5)
UH a(θ )2

3 Fractal Array

The Fractal-based array geometry can be arranged by following the Cantor set [9,
11]. It is formed by a generating sub-array for the first stage is ‘1 0 1’, each position
is maintaining one-quarter of wavelength. Figure 1 shows the schematic diagram of
the first stage of this generating array. Here dark circle shows the existence of the
antenna element and the grey circle shows the absence of the element in that position.
The new array geometry can be found from only the existence of antenna elements
as multiple of half of the wavelength. It means the inter-element spacing between
two existing elements is one-half wavelength and we consider the element position
as {0, 1}. The second stage is formed by considering 1 as 101 and 0 as 000 and it
becomes 101000101. We consider the existing elements only, i.e., 1 and we maintain
the spacing as one-half wavelength and the element position becomes {0, 1, 3, 4}.
Similarly, numbers of element positions can be recursively found. The array factor
(A) of Fractal Array is expressed as Non-uniform Linear Array.

Fig. 1 The generating sub-array geometry

780 P. Raiguru and R. K. Mishra

4 Result Analysis

The performance of DOA estimation is examined by using Fractal Array geometry.

In the numerical examples, some fixed parameter is SNR = 0 dB and K = 8, with
the element position [0 1 3 4 9 10 12 13]. In the following examples, the MUSIC
algorithm is used to estimate DOA using MAT LAB. The performance is analyzed
by comparing the results of ideal and non-ideal conditions.

4.1 FA Radiation Pattern

The radiation pattern of the Normalized Array factor of Triadic Cantor array is plotted
in Fig. 2. It shows that as the number of stages increases the array factor goes on
sharper. The radiation pattern depends on the expansion factor which will decide the
aperture length of the array and hence the performance.

1 1
Array Facter
Array Facter

0.5 0

0 -1
0 50 100 150 200 0 50 100 150 200

Angle in degrees Angle in degrees

(a) (b)
1 1
Array Facter
Array Facter

0 0

-1 -1
0 50 100 150 200 0 50 100 150 200

Angle in degrees Angle in degrees

(c) (d)

Fig. 2 The radiation pattern of triadic cantor array a 1st stage, b 2nd stage, c 3rd stage, d 4th stage
DOA Estimation on Fractal-Based Array 781

MUSIC Spectrum -10

-20

-30

-40

-50
0 10 20 30 40 50 60
Angle in degree

Fig. 3 DOA estimation of eight elements FA without MC

4.2 MUSIC Spectrum Analysis in the Absence of MC

The eight elements Fractal Array can estimate the maximum of seven signal sources
ideally. So we observe the performance by considering seven signal sources between
5° and 55° with 8° differences. Here we consider the third stage and expansion factor
as 3. Figure 3 shows it can accurately estimate the signal’s DOAs in the absence of
Mutual Coupling or ideal condition.

4.3 Mutual Coupling Matrix (MCM) and Its Effect

In this section, the MC model [6] of eight elements FA is compared with eight
elements ULA. Figure 4 shows the strength of the magnitude of the mutual coupling

Fig. 4 Magnitude of mutual coupling matrix a 8 elements ULA, b 8 elements FA

782 P. Raiguru and R. K. Mishra

MUSIC Spectrum -10

-20

-30

-40

-50
0 10 20 30 40 50 60
Angles in degree

Fig. 5 DOA estimation of eight elements FA with MC

matrix. Here the magnitude of the coupling coefficient is 0.3 is considered. The
diagonal of the matrix which is self-coupling is visualized as a yellow colour. As the
mutual coupling strength between the elements is less the colour gradually becomes
blue. It indicates that the MC effect of FA is less as compared to ULA as shown in the
figure. Here DOA estimation performance is analyzed with the MC effect in Fig. 5.
It shows that due to the mutual coupling effect only six peaks are estimated in place
of seven and the amplitude of power spectrums is also affected. The performance of
FA is affected in the presence of Mutual Coupling.

4.4 Resolution

In this section, the resolution of two closely spaced signal sources is compared for
both ideal and non-ideal conditions of FA. Two signal sources impinge from 5° and
7°. Figure 6 shows that without MC, the FA can detect the two sources with 2°
minimum angle separation. But the two signal peaks are merged into one peak when

0
MUSIC Spectrum

-10

-20

-30

-40

-50
0 5 10 15 20
Angles in degree

Fig. 6 Resolution of eight elements fractal array without MC

DOA Estimation on Fractal-Based Array 783

MUSIC Spectrum -10

-20

-30

-40

-50
0 5 10 15 20
Angles in degree

Fig. 7 Resolution of eight elements fractal array with MC

the MC effect is considered as shown in Fig. 7. It shows that the resolution of FA is

affected in non-ideal conditions.

5 Conclusion

The DOA estimation performance of FA is analyzed with and without the effect of
MC. The results show that in the ideal condition the performance is accurate. But
practically the effect of mutual coupling cannot be neglected. So the performance
is examined in both ideal and non-ideality conditions. The performance DOA esti-
mation using FA is degraded in a non-ideal condition. However, FA has reduced the
MC effect as compared to ULA.

References

1. Orton, M., Fitzgerald, W.: A Bayesian approach to tracking multiple targets using sensor arrays
and particle filters. IEEE Trans. Signal Process. 47(10), 2644–2654 (1999)
2. Godara, L.C.: Application of antenna arrays to mobile communications. II. Beam-forming and
direction-of-arrival considerations. Proc. IEEE 85(8), 1195–1245 (1997)
3. Nielsen, U., Yan, J.B., Gogineni, S., Dall, J.: Direction-of-arrival analysis of Airborne ice depth
sounder data. IEEE Trans. Geosci. Remote Sens. 55(4), 2239–2249 (2017)
4. Dey, N., Ashour, A.S., Shi, F., Sherratt, R.S.: Wireless capsule gastrointestinal endoscopy:
Direction-of-arrival estimation based localization survey. IEEE Rev. Biomed. Eng. 10, 2–11
(2017)
5. Ogawa, H., Mizuno, A., Hoko, H., Ishikawa, H. and Fukui, Y., Ogawa, H., et al.: A 110 GHz
SIS receiver for radio astronomy. Int. J. Infrared Millimeter Waves, 11(6), 717–726 (1990)
6. Basikolo, T., Ichige, K. Arai, H.: A novel mutual coupling compensation method for underde-
termined direction of arrival estimation in nested sparse circular arrays. IEEE Trans. Antenna
Propag. 69909–69917 (2018)
784 P. Raiguru and R. K. Mishra

7. Schmidt, R.: Multiple emitter location and signal parameter estimation. IEEE Trans. Antennas
Propag. 34(3), 276–280 (1986)
8. Roy, R., Kailath, T.: ESPRIT-estimation of signal parameters via rotational invariance
techniques. IEEE Trans. Acoustics Speech Signal Process. 37(7), 984–995 (1989)
9. Werner, D.H., Haupt, R.L., Werner, P.L.: Fractal antenna Engineering: The theory and design
of fractal antenna arrays. IEEE Trans. Antenna Propag. 41, 37–59 (1999)
10. Karmakar, A., Ghatak, R., Mishra, R.K., Poddar, D.R.: Sierpinski carpet fractal-based planar
array optimization based on differential evolution algorithm J. Electromagn. Waves Appl.
247–260 (2015)
11. El-Khamy, S.E., Eltrass, A.S., El-Sayed, H.F.: Adaptive beamforming synthesis for thinned
fractal antenna arrays. Int. Con. URSI GASS, pp. 1–4 (2017)
12. Raiguru, P., Mishra, R.K.: DOA estimation considering mutual coupling with fractal based
array using model-based approach. In: Proceedings of ICCCE, pp. 585–590 (2019)
Output Voltage Regulated CUK
and SEPIC Converter with High Input
Power Factor

Alok Kumar Mishra, Akshaya Kumar Patra, Ramachandra Agrawal,

Nabajyoti Swain, Debadutta Dash, Shahil Sharma, and Pranav Mohapatra

Abstract This paper presents a comparative analysis between the two converter
topologies namely SEPIC and CUK converter used for power factor correction. MAT-
LAB/SIMULINK models of SEPIC and CUK converter are developed to improve
the input power factor and output voltage regulation. Two different control tech-
niques, namely (Average current control technique and Hysteresis current control
technique) are used for input power factor correction and to get a regulated voltage
at the output PI or fuzzy logic controller (FLC) is adopted. The system is tested at
both steady-state, transient condition and its performance is then estimated and com-
pared in terms of various parameters like Total Harmonic Distortion (THD), input
power factor, output voltage ripple for PI and FLC in both Average and Hysteresis
current control method.

A. K. Mishra (B) · A. K. Patra · N. Swain · D. Dash · S. Sharma · P. Mohapatra

Department of Electrical and Electronics Engineering, Institute of Technical Education and
Research, Siksha ‘O’ Anusandhan (Deemed to be University), Bhubaneswar, Odisha 751030,
India
e-mail: [email protected]
A. K. Patra
e-mail: [email protected]
N. Swain
e-mail: [email protected]
D. Dash
e-mail: [email protected]
S. Sharma
e-mail: [email protected]
P. Mohapatra
e-mail: [email protected]
R. Agrawal
Department of EE, Institute of Technical Education and Research, Siksha ‘O’ Anusandhan
(Deemed to be University), Bhubaneswar, Odisha 751030, India
e-mail: [email protected]

© Springer Nature Singapore Pte Ltd. 2020 785

G. Pradhan et al. (eds.), Advances in Electrical Control and Signal Systems,
Lecture Notes in Electrical Engineering 665,
https://doi.org/10.1007/978-981-15-5262-5_59
786 A. K. Mishra et al.

Keywords PID · FLC · SEPIC converter · CUK converter · Average current

control · Power factor correction circuits

1 Introduction

According to the international standards and for the best power transfer and utiliza-
tion, power factor correction has become a necessity. To get a dc output voltage, we
use a rectifier and a parallel capacitor [1, 2] as shown in Fig. 1a. The input voltage
and current are shown in Fig. 1b. Figure 1b depicts a pulse shaped input current
with much harmonics content causing a very poor power factor. The IEEE and IEC
are some international entities to define or standardize the permissible limits of the
harmonic content inline currents such as IEEE 519 and IEC 61000-3-2 [3]. The
design of an AC to DC power converter to overcome these power quality issues like
obtaining UPF at the AC input mains and close regulation of the DC output volt-
age has been discussed in [4–9]. So, for small power uses in single-phase supplies,
DC-DC converters are used which are switch-based and controlled accordingly to
ensure HPF at the main side. The main motive is to emulate a resistive circuit when
seen from main side and improve power factor. Theoretically, there exist three fam-
ilies of non-isolated power factor correction topologies which are buck, boost and
buck-boost topology [1, 10–13]. The buck topology is generally used when low out-
put voltage is required but has high-frequency commuted current at input. Due to
commuted nature, it exhibits a discontinuous nature, and hence should be connected
with high-speed recovery circuit. The major drawback faced by buck topology is the
inclusion of a filter of high-frequency range within the diode bridge and the power
source [14]. The filtering inductor causes a soft variable current in the input end in
the topology of a boost converter. This topology provides high output voltage and
leads to over-voltage stress at the switches [5, 15]. The third family is buck-boost
topology and finds its applications in power factor correction circuits. It includes
CUK and single-ended primary inductance converter (SEPIC) converter [16, 17].
The output voltage polarity makes these two converters different from each other.
In case of CUK converter, the position of free-wheeling diode and the inductor is
reversed from that of the SEPIC converter to obtain reversed output voltage polarity
[18, 19]. This paper presents two different topologies (namely SEPIC and CUK)

is
i s ,Vs
L
AC O
Vs A
Source Co
D

Diode Bridge
(a) (b)

Fig. 1 a AC-DC rectifier. b Source voltage and current waveform

Output Voltage Regulated CUK and SEPIC Converter with High … 787

with two different current control methods (namely Hysteresis and Average) type
PFC circuit to improve the input side power factor of the diode bridge rectifier, lower
the THD factor with decreased ripple contained voltage at output [1, 20, 21]. The
entire modelling and the design scheme of pf rectified AC to DC power converters
are carried out in the MATLAB/SIMULINK environment.

2 Circuit Configuration and Design Equation

To improve the power factor at the input and to get a regulated voltage at the output
we have to interface a DC to DC (SEPIC or CUK) converter in between the bridge
rectifier and the load as shown in Fig. 2 (SEPIC) Fig. 3 (CUK) respectively. By
adopting these topologies a low current ripple can found at input for a low level DC-
bus voltage, which is rare in case of conventional buck/boost converter. With adequate
coupling coefficient design in a CUK or SEPIC converter, the same magnetic core
can accommodate the output and input inductors. In both the converter when switch
is on inductor L 1 and L 2 stores energy, inductor current increases linearly and output
capacitor supply power to the load and diode is reverse biased. When switch is off
diode gets forward biased and stored energy of the inductor is supplied to the load.
The different parts of all the converter systems are modelled using basic equations
and all equations are summarized in Table 1. where vs is the rms source voltage, d
is the duty cycle, i L represents inductor current ripple, vc , vo are the voltage

is L1 C1 D
Io

SW L2 Vo
Vs Co RL

SEPIC Converter
Diode Bridge

Fig. 2 SEPIC converter topology for power factor correction

is L1 C1
Io
L2

Vs SW D Co RL Vo

CUK Converter
Diode Bridge

Fig. 3 CUK converter topology for power factor correction

788 A. K. Mishra et al.

Table 1 Designed equation

Parameters SEPIC CUK
for SEPIC and CUK converter
vs ∗d vs ∗d
L1 i L 1 ∗ f s i L 1 ∗ f s
vs ∗d vs ∗d
L2 2∗i L 2 ∗ f s i L 2 ∗ f s
C1 i L 2 ∗d I0 ∗(1−d)
vc1 ∗ f s vc1 ∗ f s
Po vs ∗d
Co 4π ∗ f s ∗Vo ∗vo
8∗ f s2 ∗L 2 ∗vo

ripple of the output voltage and capacitor voltage, respectively. f s is the switching
frequency. P0 is the output power, Vo is the output voltage, I0 is the output current.

3 Different Methods of Power Factor Correction

There are two objectives of power factor correction: (1) To get a regulated voltage at
the output. (2) The wave shape of current at the input must be a sine wave. To achieve
the first objective we have to use a feedback loop at the output. There are two methods
to achieve the second objective. First method is called “Multiplier Approach” and the
second is called “Voltage follower Approach”. In Multiplier Approach, a feedback
loop of input current is used to govern the DC to DC power converter to function as an
input voltage programmed current sink. In this paper Multiplier Approach (namely
Hysteresis and Average) current control technique is, used. Multiplier Approach
Control is further sub-divided into four different methods for generating the gate
pulse for the SEPIC and CUK converter, which areas follows:
1. Hysteresis current control
2. Peak current control
3. Average current control
4. Borderline current control.

3.1 Average Current Control Method

In Average current control method, the inductor current is sensed and filtered by a
current error amplifier whose output drives a PWM modulator as depicted in Fig. 4.
In this technique, the sensed inductor current is to be filtered with the help of a current
error amplifier and the output of the same drives the PWM modulator. Hence the error
between input current I g and its reference gets minimized with the help of inner
current loop. The reference current is generated by a voltage error amplifier (PI or
Fuzzy Logic Controller). Due to the PWM modulator Average current control method
gives a constant switching frequency because of current filtering commutation noises
Output Voltage Regulated CUK and SEPIC Converter with High … 789

is I actual Io
DC-DC
Converter

Vs Vin R L Vo

SW

Diode Bridge
1
K +
Gate
Sinusoidal -
Pulse
Reference
PWM
- Modulator
+

X I ref
Current error Amplifier
PI Controller
-
Multiplier +
V o,ref

Voltage error Amplifier

PI or Fuzzy Logic Controller

Fig. 4 Average current control technique

get eliminated. This method does not need any compensation ramp. Inductor current
must have to sense which is the demerits of this method.

3.2 Hysteresis Current Control Method

Comparing with the above four methods, Hysteresis current control provides very
low distorted waveforms of input current and does not need any ramp compensation.
In Fig. 5, two sinusoidal current references I P,ref and I V,ref are generated from Iref ,
which shows the current peak and valley of the inductor, respectively. In this control
technique, if the current of inductor falls below the lower reference IV,ref then the
switch is ON and if the current of inductor rises above the upper reference I P,ref switch
is OFF. This control technique provides a variable frequency control structure which
is the demerits of this method.
790 A. K. Mishra et al.

is
Io
DC-DC
Converter

Vs Vin RL Vo

SW

Diode Bridge
I actual
Hysteresis Gate
1 Control Pulse
I ref
K
PI or

Sinusoidal X Fuzzy Logic

Controller +
- Vo
Reference Multiplier
V o,ref

Fig. 5 Hysteresis current control technique

4 Controller Used for Output Voltage Regulation

In this paper, two different controllers are used to regulate the output voltage namely
PID Controller and Fuzzy Logic Controller and the performance of the same is
compared.

4.1 PID Controller

The name itself Proportional-Integral-Derivative controller is a type of control loop

mechanism adopted in continuous modulated operation of control. A PID controller
repeatedly calculates the error in between a set point(SP) value and measured process
value(PV) and gets a correction accordingly in PID base. In real-time operation, it
accordingly provides an accurate and respective change of correction to a control
function. For example furnace temperature control that it applies a derivative term
to effectively correct the error despite a huge change.
Output Voltage Regulated CUK and SEPIC Converter with High … 791

Vo - LPF
+
Rule Base
Integrator Imax,ref
Vo,ref
e(n) ce(n)

Fuzzification Rule Evaluator Defuzzification

Data Base

Fig. 6 Basic block diagram of FLC

4.2 Fuzzy Logic Controller Controller

The basic block diagram of the proposed fuzzy logic controller for output voltage
regulation of SEPIC and CUK converter is shown in Fig. 6. There are two separate
meanings of fuzzy logic. Generally, fuzzy logic is the further application of multi-
valued logic and is also known as logic system. We can also say fuzzy logic is the
same as fuzzy sets theory, which relates to the collection of objects with unsharp
boundaries and the membership is a point of degree. There is a specific object which
degree of membership in a given set which can vary between the range 0 and 1
in fuzzy set theory. Fuzzy logic is based on sound quantitative and also deals with
imprecise information and data. Mathematical theory, the values of fuzzy variables
are expressed by proper English language. Error in voltage of a power converter can
be defined in linguistic variables like Negative Big (NB), Negative Medium (NM),
Negative Small (NS), Zero (ZE), Positive Small (PS), Positive Medium (PM), Pos-
itive Big (PB), and each variable can be defined by varying triangular membership
function. Seven fuzzy levels were chosen and were defined by fuzzy set library which
value of the error is e and change in error is de. The larger the number of fuzzy levels,
the higher the input resolution. We know that a rule is n-dimensional and n is the
number of variable included in the rule. The sum of rules is known as rule R. FIS
editor edits the input and output variables, which are e, de and output. After editing
we have to design the membership function for each variable. The final step involves
writing rules in rule editor using the rule given in Table 2.
792 A. K. Mishra et al.

Table 2 Fuzzy rules for closed-loop control of CUK or SEPIC converter

e\de NB NM NS ZE PS PM PB
NB NB NB NB NB NM NS ZE
NM NB NB NB NM NS ZE PS
NS NB NB NM NS ZE PS PM
ZE NB NM NS ZE PS PM PB
PS NM NS ZE PS PM PB PB
PM NS ZE PS PM PB PB PB
PB ZE PS PM PM PB PB PB

5 Results and Discussion

To investigate the performance of the proposed system, a simulink model of a CUK

and SEPIC converter used for power factor correction is developed. Out of four
simulink models, two of them are given in Figs. 7 and 8. The designed parameter
used in the simulation for both SEPIC and CUK converter is given in Table 3.
The system performance is investigated in six steps. Step: 1 When none of these
converters are not connected to the system. Step: 2 When CUK and SEPIC converter
is connected to the system with inner current loop only and without the output voltage
feedback loop. Step: 3 Average current-controlled CUK converter is connected to the
system with Fuzzy Logic Controller in the feedback loop. Step: 4 Hysteresis current-
controlled CUK converter with Fuzzy Logic Controller in the feedback loop. Step:
5 Average current-controlled SEPIC converter with Fuzzy Logic Controller in the
feedback loop. Step: 6 Hysteresis current-controlled SEPIC converter with Fuzzy
Logic Controller in the feedback loop. To convert AC to DC we generally connect a

Fig. 7 Simulink model of average current-controlled CUK converter

Output Voltage Regulated CUK and SEPIC Converter with High … 793

Fig. 8 Simulink model of hysteresis current-controlled SEPIC converter

Table 3 Designed parameter of the system

Parameters Values
Input voltage RMS value 120 V
Supply frequency 50 Hz
L 1, L 2 1 mH/6 mH, 10 mH
C1 , Co 10 mF, 10 mF
Load power 1 kW
Reference output voltage 100 V
Switching frequency 20 kHz/40 kHz
Voltage and current ripple 0.05
Duty cycle 0.45

bridge rectifier and filter capacitor across the load as shown in Fig. 1, the simulation
result of the same along with its harmonic spectrum is shown in Fig. 9. To improve
the performance of the input current and output voltage, CUK or SEPIC converter is
used as discussed in Figs. 2 and 3 with both the control strategy as shown in Figs. 4
and 5. Various waveforms at steady-state in open loop (average current-controlled)
are shown in Fig. 10 (Fig. 10a–c for SEPIC and Fig. 10d–f for CUK). From (Fig. 10 b
or e) it is clear that when the load is increased at t = 1 s, load voltage is not a regulated
one. To get a regulated output voltage and improved input current wave PI or Fuzzy
794 A. K. Mishra et al.

Fig. 9 Simulation results of the proposed system without any converter

Logic Controller is used in the output feedback loop, all the result obtained with PI
or Fuzzy Logic Controller is given in Table 4 for comparison but the results given in
Figs. 11, 12, 13 and 14 are with Fuzzy Logic Controller as the same performs better
than PI. To test the validity of the system with PI and Fuzzy Logic Controller a load
change and reference voltage change is created at t = 1 s and at t = 2 s, respectively.
With this dynamic change, the regulated output voltage is obtained which can be
clearly seen from Fig. 11e, 12e, 13e and 14e.
Output Voltage Regulated CUK and SEPIC Converter with High … 795

Fig. 10 Simulation results of the system in open loop [(a–c) for SEPIC and (d–f) for CUK]

Table 4 Performance parameters of the system under different condition

Performance parameter THD % Power factor Output voltage ripple %
Different condition
Without any converter 191.36 0.4479 9.2
Open loop SEPIC-ACC 4.37 0.9986 2.0
Open loop SEPIC-HCC 2.59 0.9992 3.1
Open loop CUK-ACC 4.70 0.9986 2.2
Open loop CUK-HCC 2.55 0.9995 1.8
SEPIC-ACC-PI 4.17 0.9983 2.1
SEPIC-HCC-PI 1.31 0.9995 6.0
CUK-ACC-PI 3.13 0.9990 2.1
CUK-HCC-PI 1.44 0.9997 2.2
SEPIC-ACC-FLC 4.16 0.9997 1.5
SEPIC-HCC-FLC 1.22 0.9998 1.5
CUK-ACC-FLC 2.97 0.9999 1.8
CUK-HCC-FLC 1.36 0.9997 1.5
796 A. K. Mishra et al.

Fig. 11 Simulation results of hysteresis current-controlled CUK converter with FLC

6 Conclusion

The design, modelling and simulation of CUK and SEPIC converter for power factor
correction purposes considered in MATLAB/Simulink environment. The outcomes
of the simulation work provided low THD less than 5% of the supply current with
upgraded AC mains pf and reduced output voltage ripple. Comparing with PI and
Fuzzy Logic Controller used in the output feedback loop, Fuzzy Logic Controller
gives better results in terms of THD of supply current, input power factor and output
voltage ripple which is given in Table 4. From Figs. 11, 12, 13 and 14 and Table 4,
it is clear that fuzzy logic Controlled SEPIC converter gives better results when
load disturbance and reference change occur. The prototype of the proposed SEPIC
converter can be developed, which would be attempted as future work.
Output Voltage Regulated CUK and SEPIC Converter with High … 797

Fig. 12 Simulation results of average current-controlled CUK converter with FLC

Fig. 13 Simulation results of hysteresis current-controlled SEPIC converter with FLC

798 A. K. Mishra et al.

Fig. 14 Simulation results of average current-controlled SEPIC converter with FLC

References

1. Mishra, A.K., Pathak, M.K., Das, S.: Isolated converter topologies for power factor correction—
a comparison. In: IEEE International Conference on Energy, Automation and Signal, pp. 1–6,
28 Dec 2011
2. Kuiyuan, W.: The comparison and choice of several power factor correction methods. IEEE
Vehicle Power and Propulsion Conference, pp. 1–5. IEEE (2006, September)
3. 519–2014: IEEE recommended practice and requirements for harmonic control in electric
power systems, pp. 1–29, 11 June 2014
4. Pressman, I.: Switching Power Supply Design. McGraw-Hill, New York (1991)
5. Patra, A.K., Rout, P.K.: Adaptive continuous-time model predictive controller for implantable
insulin delivery system in Type I diabetic patient. Optimal. Control Appl. Methods 38, 184–204
(2017)
6. Patra, A.K., et al.: Optimal H-infinity insulin injection control for blood glucose regulation in
IDDM patient using physiological model. Int. J. Autom. Control 8, 309–322 (2014)
7. Singh, B., et al.: A review of single-phase improved power quality AC-DC converters. IEEE
Trans. Ind. Electron. 50(5), 962–981 (2003)
8. Batarseh, I.: Power Electronics Circuits, 3rd ed. Wiley
9. Richard, R.: The fundamentals of power factor correction. Int. J. Elect. Eng. Educ. 31, 213–229
(1994)
10. Dah, D., et al.: Light-load efficiency improvement in buck-derived single-stage single-switch
PFC converters”. IEEE Trans. Power Electron. 28(5), 2105–2110 (2013)
Output Voltage Regulated CUK and SEPIC Converter with High … 799

11. Patra, A.K., Mishra, A.K., Rout, P.K.: Backstepping model predictive controller for blood
glucose regulation in Type-I Diabetes Patient. IETE J. Res. https://doi.org/10.1080/03772063.
2018.1493404
12. Patra, A.K., Rout, P.K.: Adaptive sliding mode Gaussian controller for artificial pancreas in
TIDM patient. J. Process Control 58, 23–27 (2017)
13. Liu, X., Xu, J., Chen, Z., et al.: Single-inductor dual-output buck–boost power factor correction
converter. IEEE Trans. Ind. Electron. 62(2), 943–952 (2015)
14. Kanaan, H.Y., Al-Haddad, K., Fnaiech, F.: Switching-function-based modeling and control
of a SEPIC power factor correction circuit operating in continuous and discontinuous current
modes. In: IEEE International Conference on Industrial Technology, vol. 1, No. 8, pp. 431–437
(2004, December)
15. Umamaheswari, M.G., Uma, G.: Analysis and design of reduced order linear quadratic regulator
control for three phase power factor correction using Cuk rectifiers. Electr, Power Syst. Res.
1(96), 1–8 (2013)
16. Umamaheswari, M.G., et al.: Comparison of hysteresis control and reduced order linear
quadratic regulator control for power factor correction using DC–DC Cuk converters. J. Circ.
Syst. Comput. 21(1), 1250002 (2012)
17. Umamaheswari, M.G., et al.: Analysis and design of digital predictive controller for PFC Cuk
converter. J. Comput. Electron. 13(1), 142–154 (2014)
18. Poorali, B., Adib, E.: Analysis of the integrated SEPIC-flyback converter as a single-stage
single-switch power-factor-correction LED driver. IEEE Trans. Ind. Electron. 63(6), 3562–
3569 (2016)
19. Melo, P.F., Gules, R., Ribeiro, E.F., et al.: A modified SEPIC converter for high-power-factor
rectifier and universal input voltage applications. IEEE Trans. Power Electron 25(2), 310–321
(2010)
20. Sudhakarababu, C., Veerachary, M.: Zeta converter for power factor correction and voltage
regulation. In: 2004 IEEE Region 10 Conference TENCON 2004, vol. 500, pp. 61–64. 24 Nov
2004. IEEE
21. Patra, A.K., Rout, P.K.: Backstepping sliding mode Gaussian insulin injection control for blood
glucose regulation in TIDM patient. J. Dyn. Sys., Meas., Control. 140(9), 091006–091006-15
(2018)
Energy Audit of Auxiliary Power
Consumption: A Case Study

Avijit Nayak, Lalit Mohan Satapathy , Niranjan Nayak,

and Sudhansusekhar Dash

Abstract In the present scenario, India faces a great challenge in availability of

electrical energy sources. Almost 50% of total energy generation is consumed by
the industries alone. Outdated technology is one of the major reasons for excess
energy consumption in Indian industries. Energy demand increases 6–8% annually.
But the energy generation does not change as per the same ratio. Thus, there is a
gap between the supply and demand; as a result, power balance mismatch occurs.
In order to minimize these gaps, two alternative methods can be followed, such
as electricity conservation and more power generation. The second method is not
cost effective. Even if 5% of energy saving may prevent the installation of plants in
MW. In this regard, “energy audit” is an effective tool for all industrial consumers
to identify conservation opportunities and methods of energy saving. Utilization
effectively along with reduction of wastage of electric energy is known as the energy
conservation. Efficient energy maintenance needs a basic tool called energy audit
in which the wastage of energy is identified without affecting the production. This
paper presents the energy audit in the auxiliary systems like fans and air conditioners
in a captive power plant of steel industry located in Odisha, India. The real-time
energy audit is conducted, and the results are depicted in the tables, including the
energy savings.

Keywords Power scenario · Energy audit · Power demand and generation ·

Auxiliary loads in industry

A. Nayak
National Productive Council of India, New Delhi, India
e-mail: [email protected]
L. M. Satapathy (B) · N. Nayak
Department of EEE, SOA Deemed to be University, Bhubaneswar, India
e-mail: [email protected]
N. Nayak
e-mail: [email protected]
S. Dash
Capital Institute of Management and Science, Bhubaneswar, India
e-mail: [email protected]

© Springer Nature Singapore Pte Ltd. 2020 801

G. Pradhan et al. (eds.), Advances in Electrical Control and Signal Systems,
Lecture Notes in Electrical Engineering 665,
https://doi.org/10.1007/978-981-15-5262-5_60
802 A. Nayak et al.

1 Introduction

The power consumption of a country plays vital role in its socioeconomic growth.
From 1950, Indian power system scenario has gone to a higher level. As of 2018,
total installed capacity of India is 360,456 MW. Country today faces a peak shortage
of power around 11–18% and an energy shortage of about 7–11% [1]. The electric
energy consumption through different sectors is as follows: Industrial load con-
sumption is 26.64%, and the power consumption of domestic and commercial load
is 30.78% approximately. The transport, agriculture, and other head consumption is
29.35%, 1.66%, and 11.57%, respectively. In the year of 2017, India has generated
1160.1 billion kWh of electricity with an increase of 4.72% than the previous year.
Various developments, like industrial growth and increase of per capital income
is the facts which indirectly lead the power demand. As per the report of India Brand
Equity Foundation (IBEF), an arm of the Indian government’s ministry of commerce,
the foreign investment in this sector reached $12.97 billion, accounting for 3.52%
of all FDI inflows into the country. Over the last five years, India put up 99.21 GW
of additional capacity. Of this, 91.73 GW came from thermal sources, 5.48 GW
from hydro, and 2 GW from nuclear sources. However, generation levels are still
insufficient to meet the growing demand which has outstripped supply by about
7.5%, the report said. Thus, India now plans to strike the $14.94 billion prospect
in the power transmission market, according to the IBEF. Over the last 17 years,
foreign direct investment (FDI) in the sector has reached $12.97 billion, accounting
for 3.52% of all FDI inflows into the country. However to overcome such scarcity,
the methods like “energy audit” must be adopted to identify the loss factors of the
power system and to adopt the methods how to improve the loss minimization and
revenues. Energy audit is broadly used, and many have different meanings depending
on energy service companies.
In reference [1], the authors have been adopted the energy audit and energy saving
methods of a medium-scale apparel industry in Sri Lanka. The energy spending and
reserves assessed in terms of equipment used and practical areas engaged. Investing
to recover the energy efficiency of a textile industry provides an instant and com-
paratively predictable cash flow resulting from lower energy bills. A large office
building in Riyadh City was considered and the variety of apparatus of the building
are analyzed such as specific actions of the walls, the roofs, the windows, the light-
ing fixtures, the office equipment, the chillers, HVAC, the additional air conditioning
units known as split types, and the other equipment in the building [2–5]. The high
power utilization of the building division is a topic discussed and presented at the
global level, and the prime energy consumed by buildings amounts about 40% and
the greenhouse gases emissions nearly 30%. The consciousness of the ecological,
financial, and social risks created by these emissions was suggested to replace by
the engine to improve the building performance [6]. The paper [7] presents detailed
study of the performance of renewable energy sources HVAC systems for buildings.
A method adopts a technology of the components like biomass boiler, aero ther-
mal, geothermal, and absorption heat pumps. In the literature, genuine effort has
Energy Audit of Auxiliary Power Consumption: A Case Study 803

been made to carry out energy audit at Nandi Institute of Technology and Man-
agement Sciences, Bangalore, India, to estimate daily, weekly, and monthly energy
consumption. Detection of energy wastage and inference of energy saving potential
in the canteen, all departments, and Institute Central Facilities have been made by
walk-through energy audit [8–10].
In this paper, the energy audit of the captive power plant of a steel plant, which is
situated in the Eastern India, is conducted which includes the consumption of electric
energy, energy wastage, loss due to improper maintenance at the auxiliaries used in
the plant, like boiler feed pump (BFP), condensate exciter pump, induced draft fans,
primary and secondary air fans, and HP blowers. The energy audit technique is
applied, and the result depicts the energy savings and cost benefits.
The rest part of the paper is organized as follows: The description of plant under
audit is presented along with the different auxiliaries in Sect. 2. The performance
analysis of the individual auxiliaries is calculated in Sect. 3. In Sect. 4, conclusion
is drawn.

2 The Plant Structure

The integrated steel plant is situated in Eastern India. It is installed with modern and
largest blast furnace. The plants have basic oxygen furnace (BOF), for steelmaking,
and DRI plant. The plant has its own captive power plant, bar mill, plate mill, etc.
The plant is completely captive dependent for meeting the electricity demand. In the
captive power plant apart from utilizing the coal as fuel, the blast furnace and coke
oven by-products are used for generating electricity.

2.1 Major Auxiliaries of Captive Power Plan

Here in this study, energy audit for five auxiliaries like boiler feed pump (BFP),
condensate extraction pump (CEP), induced draft fans (IDF), primary and secondary
air fans is conducted as per the following procedure.

2.2 General Procedure

The general approach adopted for energy audit of the unit auxiliaries of the system
is detailed below:
• Electrical measurements were conducted on all HT drives (BFP, CEP, ID Fan, PA
fan and SA fan, and HP blowers).
804 A. Nayak et al.

• Simultaneous observation/measurements of system equipment condition and per-

formance by measuring performance parameters through use of portable and panel
mounted instruments.
• The above measurements, together with data and information from logbooks
maintained by plant, were utilized in ascertaining the energy-related control
parameter, like “specific energy consumption (SEC).”
• Specific energy consumption figures were worked out for individual key equip-
ment, like BFP, CEP, ID fans, SA fans, PA fans, and blowers.
• Equipped with plant operating data, performance data of relevant equip-
ment/systems, areas of improvements, modifications, and retrofits were identified
for optimization of system loads—toward lowering of station APC.
• The other operating parameters were monitored at the available operating load
from panel after ensuring that the reference data is reasonably accurate from C&I
department, wherever possible the same was checked with the available local
gauges.

3 Performance Assessment of Unit Auxiliaries

3.1 Boiler Feed Pump (BFP)

Boiler feed water pump (BFP) is one of the most important components of the unit
and is the single largest contributor to the auxiliary power consumption. Generally,
one of the BFPs is in service during boiler operation (Table 1).

Observations
As the full load motor loading % was not available, the full load motor loading has
been assumed as 100% at pump full load operation. A calculation at full load was
exercised backward integration for value of expected combined efficiency for pump
operating at rated conditions.
The calculations are as follows:

Pump eff. at full load 81%

Pump in kW 2785.348
HC in kW (eff. 94.7%) 2941.233
Motor in kW 3032.199
Expected % motor loading 79%
Max expected combined eff. 74%
Energy Audit of Auxiliary Power Consumption: A Case Study 805

Table 1 Performance evaluation of BFP

S. No. Item ref Units Design BFP-1 BFP-2
1 Unit load MW 135 135 135
2 Feed water flow (booster pump) T/HR 468.5 450 413
3 Booster pump suc. pressure mWC 312.1 86 87
4 Booster pump dis. pressure mWC 123.4 159 159
5 Total dev head by booster pump mWC 80 73 72
6 Feed water flow (BFP) T/HR 440 430.0 391.7
7 BFP suc. pressure mWC 130 158 158
8 BFP dis. pressure mWC 1740 1384 1378
9 Total dev head by BFP mWC 1610 1226 1220
10 Suction temp. °C 166.4 165 165
11 Density Kg/m3 901 903 903
12 HC engagement % – 55 53
13 Liquid kW of booster pump kW 113.4 89.5 81.0
14 Liquid kW of BFP kW 2142.8 1665.8 1521.4
15 Total liquid kW kW 2256.1 1755.3 1602.4
16 BFP (motor input) kW 3814.4 2837.32 2810.89
17 Motor eff. % 97 96 96
18 Motor output power kW 3700 2724 2698.5
19 Combined efficiency % 59.1 61.9 57.0
20 BFP efficiency (curve) % 81.3 80 80
Note Combined efficiency = efficiency of (pump * motor * fluid coupling)
At a particular time, only two numbers of pumps are operational. The trials of three pumps were
taken at different point of time

The comparison of BFPs, against the expected combined efficiency, reflects

shortfall of 13–15%.

Expected combined eff. 74%

Combined eff. of 3B 62%
Combined eff. of 3B 57%

At the operational flow and head, the pump efficiency from the curve reflects a
value of 80%; operational HC engagement is found to be in range of 55%; however,
due to unavailability of HC losses at part load operation, the pump efficiency could
not be evaluated.
Therefore at this stage, it is inappropriate to comment whether the performance
degradation is due to pump performance degradation or loss in HC. It is strongly
recommended to check the hydro coupling performance at part load condition as
well as to check the pump internals for efficiency improvement.
806 A. Nayak et al.

While pump loading in respect of head and flow is found to be satisfactory, the
performance of pump 3B is found to be marginally in higher side than 5B. Moreover,
similar behavior in terms of efficiency level is being observed for both the units, and
it is unlikely that both the cartridge conditions have deteriorated. Therefore, the HC
performance may be checked as per the design characteristics and if required OEM
may be consulted.
While no recirculation valves passing was observed, the effort of the JSPL officials
for excellent housekeeping is commendable. Periodic as run efficiency assessment
along similar lines of NPC methodology is recommended (at monthly intervals) to
enable identification of any gaps in performance and optimization of auxiliary power
consumption in BFPs.

Encon Option
Combined Efficiency improvement
Present operational condition indicates a lower combined efficiency of BFP sys-
tem. The reasons identified may be either increased power loss in HC at part load
condition or degradation of pump internal cartage condition. While it is recom-
mended to study the HC characteristic and pump internals for pinpointing the cause,
rectification of the same will envisage the following benefits:
Cost–Benefit Analysis
Present condition (3B & 5B)

Pump ref 3B 5B
Duty conditions: BP and BFP 450 TPH, 73 mWC 413 TPH, 72 mWC
430 TPH, 1226 mWC, 391 TPH, 1220 mWC
Liquid kW 1755 kW 1602 kW
Combined eff. (BFP, booster pump FC, and 62% 57%
motor eff.)
Motor power (I/P kW) 2837 kW 2810 kW

Proposed condition

Pump ref 3B 5B
Duty conditions: BP and BFP 450 TPH, 73 mWC 413 TPH, 72 mWC
430 TPH, 1226 mWC 391 TPH, 1220 mWC
Liquid kW 1755 kW 1602 kW
Least expected combined eff. 65% 65%
Motor power (I/P kW) 2700 kW 2465 kW
Reduction in kW input 137 kW 345 kW
Annual energy savings (@ 7615 h/yr) 1,043,255 2,627,175
Annual monetary savings Rs (@ Rs 2.5/unit) 2,608,137.5 6,567,937.5
Energy Audit of Auxiliary Power Consumption: A Case Study 807

The drive and the driven equipment are coupled by hydraulic coupling which also
acts as a control mechanism. At lower load operation, the pump delivery is controlled
by hydraulic coupling. The full load HC efficiency is indicated as 94.7%. Moreover,
part load inefficiencies of fluid coupling are being well known; it is recommended
to adopt VFD (LT) for BFPs instead of fluid coupling. In case the problem is asso-
ciated with pump internals, the same may be discussed with OEM for necessary
modification.

3.2 Condensate Extraction Pump (CEP)

CEPs constitute a key auxiliary in terms of criticality, in Rankine cycle as well as their
contribution to overall auxiliary power consumption. It serves to evacuate condensate
from the condenser and pushes through a train of heat exchangers before discharging
into the de-aerator. Individual CEP flow has been measured.

Observations
While the pump loading is in respect of flow, head is close to design parameters.
There is a considerable deviation in terms of efficiency slippage. It is evident from the
table above that the pump is almost underperforming in comparison with the design
conditions by 10–12%. While minor gland leakage was observed in 3B, the internals
of both the pumps are to be checked to address the reason for underperformance.
Since this is a new unit, the OEM may be consulted for the underperformance of
the pumps. Similar to BFP, CEP of both the units is presenting similar behavior in
terms of efficiency, while operating at similar operating parameters. Therefore, the
OEM may be consulted for such slippage in efficiency with respect to design level.
Individual CEP flow has been measured (Table 2).

Encon Option
Combined Efficiency improvement:
Present operational condition indicates a lower combined efficiency of CEP. The
operational factors influencing efficiency variation are felt to be in-line with design
level on head, flow, and drive motor loading, apart from the intrinsic efficiency levels
of the pumps and the drive motor. On rectification, the following savings or reduction
in terms of auxiliary power consumption is expected.
Cost–Benefit Analysis:
Present condition (3B and 5B):

Pump ref 3B 5B
Duty conditions 332 TPH, 292 mWC 337 TPH, 187.5 mWC
Liquid 173.6 kW 172.2 kW
combined eff. 60.4% 59.2%
Motor power (I/P kW) 287.59 291.05
808 A. Nayak et al.

Table 2 Performance evaluation of expected part load efficiency 77%

S. No. Item ref Units Design CEP-1 CEP-2
1 Unit load MW 135 136 134
2 CEP flow m3 /HR 394 332 337
3 Suc. pressure mWC – −0.93 −0.958
4 Dis. pressure of CEP mWC – 191 186.6
5 Total dev head by CEP mWC 197 191.93 187.558
6 Suction temp. °C 46.3 45.6 46.7
7 Density Kg/m3 989 982.0 989
8 Liquid kW of CEP kW 211.5 173.64 172.24
9 CEP (motor input) kW 287.9 287.59 291.05
10 Motor eff. % 93 92 92
11 Motor O/P power (rated 315 kW) kW 267.75 264.6 267.8
12 Combined efficiency % 72.5 60.4 59.2
13 Pump efficiency (curve) % 78.0 65.6 64.3
14 % Load on motor % 85 84.00 85.01
15 % Load on flow % 100 84.26 85.53
16 % Load on head % 100 97.43 95.21
17 Sp. energy consumption of CEP kWh/m3 0.73 0.87 0.86
Note Combined efficiency = efficiency of (pump * motor) as motor and pump are directly coupled

Proposed condition:

Pump ref 3B 5B
Duty conditions: BP and BFP 332 TPH, 292 mWC 337 TPH, 187.5 mWC
Liquid 173.6 kW 172.2 kW
Least expected combined eff. 65% 65%
Motor power (I/P kW) 267.13 kW 264.98 kW
Reduction in kW input 20.46 kW 26.07 kW
Annual energy savings kWh (@ 7615 h/yr) 155,779.23 198,495.62
Annual monetary savings Rs (@ Rs 2.5/unit) 389,448.07 496,239.05
Expected investment Minor Minor
Energy Audit of Auxiliary Power Consumption: A Case Study 809

3.3 Induced Draft Fans

Induced draft fans, evacuating the boiler flue gases, constitute a key HT auxiliary
from both functional point of view and energy intensity.

Observations
In the absence of provision for flow measurement, the fan flow was assessed from fan
curve and the same was matched with the flow of SA and PA fans. The operational
factors influencing efficiency variation are felt to be part loading on head, flow, apart
from the intrinsic efficiency levels of the fan and losses in HC. The HC performance
characteristic is to be referred for actual losses in HC. For identification of actual
losses alternatives like VFD installation or in case, permanent reduction in speed two
motors may be explored. Inter-comparison of ID fans indicates the performance of
1 in lower side. It is recommended to look into the fan profile and HC performance
during the next overhauling. The flow balance from curve and PA and SA reflects
minimal ingress to the tune of 20%; however, it is strongly recommended to make
necessary arrangements for flow measurement during the next opportune moment.
Periodic as run efficiency assessment along similar lines of NPC methodology is rec-
ommended (at monthly intervals) to enable identification of any gaps in performance
and optimization of auxiliary power consumption in ID fan (Table 3).

3.4 Secondary Air Fans

The trial observations on SA fans of units are presented in Table 4.

Observations
The SA fan flow measurement was done at the fan suction side. And the same was also
cross-checked with available instrumentation at unit control station (Table 4). While
the power consumption of the two fans is almost similar, there is little variation in
head developed and it is recommended to check the suction filter conditions. While
the operational parameters of both the fans are similar, deviation in tune of 6%
is being observed in terms of combined efficiency of both the fans. The expected
reasons may be deterioration of fan intrinsic efficiency apart from IGV opening.
The observed IGV engagements were found to be 36% and 39%, respectively. The
inefficiency of IGV control is well known below 70%, and below 70% IGV acts as a
damper. In the absence of actual performance curve w.r.t. IGV operation, the actual
power loss against IGV could not be quantified. The reserve margin of fan is found
to be in higher side leading to IGV operation. In such situation, operation at reduced
speed will result in performance enhancement of the fan; however, if the savings for
combined efficiency of 65% will be in tune of Rs 15,000–100,000 (60–135 kW) per
year; hence, cost benefit for VFD installation will be uneconomical; however, option
for 02 speed motor may be explored.
810 A. Nayak et al.

Table 3 Performance analysis of ID fans

S. No. Item ref Units Design ID # 1 ID # 2 ID # 3 ID # 4
1 Unit load MW 135 135 135 135 135
2 Total fan TPH 495.6 406.81 319.63 327.27 354.54
flow
3 Density Kg/m3 0.82 0.86 0.88 0.82 0.82
4 Total fan m3 /S 167.87 131.40 100.89 110.86 120.10
flow
5 FG temp °C 140 125 130 137 137
6 Head mmWC 850.0 521.4 531.1 521.4 531.1
developed
7 Air power kW 1398.93 671.68 525.35 566.71 625.35
8 IGV/FC % – 100/48 100/38.4 100/49 100/39
9 RPM rpm 720.00
10 Motor kW 1855.67 1292.67 918.91 1014.19 1084.79
input
power
11 Motor % 97.0 95.0 95.0 95.0 95.0
efficiency
12 Motor O/P kW 1800 1228.04 872.97 963.48 1030.55
power
13 Combined % 51.96 57.17 55.88 57.65
eff. (incl.
HC)
14 Fan % 84.6% 54.7% 60.2% 58.8 60.7
efficiency (fan) Fan:83% Fan:83%
(incl. HC)
15 % Load on % – 69.7 49.5 54.7 58.5
motor
16 % Load on % – 78.3 60.1 66.0 71.5
flow
17 % Load on % – 61.3 62.5 61.3 62.5
head
18 SEC kWh/TPH – 3.18 2.87 3.10 3.06

Similar to previous unit, the SA fan flow measurement was done at the fan suction
side. And the same was also cross-checked with available instrumentation at unit
control station. While the power consumption of the two fans is almost similar, there
is little variation in head developed and this is expected to be because of header
connection at unequal distance from both the fans. The expected reasons may be
deterioration of fan intrinsic efficiency apart from IGV opening. The observed IGV
engagements were found to be 29% and 26%, respectively. The inefficiency of IGV
control is well known below 70%, and below 70% IGV acts as a damper. The IGV
engagement is much lower than that of unit reason being the head developed is
Energy Audit of Auxiliary Power Consumption: A Case Study 811

Table 4 Performance analysis of SA fans

S. No. Item Ref Units Design SA # 1 SA # 2 SA #3 SA # 4
1 Unit load MW 135 135 135 135 135
2 Total fan flow TPH 157.8 136.8 140.1 128.3 127.9
3 Density Kg/m3 1.106 1.16 1.16 1.16 1.16
4 Total fan flow m3 /s 39.62 32.76 33.55 30.72 30.62
5 Cold air temp °C 43 31 31 31 31
6 Head developed mmWC 1565.5 1172.0 990.0 662.0 652.0
7 Air power kW 608.09 376.42 325.65 199.38 195.75
8 IGV/VFD % – 36 39 29 26
9 Motor input kW 842.11 638.33 634.43 592.49 602.54
power
10 Motor efficiency % 95.0 94.0 94.0 94.0 94.0
13 Motor O/P power kW 800 600.03 596.37 556.94 566.39
14 Combined eff. % 80.5 58.97 51.33 33.65 32.49
15 Fan efficiency % 84.7 62.7 56.3 35.8 35.6
16 % Load on motor % 100.0 75.8 75.3 70.4 71.6
17 % Load on flow % 100.0 82.7 84.7 77.5 77.3
18 % Load on head % 100.0 74.9 63.2 42.3 41.6
19 SEC kWh/TPH 5.34 4.67 4.53 4.62 4.71
Note Combined efficiency = efficiency of (fan * motor * coupling * IGV)

lower by 300 mmWC (appx.). In the absence of actual performance curve w.r.t. IGV
operation, the actual power loss against IGV could not be quantified. The reserve
margin of fan is found to be in higher side leading to IGV operation. As discussed,
the IGV is now operating as a damper, and thereby the combined efficiency has come
down to around 33%. Since the reserve margin is higher it is felt that fan operation
at reduced speed will enhance combined efficiency of fan.

3.5 Primary Air Fan

Observations
Both the PA fans of two units are operating at a very lower combined efficiency
against the design level (Table 5). Though the operating point of the fan has deviated
from the design in respect of flow as well as the head, the IGV is above 80% at full load
condition. However the combined efficiency is in lower side. In comparison of fan
performance of PA fans 1, 2 the 3, 4 PA fans performance are lower in range of 7–10%
while much difference are not observed in terms of flow or pressure developed. While
it is strongly felt that the fan intrinsic efficiency has come down it is recommended to
812 A. Nayak et al.

Table 5 Performance analysis of PA fans

S.No. Item Ref Units Design PA #1 PA # 2 PA # 3 PA # 4
1 Unit load MW 135 137 137
2 Total fan flow TPH 192.7 126.8 127.7
3 Density Kg/m3 1.106 1.12 1.12 1.12 1.12
4 Total fan flow m3 /s 48.39 33.00 34.20 31.44 31.68
5 Cold air temp °C 43 30 30 30 30
6 Head mmWC 2343.3 1514.0 1458.0 1478.0 1460.0
developed
7 Air power kW 1111.69 489.82 488.86 455.57 453.46
8 IGV/VFD % – 75 77 87 84
10 Motor input kW 1473.68 1124.29 1197.55 1304.35 1332.29
power
11 Motor % 95.0 94.0 94.0 94.0 94.0
efficiency
12 Motor O/P kW 1400 1056.83 1125.70 1226.08 1252.35
power
13 Combined eff. % 81.0 43.57 40.82 34.93 34.04
14 Fan efficiency % 85.2 46.3 43.4 37.2 36.2
curve/calc)
15 % Load on % 100.0 76.3 81.3 88.5 90.4
motor
16 % Load on % 100.0 68.2 70.7 65.0 65.5
flow
17 % Load on % 100.0 64.6 62.2 63.1 62.3
head
18 SEC kWh/TPH – 8.45 8.68 10.29 10.43
Note Combined efficiency = efficiency of (fan * motor * coupling * IGV)

check the fan internals and blade profile before taking up the matter with the Original
Equipment Manufacturer. The reduced fan efficiency is adding up to the auxiliary
power consumption of the plant, therefore it is recommended to draw immediate
attention to the lowered performance of PA fans.

4 Conclusion

The results depicts that a steel plant auxiliary units like boiler feed pump (BFP),
Condensate Extraction Pump (CEP), Induced draft fans (IDF), Primary (PAF) &
secondary air fans (SAF) have a significant reserves on energy and cost also. There
is a significant amount of energy can be saved by following the audit suggestion. It is
evident from the table above that the pump is almost underperforming in comparison
Energy Audit of Auxiliary Power Consumption: A Case Study 813

to the design conditions by 10–12% in case of pump loading. Similar to BFP, CEP
of both the units presenting similar behavior in terms of efficiency, while operating
at similar operating parameters. Therefore the OEM may be consulted for such
slippage in efficiency with respect to design level. The flow balance from curve
and PA and SA reflects minimal ingress to the tune of 20%, however it is strongly
recommended to make necessary arrangements for flow measurement during the
next opportune moment. In such situation, operation at reduced speed will result in
performance enhancement of the fan; however, if the savings for combined efficiency
of 65% will be in tune of Rs 15,000–100,000 (60–135 kW) per year; hence, cost
benefit for VFD installation will be uneconomical; however, option for 02 speed
motor may be explored. As discussed, the IGV is now operating as a damper, and
thereby, the combined efficiency has come down to around 33%. Since the reserve
margin is higher, it is felt that fan operation at reduced speed will enhance combined
efficiency of fan. The performance level of both the HP blowers is satisfactory;
however, the flow, head, and power consumption does not match with that of the
provided characteristics. Therefore, it is recommended to recheck the same. Also,
the graph indicated a power requirement above 450 kW, while the motor rating is
found to be 450 kW; the same may be rechecked.

References

1. Mendis, N.N.R., Perera, N.: Energy audit: A case study. In: 2006 International Conference on
Information and Automation, Shandong, pp. 45–50 (2006)
2. Dongellini, M., Marinosci, C., Morini, G.L.: Energy audit of an industrial site: A case study.
Energy Procedia 45, 424–433 (2014)
3. Doukas, D.I., Bruce, T.: Energy audit and renewable integration for historic buildings: The
case of craiglockhart primary school. Procedia Environ. Sci. 38, 77–85 (2017)
4. Kumar, A., Ranjan, S., Singh, M.B.K., Kumari, P., Ramesh, L.: Electrical energy audit in
residential house. Procedia Technol. 21, 625–630 (2015)
5. Al-Ajlan, S.A.: Energy audit and potential energy saving in an office building in Riyadh, Saudi
Arabia. J. King Saud Univ. Eng. Sci. 21(2), 65–74 (2009)
6. Belussi, L., Danza, L., Salamone, F., Meroni, I., Galli, S., Svaldi, S.D.: Integrated smart system
for energy audit: Methodology and application. Energy Procedia 140, 231–239 (2017)
7. Lucentini, M., Naso, V., Borreca, M.: Parametric performance analysis of renewable energy
sources HVAC systems for buildings. Energy Procedia 45, 415–423 (2014)
8. Ajapunje, A.K., Apte, A.: Analytical study of energy conservation and auditing. Int. J. Electr.
Eng. Technol. 7(2), 79–87 (2016)
9. Sizganova, E.Y., Antonenkov, D.V., Solovev, D.B.: Energy-saving efficiency and potential in
educational establishments of Neryungrinsky Region. Int. J. Energy Technol. Policy 15(2),
180–193 (2019)
10. Akadiri, S., Bekun, F.V., Taheri, E., Akadiri, A.: Carbon emissions, energy consumption and
economic growth: A causality evidence. Int. J. Energy Technol. Policy 15, 320–336 (2017)
Atomistic Level Process to Device
Simulation of GaNFET Using TNL
TCAD Tools

R. K. Nanda, E. Mohapatra, T. P. Dash, P. Saxena, P. Srivastava,

R. Trigutnayat, and C. K. Maiti

Abstract An atomistic level process to device simulation tools developed by Tech

Next Lab (TNL) is reported. Modeling of the deposition of high-quality ultrathin
AlGaN epitaxial films grown on GaN substrates by molecular beam epitaxy (MBE)
has been performed. The surface morphology, crystalline quality, and interfacial
property of as-grown AlGaN epitaxial films on GaN substrates are studied using
simulation. The epitaxial layer characterization for extract of exact carrier mobility
and use of epitaxially grown material for GaN-FET device application has been
demonstrated. Results obtained on the basis of process to device simulation have
been calibrated with reported results.

Keywords GaN-FET · Epi-growth · Full band simulation · TNL

R. K. Nanda (B) · E. Mohapatra · T. P. Dash · C. K. Maiti

Department of Electronics and Communication Engineering, Siksha ‘O’ Anusandhan (Deemed to
be University), Bhubaneswar, Odisha 751030, India
e-mail: [email protected]
E. Mohapatra
e-mail: [email protected]
T. P. Dash
e-mail: [email protected]
C. K. Maiti
e-mail: [email protected]
P. Saxena · P. Srivastava · R. Trigutnayat
Tech Next Lab (P) Limited, Niwaz Ganj, Lucknow 226003, India
e-mail: [email protected]
P. Srivastava
e-mail: [email protected]
R. Trigutnayat
e-mail: [email protected]

© Springer Nature Singapore Pte Ltd. 2020 815

G. Pradhan et al. (eds.), Advances in Electrical Control and Signal Systems,
Lecture Notes in Electrical Engineering 665,
https://doi.org/10.1007/978-981-15-5262-5_61
816 R. K. Nanda et al.

1 Introduction

Transistors based on GaN significantly extend the capabilities of power electron-

ics and microwave devices [1, 2]. GaN being a wide-gap semiconductor material is
normally used for constructing the base material of a blue light emitting diode [3].
Furthermore, material properties such as high withstand voltage and low conducting
resistance are well known [4]. It is a semiconductor material that is strongly expected
to be applied to the next-generation power devices by taking advantage of these char-
acteristics. In particular, the AlGaN/GaN-HEMT (high-electron-mobility transistor)
structure based on GaN is expected as a high-speed switching element with high gain
due to downsizing of device size and reduction of parasitic capacitance [5–8]. A high
electron density in the region of the two-dimensional electron gas in combination
with acceptable electron mobility makes it possible to implement high current den-
sities in the transistor-channel cross section and high gains. However, experimental
optimization of heterostructure transistors still remains a complicated and expensive
procedure [9, 10].
The atomistic simulation suite developed at Tech Next Lab (TNL) consists of the
following modules [11–13]:
• EpiGrow Simulator
• Full Energy Band Simulator
• Electron Mobility Simulator
• Monte Carlo Particle Device Simulator.
Based on the reactor process conditions such as partial pressures of elements of
which thin film has to be grown, substrate orientation, and temperature, process sim-
ulation of the film deposition is performed. Extraction of the physical parameters of
the deposited films such as lattice parameter, strain, deformation, and surface rough-
ness of films is calculated. Full Energy Band Simulator provides various parameters
for carrier’s transportation on full band structure associated with epi-grown AlGaN
monolayers.
Electron Mobility Simulator is capable to simulate carrier’s mobility under the
influence of external electric field and effect of same on the carrier transport on full
energy band structure, and lastly the Monte Carlo Particle Device Simulator module
[13] is used for particle-based device simulation.

2 Atomistic Process Simulation

A thin film of Ga1-x Alx N (where x = 0, 0.15, 0.3, and 0.39) is grown on 3C-SiC
substrate by EpiGrow simulator. The input parameters needed to grow thin film
by EpiGrow simulator are energy values such as surface energy, film energy, and
Schwoebel energies and reactor conditions as partial pressure substrate temperature.
Surface diffusion energy barrier, E s , is also called activation energy. The calculation
Atomistic Level Process to Device Simulation … 817

of it is based on the assumption that bonds only exist between nearest neighboring
atoms. E n is nearest binding energy on the horizontal plane. If n is the number of
nearest neighbors on the surface, the total activation energy is given as

E = Es + n En (1)

Epi-growth has been done with the following process parameters:

Parameters Values Unit

Time 30 s
Temperature 800 °C
Surface energy 2 eV
Desorption barrier energy 4 eV
Schwoebel barrier 0.002 eV
Incorporation barrier 0.05 eV
Nearest neighbor attraction 0.05 eV

Precursors and gas ambience used during simulation are

Materials Partial pressure

Ga (mbar) Al (mbar) N2 (mbar)
GaN 0.3 0.0 3.0
Ga0.85 Al0.15 N 0.3 0.03 3.0
Ga0.7 Al0.3 N 0.28 0.05 3.0
Ga0.61 Al0.39 N 0.25 0.10 3.0

Figure 1 depicts the atomic lattice structure. On the lattice structure, the lattice
constant, roughness, strain, and defects can be mapped and traced. Figure 2 represents
the variation of lattice constant with respect to Al contents in the AlGaN material.
One can establish a conclusion from Fig. 2 that increasing the Al contents, the lattice
constant will be decreased, i.e., showing inversely dependence of lattice constant
on the Al contents in the AlGaN alloy during epi-growth. Figure 3 shows different
structure roughness at different operating temperature conditions. Roughness in the
epi-growth layer increases with the rise in temperature due to dominance of diffusion
process of atoms on the substrate.
818 R. K. Nanda et al.

Fig. 1 Epitaxial growth of AlGaN over GaN using TNL EpiGrow simulator

Fig. 2 Variation of lattice constant with Al mole fraction

Atomistic Level Process to Device Simulation … 819

Fig. 3 Surface roughness at the interface of AlGaN/GaN

3 Full Band Simulation

Full Band Simulator [11] is used to simulate full electronic band structures for AlGaN
epi-grown layer on GaN, the form factors interpolated from GaN and AlN form
factors. The full electronic band structure includes bowing of band energies and
their deformation potentials. The lattice constant extracted from epi-grown layers of
GaN and AlGaN is used and given below:
Figures 4 and 5 depict the full electronic band structures for epi-grown GaN
and AlGaN materials, respectively. The different physical parameters, e.g., carrier
group velocity, different energies associated with different bands, effective mass,
deformation potentials, and density of states for carriers have been traced on the full
electronic band structures for simulation of carrier transport under external forces.
The band-gaps at gamma valley are 3.2 and 3.41 eV for GaN and Al0.37 Ga0.63 N,
respectively.
820 R. K. Nanda et al.

Fig. 4 Full band structure of GaN

4 Mobility Characterization Simulation

Epi-grown layers are characterized with the help of Electron Mobility Simulator
[12] to simulate the electron transport in AlGaN/GaN materials including different
scattering mechanisms under the influence of electric field.
Figures 6 and 7 show the scattering rates at gamma valley and electrons occupation
at different valleys under the application of external forces in the form of electric
field. Initially, all electrons remain at gamma valley, as the electric field increases the
transition of electrons starting from gamma valley to L- and X-valley. Since electrons
effective mass is much higher in L-valley, electrons cannot transit to X-valley.
Atomistic Level Process to Device Simulation … 821

Fig. 5 Full band structure of AlGaN

Fig. 6 Carrier scattering at gamma valley

822 R. K. Nanda et al.

Fig. 7 Valley occupation variation with electric field

Figure 8 demonstrates the drift velocity of carriers under the application of electric
field; the mobility of electrons increases as electric field increases and starts decreas-
ing after attending the saturation values after which mobility starts to decrease due
to different scattering mechanism and showing negative differential mobility. The
mobility of carriers in epi-grown sample of AlGaN/GaN is coming out to almost
200 cm2 /V s which is reasonably agreed with the reported standard mobility of
electrons in GaN/AlGaN sample ~300 cm2 /V s.

5 Device Simulation

A GaNFET with 190 nm gate length is chosen for device simulation as shown in
Fig. 9. The device simulation includes the physical and electrical property study. The
electrical performance has been studied in terms of I d –V g and I d –V d .
The I d –V g curve represents the standard current voltage characteristics for
AlGaN/GaN on Si substrate. Under high reverse bias conditions, the current is almost
negligible; in moderate to low reverse bias conditions, current starts rising linearly,
and under forward bias conditions, it attains almost saturation as shown in Fig. 10
(Fig. 11).
Atomistic Level Process to Device Simulation … 823

Fig. 8 Drift velocity versus electric field

The I d –V d curve matches with the standard I–V characteristics of GaN-FET

devices under same operating conditions. At low forward bias condition, the current
increases linearly showing ohmic type of dependence on the voltage, while as device
starts operating at high forward bias conditions, the current attends saturation due to
different scattering mechanisms, and hence, the mobility of carriers decreases.

6 Conclusion

The present paper shows the capabilities of innovative TNL TCAD simulators with
atomistic level simulation starting from epi-growth, material characterization to
device simulation. The TNL TCAD simulators use solution Boltzmann equation with
Monte Carlo technique for material characterization and device simulation purposes.
The carriers transport on full band structure under appropriate biasing conditions can
provide valuable information of inside physical mechanisms which are not possible
with the conventional drift-diffusion and hydrodynamic models.
824 R. K. Nanda et al.

Fig. 9 Device structure with 25 nm thick AlGaN/GaN layer grown on SiC (100) substrate

Fig. 10 Output (I d –V g )
0.035
characteristics
0.030
Drain Current (Amp)

VD =1V
0.025

0.020

0.015

0.010

0.005

0.000

-10 -8 -6 -4 -2 0 2
Gate Voltage (V)
Atomistic Level Process to Device Simulation … 825

Fig. 11 Output (I d –V d )
characteristics VD=-1V
0.15 VD =-2V
VD=-3V

Drain Current (A)

0.10

0.05

0.00

0 2 4 6 8 10 12 14 16
Drain Voltage (V)

References

1. Aleksandr, G., et al.: Dependence analysis of the GaN HEMT parameters for space application
on the thickness AlGaN barrier layer by numerical simulation. In: 2017 IEEE 2nd International
Conference on Opto-Electronic Information Processing (ICOIP), pp. 79–82 (2017)
2. Brocato, R.: High power-density photovoltaic boost converter and inverter using GaN and
AlGaN devices housed in 3D printed packages. In: 2017 IEEE 5th Workshop on Wide Bandgap
Power Devices and Applications (WiPDA), pp. 220–224 (2017)
3. Nakamura, S., Mukai, T., Senoh, M.: Candela-class high-brightness InGaN/AlGaN double-
heterostructure blue-light-emitting diodes. Appl. Phys. Lett. 64, 1687–1689 (1994)
4. Wu, Y., Jacob-Mitos, M., Moore, M.L., Heikman, S.: A 97.8% efficient GaN HEMT boost
converter with 300 W output power at 1 MHz. IEEE Electron Device Lett. 29(8), 824–826
(2008)
5. Kimball, D.F., Jeong, J., Hsia, C., Draxler, P., Lanfranco, S., Nagy, W., Linthicum, K., Larson,
L.E., Asbeck, P.M.: High-efficiency envelope-tracking W-CDMA base-station amplifier using
GaN HFETs. IEEE Trans. Microw. Theory Tech. 54, 3848–3856 (2006)
6. Mishra, U.K., Parikh, P., Wu, Y.-F.: AlGaN/GaN HEMTs-an overview of device operation and
applications. Proc. IEEE 90(6), 1022–1031 (2002)
7. Palacios, T., Chakraborty, A., Rajan, S., Poblenz, C., Keller, S., DenBaars, S.P., Speck, J.S.,
Mishra, U.K.: High-power AlGaN/GaN HEMTs for Ka-band applications. IEEE Electron
Device Lett. 26, 781–783 (2005)
8. Mizutani, T., Ohno, Y., Akita, M., Kishimoto, S., Maezawa, K.: A study on current collapse in
AlGaN/GaN HEMTs induced by bias stress. IEEE Trans. Electron Devices 50(10), 2015–2020
(2003)
9. Vitanov, S., Palankovski, V., Murad, S., Roedle, T., Quay, R., Selberherr, S.: Predictive simula-
tion of AlGaN/GaN HEMTs. In: IEEE Compound Semiconductor IC Symposium, pp. 131–4
(2007)
10. Tikhomirov, V., Zemlyakov, V., Volkov, V., Parnes, Y., Vyuginov, V., Lundin, W., Sakharov,
A., Zavarin, E., Tsatsulnikov, A., Cherkashin, N., Mizerov, M., Ustinov, V: Semiconductors
50(2), 244–248 (2016)
11. Full Band Simulator User Manual. Tech Next Lab (P) Ltd., Lucknow (2019)
826 R. K. Nanda et al.

12. Electron Mobility Simulator User Manual. Tech Next Lab (P) Ltd., Lucknow (2019)
13. Monte Carlo Particle Device Simulator User Manual. Tech Next Lab (P) Ltd., Lucknow (2019)
Modeling, Analysis, and Control
of Vehicle Suspension System Based
on Self-Tuned PI Control

Akshaya Kumar Patra, Alok Kumar Mishra, Anuja Nanda,

Lalit Mohan Satapathy, Ramachandra Agrawal, and Narayan Nahak

Abstract The aim of this paper is to design a self-tuned proportional and integral
controller (STPIC) for a vehicle suspension (VS) system to improve the ride comfort
by absorbing the shocks due to a rough and uneven road. In this control strategy, the
conventional proportional and integral controller (CPIC) is re-formulated with an
automated self-tuned approach to improve the control performance. The STPIC is a
novel approach whose gains dynamically vary with respect to the error signal. The
validation of the improved control performance of STPIC is established by compar-
ative result investigation with other published control algorithms. The comparative
results clearly reveal the better response of the suggested approach to control the
oscillation of the VS system within a stable range with respect to the accuracy,
robustness, and capability to control uncertainties.

Keywords Ride comfort · Road profile · Vehicle suspension system · STPIC

A. K. Patra (B) · A. K. Mishra · A. Nanda · L. M. Satapathy

Department of EEE, Institute of Technical Education and Research, Siksha ‘O’ Anusandhan
(Deemed to be University), Bhubaneswar 751030, India
e-mail: [email protected]
A. K. Mishra
e-mail: [email protected]
A. Nanda
e-mail: [email protected]
L. M. Satapathy
e-mail: [email protected]
R. Agrawal · N. Nahak
Department of EE, Institute of Technical Education and Research, Siksha ‘O’ Anusandhan
(Deemed to be University), Bhubaneswar 751030, India
e-mail: [email protected]
N. Nahak
e-mail: [email protected]

© Springer Nature Singapore Pte Ltd. 2020 827

G. Pradhan et al. (eds.), Advances in Electrical Control and Signal Systems,
Lecture Notes in Electrical Engineering 665,
https://doi.org/10.1007/978-981-15-5262-5_62
828 A. K. Patra et al.

1 Introduction

In recent times, many researchers in the field of vehicle suspension (VS) system
dynamics have devoted to arrive at an optimal solution with a compromise between
vehicle handling, ride comfort, and stability. There is a need for better approach today
as the above problems are very much evident in the modern vehicle cases. Specific
to the large sedan and luxury cars, even if excellent ride qualities are achieved, it is
limited to acquiring adequate handling behavior. With reference to sports vehicles,
although it is provided with very good handling capability, it fails to provide desir-
able ride quality. There are many options in between for variations in the designing
stage of the vehicle manufacturers to meet the customer needs. From the designing
point of view, passenger comfort and vehicle control are the two primary objectives
to be considered. Road disturbances such as bumps or potholes are to be handled
not to sacrifice the passenger comfort. At the control stage, these factors are gener-
ally considered either through keeping the vehicle body from rolling and pitching
excessively, or by maintaining good contact between the tire and the road.
Nowadays, hydraulic dampers (shock absorbers) and springs are extensively used
for vehicle suspensions. As a principle, these are charged with the job of absorbing
bumps, minimizing the vehicle’s body motions during acceleration, braking and
turning of the vehicle, and keeping the tires in contact with the road surface. However,
from the designing point of view, these objectives are contradictory to each other to
achieve all simultaneously at their optimum level.
The spring and the damper are the two essential components in VS design. The
spring design mostly depends on the weight of the vehicle. The damper design is
based on the suspensions placement on the compromise curve and so it is essential to
be perfectly chosen to make the optimal vehicle performance for any type of vehicle.
For ideal performance, the damper should act such that passengers isolate from
low-frequency road disturbances and absorb high-frequency road disturbances. High
damping is essential to achieve for best isolation of passengers from low-frequency
disturbances.
However, even though it is desirable to design a high damping system, it is on
the other degrading the high-frequency absorption rate. In other way providing low
damping, the damper offers adequate high-frequency absorption in terms of scari-
fying low-frequency isolation. To meet these contradictory objectives, it is essential
to design and focus on automotive suspensions without compromising any of the
factors mentioned above. As a solution to the above, three types of suspensions can
be improved. These are passive, fully active, and semi-active types of suspensions.
The spring and damper are the two basic components of the conventional passive
suspension. Both the components are considered and fixed at the design stage. The
suspension stores energy in the spring. Later, it dissipates energy through the damper.
It is very much needed for further research to develop robust control algorithms
to enhance the performance of VS system. This in turn enhances the vehicle capa-
bilities to handle the aforementioned issues. An ideal design of VS system needs to
achieve many performance characteristics such as (1) control of body movement and
Modeling, Analysis, and Control of Vehicle … 829

(2) control of suspension movement and force distribution. With the performance
point of view, the VS should able to isolate the body for comfort against the road
impact and inertial disturbances. These are generally associated with cornering and
braking or acceleration of the vehicle system [1]. As discussed above, many per-
formance objectives are conflicting in nature, that to enhance one the other factor
degrades. Considering all the objectives, as a goal for the designing of a suspension
system, is difficult to meet [2]. Minimization of vertical force to the passengers can
be achieved by minimizing the vertical vehicles body acceleration of the suspension.
Another factor which plays a vital role in passengers comfort is the optimal contact
between wheel and road surface, and this is essential in various driving conditions
in order to maximize safety factor [3]. Among few designs in the past, the system
presented in [4] based on unconstrained optimizations for passive suspension (PS)
system case is widely accepted and used. This successfully performs to achieve
the desirability of low suspension stiffness, reduced unsprung mass (UM), and an
optimum damping ratio for the better controllability. As the PS system performs sat-
isfactorily to some extent, it is considered in many applications for the VS system.
However, both the spring and damper do not supply energy to the suspension system
(SS). They only control the motion of the vehicles body and wheel by limiting the
suspension velocity. This is computed according to the rate specified by the designer.
To overcome this issue, the active suspension (AS) system is considered as an effi-
cient option for this application. The AS systems have the ability to dynamically
respond to changes in the road profile. It is due to this fact that it can supply energy
to produce relative motion between the body and the wheel. Sensors are provided
in suspension system to measure the parameters dynamically. The parameters such
as body velocity, suspension displacement, and wheel velocity and wheel and body
acceleration are sensed and computed for the controller as input parameters [5]. An
AS can be thought as the integration of the passive components to actuators that sup-
ply additional forces. These additional forces are computed by a control law based
on the input data from the sensors fixed to the vehicle. For modeling the real-time
dynamic conditions, the uncertainties due to system design and other external dis-
turbances are needed to be considered for the controller. This motivates for a better
control design to increase the robustness and controllability under uncertainties and
disturbances.
During the past three decades, so many control strategy techniques are suggested
and tested by absorbing the shocks due to a rough and bumpy road in case of VS
system. Time-discrete and switching PID control strategy is implemented in VS
problems with variable control gains based on the measured suspension variables
[6, 7]. However, the optimal gain parameter setting, a lesser range of robust control,
and need of change of gain setting with varying conditions are the major limitations
to limit the real-time application of these controllers. Among other projected robust
control algorithms applied for limiting the oscillation and velocity of the VS system
are fuzzy logic control [8–13], fuzzy PID control [14], genetic algorithm [15], neural
network [16], neuro-Fuzzy (NF) control [17], linear quadratic regulator (LQR) [18],
H-infinity control [19], and sliding mode (SM) control [20, 21]. However, even if
these control techniques are implemented effectively by absorbing the shocks due to
830 A. K. Patra et al.

the rough and bumpy road in case of VS system with enhanced accuracy and damp-
ing of oscillation, they still fail to handle various constraints and random change
found in a suspension environment. These approaches are not fully insensitive to
the disturbances and the uncertainties of the model in spite of the improved perfor-
mance. Hence, optimal control parameters setting for still better performance and
for avoiding slow response following road disturbance (road impact), the current
work suggests an alternative novel technique based on the self-tuned PI control con-
cept. Application of the suggested approach to control the oscillation and velocity in
vehicle suspension system results to ensure a better robust controller in comparison
with other contemporary well-established approaches under both harmonized and
incompatible uncertainties.
The highlights of this manuscript are as follows:
• Development of a Simulink model of a VS system.
• Design of a novel STPIC to regulate the oscillation of the VS system within a
stable range.
• Evaluation of the control actions of the STPIC under huge deviation of road
disturbance.
• Comparative investigation to certify the better response of the STPIC.
This manuscript is organized as follows: Sect. 2 concisely illustrates the VS model
with mathematical details reflecting to its dynamic characteristics of its suspension
vibration process. Also, it clearly demonstrates the simulation execution of the sys-
tem on MATLAB environment. A detailed presentation on how the control technique
is formulated and how it is implemented for this problem is presented in Sect. 3.
Comparative results of the proposed approach with other published control tech-
niques and the related analysis are provided in Sect. 4. The concluding comments
are summarized in Sect. 5.

2 Problem Formulation and Modeling

2.1 System Overview

The overall closed-loop model of VS system is depicted in Fig. 1a. The road distur-
bances w(t) and v(t) are reflected as the process disturbance and the sensor noise,
respectively, in this study. The controller receives information about the oscillation
of the VS system as input to provide the optimal control force u(t), and it is applied
in between the UM and the sprung mass (SM) to reduce the relative motion between
them.
Modeling, Analysis, and Control of Vehicle … 831

Fig. 1 a VS model with STPIC; b schematic model of the VS system; c simulation model of the
nonlinear VS system; and d simplified diagram of VS system

2.2 Modeling of VS System

In the year 1958, the concept of VS system emerged into limelight. It has been
found in the later period that the vibration suppression capability is limited in case
of traditional PS and semi-active suspension systems. This leads to focus on an AS
832 A. K. Patra et al.

Table 1 VS model states and parameters

Symbol Description
w(t) Road impact acting on the UM
x 1 (t) Displacement of the SM in vertical direction
x 2 (t) Displacement of the UM in vertical direction
x 1 (t) Velocity of the SM in vertical direction
x 2 (t) Velocity of the UM in vertical direction
[x 1 (t) – x 2 (t)] Oscillation of suspension system
u(t) Applied control force between the SM and UM

system with additional control forces as a better alternative to the above drawbacks
to suppress the oscillations and used in modern vehicle industry. The major factor
that differentiates the active SS performance-wise is the ability to inject energy into
vehicle via actuators unlike dissipates energy in case of conventional suspension
system. To produce a desirable control force to handle the variety of road disturbances
(road impact) in real-time applications, actuators are placed in between the UM and
the SM in AS system. The most challenging task of designing an AS system is
to enhance the ride comfort by absorbing the shocks due to the rough and bumpy
road. The force actuator in the case of the AS system is capable to add and also
dissipate energy from the system. This results in an increase in the ride comfort and
vehicle handling due to the capability of the SS to regulate the vehicle altitude and
to eliminate the adverse effects of braking and vehicle roll during cornering and
braking. The schematic model of the VS system is illustrated in Fig. 1b.
All the physical activities of the VS system are mathematically expressed and
specified by Eqs. (1) and (2). All states and parameters of the model are represented
in Table 1. All specifications of the VS model are documented in Table 2. The
Simulink diagram of the VS model is established with respect to Eqs. (1) and (2) as
displayed in Fig. 1c.
   
d 2 x1 (t) dx1 (t) dx2 (t)
m1 = −b1 − − k1 (x1 (t) − x2 (t)) + u(t) (1)
dt dt dt

Table 2 VS model specification values

Symbol Quantity/meaning Value
m1 SM (mass of the vehicle supported by the suspension) 2500 kg
m2 UM (mass of the axle and wheel) 320 kg
k1 Spring constant of SS 80,000 N/m
k2 Spring constant of wheel and tire 500,000 N/m
b1 Damping coefficient of SS 350 N s/m
b2 Damping coefficient of wheel and tire 15,020 N s/m
Modeling, Analysis, and Control of Vehicle … 833

   
d 2 x2 (t) dx1 (t) dx2 (t)
m2 = b1 − − k1 (x1 (t) − x2 (t))
dt 2 dt dt
 
dw(t) dx2 (t)
+ b2 − + k2 (w(t) − x2 (t)) − u(t) (2)
dt dt

2.3 Performance Analysis of VS System

There are four poles in VS system. One of them lies in right-hand side of the complex
plane. As a result, the system becomes unstable. This needs the design of an adaptive
controller for improving the stability of the system by means of shifting the poles
into the left-hand side of the complex plane. The VS system Simulink model in the
open-loop form is depicted in Fig. 1c. The VS system has two inputs and two outputs.
The control force u(t) and road disturbance w(t) are the two inputs of the VS system.
The oscillation and velocity of the SS are the two outputs of the VS system. An
uncontrolled oscillation and velocity of the suspension system are being observed,
owing to the application of 0.1 m impulsive road disturbance on the unsprung mass
at the simulation time of 1.0 s. The irregular oscillation and velocity of the SS are
illustrated in Fig. 2a, b. Figure 2a, b illustrates the unstable dynamics under various
model uncertainties and disturbances. These unstable dynamics can be reduced by
applying the suitable control algorithms. In this case, the oscillation of the SS is the

Fig. 2 a Response of oscillation of the SS with road impact; and b response of velocity of the SS
system with road impact
834 A. K. Patra et al.

most essential outcome needs to be controlled within a stable range through suitable
control techniques, and velocity of the SS is analyzed in order to view the motion
trajectory.

3 Control Algorithm

The STPIC control algorithm is demonstrated in this section. The closed-loop system
response with respect to robustness, accuracy, and stability is analyzed. The control
specifications such as settling time t s , steady-state error ess , maximum overshoot
OMax and maximum undershoot U Max are also evaluated and examined with proper
validation of the controller action.

3.1 STPIC Design

The VS model with STPIC is shown in Fig. 1a. In STPIC control technique, the
error signal e(t) is used to generate control signal u(t). The u(t) to the VS model
mathematically can be described in terms of the STPIC as [22–26]:

u(t) = K p (t)e(t) + K i (t) e(t)dt (3)

where K p (t) and K i (t) are denoted for the proportional and integral gains of the
suggested STPIC, respectively. The K p (t) can be expressed in terms of error signal
e(t) as follows:
 
K p (t) = K p(Max) − K p(Max) − K p(Min) e−[K e(t)] (4)

where K p(Max) and K p(Min) are denoted for the pre-specified maximum and minimum
values of K p (t), respectively. The K is a constant that regulates the computation of
K p (t) within the specified range according to the existing error. In case of large error
signal e(t), the corresponding computed K p (t) value is also large to speed up transient
response damping. Similarly, in case of small error signal e(t), the corresponding
computed K p (t) is the minimum value and is responsible to eliminate overshoots
and undesirable oscillations. The K i (t) for the STPIC is expressed in terms of e(t) as
follows:

K i (t) = K i(Max) e−[K e(t)] (5)

where K i(Max) is denoted for the maximum values of K i (t). This is because under
the approaching stage of steady-state conditions when the value of the error signal
e(t) is small, the controller uses K i(Max) and that overcomes the existing associated
Modeling, Analysis, and Control of Vehicle … 835

Table 3 STPIC parameter values

K p(Max) K p(Min) K i(Max) K i(Min) K
14 12 1 0 0.0001

Fig. 3 Simulink diagram of STPIC

steady-state error quickly. Similarly, in case of large error signal e(t), the computed
value of K i (t) is small and that eliminates the associated overshoots and undesirable
oscillations. The two controller gains K p (t) and K i (t) are optimally varied according
to the error signal e(t) as mentioned in Eqs. (4) and (5), respectively. It is observed
from Eq. (4) that when an error signal e(t) is large, the exponential term approaches
zero (e−[ke(t)] → 0), and hence, K p (t) = K p(Max) . In other way when the error e(t)
is small, the exponential term approaches one (e−[ke(t)] → 1), and hence, K p (t) =
K p(Min) . From Eq. (5), it is found that the integral gain is allowed to vary within
the range 0 ≤ K i (t) ≤ K i(Max) . The controller parameter values are represented
in Table 3. The Simulink structure of the STPIC as mathematically described in
Eqs. (3)–(5) is illustrated in Fig. 3.

4 Results and Discussion

The performance of the closed-loop VS system with STPIC is described in detail in

this section. The suggested control approach is compared with other popular control
algorithms to justify its enhanced performance.
836 A. K. Patra et al.

Fig. 4 a Response of oscillation of the SS with road impact based on STPIC; and b response of
velocity of the SS with road impact based on STPIC

4.1 Performance Analysis of VS System with STPIC

In this section, all physical activities of the closed-loop VS model with suggested
STPIC are examined under different conditions and the huge deviation of road distur-
bance. The time domain response of the oscillation and the velocity of the suspension
system with 0.1 m impulsive road disturbance at the simulation time of 1.0 s are dis-
played in Fig. 4a, b. The output results clearly specify that the suspension system
attains the zero oscillation and zero velocity with less settling time and also attains
the balance position where the system is absolutely steady. To achieve the enhanced
system response, the required control force u(t) is generated by the suggested STPIC
and is demonstrated in Fig. 5.

4.2 Robustness of the STPIC

Figure 6a, b illustrates the oscillation and the velocity of the VS system with suggested
STPIC under the huge deviation of road disturbance w(t). The time domain outcomes
under huge deviation of road disturbances show the enhanced performance of the
closed-loop VS model with STPIC. Overall in each case, the VS system achieves
finally zero oscillation and zero velocity with less settling time. As indicated by the
results, the suggested controller’s robust performance under the huge deviation of
road disturbances compared to other published control techniques is much better.
Modeling, Analysis, and Control of Vehicle … 837

Fig. 5 Control force generated by STPIC

Fig. 6 a Oscillation of the SS with deviation of ±40% road impact based on STPIC; and b velocity
of the SS with deviation of ±40% road impact based on STPIC

4.3 Verification of Stability

There are four poles in VS system. One of them lies in the right-hand side of the
complex plane as displayed in Fig. 7a. It signifies the system dynamics that are
838 A. K. Patra et al.

Fig. 7 a Location of VS system poles in complex plane; and b location of closed-loop VS system
poles in the complex plane

unstable in nature. For enhancement of stability of the VS system, the STPIC is

developed and implemented in the VS system. As a result, four poles of the VS
system are dragged toward the left-hand side of the complex plane as displayed in
Fig. 7b. Consequently, the system stability is enhanced due to the suggested STPIC.

4.4 Comparative Study

The proposed STPIC control approach is compared with other popular control
approaches such as PID, fuzzy, NF, LQR, H ∞, and SM to justify its enhanced
performance as a controller. Figure 4a illustrates the effect of road impact in the
oscillation of the suspension system with the suggested KFSMGC approach. Table 4

Table 4 Comparative result analysis related to oscillation of the suspension system

Controller PID [6] Fuzzy [11] NF [17] H ∞ [19] SM [21] STPIC
(proposed)
Road 0.1 0.1 0.1 0.1 0.1 0.1
disturbance (m)
t s (s) 4.2 4.1 3.7 3.1 3.0 2.5
OMax (m) 0.0050 0.0049 0.0047 0.0043 0.0040 0.0036
U Max (m) 0.0042 0.0045 0.0046 0.0041 0.0048 0.0041
Noise (%) 10 10 5 5 5 1
ess (%) 0 0 0 0 0 0
Modeling, Analysis, and Control of Vehicle … 839

presents a comparative analysis with respect to t s (s), OMax (m), U Max (m), noise
(%), and ess (%). The effect of road disturbance in the oscillation of the suspension
system applying different control approaches such as PID, fuzzy, NF, H ∞, and SM
is also documented in Table 4 based on the references, and [6, 11, 17, 19] and [21],
respectively. Similar working conditions are followed with the same level of sensor
noise in all control techniques application for comparison.
The oscillations of the suspension system under 0.1 m impulsive road disturbance
are tested. The corresponding results are presented for the various control approaches
along with the proposed STPIC with respect to time domain specifications such as
OMax (m), U Max (m), and t s (s). The results signify the better controllability of the
STPIC. The simulation results also demonstrate the high noise and chattering elimi-
nation capability with high robustness for the proposed approach. Overall, by looking
to the above comparative analysis, the findings of suggested approach advantages are
the higher accuracy and stability, more robustness, high noise and chattering elimina-
tion capability, and better capability to handle uncertainty under various conditions
and huge deviation of road disturbance.

5 Conclusions

In this manuscript, a novel control strategy STPIC is proposed based on self-tuned

approach. To justify its enhanced performance, it has been applied and tested to
control the oscillation and the velocity of the suspension system in the vehicle. In
suggested STPIC, self-tuned algorithm, is utilized to enhance the control perfor-
mance. The comparative results clearly reflect that the suggested STPIC is arrived
at better performance than the other control approaches such as PID, fuzzy, NF,
LQR, H ∞, and SM with respect to stability, reliability, and robustness under vari-
ous abnormal conditions and disturbances. The better performance of the suggested
approach in terms of improved accuracy and stability, enhanced robustness, high
noise elimination capability, and better ability to control uncertainty justifies its
real-time application.

References

1. Wright, P.G.: The application of active suspension to high performance road vehicles. In:
Microprocessors in Fluid Engineering IMechE Conference Publications (1984)
2. Fateh, M.M., Alavi, S.S.: Impedance control of an active suspension system. J. Mechatron. 19,
134–140 (2009)
3. Lin, J.S., Kanellakopoulos, I.: Nonlinear design of active suspension. In: 34th IEEE Conference
1995, vol. 17, pp. 45–59 (1995)
4. Alleyen, A., Hedrick, J.K.: Nonlinear adaptive control of active suspensions. IEEE Trans.
Contr. Syst. Technol. 3(1), 845–860 (1995)
5. Esmailzadeh, E., Taghirad, H.D.: Active vehicle suspensions with optimal state feedback
control. J. Mech. Sci. 200(4), 1–18 (1997)
840 A. K. Patra et al.

6. Kumar, M.S.: Development of active suspension system for automobiles using PID controller
(2008)
7. Talib, M.H.A., Darns, I.Z.M.: Self-tuning PID controller for active suspension system with
hydraulic actuator, In: 2013 IEEE Symposium on Computers & Informatics (ISCI), pp. 86–91
(2013)
8. Pekgökgöz, R.K., Gürel, M.A., Bilgehan, M., Kısa, M.: Actıve suspension of cars using fuzzy
logic controller optimized by genetic algorithm. Int. J. Eng. Appl. Sci. (IJEAS) 2(4), 27–37
(2010)
9. Li, H.: Reliable fuzzy control for active suspension systems with actuator delay and fault. IEEE
Trans. Fuzzy Syst. 20(2), 342–357 (2012)
10. Lin, Y.-J., Lu, T.-Q., Padovan, J.: Fuzzy logic control of vehicle suspension systems. Int. J.
Veh. Des. 14(5), 457–470 (1993)
11. Yoshimura, T., Nakaminami, K., Kurimoto, M.: Active suspension of passenger cars using
linear and fuzzy logic controls. Control. Eng. Pract. 1, 41–47 (1998)
12. Yeh, E.C., Tsao, Y.J.: A fuzzy freeview control scheme of active suspension for rough road.
Int. J. Veh. Des. 15(1), 166–180 (1994)
13. Yang, C.: Active suspension system of automobile based on fuzzy control. Northeastern
university (2007)
14. Yang, Q., Zhou, K., Zhang, W., Xing, X., Yuan, C.: Fuzzy-PID control on semi active air
suspension. Trans. Chin. Soc. Agric. Mach. 39(9), 24–29 (2008)
15. Du, H., Lam, J., Sze, K.Y.: Non-fragile output feedback H1 vehicle suspension control using
genetic algorithm. Eng. Appl. Artif. Intell. 16, 667–680 (2003)
16. Wang, Y.J.: Analysis of vehicle suspension control using neural networks. Northeastern
University (2007)
17. Aldair, A.A., Wang, W.J.: A neurofuzzy controller for full vehicle active suspension systems.
J. Vib. Control 18(12), 1837–1854 (2012)
18. ElMadany, M.M., Abduljabbar, Z.S.: Linear quadratic Gaussian control of a quarter-car
suspension. Veh. Syst. Dyn. 32(6), 479–497 (1999)
19. Kaleemullah, M., Faris, W.F., Hasbullah, F.: Design of robust H∞, fuzzy and LQR controller for
active suspension of a quarter car model. In: 2011 4th International Conference on Mechatronics
(ICOM), pp. 1–6. IEEE (2011)
20. Li, H., Yu, J., Hilton, C., Liu, H.: Adaptive sliding-mode control for nonlinear active suspension
vehicle systems using T-S fuzzy approach. IEEE Trans. Ind. Electron. 60(8), 3328–3338 (2013)
21. Chen, P.C., Huang, A.C.: Adaptive sliding mode control of nonautonomous active suspension
systems with time-varying loading. J. Sound Vib. 282, 1119–1135 (2005)
22. Mohamed, S.Z.: A self-tuning PI controller for the speed control of electrical motor. Electr.
Power Syst. Res. 110, 293–303 (2015)
23. Patra, A.K., Mishra, A.K., Rout, P.K.: Backstepping model predictive controller for blood
glucose regulation in type-I diabetes patient, IETE J. Res. (2018). https://doi.org/10.1080/
03772063.2018.1493404
24. Patra, A.K., Rout, P.K.:, Backstepping linear quadratic gaussian controller design for balancing
an inverted pendulum. IETE J. Res. 1, 1–15 (2019)
25. Patra, A.K., Rout, P.K.: Adaptive continuous-time model predictive controller for implantable
insulin delivery system in type I diabetic patient. Optim. Control. Appl. Methods 38(2), 184–204
(2017)
26. Patra, A.K., Rout, P.K.: Adaptive sliding mode Gaussian controller for artificial pancreas in
TIDM patient. J. Process Control 59, 13–27 (2017)
IoT-Based Automatic Irrigation Control

Akkenaguntla Karthik, Anumula Amarnath, T. M. Manohar Reddy,

Veldhanda Tulasi Krishna, and A. V. Pavan Kumar

Abstract Modern world is facing problems to secure the basic needs such as food,
water, shelter, and electricity for people. Among these, the most focused need is water
as the need of water is increasing immensely. Agriculture sector plays a vital role in
the economic development of any country. It is also a major source of raw materials
to industries like cotton, tobacco, sugar, etc. Currently, the agriculture sector is said
to be facing water problem because of water scarcity. So, a need has raised to use
water carefully without wasting it. Human cannot monitor the water availability
in fields throughout the day. The cons of manual monitoring can be reduced by
using this technology. This paper presents a method to monitor the water level in
fields using soil moisture sensor. The proposed method is integrated with GSM and
IoT technology for getting the status of pump and operating them from anywhere
in the world. The electricity required for the operation of components is less. The
water consumed by plants and soil is continuously monitored by using soil moisture
sensor; based on the moisture, the motor is automatically controlled to feed water.
The operational status of the entire system is communicated through SMS.

Keywords GSM · IoT technology · Soil moisture sensor

A. Karthik · A. Amarnath · T. M. Manohar Reddy · V. T. Krishna · A. V. Pavan Kumar (B)

Department of EEE, Madanapalle Institute of Technology & Science, Madanapalle,
AP 517325, India
e-mail: [email protected]; [email protected]
A. Karthik
e-mail: [email protected]
A. Amarnath
e-mail: [email protected]
T. M. Manohar Reddy
e-mail: [email protected]
V. T. Krishna
e-mail: [email protected]

© Springer Nature Singapore Pte Ltd. 2020 841

G. Pradhan et al. (eds.), Advances in Electrical Control and Signal Systems,
Lecture Notes in Electrical Engineering 665,
https://doi.org/10.1007/978-981-15-5262-5_63
842 A. Karthik et al.

1 Introduction

Human is living in automated world. The technology invented yesterday is getting

obsolete today. So, there is need to update ourselves according to the latest technolo-
gies [1–4]. Many conventional methods are followed by farmers for cultivation of
crops in their fields which requires more labor work, cultivation time, and water. In
agriculture sector, irrigation system is a major foundation for cultivation of crops.
But, the usage of water is more than what is necessary; due to this, water gets wasted
and crops may damage. Nowadays, water is a major resource for existence of living
things. But water scarcity is more in world, and hence, a need has raised to use water
wisely [5–7]. In this regard, we are proposing a method which helps the agricul-
ture sector named as “IoT-Based Automatic Irrigation Control Using Soil Moisture
Sensor and GSM Module” which continuously checks the moisture content in soil
using soil moisture sensor, and when required, water is supplied by turning on the
pump. If the moisture content is enough good, then motor gets turned off. Thus, the
amount of water used will be less compared to popular conventional methods. Any
proposed method or system is said to be efficient when it saves money and electric-
ity and reduces labor work. Our proposed system has all these characteristics, and
hence, it is efficient and has great social impact [8–10]. If this technology is used
for domestic purposes like gardening in homes, it requires soil moisture sensor to
detect the moisture content in soil and a IoT device to control it from remote/far
places [11, 12]. The automatic performance in the proposed system is brought using
soil moisture sensor, 12 V SPDT relay, GSM module, ESP8266 Wi-fi module, and a
bulb considered as load (motor/pump). The moisture content in soil is continuously
checked; when the moisture content is less, the motor (pump) is turned on by SPDT
relay, and when the moisture content is more, the motor (pump) is turned off by SPDT
relay. The status of motor is updated to user by using GSM and IoT technology. IoT
technology is considered to operate the motor and know its status from anywhere
in the world. This technology can be implemented at homes for domestic purpose
and also be used in fields. Thus, the operation of motor can be automated; water and
electricity can be saved. This technology reduces human errors, water wastage, and
electricity. This technology can be useful in countries like India where agriculture
sector is considered as a backbone of the country.

2 Methodology

The technologies and methodologies adopted for the proposed project are discussed
below. The hardware components utilized in the proposed work are
1. Soil moisture sensor (FC-28).
2. GSM module (SIM 800A).
3. 12 V SPDT relay.
4. 230 V Single-phase bulb considered as load.
IoT-Based Automatic Irrigation Control 843

Fig. 1 Soil moisture sensor

(FC-28)

5. Two ESP 8255 Wi-fi modules.

6. Selector switches.

2.1 Soil Moisture Sensor (FC-28)

It consists of 2 probes and LM393 comparator as shown in Fig. 1. The probes help in
detecting the soil moisture, and LM393 comparator helps in comparing the threshold
value set by the potentiometer which is present in circuit. The sensor can be connected
in 2 modes: analog and digital modes. The input voltage is 3.3–5 V, and output voltage
is 0–4.2 V. The input current is 35 mA. It has four pins: power pin (VCC), ground
(GND), analog output (AO), and digital output (DO). When moisture content in soil
is more, the probes conduct because of low resistance between probes; when the
moisture content is less, the probes do not conduct because of high resistance. Thus,
the probes help in detecting the moisture content in soil.

2.2 GSM Module (SIM 800A)

SIM800A is a quad-band GSM module with RS232 interface shown in Fig. 2. With
low power consumption voice, SMS and data information can be sent. Using AT
commands, one can be able to communicate and monitor the appliances.

2.3 12 V SPDT Relay

A 12 V SPDT relay is considered when one circuit is needed to be turned on/off

shown in Fig. 3. Basically, it consists of a solenoid which acts when it is energized.
It consists of three terminals, namely normally open (NO), normally close (NC), and
common (COM).
844 A. Karthik et al.

Fig. 2 GSM module (SIM

800A)

Fig. 3 12V SPDT relay

2.4 Electrical Load

Electric motor is a device which converts electrical energy into mechanical energy. A
230 V single-phase bulb is considered instead of agricultural pump for demonstrating
the operating principle of proposed project.

2.5 Two Node MCU

It is a firmware for the ESP8266 as shown in Fig. 4. It is a System on Chip (SoC)

that integrates all the components of computers or other electronic systems. It is used
to connect the electrical and mechanical systems to the Internet in order to control
them.
IoT-Based Automatic Irrigation Control 845

Fig. 4 Two Node MCU

3 Architecture and Working of Proposed System

Figure 5 provides the information about the architecture of the proposed system. The
proposed system consists of power unit, two ESP 8266 modules (one module is used
for sensing of input power which is required to ON the pump in the field and another
module for main controller unit), a 5 V (DC) SPDT relay, 2 selection switches (one
for manual mode operation and another for automatic mode operation), one 230 V
single-phase 50 Hz bulb considered as water pump and a soil moisture sensor to
sense the moisture content in soil.
When the system is activated at first the Node MCU, the controller checks the
availability of power. If the power is available, the module sends information to user

Fig. 5 Block diagram of the proposed system

846 A. Karthik et al.

Fig. 6 Power supply

detection in mobile
application

stating “Power is activated” through GSM and status is also displayed in the mobile
application as shown in Fig. 6.
Then, the user will select the mode of operation either manual or automatic. This
depends upon the type of crop cultivated by user. These modes can be selected by
using mobile application as shown in Fig. 7 and also by using selection switches
which are located at fields as shown in Fig. 8 (The selecting switches will act as
alternating option to select the mode of operation for user). The two selector switches
are used for initializing manual and automatic modes by farmer/user itself, when there
is no availability of mobile or if any error caused in the IoT modules.
Here, one selecting switch is for a manual and another for automatic operations.
In manual mode, the water pump will be in ON condition as per user application in
the field and in automatic mode water pump will ON or OFF as per the water content

Fig. 7 Automatic and

manual mode selection
through mobile application
IoT-Based Automatic Irrigation Control 847

Fig. 8 Automatic and manual mode selection switches in the field

Fig. 9 Power control to the load in automatic mode

in the field which is measured using soil moisture sensor (The value of water content
can be adjusted as per the required crops cultivation) as shown in Fig. 9.
The selection of manual mode and automatic mode can be done by using both
mobile application or by using selecting switches as per the user convenient. The
system will also provide the status of the soil moisture value and the water pump
condition through mobile application, which will help to monitor the condition of
the field. If in case main power is disconnected in the middle of operation or power
not available at the field, user will be updated though SMS as the “Power system is
deactivated” and also displayed in mobile application as shown in Fig. 10.
The proposed system is capable of sending SMS during power failure with the help
of backup battery present in the system. If in case the user selects both the modes
using mobile application or selection switches (automatic and manual), the user
gets alerted by buzzer. The differences between proposed system and other existing
systems are: the proposed system can update the status of motor and moisture levels
in soil; it has two modes of operation which are selected depending on the type of
crop cultivated; less human intervention required; able to update user even when
power is not available at the field through GSM and mobile application. The entire
proposed system is shown in Fig. 11.
848 A. Karthik et al.

Fig. 10 Power supply cutoff detection

Fig. 11 Hardware implementation of the proposed system

4 Conclusion and Future Work

The designed system can be programmed in such a way that the farmer can select
the type of crop that is cultivated, and according to it, the water is supplied to crops.
The amount of nutrients required for the crop can be also found, and accordingly,
nutrients are also supplied automatically. This can be achieved by collecting required
data from the irrigation department.
IoT-Based Automatic Irrigation Control 849

References

1. Liu, J., Chai, Y., Xiang, Y., Zhang, X., Gou, S., Liu, Y.: Clean energy consumption of power
systems towards smart agriculture: roadmap, bottlenecks and technologies. CSEE J. Power
Energy Syst. 4(3), 273–282 (2018). https://doi.org/10.17775/cseejpes.2017.01290
2. Daskalakis, S.N., Goussetis, G., Assimonis, S.D., Tentzeris, M.M., Georgiadis, A.: A uW
backscatter-morse-leaf sensor for low-power agricultural wireless sensor networks. IEEE Sens.
J. 18(19), 7889–7898 (2018). https://doi.org/10.1109/JSEN.2018.2861431
3. Lucas Gomes Salmento, M., Camponogara, A., Manhaes de Andrade Filho, L., Vidal Ribeiro,
M.: A novel synchronization scheme for impulsive UWB-based PLC systems. IEEE Lat. Am.
Trans. 15(11), 2050–2058 (2017). https://doi.org/10.1109/tla.2017.8070407
4. Czajkowski, K., Fitzgerald, S., Foster, I., Kesselman, C.: Grid information services for dis-
tributed resource sharing. In: 10th IEEE International Symposium on High Performance Dis-
tributed Computing, pp. 181–184. IEEE Press, New York (2001). https://doi.org/10.1109/hpdc.
2001.945188
5. Viani, F., Bertolli, M., Salucci, M., Polo, A.: Low-cost wireless monitoring and decision support
for water saving in agriculture. IEEE Sens. J. 17(13), 4299–4309 (2017). https://doi.org/10.
1109/jsen.2017.2705043
6. Jagüey, J.G., Villa-Medina, J.F., López-Guzmán, A., Porta-Gándara, M.Á.: Smartphone irri-
gation sensor. IEEE Sens. J. 15(9), 5122–5127 (2015). https://doi.org/10.1109/jsen.2015.
2435516
7. Maheswararajah, S., Halgamuge, S.K., Dassanayake, K.B., Chapman, D.: Management of
orphaned-nodes in wireless sensor networks for smart irrigation systems. IEEE Trans. Signal
Process. 59(10), 4909–4922 (2011). https://doi.org/10.1109/tsp.2011.2160258
8. Vaishali, S., Suraj, S., Vignesh, G., Dhivya, S., Udhayakumar, S.: Mobile integrated smart
irrigation management and monitoring system using IOT. In: International Conference on
Communication and Signal Processing (ICCSP), pp. 2164–2167. Chennai (2017). https://doi.
org/10.1109/iccsp.2017.8286792
9. Namala, K.K., Prabhu, AV, K.K., Math, A., Kumari, A., Kulkarni, S.: Smart irrigation with
embedded system. In: IEEE Bombay Section Symposium (IBSS), Baramati, pp. 1–5 (2016).
https://doi.org/10.1109/ibss.2016.7940199
10. Ghosh, S., Sayyed, S., Wani, K., Mhatre, M., Hingoliwala, H.A.: Smart irrigation: a smart drip
irrigation system using cloud, android and data mining. In: IEEE International Conference
on Advances in Electronics, Communication and Computer Technology (ICAECCT), Pune,
pp. 236–239 (2016). https://doi.org/10.1109/icaecct.2016.7942589
11. Pernapati, K.: IoT based low cost smart irrigation system. In: Second International Confer-
ence on Inventive Communication and Computational Technologies (ICICCT), Coimbatore,
pp. 1312–1315 (2018). https://doi.org/10.1109/icicct.2018.8473292
12. Rau, A.J., Sankar, J., Mohan, A.R., Das Krishna, D., Mathew, J.: IoT based smart irrigation sys-
tem and nutrient detection with disease analysis. In: IEEE Region 10 Symposium (TENSYMP),
Cochin, pp. 1–4 (2017). https://doi.org/10.1109/tenconspring.2017.8070100
Adaptive Controller Design for SMIB
System Using Sliding Mode Control

Subhranshu Sekhar Puhan, Sobhit Panda, and Animesh Kumar

Abstract Sliding mode control (SMC) is one of the robust control techniques with
discontinuous control action. The fixed high gain in conventional SMC may lead to
high-frequency oscillations in the control input known as chattering. The chattering
can excite the unmodeled dynamics of the system and results in instability. Adaptive
sliding mode control is one of the methods to attenuate the effect of chattering in
which the gain of controller is changed based on the error dynamics. In advance
SMC requires the knowledge of the time derivative of the sliding variable, which
is discontinuous, so it is not measurable. This paper proposes an modified adaptive
sliding mode controller based on reachability condition of sliding mode control,
which can reduce the gain even in the reaching phase of the sliding mode, and also
the gain is varied according to the estimated disturbance using a disturbance observer.
The novelty lies to analyze the disturbances in the input dynamics and estimate the
disturbance through a Leunberg observer design. The proposed control action is
then applied to a single machine infinite bus, having constant voltage and frequency
(SMIB) power system having DC1A exciter and power system stabilizer associated
with it to evaluate the performances of the controller. At last, numerical simulation
results are presented for validation purpose.

Keywords SMC · Adaptive SMC · Nonlinear control · SMIB · Disturbance

observer nonlinear control · Chattering

S. S. Puhan (B) · A. Kumar

Department of Electrical Engineering, Siksha ‘O’ Anusandhan (Deemed to be University),
Bhubaneswar, Odisha, India
e-mail: [email protected]
A. Kumar
e-mail: [email protected]
S. Panda
Department of Electronics Instrumentation, CET, Bhubaneswar, Odisha, India
e-mail: [email protected]

© Springer Nature Singapore Pte Ltd. 2020 851

G. Pradhan et al. (eds.), Advances in Electrical Control and Signal Systems,
Lecture Notes in Electrical Engineering 665,
https://doi.org/10.1007/978-981-15-5262-5_64
852 S. S. Puhan et al.

1 Introduction

Sliding mode controller (SMC) is having a nonlinear control strategy that can deal
with the vulnerabilities in the systems. The control activity in SMC is irregular which
powers the system to take after the coveted reaction. The principle favorable position
of SMC is power against parameter vulnerability and results in decreased request
progression. Also, SMC has a high level of adaptability in its outline decisions
and the control strategy is generally simple to actualize when contrasted with other
nonlinear control techniques. Such properties make SMC exceptionally reasonable
for applications in nonlinear systems, and it is actualized in modern applications, for
example, electrical drives, car control, process control, and so forth.
The outline system of SMC can be part into two goals. (i) Design of sliding
exterior to such an extent that shut circle system is steady on this surface. (ii) Design
of an intermittent controller which powers the system to achieve the sliding surface
in a limited time remains there in entirely future time (positive invariance). Sliding
surface is to be outlined ideally to fulfill all limitations and required particulars.
The SMC has two stages. The underlying stage amid which the state direction is
coordinated towards the sliding surface is called as “achieving stage,” and the span
in which the state direction moves towards the source along the sliding surface is
called as “sliding stage.” Amid the achieving stage, the shut circle system is touchy
to aggravations and thus the most critical assignment is to outline a controller that
drives the system directions to the sliding surface in a limited time and keep up
it at first glance even within the sight of vulnerabilities and unsettling influences.
Henceforth, SMC is uncaring to coordinated vulnerabilities and aggravations on the
sliding surface and the shut circle system takes after a lessened request flow. If there
should arise an occurrence of ordinary SMC, the pickup of irregular control activity is
settled and the execution of traditional SMC continuously is influenced by the marvel
known as “prattling” which is because of the high recurrence broken control activity
close to the sliding surface. Prattling is unfortunate since it can energize unmodeled
system elements and prompt precariousness of the system. Numerous techniques
have been created to diminish prattling in SMC which incorporate limit layer control
and higher request sliding mode control (HOSM). Sliding mode controller design
objective generally lies in twofold (i) To select a sliding surface (ii) The trajectories
of the closed-loop controller will follow the reaching phase of controller in finite time
and be there is discussed [1–3]. Most of time during the controller design the value of
controller gain is high to overcome effect of disturbances, but by doing this we have to
compromise for chattering; again, high gain chattering is added when the closed-loop
control trajectories have to reach the reaching phase S(x) = 0. Due to high-frequency
dynamics and excitement because of chattering, the control input should be free
from chattering as discussed in [4]. The problem of finite-time-based constrained
stabilization with uncertainties is discussed in [8]. To avoid high-frequency chattering
previously researcher has followed various methods such as designing controller
based on boundary layer [5], appropriate switching function choosing [6, 7], higher-
order sliding mode control (HOSMC) [8] and adaptive SMC design with adjustable
Adaptive Controller Design for SMIB System … 853

controller gain [9, 10]. Increasing the gain in a small range around the switching
region is provided in [10]. The sliding mode controller application for closed-loop
tracking control of a DC machine along with improved chattering performances is
discussed in [10].
In this work, we have proposed an adaptive SMC and gain of the SMC is adapted
according to error minimization of estimated disturbance, with the help of linear
observer design. The novel technique is next applied to a SMIB power system having
nonlinear exciter and also having power system stabilizer. The rest of the paper
is organized as follows: Sect. 2 represents the design of existing adaptive sliding
mode controller, and Sect. 3 deals with observer design for input disturbances in the
controller. To validate the proposed controller, simulation results are explained in
Sect. 5 that followed by a conclusion in Sect. 6.

2 Existing Adaptive SMC

A noteworthy disadvantage of customary SMC is settled pickup which is gotten

through the bound of disturbances. This may bring about overestimation and prompts
chattering and may lead to chattering. The chattering could energize the unmodeled
flow of the system and may prompt unstability. The chattering can be diminished
by fluctuating the pickup of SMC in light of the immediate disturbances. In the
following area, diverse kinds of adaptive sliding mode controller of this specific
nature are talked about.

2.1 Design of Adaptive SMC

Consider a nonlinear signal input of any system

ẋ = f (x) + g(x) ∗ u + d (1)

where x is the state vector, u is the control input, and d is the external disturbance.
The functions f (x) and g(x) are the control inputs which are drifted, so the exact
value is not known. Equation (1) can be written as

ẋ = ( f (x) + ∂ x) + (g(x) + ∂ x) ∗ u + d (2)

where in f (x) and g(x) are nominal parts of the system and the variant in f (x) and
g(x) are parameter uncertainties.
Let us consider, there exist a sliding Surface-

S(x) = R n → R (3)
854 S. S. Puhan et al.

The sliding function described in Eq. (3) in such a manner that the system
described in (1) is stable in the switching manifold. The objective of the controller
design is to determine a control input such that to reach the switching manifold and
stay there for infinite time, which can be achieved by the help of Lyapunov condition
such as

S Ṡ = −β(t)|s| (4)

This equation will provide negative energy function, and absolute stability is
achieved according to Lyapunov theory.
Let us define a control input as

u ≡ u0 + us (5)
 
∂S −1 ∂ S
u0 = − (g0 (x)) ( f 0 (x)) (6)
∂x ∂x
 
∂S −1
us = − (g0 (x)) K sign(s(x)) (7)
∂x

Generally, in case of SMC design, the gain of the controller is adjusted in such a
way that it may compromise either with robustness or with chattering. So the best
method is to adopt the value of K to avoid either underestimation or overestimation.

3 Observer Design for Disturbance Estimation

Estimating the unknown disturbances in the control input is quite challenging task.
Designing a Leunberg observer for estimation purpose will really help the purpose.
Estimating the unknown disturbance d, which itself a function of state variable,
control input and disturbance in such a fashion that controller gain K will be adjusted
itself to satisfy Lyapunov condition.
Consider the dynamics of sliding function such as

S = d + us (9)

Again let the observer designed have dynamics such as

ṗ = l(l Ṡ − p + u s ) (10)

d̃ = l S − p (11)

d = l(d + u s ) − l(S − p + u s ) (12)

Adaptive Controller Design for SMIB System … 855

d = ld − l d̃ = −le (13)

than the dynamics of error will be

ė = le − l d̃ (14)

By choosing the higher value of observer gain ‘l’, the error dynamics will reduce
to ‘0’, iff and only if the estimated error magnitude reduces to ‘0’.

4 Adaptive SMC for SMIB System

A SMIB system having nonlinear exciter can be modeled as follows with the help
of exact linearization design approach

δ̇ = ω − ω0 (15)

ω0 ω0 D ω0 E q V f
ω̇ = PH + PM L0 − (ω − ω0 ) −   sinδ (16)
H H H H Xd

Xd 1 X d − X d 1 dωr 1
Ė q =   E q +   V f cos δ + V f + K Stab (17)
Xd TD0 TD0 Xd TD0 dt TD
d Eq
V f = Eq + =u (18)
dt
where the angle between EMF and terminal voltage of the synchronous generator,
the rotor speed of the generator, and voltage behind the transient reactance are δ,
ω, ω0 , Eq, respectively. PM L0 and PH are the turbine lower limit and higher limit,
and stabilization constant is represented by K Stab . The value of stabilization constant
should be wisely declared to avoid high-frequency attenuation.
Here, in this paper, we have considered a power system stabilizer (PSS) to add
damping to generator rotor oscillation for excitation voltage control. The state space
modeling of the controller is shown below
 T
[ẋ] = E q δω (19)

ẋ = f (x) + g(x) ∗ u + d (20)

⎡ d −X d
 ⎤
1
TD0
E q +1 X
TD0 X d
V f cos δ + K Stab ωr
⎢ ω0 ⎥
f (x) = ⎣ P
H M L0
D E V
− H (ω − ω0 ) + q X f sinδ ⎦ (21)
d
ω − ω0
856 S. S. Puhan et al.

⎡ ⎤
1
⎢ ⎥ TD0
[g(x)] = ⎣ 0 ⎦ (22)
0

where H represents inertia constant of the synchronous generator, D specifies damp-

ing coefficient associated with the machine, X d is the direct axis transient reactance,
and X s specifies for steady-state reactance.
The equivalent state space modeling of the SMIB system has done with assuming
the state of the system as represented in Eq. 19.
Time constant of the SMIB system has very significant role in the modeling
process; high value of time constant will have less settling time but with settled at
any unwanted value.

K Stab ωr
ẋ1 = f (x1 ) + g(x1 ) ∗ u + (23)
TD TD

ẋ2 = x3 (24)

d D
ẋ3 = f (x2 ) + g(x2 ) ∗ u − x2 (25)
dt H
The switching function and sliding function have been taken according to the
significance as discussed previously.

S = C1 ∗ x1 + C2 ∗ x2 + C3 ∗ x3 (26)

u switching = K t sign(S) (27)

The uncertainty in the control input can be balanced by regulating the value of
K t , which adapts itself by the error dynamics.

5 Numerical Results

This section represents the validation results of designed adaptive SMC for SMIB
system having a nonlinear exciter and a power system stabilizer associated with it.
The following data have been taken for simulation purpose D = 3.1 sint + 3t + 5 pu,
ωr = t + 0.98 pu, and other data are in Table 1.

Table 1 Numerical data for validation

 
H TD0 Xd Xd ω0 E q0 C1 C2 C3
10 5s 0.962 pu 0.344 pu 314 rad/s 1.405 pu 10 1 0.5
Adaptive Controller Design for SMIB System … 857

The simulation results of all state variables along with controller gain are shown
above. The controller gain is adapted itself by minimizing the error in dynamics,
which can be possible by observer gain analysis.
From the above simulation results, it is conforming that all dynamics of state
variable are approaching to zero as time progress and the controller gain ‘K’ adapted
accordingly to maintain the reachability condition and the controller input should be
within the manifold. In an observer-based controller design, change in state variable
amplitude will be less and hence less chattering.

6 Conclusion

This paper discusses the estimation of the disturbances and also tuning of the value
of controller gain ‘K’ such that switching manifold is reached in finite time and stays
there for longer. The same technique is applied to a SMIB power system with non-
linear DC1A exciter modeling along with (PSS), and the dynamics of state variable
are approaching to zero with lesser chattering as per the specified objective. The
same techniques can be applicable to multimachine system angle stability analysis
and voltage stability analysis for power system.
858 S. S. Puhan et al.

References

1. DyLiacco, T.E.: The adaptive reliability control system. IEEE Trans. Power Appar. Sys. 86,
517–531 (1967)
2. Young, K.D., Utkin, V.I., Ozguner, U.: A control engineers guide- to, sliding mode control.
IEEE Trans. Control Syst. Technol. 7(3), 328–342 (1999)
3. Utkin, V.I.: Sliding Modes in Control and Optimization (original Russian language ed.).
Springer, Berlin (1992)
4. Bandyopadhyay, B., Deepak, F., Kim, K.: Sliding Mode Control using Novel Sliding Surfaces.
Springer, Berlin (2009)
5. Li, S., Yang, J., Chen, W.-H., Chen, X.: Disturbance Observer Based Control: Methods and
Applications. CRC Press, Boca Raton (2014)
6. Edwards, C., Spurgeon, V.: Sliding Mode Control: Theory and Applications. Taylor Francis,
Abingdon (1998)
7. Boiko, I., Fridman, L., Pisano, A., Usai, E.: Analysis of chattering in systems with second-order
sliding modes. IEEE Trans. Autom. Control. 52(11), 20852101 (2007)
8. Chan, S.P.: A disturbance observer for robot manipulators with application to electronic
components assembly. IEEE Trans. Industr. Electron. 42(5), 487493 (1995)
9. Yang, J., Chen, W.-H., Li, S.: Non-linear disturbance observer-based robust control for systems
with mismatched disturbances/uncertainties. IET Control Theory Appl. 5, 20532062 (2011)
10. Chang, F.-J., Twu, S.-H., Chang, S.: Tracking control of dc motors via an improved chattering
alleviation control. IEEE Trans. Industr. Electron. 39(1), 2529 (1992)
Classification of EEG Signal Using SVM

Sandhyalati Behera and Mihir Narayan Mohanty

Abstract Biomedical signals are basically contaminated with noise due to differ-
ent environments such as machine fault and position of physical change of human
subject. Since the signals are collected from the human subject are of physical func-
tionalities, the complexity is more to analyze. Artifact is one of such interference
and seems to be a signal with high amplitude. Detection of these artifacts can be
performed with the classification task. In this paper, authors are attempted to classify
the signals and artifacts by using neural networks to detect and remove. Though the
models of neural networks are used earlier for classification tasks only, detection of
artifacts is the novelty in this work. Different models like MLP, RBFN, SVM are used
and the results are compared. It is found that the cubic SVM outperforms all other
models. Time-domain and frequency-domain features have been obtained to feed to
the model. Further, the combination of features is made as another novel technique
which provides better accuracy as compared to only either time-domain features or
frequency-domain features in cubic SVM model. The accuracy found 95.1% and is
given in the result section.

Keywords EEG-electroencephalogram · EOG-electrooculogram · Artificial neural

network · SVM-support vector machines

S. Behera
Department of Electrical and Electronics Engineering, Institute of Technical Education and
Research, Siksha ‘O’ Anusandhan (Deemed to be University), Bhubaneswar, Odisha 751030,
India
e-mail: [email protected]
M. N. Mohanty (B)
Department of Electronics and Communication Engineering, Institute of Technical Education and
Research, Siksha ‘O’ Anusandhan (Deemed to be University), Bhubaneswar, Odisha 751030,
India
e-mail: [email protected]

© Springer Nature Singapore Pte Ltd. 2020 859

G. Pradhan et al. (eds.), Advances in Electrical Control and Signal Systems,
Lecture Notes in Electrical Engineering 665,
https://doi.org/10.1007/978-981-15-5262-5_65
860 S. Behera and M. N. Mohanty

1 Introduction

Imaging techniques of the human body are electrocardiography (ECG, heart),

electromyography (EMG, muscular contraction), magnetoencephalography (MEG,
brain), electrogastrography (EGG, stomach), electroencephalography (EEG, Brain),
electrooculogram (EOG, eye dipole field) [1]. In this article, EEG signal is
considered.
Generally, The amplitude of the EEG signal is small and in the range of microvolt.
The EEG signal undergoes different types of artifacts during the time of its recording.
Due to artifact, the shape of the original signal changes, and the important information
in EEG may be lost. EEG artifacts are maybe generated from the external source
or maybe from the human body [1–3]. So, artifact detection and removal play an
important role in EEG analysis.
Different methods such as time-domain methods and frequency-domain methods
are used for the detection and removal of artifacts from EEG signals [4–6]. The time-
domain methods such as ICA, RLS (recursive least square) with the help of adaptive
filtering used to take apart the artifacts from the brain signal. The adaptive filtering
method estimates the interference in the brain signal and is subtracted from the EEG
signal to get clean EEG [7–10]. To estimation of interference requires EOG reference
channel. The analysis of EEG signal becomes very difficult due to the presence of
artifact. So, it is necessary to remove the artifacts from EEG records. The statistical
techniques with the help of wavelet transform used for the removal of artifacts [9,
11–13]. The modified multiscale entropy [14], sample entropy, kurtosis are some
statistical techniques used for detecting artifacts. To increase the processing speed
the dimension of EEG signal is reduced with the help of ICA, PCA, Wavelet ICA.
The process of artifact removal includes artifact identification and artifact removal
without distorting the original signal of interest. The simple filtering and straight
forward signal processing methods fail to remove artifact due to overlapping of arti-
facts with EEG signal both in time and spectral domains [15]. In adaptive filtering,
the desired output compared with observed output [16]. The Empirical mode decom-
position method uses fractional Gaussian noise (fGn) as a reference signal to detect
the distinguishing feature of EOG signal. The Ocular artifacts are also removed by
spatial decomposition methods in [17].
The Linear Discriminant Analysis (LDA), Support vector machine (SVM), Artifi-
cial Neural Network (ANN), Nearest neighbor classifiers are used for classification of
EEG signals [18]. Out of these classifiers, some of the classifiers are linear and some
are nonlinear. The nonlinear classifiers give better result in the case of nonstationary
and noisy environment. There are many applications of EEG signal classification.
The artifact detection is one of them, etc. The feature extraction performs an essential
part in classification. The performance of a classifier fully depends on the selection
of appropriate features [15]. The features extracted from the wavelet transform and
classified with the help of an artificial neural network (ANN) and logistic regression
(LR) gives effective results. The features extracted from the coefficients of lifting
based wavelet transform (LBDWT) increases the computational speed [15, 19]. The
Classification of EEG Signal Using SVM 861

adaptive classification algorithms also used for the classification of EEG signals.
The weighted distance nearest neighbor (WDNN) classification algorithm, assigns a
weight to each training sample. The features extracted from the independent compo-
nents and classified with the help of the Bayesian classifier used to detect artifactual
EEG signals [20–23].
Support vector machine is an efficient popular classifier as compared to other
machine learning classifiers and is used since some decades. The recent work based
on this model is cited as follows.
The nonlinear SVM classifier model used for automatic seizure detection in [24].
The entropy derived from a discrete wavelet transform is used as a feature to detect the
seizure in EEG signal. For classification of a mental task such as thinking backward,
forward, left and right the features extracted from the wavelet transform are used [25].
The features extracted from the wavelet transform are classified with an accuracy of
75% using an SVM classifier. The fractures extracted from time-domain, spectral-
domain, and wavelet domain are classified using SVM classifier for detection of
seizure from the nonseizure EEG signals. The accuracy of the SVM classifier is
compared with the LDA and QLDA classifier. It is given that SVM classifier gives
an accuracy of 97.05% which is more than the accuracy of LDA and QLDA classifier
[26].
However, this model is used nowhere yet for the application of artifact detection
in biomedical signals. Therefore, the authors have taken the approach to detect the
artifact from EEG signal by modifying the kernel.
In this paper, a brief overview of artifact detection, removal, and classification
given in Sect. 1. The features used for classification purposes are given in Sect. 2.
The classifier taken are given in Sect. 3. Results obtained are given in Sect. 4 and
Sect. 5 is about the conclusion and future work.

2 Methods

The proposed method uses both time and frequency-domain features for the classifi-
cation. For the classification, SVM classifier is taken. To obtain the features database
used in [27] is considered. This EEG database has four types of datasets these are
pure EEG, EEG with horizontal eye movement artifact (HEOG), EEG with vertical
eye movement artifact (VEOG), and the EEG with both vertical and horizontal eye
movement artifacts. It is clearly visible from Fig. 1. That the clean EEG signals taken
for the proposed method are of 6001 samples and the amplitude ranges in between
−60 and 60 µV. The contaminated EEG signal taken is of 6001 samples and is clearly
shown in Fig. 2. The amplitude of the contaminated EEG signal ranges between
−50 and 200 µV. The amplitude of the EEG signal in the presence of artifact becomes
high.
862 S. Behera and M. N. Mohanty

Fig. 1 Plot of pure EEG signal

Fig. 2 Plot of contaminated EEG signal

2.1 Time-Domain Feature

For extracting features both time and frequency-domain features are considered. The
time-domain features are obtained from the cross-correlation of EEG signals [28].
The cross-correlation Y X 1 X 2 between two signal X 1 and X 2 is given by Eq. (1).

−n−1
N
Yx1 x2 (n) = x1m−1 x2m n ≥ 0 (1)
m=0

= Yx2 x1 (−n) n < 0

where N is the length of x1 or length of x2 and m is the shift parameter. Here x1

and x2 are considered as two EEG signals. For cross-correlation one EEG signal
is taken as reference and the reference EEG signal is correlated with another EEG
signal. This procedure is applied to contaminated signals also. The cross-correlation
Classification of EEG Signal Using SVM 863

Fig. 3 Plot of cross co-relation of two pure EEG signal

Fig. 4 Plot of cross co-relation of two contaminated EEG signal

of pure EEG and contaminated EEG is given in Figs. 3 and 4 respectively. It is clearly
visible in Fig. 3 that the peak occurs at 6000 sample and the peak value is high. For
the contaminated signal the peak occurs near to 10,000 samples. The peak value
obtained from the cross-correlation of two pure EEG signal is higher as compared to
the peak value obtained from the cross-correlation of two contaminated EEG signal.
Features derived from cross-correlation Y (n) are as follows:
1. Peak value (PV): Peak value is the peak value of the cross-correlation.
2. Ins value: Instant value corresponding to peak value.
3. Centroid: Centroid is calculated by Eq. 2 which is given below
864 S. Behera and M. N. Mohanty


M
n · Y (n)
n=1
Centroid = (2)

M
Y (n)
n=1

where the length of cross-correlation is M.

4. EW: Equivalent width is calculated by Eq. 3 as given below


M
Y (n)
n=1
EW = (3)
pv

5. MSA: Mean Square Abscissa is calculated by Eq. 4.


M
n 2 Y (n)
m=1
MSA = (4)
M
Y (n)
n=1

For the time-domain feature after cross-correlation first a pure EEG signal is taken
as a reference after that the reference signal is correlated with another pure EEG
signal. Similarly, one contaminated EEG is taken as a reference it is cross-correlated
with another contaminated EEG signal.

2.2 Frequency-Domain Features

First of all, the DFT of the Pure and Contaminated signal is obtained. After calcu-
lation of DFT of EEG segments, the total power of each frequency band and whole
frequency bands are calculated.
The DFT of the EEG signal is calculated by Eq. 5.

N −1
 j2πkn
X (k) = x(n) · e k k = 0, 1, 2, 3, . . . , N − 1 (5)
n=0

From Eq. (5), x(n) is considered as EEG signal and X (k) is its DFT and N
represents the sample size. Features are the ratios of each sub-band power to the
whole band (0–40 Hz) power. So, for five sub-bands five features are obtained. In the
proposed method the frequency range for different sub-bands are taken as follows:
Delta = 1–4 Hz, Theta = 4–8 Hz, Alpha = 8–13 Hz, Beta = 13–25 Hz, Gamma =
25–40 Hz.
Classification of EEG Signal Using SVM 865

For each sub-band features are calculated. For sub-band delta, the feature cal-
culated is prdelta, for theta feature is prtheta, for alpha feature is pralpha, for beta
feature is prbeta and for gamma-band the feature is prgamma. For a sub-band the
power ratio is calculated by Eq. (6) is given by

PT
prtheta = (6)
TP
where PT is the power in theta sub-band and TP is the total power in whole band.
Similarly, the power ratio in other bands is calculated by using Eq. (6).

3 Detection Models

In the proposed method, for the detection of EEG artifacts SVM classifier, MLP
classifier and RBFN classifier are used. The classifiers used for this proposed work
are supervised classifiers. The EEG signal having artifact is assigned with one level
and the clean EEG signal is assigned with another level.

3.1 MLP Classifier

Multilayer Perception (MLP) is a supervised multilayer feedforward artificial neural

network model. The structure of MLP is shown in Fig. 5. It has an input layer to
receive the features, an output layer for prediction, and makes a decision about the
input. The middle layer is called a hidden layer. Each neural in hidden layer takes
the weighted sum of input from the previous layer and transforms the input through

Fig. 5 Structure of MLP

classifier
866 S. Behera and M. N. Mohanty

an activation function. The activation function maps the input-output relations in

nonlinear way. MLP classifier assigns each feature input to a class level.

3.2 RBFN Classifier

RBFN divides the feature to the training and testing set. To classify a new input,
the RBFN classifier computes the Euclidean distance between new input and its
training prototype. The training vectors are stored in RBF neuron and each RBF
neuron compares the input vector with the stored training vector. Each output node
computes the weighted sum of values from every RBF neuron. A class is assigned
depending on the weighted sum. If the weight is positive, then the FBF neuron
belongs to its category and for negative weight, the RBF neuron belongs to others.

3.3 SVM Classifier

SVM classifier tries to separate the input space with the help of a hyperplane. There
are many possible hyperplanes chosen to separate data points into two classes. The
main objective is to choose a hyperplane that has maximum distance between the
data points of two classes. From the number of features, the dimension of hyperplane
is obtained. If the feature has two classes, then SVM used is called Binary SVM
[29, 30].
Depending on the hyperplane used, SVM can be considered is of two types: Linear
SVM and nonlinear SVM. Linear SVM predicts a hyperplane to be a straight line
but for nonlinear SVM the hyperplane is other than a straight line.
For a two-class linearly separable classifier problem the hyperplane is found by
Eqs. (6) and (7)

ai · xi + b ≥ +1 if yi = +1 (6)

ai · xi + b ≤ +1 if yi = −1 (7)

where xi is the feature vector and yi is a class for ith the feature. ai is a weight normal
to the hyperplane and b is used as bias.
The distance margin to the hyperplane found by Eq. (8)

w · xi + b ≤ ±1 (8)

The margins are determined by the input vectors which are called support vectors.
If the input vectors are not linearly separable, it is transformed into the linearly
Classification of EEG Signal Using SVM 867

separable case with the help of kernel function. In the proposed work, linear SVM,
quadratic SVM, and cubic SVM classification models are taken for classification.

4 Results and Discussion

The classification of features is performed by using MATLAB2015a. The time-

domain and frequency-domain features are classified by using the SVM classifier.
The time-domain features are obtained from the cross-correlation are given in Table 1
and the frequency-domain features are given in Table 2.
For the proposed classification work, five time-domain features and five
frequency-domain features are considered. Time-domain features are represented
as TD features and frequency-domain features are represented as FD features. TD
features and FD features are classified individually. Combining time-domain and
frequency-domain features which are known as TDFD features also classified. Accu-
racy obtained through different classifiers with varioation of features is presented in
Table 3.
For the classification, three types of models are considered in the proposed method.
With TD features, cubic SVM classifier shows better accuracy than MLP and RBFN
whereas, with FD feature the RBFN shows better accuracy. However, combined
features outperform in cubic SVM model.

Table 1 Features obtained in time-domain

Features T1 T2 T3 T4 T5
Feature name PV Ins. value Centroid EW MSA

Table 2 Features obtained from frequency-domain

Features F1 F2 F3 F4 F5
Feature name prdelta prtheta pralpha prbeta prgamma

Table 3 Accuracy
Detection accuracy (%)
comparison among different
NN models Features MLP RBFN Cubic SVM
TD features 78.1 78.9 82.5
FD features 85.1 86.0 81.6
TDFD features 86.6 82.5 95.1
868 S. Behera and M. N. Mohanty

5 Conclusion

In this paper, the EEG signals are analyzed in detail. MLP, RBFN, and cubic SVM
models are successfully implemented for detection of artifact. The combined features
provide better results as compared to single type of features as well as use of cubic
SVM. In future, the artifacts are to be removed by use of suitable model to find the
clean signal for diagnosis. For further improvement, the choice of suitable features
is important future work.

References

1. Teplan, M.: Fundamentals of EEG measurement. Meas. Sci. Rev. 2, 1–11 (2002)
2. Aydemir, O.: Classifying various EMG and EOG artifacts in EEG signals. Prz. Elektrotech-
niczny 88, 218–222 (2012)
3. Parvinnia, E., Sabeti, M., Jahromi, M.Z., Boostani, R.: Classification of EEG signals using
adaptive weighted distance nearest neighbor algorithm. J. King Saud Univ. Comput. Inf. Sci.
26, 1–6 (2014)
4. Croft, R.J., Barry, R.J.: Removal of ocular artifact from the EEG: a review. Neurophysiol. Clin.
Clin. Neurophysiol. 30, 5–19 (2000)
5. Delorme, A., Sejnowski, T., Makeig, S.: Enhanced detection of artifacts in EEG data using
higher-order statistics and independent component analysis. Neuroimage 34, 1443–1449 (2007)
6. Dhiman, R., Saini, J., Priyanka, A.: Artifact removal from eeg recordings—an overview. In:
Proceedings of the NCCI, pp. 1–6 (2010)
7. Ghandeharion, H., Erfanian, A.: A fully automatic ocular artifact suppression from EEG data
using higher order statistics: Improved performance by wavelet analysis. Med. Eng. Phys. 32,
720–729 (2010)
8. Kumar, P.S., Arumuganathan, R., Sivakumar, K., Vimal, C.: A wavelet based statistical method
for de-noising of ocular artifacts in EEG signals. IJCSNS Int. J. Comput. Sci. Netw. Secur. 8,
87–92 (2008)
9. Mammone, N., La Foresta, F., Morabito, F.C.: Automatic artifact rejection from multichannel
scalp EEG by wavelet ICA. IEEE Sens. J. 12, 533–542 (2011)
10. Schlögl, A., Keinrath, C., Zimmermann, D., Scherer, R., Leeb, R., Pfurtscheller, G.: A fully
automated correction method of EOG artifacts in EEG recordings. Clin. Neurophysiol. 118,
98–104 (2007)
11. Behera, S., Mohanty, M.N.: A statistical approach for ocular artifact removal in brain signals.
In: 2018 2nd International Conference on Data Science and Business Analytics (ICDSBA),
pp. 500–503. IEEE (2018)
12. LeVan, P., Urrestarazu, E., Gotman, J.: A system for automatic artifact removal in ictal
scalp EEG based on independent component analysis and Bayesian classification. Clin.
Neurophysiol. 117, 912–927 (2006)
13. Mahajan, R., Morshed, B.I.: Unsupervised eye blink artifact denoising of EEG data with mod-
ified multiscale sample entropy, kurtosis, and wavelet-ICA. IEEE J. Biomed. Health Inform.
19, 158–165 (2015)
14. Wu, S.-D., Wu, C.-W., Lee, K.-Y., Lin, S.-G.: Modified multiscale entropy for short-term time
series analysis. Phys. A 392, 5865–5873 (2013)
15. Winkler, I., Haufe, S., Tangermann, M.: Automatic classification of artifactual ICA-components
for artifact removal in EEG signals. Behav. Brain Funct. 7, 30 (2011)
16. Molla, M.K.I., Islam, M.R., Tanaka, T., Rutkowski, T.M.: Artifact suppression from EEG
signals using data adaptive time domain filtering. Neurocomputing 97, 297–308 (2012)
Classification of EEG Signal Using SVM 869

17. Sameni, R., Gouy-Pailler, C.: An iterative subspace denoising algorithm for removing
electroencephalogram ocular artifacts. J. Neurosci. Methods 225, 97–105 (2014)
18. Lakshmi, M.R., Prasad, T., Prakash, D.V.C.: Survey on EEG signal processing methods. Int.
J. Adv. Res. Comput. Sci. Softw. Eng. 4 (2014)
19. Nandish, M., Stafford, M., Kumar, P.H., Ahmed, F.: Feature extraction and classification of
EEG signal using neural network based techniques. Int. J. Eng. Innov. Technol. (IJEIT) 2, 1–5
(2012)
20. Abootalebi, V., Moradi, M.H., Khalilzadeh, M.A.: A new approach for EEG feature extraction
in P300-based lie detection. Comput. Methods Programs Biomed. 94, 48–57 (2009)
21. Hossin, M., Sulaiman, M.: A review on evaluation metrics for data classification evaluations.
Int. J. Data Min. Knowl. Manag. Process. 5, 1 (2015)
22. Kocadagli, O., Langari, R.: Classification of EEG signals for epileptic seizures using hybrid
artificial neural networks based wavelet transforms and fuzzy relations. Expert Syst. Appl. 88,
419–434 (2017)
23. Subasi, A., Ercelebi, E.: Classification of EEG signals using neural network and logistic
regression. Comput. Methods Programs Biomed. 78, 87–99 (2005)
24. Raghu, S., Sriraam, N., Temel, Y., Rao, S.V., Hegde, A.S., Kubben, P.L.: Performance evaluation
of DWT based sigmoid entropy in time and frequency domains for automated detection of
epileptic seizures using SVM classifier. Comput. Biol. Med. 110, 127–143 (2019)
25. Caesarendra, W., Lekson, S.U., Agung, M.: Mental tasks EEG signal classification using
support vector machine. J. Infrastruct. Facil. Asset Manag. 1 (2019)
26. Nandy, A., Alahe, M.A., Uddin, S.N., Alam, S., Nahid, A.-A., Awal, M.A.: Feature extraction
and classification of EEG signals for seizure detection. In: 2019 International Conference on
Robotics, Electrical and Signal Processing Techniques (ICREST), pp. 480–485. IEEE (2019)
27. Klados, M.A., Bamidis, P.D.: A semi-simulated EEG/EOG dataset for the comparison of EOG
artifact rejection techniques. Data Brief 8, 1004–1006 (2016)
28. Iscan, Z., Dokur, Z., Demiralp, T.: Classification of electroencephalogram signals with
combined time and frequency features. Expert Syst. Appl. 38, 10499–10505 (2011)
29. Bayram, K.S., Kızrak, M.A., Bolat, B.: Classification of EEG signals by using support vector
machines. In: 2013 IEEE INISTA, pp. 1–3. IEEE (2013)
30. Kumar, A., Mohanty, M.N., Routray, A.: Design of support vector machines with time frequency
kernels for classification of EEG signals. In: 2010 IEEE Students Technology Symposium
(TechSym), pp. 330–333. IEEE (2010)
Dynamic Selection of the Best Server Out
of Multiple Servers Available Using Bully
Algorithm in Any Geographical Area

Kartikeya Shukla and Bhaswati Sahoo

Abstract This paper is about selecting the best server out of many servers available
for a website. In the present scenario, the number of users is increasing day by day
on the Internet and the services provided by this website are facing heavy traffic.
Services like downloading and uploading fail due to many reasons like geographical
location, heavy traffic or server not responding. Sometime even the best server might
fail to provide the service. This can be reduced by implementing bully algorithm
of the distributed algorithm. This paper provides an idea and extended use of this
algorithm in various scenarios. This algorithm deals with a large number of servers
that are working and providing the best server (depending upon the priority and
working of the server) out of multiple servers, which is similar to electing a leader
in an election. This algorithm can be programmed and computed individually on
each server. It can save time and even reduce the work load on each server. It can
further be implemented in routers, printers and other distributed computing devices.
The algorithm can be considered successful and is ready to be implemented on the
system.

Keywords Bully algorithm · Distributed algorithm · Geographical · Routers ·

Programmed · Traffic

1 Introduction

Distributed algorithm [1] can be defined as an algorithm which is used to run on a

distributed network of computer systems, i.e. each individual system is independent
of other systems and communicates via only communication networks. Each proces-
sor has its own memory. This algorithm is designed in such a way so that they can
run simultaneously on independent processors, with separate parts of algorithm and
having limited information about the other parts of the algorithm or system. Hence,
these are the sub-type of parallel algorithm, running concurrently. There are many

K. Shukla (B) · B. Sahoo

Kalinga Institute of Industrial and Technology, Bhubaneshwar, Odisha 751024, India
e-mail: [email protected]

© Springer Nature Singapore Pte Ltd. 2020 871

G. Pradhan et al. (eds.), Advances in Electrical Control and Signal Systems,
Lecture Notes in Electrical Engineering 665,
https://doi.org/10.1007/978-981-15-5262-5_66
872 K. Shukla and B. Sahoo

reasons which act as obstacle in this algorithm such as link failure, process failure,
unreliable network and timing synchronization. Hence, an appropriate algorithm is
necessary according to the requirements. Many algorithms that are used in distributed
computing require a coordinator that provides functions needed by other processes
in the system.
Usually, the network model consists of client–server, peer-to-peer or hybrid
approaches, but when there is heavy traffic on the web page many features like
downloading, uploading, streaming, etc., require large servers to maintain the work-
load of the traffic. Many approaches like cloud computing are offering access to the
remote servers. But even they require large servers and maintenance. There are many
servers that provide access to the data but they can be improved if there is an algo-
rithm that checks the user’s location from the server and current status of it, and in
some cases, if the server fails then it would be better to select the best server among
the available servers. Similarly, consider another scenario where a large organization
has many printer devices and some might fail to print a page. This can be sorted down
by using a popular distributed algorithm called bully algorithm [2], which is used to
select the best criteria independently on working condition. It is similar to the bully
election algorithm. The middleware of the servers can be programmed in order to
tackle these conditions and selecting the best option available at that moment.

2 Related Work

The bully algorithm has been implemented and modified in many scenarios of dis-
tributed computing. The modifications among these are reducing message passing
[3]. In this work, the efficiency of the algorithm is increased by reducing message
passing. Modified bully algorithm [4] implements an efficient approach to minimize
the redundancy in electing the coordinator and improving the effectiveness. Improved
bully election algorithm for distributed computing [5] is based upon reducing the
complexity of the message. The election in distributed computing [6] has considered
a scenario to restart the election of leader after a failure occurs in the nodes. Paral-
lel computing is used to compare the performance which is increased by using large
number of processors, and clock synchronization [7] has considered the implementa-
tion and improvement in various scenarios of increasing performance. Other works in
distributed computation, such as improved algorithm for decentralized algorithm [8]
for decentralized systems, use the technique of selective message passing and reduc-
ing the time complexity and electing good leader in parallel and distributed systems
[9]. An emergent and type of approach in this election algorithm is described and its
performance is evaluated. The design incorporates non-deterministic behaviour. This
resulting algorithm has very low communication complexity and is simultaneously
very stable, scalable and robust [10]. A more efficient approach has been described in
paper, an efficient overhead-aware leader election algorithm [11], which effectively
and efficiently selects the leader.
Dynamic Selection of the Best Server … 873

3 Implementation of Bully Algorithm

Consider a scenario in which there are six servers available with user present within
the range of these servers and priority of the server. When the user requests for
operation like downloading (large file), the web page must return the best working
server among these servers. The best working server can be described as the server
which provides continuous high speed, no drop-down connection.
Let Server 1 and Server 3 are out of service and Server 5 is too busy to respond;
see Fig. 1. Now we have to select and return the best server to the client, so we
select the best priority of the available servers. Our objective is to get the maximum
priority of the working server at any instant of time. Priority of the server here can be
defined as the value which can be calculated or considered from all the features like
(traffic on the server, geographical location from the client, performance in speed,
busy state, etc.) we mathematically calculate these factors and assign the priority
value of the individual servers. Each server does not know the priority of the other
servers (real-time scenario). If a server is in idle condition, then its priority can be
increased so that it will be working.
Just like in bully algorithm of leader selection in a large number of processes, here
same operation is going to happen but an extended implementation. Let us say when
client request from the master–server about file, it selects the maximum priority of
the server using this algorithm.

3.1 Constraints

If no servers are working, then we could never find the current best server, therefore
at least one server should be in working condition.

Fig. 1 Client–server
connection network
874 K. Shukla and B. Sahoo

3.2 Result with Explanation

The real-time working can be done through a situation in which the communication
of the server begins with three response states SELECT, NO and YES states. This
working can be implemented on heavy loaded servers also. This can be explained
through a particular scenario. Consider there are six servers and one master–server.
All these servers have been implemented with this algorithm. See Fig. 1. Let us
assume the priority values as
Server 1 = 10
Server 2 = 40
Server 3 = 50
Server 4 = 30
Server 5 = 80
Server 6 = 52
Master–Server = 10.
Note that the master–server is just like the other servers but the main difference
is that whenever a client requests and connects to the first random server, then that
server becomes a master–server and it is the duty of that master–server to find the
higher priority server greater than itself, within a timeout. If greater priority is found,
then redirect the client to that server else remain connected to the client.
These states can be a call to a defined function of the algorithm. Let us say
randomly Server 4 is called by the master–server, when the client had requested
the master–server. The Server 4 responds to every other available servers which
have greater priority by sending state signal SELECT. In response to this, the higher
priority server, like Server 2 and Server 6, responds YES back to Server 4 while
Server 1, Server 3 and Server 5 do not respond which have default value returned as
NO state. Hence, only Server 2 and Server 6 are left. Now randomly (master–server)
the Server 2 sends YES state to the greater priority value of the available servers. In
response, the lower priority servers and dead servers respond as NO, and the Server
6 responds back to Server 2 as YES.
Now the Server 6 responds as SELECT to every other servers, and when it does
not get any greater response (YES state) from any other server then it declares itself as
the greatest priority (bully) to all the servers and responds back to the master–server.
And master–server reconnects this Server 6 to the client.
It is also to be noted that the priorities of the servers are dynamic (randomly
changing) in nature, because these priorities depend upon various factors like client
location at any instance, workload on a server, heavy traffic, server failure, etc. Hence,
a timeout function can be called in a regular interval or whenever there is heavy load
on a server. A timeout is a function call which initiates this algorithm again, so as to
find out the best server in a given duration. If timeout is very low, then we could never
find the best server and it results in a longer waiting to client; whereas, if timeout
is very large, then we fail to achieve the best server in dynamic condition. Hence, a
precise average value should be taken, and one should monitor the changes in server
Dynamic Selection of the Best Server … 875

as the value of timeout changes and graph the best value of it. The server selected can
be taken in an organized manner through a technique of well-organized bully leader
election algorithm for distributed system [12]. For time constraint improvement, a
timer-based leader election algorithm [13] is suitable to be considered. The unique
and precise server, at any instant, can be found through novel leader election algorithm
using buffer [14]. Various other distributed algorithms can be taken in approximation
according to the need like comparison of various election algorithms in distributed
system [15], which gives a brief idea about picking a particular algorithm in an
optimized way.

3.3 Pseudo Code/Algorithm

This pseudo code provides an idea about the implementation on a device.

START:
Read: Input the number of the Servers. Let Variable->n.
Create Priority (pro) and Status (sta) Array of the Servers with size n.
Read: Input the Status of n Servers(1 for Active, 0 for Inactive)
Read: Input the Priority of n Servers(Priority Values)
Declare dummy variables like co, initially 0, which is the Final Coordinator.
Read: Input a variable (ele) which will initiate the Election
Create: A function (select) that will take the argument (ele)
BEGIN FUNCTION:
ele->ele-1
co->ele+1 //Final Server
LOOP: with i=0 till i=n
IF: pro[ele] is less than pro[i]
PRINT: Selection message to the other Server.
IF: sta[i] is 1 //Status of that Server is 1
CALL Function select(i+1) //Recursive Call to the above function
END IF:
END IF:
END LOOP:
END FUNCTION:
END

3.4 Calculation of Time Complexity

The SELECT function takes N number of arguments in server

void select(N) {
876 K. Shukla and B. Sahoo

for(i=0; i<N; i ++){

//O(n) time
if(pro[ele]<pro[i]) {
if(sta[i]==1)
select(i+1); //Recursive Call in O(n-1)
}
}
}
Worst Case (O(n2 )) : If the initiated server has least priority or not working, then
iteration takes O(n) time in LOOP and O(n − 1) in recursive function call.
Total Time Taken: O(n (n − 1)) ~ O(n2 ).
Average Case (O(n2 )): If the initiated server has less priority value or the recursive
function call returns highest priority value at the first iteration.
Total Time taken: θ (n(n−1)) ~ θ (n2 ).
Best Case (O(n)): If the initiated server has highest value priority or other servers
are not working to respond, then only one loop iteration is required.
Total Time Taken: WÄ (n).

4 Advantages

• In real-time scenario, this algorithm avoids the heavy working server. So it gives
an efficient utilization of server resources.
• It can give the dead status of the servers to the master–servers, so that it can be
rectified.
• It reduces the load on the servers, by dividing them among their priorities.

5 Other Implementations Scenario

This algorithm can be implemented in devices like printers. If an organization where

there are a large number of printers, then this algorithm could effectively select the
best printer out of many other printers in real-time scenario.
Suppose there are many printers and there are many pages for different clients to
print, this algorithm on printers can divide the work load. And if some printer might
be low on INK, it can refuse to print the page and allow other printers to do the job.
This saves a lot of time, at the same time it can report to others.
Dynamic Selection of the Best Server … 877

6 Conclusion

Thus, the above context gives an idea and approach about the popular distributive
algorithm (bully algorithm) and showed that a server or a device can be programmed
to overcome the problems faced in the network. It could save a lot of time and also
can distribute the load on heavy servers or devices. Multiple servers can be handled
together in any geographical location, and replacement of any server can also become
easy. This algorithm can further be improved if we could mathematically find the best
formula to calculate priority value and also by further reducing the time complexity
of algorithm in the worst case scenario.

References

1. Coulouris, G.F., Dollimore, J., Kindberg, T.: Distributed Systems: Concepts and Design.
Pearson Education, London (2005)
2. Lynch, N.: Distributed Algorithms. Morgan Kaufmann Publishers, San Francisco, CA (1996).
ISBN 978-1-55860-348-6
3. Basim, A., Laith, H.B., Mohammad, A.: Reducing massage passing and time complexity in
bully election algorithms using two successors. Int. J. Electron. Electr. Eng.(2013)
4. Mamun, Q.E., Masum, S.M., Mustafa, M.A.: Modified bully algorithm for electing coordinator
in distributed systems. WSEAS Trans. Comput. 3(4), 948–953 (2004)
5. Soundarabai, P.B., et al.: Improved bully election algorithm for distributed systems. Preprint
at arXiv:1403.3255 (2014)
6. Garcia-Molina, H.: Elections in a distributed computing system. IEEE Trans. Comput. 1, 48–59
(1982)
7. Gusella, R., Zatti, S.: An election algorithm for a distributed clock synchronization program.
No. UCB/CSD-86/275. California Univ Berkeley (1985)
8. Chang, E., Roberts, R.: An improved algorithm for decentralized extrema-finding in circular
configurations of processes. Commun. ACM 22(5), 281–283 (1979)
9. Singh, S., Kurose, J.F.: Electing “good” leaders. J. Parallel Distrib. Comput. 21(2), 184–201
(1994)
10. Anthony, R.J.: Emergence: A paradigm for robust and scalable distributed applications. In:
International Conference on Autonomic Computing, 2004. Proceedings of the IEEE (2004)
11. Yassein, M.B., Ala’a, N.A., Sana’a, A.A.: An efficient overhead-aware leader election algorithm
for distributed systems. Int. J. Comput. Appl. 49(6), 10–15 (2012)
12. Numan, M., et al.: Well-organized bully leader election algorithm for distributed sys-
tem. In: 2018 International Conference on Radar, Antenna, Microwave, Electronics, and
Telecommunications (ICRAMET). IEEE (2019)
13. Biswas, A., Dutta, A.: A timer based leader election algorithm. In: 2016 International IEEE
Conferences on Ubiquitous Intelligence & Computing, Advanced and Trusted Computing,
Scalable Computing and Communications, Cloud and Big Data Computing, Internet of People,
and Smart World Congress (UIC/ATC/ScalCom/CBDCom/IoP/SmartWorld). IEEE (2016)
14. Ranwa, M., Sharma, P., Mehrotra, G.: Novel leader election algorithm using buffer. In: 2017
2nd International Conference on Telecommunication and Networks (TEL-NET). IEEE (2017)
15. Gupta, P., Vishwakarma, R.G.: Comparison of various election algorithms in distributed system.
Int. J. Comput. Appl. 53(12) (2012)
Heart Disease Prediction Using Machine
Learning Techniques

Shekharesh Barik, Sambit Mohanty, Deepankar Rout, Subhra Mohanty,

Akshaya Kumar Patra, and Alok Kumar Mishra

Abstract Heart-related diseases or cardiovascular diseases are the primary purposes

behind a large number of deaths on the planet in the course of the most recent couple
of decades. It has risen as the most terrifying ailment around the world. Actually, in
India, these issues are progressively awful; according to the Journal of the American
College of Cardiology in India, the demise rate because of cardiovascular maladies
increments around 34% in the middle of 1990–2016. Presently, we are in a time
of data age where a huge quantity and variety of information is stored in different
enterprises like retail, producing, medical clinic, and online networking. We can
gather the information and break down the information to foresee the components
and reasons for heart diseases so that safety measures can be taken to decrease
the demise rate. There exists various types of information investigation instrument
and procedure which requires an ideal informational collection; at that point, we
can apply distinctive sort of machine learning strategies to anticipate whether the
patient can be influenced by heart diseases or not by utilizing the recently gathered
datasets. In this paper, we will exhibit how to utilize various kinds of machine learning
models like K-nearest neighbor, decision tree classifier, and random forest classifier,

S. Barik (B) · D. Rout · S. Mohanty

CSE Department, DRIEMS (Autonomous), Cuttack 754022, India
e-mail: [email protected]
D. Rout
e-mail: [email protected]
S. Mohanty
e-mail: [email protected]
S. Mohanty
Software Developer, SLFS Lab, Bhubaneswar 754001, India
e-mail: [email protected]
A. K. Patra · A. K. Mishra
Department of Electrical and Electronics Engineering, Institute of Technical Education and
Research, Siksha ‘O’ Anusandhan (Deemed to be University), Bhubaneswar 751030, India
e-mail: [email protected]
A. K. Mishra
e-mail: [email protected]
© Springer Nature Singapore Pte Ltd. 2020 879
G. Pradhan et al. (eds.), Advances in Electrical Control and Signal Systems,
Lecture Notes in Electrical Engineering 665,
https://doi.org/10.1007/978-981-15-5262-5_67
880 S. Barik et al.

and furthermore make a presentation correlation among these models so that we

can get accurate precision about a patient having heart disease (Chen et al. in 2011
Computing in Cardiology IEEE, 557–560, 2011, [1]), (Kishore et al. Heart attack
prediction using deep learning, [2]).

Keywords Heart disease · Machine learning · Data analysis · K-nearest neighbor ·

Decision tree classifier · Random forest classifier

1 Introduction

The medical term for heart is cardio. So, heart diseases are also known as cardio-
vascular diseases (CVD). Heart disease is one of the most common diseases in our
modern era, and it can prompt to reduce the life span of individuals these days. There
are various reasons behind it, which leads it to increase day by day. Like in our
everyday life, heavy routine and busy schedule increase work stress and the most
significant reasons are people are addicted to having cigarette, tobacco, and habit
to taking bad food impact to various heart diseases arising in human body. A report
from World Health Organization (WHO) tells that 17.5 million people are dying
every year because of heart diseases (Fig. 1).
The vital component of human body is heart. Due to various factors, blood clots
in the vein. This may happen after a mild or severe heart attack. Clotting blocks the
transmission of blood through the vein which is connected to heart. This may lead to
a severe heart stroke or brain stroke resulting in loss of life. Parameter or factors for
developing heart diseases are an important issue. High blood pressure is the biggest
factor for heart problems. Diabetes is a very important parameter for heart disease.

Fig. 1 Diagram of blood

clotting in heart
Heart Disease Prediction Using Machine Learning Techniques 881

Person suffering from diabetes for long time is more likely to develop a heart disease.
Nowadays, hypertension, bad food habit or lifestyle and stressful life are growing
rapidly. These attributes are considered as one of the causes for heart disease. Age
can also be considered as one of the factors for falling into heart diseases. Older
people have more chance of getting into heart-related problems than younger ones.
Other factors can be smoking, high cholesterol, obesity, etc. In this paper, we have
taken some useful factors from dataset in order to predict the heart disease [3, 4].
The main challenge that lies in front of scientists and researchers is the accuracy in
prediction of diseases and optimization of results. Nowadays, a large amount of data
is available in various sectors. It can be social media, hospital, manufacturing, retail,
etc. We can collect this data and analyze this data to predict the factors and causes
of heart disease so that precautions can be taken to reduce the death rate. We can
apply different kinds of machine learning techniques to predict whether the patient
can be affected by heart diseases or not. We are using machine learning algorithms
like K-nearest neighbor, decision tree classifier, and random forest classifier [5, 6].

2 Algorithm Used

In this paper, we have classified the patient’s data to predict whether a patient has a
heart disease or not. For this classification purpose, we have used Cleveland patient
dataset available in UCI repository. The Cleveland dataset contains 303 numbers of
patient records, and each record has 14 attributes. These 14 attributes are used for
evaluating and predicting whether a patient has heart disease or not. If a patient has
heart disease, then it is treated as 1, and if no heart disease, then treated as 0. We have
used three classification algorithms for classifying and predicting whether a patient
has heart disease or not. The machine learning algorithms are K-nearest neighbors,
random forest classifier, and decision tree [7].
We have used Jupyter notebook which is a freely available software for performing
machine learning operations. For machine learning, we need to import the sklearn
module which contains all the essential algorithms and functions. We need to import
Python NumPy module and Pandas module for data analysis purpose. To plot dif-
ferent graphs, we need to import Matplotlib module which contains all the methods
related to plotting graphs. Our data is stored in a CSV file which needs to be imported
to the notebook by using Python Pandas module as a data frame. After importing
the data, we can apply various data analysis and machine learning algorithms for
classification and prediction [1].
Before applying our machine learning algorithms on the dataset, we can see how
the features of our dataset look like in plots; for this, we can use Matplotlib module
(Fig. 2).
882 S. Barik et al.

Fig. 2 Features of datasets in plots

2.1 K-Nearest Neighbor (KNN) Algorithm

KNN is a nonparametric supervised learning algorithm which creates the model

structure from the given dataset. In KNN, k is a number which decides how many
nearest neighbors are used to provide the best result for the dataset. Let ‘a’ is point
whose label is to be predicted then first we have to find the ‘k’ closest point near to
‘a’. Then, make a voting that how many nearest data points represent to a particular
class. The class which gets highest voting, the point ‘a’ will belong to that class.
For finding closet point near to our data point, we can use any of these distance
calculation methods like Euclidean distance, Hamming distance, and Manhattan
distance. Finding the best k value for the dataset is a challenging task. For every
dataset, k value is different. However, some assumption says k value is an odd number
which provides the best results. We can say that the value of k is the controlling factor
for the problem. Research shows that if we take k value a less number, then it may
Heart Disease Prediction Using Machine Learning Techniques 883

lead to overfitting problem; i.e., noise has more impact on our prediction rather than
actual prediction. If we take larger value of k, then it leads to expensive computation.
So, we need to find an optimal value of k. We can find the best k value by performing
our operation on the dataset by using different k values. The k value which gives best
result will be taken for that particular dataset [8].
To use KNN classifier in our code, we need to import a module named as sklearn
module. In our dataset, there exist some categorical variables like sex, cp, fbs, restecg,
exang, slope, ca, that. We need to convert these variables into dummy variables that
means converting to numbers. For this purpose, we need to use standard scalar class
from sklearn module. It will convert these categorical variables to dummy variables.
Now, we can apply our KNN algorithm to this dataset. Before that, we have to
separate the features and labels from the dataset. In X variable, we have taken all
the features of the dataset, and in Y variable, we have taken the label, i.e., the target
column of our dataset. We do not know which k value is best suitable for our dataset.
So, we will check every k value on our model. In this case, we have taken k value in
a range of (1, 21); i.e., all values between this range are used as k to give accurate
score. For this purpose, we have used cross-validation techniques. It gives accurate
measure of the performance of the machine learning model, that is, what we expect
from our model. Here, we have used the cross-validation score for every value of k.
The k value which gives the highest score is chosen for further operation. To know
all the cross-validation scores of all the values in the range (1, 21), we have plotted
a graph by using Matplotlib (Fig. 3).

Fig. 3 Finding the best value of ‘k’ from a range of values

884 S. Barik et al.

From this graph, it is clearly seen that at k = 12, the score is 0.85066 which is
the highest among all the k values in the range (1, 21). So, we have taken k = 12 for
further operations. Then, we have applied this k value to the KNN classifier and get
the prediction value = 0.8506637004078605.
The snapshot of the code for getting the best value for ‘k’ is given below:

2.2 Decision Tree

Decision tree is a tree-like structure. The internal node or non-leaf node represents the
features, branch represents decision rules, and each leaf node represents the result.
In the decision tree, the root of the tree is the topmost node. This root node learns to
classify the dataset based on the features present in dataset. It partitions the dataset
in a recursive manner. It gives the flowchart-like representation which is suitable
for human understandable format and helps in understanding and taking decision.
Decision tree is a white box-type machine learning algorithm model which gives
detailed information of process behind it. It is a nonparametric method. Attribute
selection measure process is used to break the dataset by using the features; it is
a heuristic splitting rule to break the data points. Some of the popular selection
measures are gain, gain ratio, Gini index, etc.
To use decision tree algorithm on our dataset, we need to import decision tree
classifier from sklearn module. We need to divide the dataset into train and test.
Train dataset is used for training our decision tree classifier, and test dataset is used
to test the prediction values. We have used the cross-validation techniques to predict
the score as it provides the most accurate results what we expect from the model.
Decision tree problem is type of NP-complete problem. Practically, this algorithm
uses heuristics which is a greedy algorithm. So, there exists local optimal solution at
each node. That means, algorithm does not guarantee about returning global optimal
tree. So, we need to use a random seed so that it can control the random choices; it
can be any number. Here, we have taken random state = 7. In this paper, we have
taken a parameter cv = 10 that is called cross-fold value. It indicates, we have divided
the entire dataset to that many parts. Here, cv = 10 means the dataset is divided into
ten parts between which nine parts are used for training the dataset and one part is
used to test the dataset. Cross-validation technique gives better result in comparison
with train and test methods; then, we calculate the mean value of the score to get
accurate prediction value for the entire dataset. In this dataset, we get prediction
value 0.755150. These values can be improved further by adjusting the parameters.
The snapshot of the code is given below:
Heart Disease Prediction Using Machine Learning Techniques 885

2.3 Random Forest Classifier

It is a type of supervised learning algorithm. As the name says, forest basically

contains trees, i.e., decision trees. Random forest gives its prediction value basing
upon the prediction of decision trees used. In random forest, many decision trees are
present, and from the dataset, few features and labels are given to each decision tree
on random basis. Then, these decision trees perform their operation on the supplied
data points and predict their score. As multiple decision trees are present, so random
forest makes a voting for classification-type problems. Majority of voting result
provided by decision trees will decide the class of given problem. Random forest
can also be used as repressors. In this case, the output is a continuous value which is
the mean or median of all predicted values of the decision tree [9] (Fig. 4).

Training TrainingSam Training ....... Training

ple 1 Sample 2 Sample n

Test set

Decision Decision ....... Decision

Tree 1 Tree 2 Tree n

Voting

Prediction

Fig. 4 Control flow diagram for random forest

886 S. Barik et al.

Random forest provides better results in comparison with decision trees as deci-
sion tree suffers from low bias and high variance problem so it may not give accurate
prediction values. To overcome such type of problem in random forest, many number
of decision trees are used so that the high variance problem can be reduced and it
will provide more accurate prediction. In fact, random forest is the most popular
machine learning algorithm used for both classification and regression problems. In
real life, it is used for design of recommendation system, credit card fraud detection,
classification of loyal loan applicants, disease prediction, and so on. Random forest
may be slow in terms of prediction time as it uses many decision trees [7]. To use
random forest classifier on our dataset, we need to import random forest classifier
class from sklearn module. The snapshot of the code is given below:

Here, we have used a parameter n_estimator which is a controlling parameter of

random forest algorithm. It indicates how many number of decision trees represent
inside the random forest classifier. The more the number of value of n_estimator,
i.e., the more the number of decision trees, then the more is the accuracy we can get.

3 Comparison Results

We have used three classification algorithms on our dataset to get the prediction for
heart disease. For KNN, with k value = 12, we get prediction value 0.85066. For
decision tree classification, we get prediction value 0.7551. For random forest clas-
sification with ten decision trees, we get prediction value 0.8012. We have compared
the above results with the results of another paper mentioned in the reference section
list. We found that paper has the similar goal as of ours but with some other set of
machine learning algorithms for heart disease prediction. It uses the machine learn-
ing algorithms like naïve Bayes classifier, support vector machine, decision tree, and
K-nearest neighbor. The comparison result is matched with the results of our case.
In the referred paper, we find that the prediction accuracy for naïve Bayes classifier
is 83.4983, for support vector machine is 84.1584, for decision tree is 77.5578, and
for K-nearest neighbor is 76.2376. Hence, it is clearly seen that KNN model predicts
more accurately in our case (Fig. 5).
Heart Disease Prediction Using Machine Learning Techniques 887

Fig. 5 Comparison model Predicon value

for the three classification 0.9
algorithms 0.85
0.8
0.75
0.7
KNN Decision Random
Tree forest
classificaon

Predicon value

4 Conclusion and Future Scope

The comparison results are matched with the results of another paper as mentioned
in the reference list. It is found that KNN in our case gives the highest accuracy in
comparison with all other machine learning algorithms. In this paper, we have applied
only three machine learning algorithms on the dataset to predict the accuracy value
but there exist other machine learning algorithms which can be applied on the dataset
which may perform better. We can also improve the performance of these algorithms
by adjusting tuning parameters and other adjustment parameters [10].

References

1. Chen, A.H., Huang, S.Y., Hong, P.S., Cheng, C.H., Lin, E.J.: HDPS: Heart disease prediction
system. In: 2011 Computing in Cardiology IEEE, pp. 557–560 (2011)
2. Kishore, A., Kumar, A., Singh, K., Punia, M., Hambir, Y.: Heart attack prediction using deep
learning (2018)
3. Almarabeh, H., Amer, E.: A study of data mining techniques accuracy for healthcare. Int. J.
Comput. Appl. 168(3), 12–17 (2017)
4. Krishnaiah, V., Narsimha, G., Chandra, N.S.: Heart disease prediction system using data mining
technique by fuzzy K-NN approach. In: Emerging ICT for Bridging the Future—Proceedings
of the 49th Annual Convention of the Computer Society of India (CSI), vol. 1, pp. 371–384,
Springer, Switzerland (2015)
5. Dinesh Kumar, G.: Prediction of cardiovascular disease using machine learning algorithms.
In: Proceeding of 2018 IEEE International Conference on Current Trends Toward Converging
Technologies, Coimbatore, India (2018)
6. Khourdifi, Y., Mohamed, B.: Heart disease prediction and classification using machine learning
algorithms optimized by particle Swarm optimization and Ant Colony optimization. Int. J.
Intell. Eng. Syst. 12(1) (2019)
7. Rairikar, A., Kulkarni, V., Sabale, V., Kale, H., Lamgunde, A.: Heart disease prediction using
data mining techniques. In: Intelligent Computing and Control (I2C2), IEEE International
Conference on 2017 Jun, pp. 1–8 (2007)
8. Santhana Krishnan, J., Geetha, S.: Prediction of heart disease using machine learning algo-
rithms. In: 1st International Conference on Innovations in Information and Communication
Technology (ICIICT), IEEE (2019)
888 S. Barik et al.

9. Xu, S., Zhang, Z., Wang, D., Hu, J., Duan, X., Zhu, T.: Cardiovascular risk prediction method
based on CFS subset evaluation and random forest classification framework. In: 2017 IEEE
2nd International Conference on Big Data Analysis (2017)
10. Kaur, A., Arora, J.: Heart disease prediction using data mining techniques: A survey. Int. J.
Adv. Res. Comput. Sci. 9(2), 569–572. Mar-Apr 2018 (2018)
Loss Allocation Strategies in Active
Power Distribution Networks: A Review

Ambika Prasad Hota, Sivkumar Mishra, and Debani Prasad Mishra

Abstract In this paper, an up to date bibliographic survey of various loss allocation

(LA) methods applied to power distribution systems are presented. It not only dis-
cusses regarding the properties to be followed for a fair LA but also explains some
different techniques used for bifurcation of the cross-terms associated with power
loss equations. The pros and cons of several LA procedures are also presented here
in order to extend the present research further to achieving a better loss management
system.

Keywords Distribution network · Loss allocation · Distributed generators

1 Introduction

Deregulated power supply is considered as one of the important aspects of Smart

Grid. Electricity sectors worldwide are undergoing major structural changes due to
the large penetration of renewable and distributed energy sources at the load ends.
At this restructured environment several generations, transmission, and distribution
sectors are being introduced in place of a single state wound vertically integrated
system with the main objective to introduce competition for bringing down the cost of
electricity and to enhance the service quality. In such an open-access environment, it
becomes extremely important that various wheeling activities must be clearly defined
and the corresponding cost must be recovered. Transmission and Distribution power
loss is an important component of the cost of electricity. Hence, it is to be recovered
from the network users by suitably and fairly allocating it among the users. The

A. P. Hota · D. P. Mishra
IIIT, Bhubaneswar, Odisha, India
e-mail: [email protected]
D. P. Mishra
e-mail: [email protected]
S. Mishra (B)
CAPGS, BPUT, Rourkela, Odisha, India
e-mail: [email protected]
© Springer Nature Singapore Pte Ltd. 2020 889
G. Pradhan et al. (eds.), Advances in Electrical Control and Signal Systems,
Lecture Notes in Electrical Engineering 665,
https://doi.org/10.1007/978-981-15-5262-5_68
890 A. P. Hota et al.

nonlinear relationship between power loss and injected power makes loss allocation
(LA) process difficult and complicated. In this paper, detailed descriptions of the
state-of-art of the loss allocation methods, applied to the radial distribution networks
(RDNs) for fair loss allocation among the consumers are presented. The different
loss allocation techniques with their respective assumptions and contributions are
discussed in this paper. A comparative study is carried out on standard RDNs using
various loss allocation techniques to provide an idea for suitable and fair LA to the
users. As placement of distributed generators (DGs) in electrical power distribution
networks (EPDNs) often lead to a substantial reduction in active power loss and it is
very essential to allocate this decrease in power loss fairly among the consumers and
simultaneously giving the incentive to the connected DGs. So, several LA problem-
solving techniques, based on mathematical and heuristic algorithms are discussed in
this paper. The contribution of various approaches to the LA problem is reviewed and
also presented in this paper, which will provide thorough knowledge to the reader
regarding proper network reconfiguration to achieve fair loss allocation. Macqueen
and Irving [1] proposed an algorithm which is a combination of graph theory and
standard load flow analysis (LFA) for assigning losses to the users according to the
burden they imposed on the power system. This LA procedure is suitable for the
allocation of both demand and energy losses. Mutale et al. [2] considered DGs in
RDNs for LA using two schemes; one based on marginal method, which required
reconciliation due to over recovery of loss and the other based on direct loss coeffi-
cients. In literature many loss allocation (LA) methods are available, most of them are
proposed exclusively for allocating transmission losses [3]. Costa and Matos [4] pro-
posed a branch-based LA scheme for distribution networks with DGs. Carpento et al.
[5] proposed a branch current decomposition-based LA scheme (BCDLA) for RDNs
with DGs. In [6], adaptions of several transmission systems based LA schemes to dis-
tribution networks with DGs are critically compared. Carpento et al. [7] extended the
BCDLA scheme to three-phase unbalanced RDNs with DGs. Similarly, LA schemes
exclusively for radial distribution networks (RDNs) are also proposed [8–10], where
it is presumed that the consumers at various buses have to pay for the active power
loss in the RDN and the substation bus, which is the natural slack bus, is not allo-
cated with any loss. The penetration of DGs into the distribution system changes
the network configuration, thus converting a network form passive to active one.
Hence, loss allocation methods are also needed to be modified to include DGs in the
loss allocation process. Atanasovski and Taleski [11] developed a power summation
method based LA (PSMLA) for RDNs with DGs using a quadratic relation between
losses and power flows. Again in 2012, Atanasovski and Taleski [12] proposed an
energy summation based LA (ESMLA) for RDNs with DGs with a statistical repre-
sentation of daily load and generation curves. A new bus identification scheme [13]
and [14] is used to formulate and implement the proposed LA method for RDNs with
DGs, where DGs are modeled as negative constant power loads. In this proposed LA
method, the active power loss in a branch of RDN is expressed as a summation of
separate terms and where each term can be related to the active and reactive power
injections at the subsequent nodes of the branch. Thus the contribution of all the
Loss Allocation Strategies in Active Power … 891

buses (or consumers) to the total active power loss is determined so that loss is allo-
cated to all the buses in a fairway. Jahromi et al. [15] proposed a new LA method
for RDNs with DGs but a normalization process is required to make the total loss
allocated equal to the actual loss. Jagtap and Khatod allocate losses in an RDN with
DGs using the geometric scheme of allocation for various load models in [16], using
a branch oriented approach with backward sweep network reduction in [17, 18],
using an exact method in [19] and using the logarithmic scheme of allocation in [20].
Kashyap and De [21] proposed a two-step proportional sharing based LA approach
for RDNs with DGs. Recently; Kumar et al. [22] proposed a cross-term decomposed
approach for LA of RDNs with DGs.

2 Methods of Loss Allocation

Power loss calculation and fair distribution are one of the major points of discussion in
a deregulated power system. In a conventional system, the total network losses were
reduced by the system operator by controlling the active and reactive power output
of the generators, as per the result obtained from optimal power flow. The costs of
losses are provided by the entire power system. But, in the deregulated environment,
as the competition is high, the allocation of losses among the consumers is not an
easy task rather a most complicated and challenging job as pointed out by Fang and
Ngan in [23]. As per literature, loss allocation problem is purely economic in nature
and is to be analyzed with the help of mathematical algorithms, but the difficulty in
solving this LA problem is the nonlinear relationship between branch power losses
and bus power injections [11]. The various transmission losses are compared by
three different techniques i.e. pro-rata, Proportional Sharing (PSLA) procedures,
and Incremental Transmission loss (ITL) method by Conejo et al. in [3]. The Pro-
rata (PR) technique is found to be simple and straightforward, as half (50%) of the
total losses are assigned to the suppliers and the rest half to the loads in the first
phase, and then the generators or loads are assigned losses individually proportional
to their active power generated or consumed irrespective of their location in the
network. Hence, the remotely located generators on loads are benefited through this
technique. ITL procedures are known as Marginal Procedures as these methods often
result in over recovery of losses. In PSLA method, the optimal results obtained from
load flow are used with a linear proportional sharing principle. This principle states
that “the power flow reaching a bus from any power line splits among the lines
evacuating power from the proportionally to their corresponding power flows” [3].
A loss allocation method must have the following properties; so as to be called as
a fair allocation method as discussed by Conejo et al. in [24] are as follows:
i. The injected current or power values should be reflected at each node.
ii. The position of the buses in a RDN should be clearly specified.
iii. The relationship between current and voltage with their respective network
topology should be reflected.
892 A. P. Hota et al.

Fig. 1 A sample RDN

iv. It should be easy to understand and implement.

v. The method should be sufficient enough to provide either incentives or penalties
to the suppliers or loads by considering their contribution to power loss with
respect to their location and ratings.
vi. It should be consistent with a solved load flow.
It is noticed from the literature that most of the LA methods are meant for trans-
mission systems and very few are analyzed for distribution systems. The LA analysis
for transmission system (TS) and distribution system (DS) varies due to the effect
of slack node. The generator at slack node is used for compensation of all the losses
in TS. The slack node (generally the root node of the RDN) is used to represent the
connection to the higher voltage network.
The equivalent current injection (ECI) at bus ‘a’ of a radial distribution network
(Fig. 1) is calculated in terms of the corresponding node power (S La ) and node voltage
(V a ) as:

(PLa − Q La )
I La = , a = 2, 3, . . . , nb (1)
(Va )∗

The current in any branch-b of a RDN can be calculated by adding the ECI values
of all the subsequent buses of the concerned branch.

I (b) = I L(ai ) (2)
i

where “i” is the subsequent buses of branch-b.

The current of any branch-b can be expressed in complex form as the sum of all
real values and complex values as:

I (b) = ILR(ai ) − j ILI(ai ) (3)
i

The absolute value or magnitude of the current of branch-b can be calculated from
the above equation as:
 2  2
 
|I (b)| =
2
ILR(ai ) + ILI(ai ) (4)
i i
Loss Allocation Strategies in Active Power … 893

Expanding the terms inside the brackets and rearranging

  2  2
|I (b)|2 = ILR(ai ) + ILI(ai )
i

 
LSB−1 LSB
+2· ILR(ai ) · ILR(ak ) + ILI(ai ) · ILR(ak ) (5)
i=FSB k=i+1

Here, FSB and LSB denote the first and last subsequent buses of the branch-b
respectively. The expression of real power becomes:

  2  2
PLoss (b) = R(b) ILR(ai ) + ILI(ai )
i

 
LSB−1 LSB
+2· ILR(ai ) · ILR(ak ) + ILI(ai ) · ILI(ak ) (6)
i=FSB k=i+1

LSB−1 LSB
where the term 2 · i=FSB k=i+1 ILR(ai ) · ILR(ak ) + ILI(ai ) · ILI(ak ) is called as the
cross-term of power loss equation.
These cross-terms make the loss allocation process difficult and complicated. The
allocation of these cross-terms can be made in several ways. The cross-terms can be
divided into two components, as explained below.

α(ai ) · ILR(ai ) · ILR(ak ) + α(ak ) · ILR(ai ) · ILR(ak ) = 2ILR(ai ) · ILR(ak ) (7)

β(ai ) · ILI(ai ) · ILI(ak ) + β(ak ) · ILI(ai ) · ILI(ak ) = 2ILI(ai ) · ILI(ak ) (8)

Here i and k have values as previously explained and α and β are the loss allocation
factors for the cross-terms of real and reactive components of load current. Now the
above two equation can be simply written as

α(ai ) + α(ak ) = 2.0 (9)

β(ai ) + β(ak ) = 2.0 (10)

2.1 Pro-Rata (PR) Scheme of Loss Allocation

It is a very simple and straight forward way of allocating active power losses in a
network. Here, active power loss is allocated to a bus which is in proportion to the
kVA load at the bus to the total kVA load of the system. Thus, the active power loss
894 A. P. Hota et al.

allocated to a bus-a can be calculated as:

kV Aa
plossalloc (a) = Tloss · , a = 2, 3, . . . , nb (11)
T kV A
It is noteworthy that it is more justifying to consider the kVA load instead of kW
for active power loss allocation as reactive powers also contribute significantly to the
active power of the system.

2.2 Proportional Sharing Scheme of Loss Allocation

The active power loss in a specific branch-b can be allocated to the consumers of sub-
sequent branches in a fair way in the presence of cross-terms with a straight forward
approach. The intuitive and straight forward approach results from the proportionality
assumption [25] are as follows:
α(ai ) α(ak )
= (12)
ILR(ai ) ILR(ak )
β(ai ) β(ak )
= (13)
ILI(ai ) ILI(ak )

From (9) and (12)

2 · ILR(ai )
α(ai ) = (14)
ILR(ai ) + ILR(ak )

Similarly, from (10) and (13)

2 · ILI(ai )
β(ai ) = (15)
ILI(ai ) + ILI(ak )

Based on the proportional sharing principle, the consumers connected beyond

branch-b (i.e., to ai , for i = FSB to LSB of branch-b) are assigned losses by the
expression (Table 1):
⎡ ⎤
{ILR (ai )}2 + {ILI (ai )}2
⎢ ⎥
⎢ 
LSB(b)
2 · ILR (ai ) ⎥
⎢ + ILR (ai ) · ILR (ak ) · ⎥
⎢ ILR (ai ) + ILR (ak ) ⎥
ploss (ai ) = R(b).⎢
⎢ k = FSB(b),k=i
⎥
⎥
⎢ ⎥
⎢ 
LSB(b)
2 · ILI (ai ) ⎥
⎣ + ILI (ai )·ILI (ak ). ⎦
k = FSB(b),k=i
ILI (ai ) + ILI (ak )
(16)
Loss Allocation Strategies in Active Power … 895

Table 1 Contribution of the reviewed loss allocation (LA) works

Authors Method Contribution
Kumar et al. [26] ECTDM This paper has implemented an exact method for
decomposition of the cross-term present in the power
loss equation. The loss allocation factor used
decomposes the mutual term fairly among the
network participants. It also uses a superposition
based DG remuneration technique which allocates
the exact amount of network loss reduction to the
DG owners. The efficiency of the method has been
tested on a 33-bus RDN
Usman et al. [27] MPLAP A multi-phase LA technique developed in this paper
provides clear information regarding LA of neutral
conductor. It uses a two-phase switching
methodology to reduce unbalance in the power
network. This method is found suitable for
assessment and controlling of unbalancing nature of
the Italian distribution system
Masoud et al. [28] Khosravi’s The algorithm is developed by considering the
method injection of both real and reactive power flow of each
line for allocation of losses in a RDN. The algorithm
is executed in three steps. In the first step, the
contribution of loads to the LA without supplier is
performed. The impact of DGs on the loss allocation
is calculated in the second step. The contribution of
each load and DG unit is computed separately in the
third step by introducing a normalization factor in
the algorithm. Further, this technique is easy to
implement in larger RDNs
Kumar et al. [29] BCDM This paper presents a branch current decomposition
algorithm for allocating losses of a reconfigured
RDN in the presence of DGs. The proposed
procedure is tested with changing loads, load power
factor, and also with different network topologies.
Virtual branch voltage drops and reactive power
transactions are considered for fair loss allocation
analysis. The superposition principle is used to
identify the DG owners for getting incentives or
penalties
Jagtap and Jagtap’s method A current summation algorithm is used for LA in a
Khatod [20] radial reconfigured distribution network with DGs.
The current of a branch is decomposed into real and
reactive components by considering the currents of
subsequent loads and DGs connected in the network.
This algorithm bifurcates the cross-term of power
loss equation using a logarithm scheme of LA. The
results obtained by the proposed method for a 33-bus
RDN is compared with other existing methods
(continued)
896 A. P. Hota et al.

Table 1 (continued)
Authors Method Contribution
Sharma and Sequential The objective of this paper is to decrease the memory
Abhyankar [30] Shapley value and computational burden for fair LA as compared to
(SSV) method the Shapley value technique. This scheme allocates
losses to the end-users sequentially by using Shapley
value technique. The loss allotted to each player by
this process is equal to the conventional SV method.
The results obtained by SSV method are tested on
three standard power networks and a comparative
study is performed with other approaches to validate
the novelty of the method
Sharma and Analytical This paper introduces a novel scheme of loss
Abhyankar [31] formulation of allocation for both radial and weakly meshed
shapley value distribution network without any assumption and
(SV) method approximation. The algorithm uses a cooperative
game approach for obtaining a fair loss allocation
result by considering the contribution of each
participant for getting rewards or penalties. An
analytical procedure is followed with Shapley value
to overcome the computational burden and to achieve
a fair LA with network data and load flow results. A
comparative analysis is performed at various DG
output levels with changed network topology for
establishing the effectiveness of the technique
Kashyap and De Proportional The aim of this paper is to implement a novel loss
[21] sharing based LA allocation technique to allocate losses in a fair way
method to the users in the presence of DGs in a RDN. A
method based on proportional sharing is used for LA
which comprises two steps. Loss is allocated to the
loads according to their contribution in the first step
and then in the next step DGs are allocated with the
help of power-sharing and contribution matrix. The
algorithm is tested and validated by considering a
6-bus RDN
Jagtap and Modified pro-rata This paper explains regarding the effect of load
Khatod [17] and branch variation to the RDN loss allocation. The proposed
current based LA technique of this paper is based on a branch current
method method that establishes a direct link between the two
end-node voltages and the apparent power of the
receiving node of a branch. It also uses a modified
PR technique for LA at different load factors. The
results are tested by considering a 28-bus and 33-bus
test network
(continued)
Loss Allocation Strategies in Active Power … 897

Table 1 (continued)
Authors Method Contribution
Ghofani-Jahromi Jahromi’s method The loss allocation is executed in the presence of
et al. [15] DGs in a radial distribution network by a branch
oriented approach. A three-step execution procedure
is followed where; losses are assigned to loads due to
real and reactive power flow separately in the first
step, then to the DGs in the next step. In the third
step, normalization technique is used for settlement
of over recovery of the losses
Savier and Das Exact method The objective of this paper is to provide a
[10] comparative study between the proposed LA method
and the existing (pro-rata, proportional, and
quadratic) methods. The LA method is explained by
considering different case studies along with
examples. In this paper, the effect of cross-term is
eliminated analytically from LA formulation
Mishra et al. [13] Circuit theory This paper presents an overview of different LA
based LA method techniques and the developed method is found
suitable for active power loss allocation in a RDN.
The proposed approach uses a new bus identification
technique which simplifies the load flow calculation
and reduces computation time. The method is tested
on two test system i.e. 30-bus and 69-bus RDN
Atanasovski and Energy The proposed method is only operative for RDNs.
Taleski [12] summation The ESMLA is an energy summation algorithm that
method based loss uses statistical analysis of daily load and generation
allocation curves. The results obtained on a modified 32-bus
(ESMLA) system are compared with the published data and are
found to be efficient and more accurate with
minimum data and calculations
Atanasovski and Power summation The PSMLA technique is used for RDNs. This
Taleski [11] method based loss algorithm establishes a direct relationship between
allocation the real and reactive power injected at nodes and
(PSMLA) losses in each branch of the power network. The
DGs are considered a negative load in the calculation
process of LA. The cross-terms are distributed by
using a quadratic relation between losses and power
flows. The result of a 32- bus test system is
compared with other published methods
Savier and Das [8] Exact method In the proposed method, no assumptions are made
for LA. The consumers having the same load
demand at different locations can be designed by the
proposed method to have a fair loss allocation. A
case study is presented on a 30-bus RDN and the
results are compared with pro-rata, proportional, and
quadratic method of LA. The consumers near to the
substation with equal load demand can be allocated
losses fairly by this method (advantages to pro-rata
method). Each branch of the RDN can be allocated
losses suitably without any assumption
(continued)
898 A. P. Hota et al.

Table 1 (continued)
Authors Method Contribution
Carpaneto et al. Branch current A circuit-based LA method (BCDM) is proposed for
[5] decomposition RDNs with DGs which is based on the data provided
method (BCDM) by the network and load flow solution. This method
effectively detects the location and time-domain
evolution of DGs and loads. The results obtained by
the proposed method are compared with other
methods along with a discussion on the SMLA
technique to escape the paradox implicitly present in
its formulation
Costa and Matos Costa’s method A branch-based LA scheme for distribution
[4] networks with DGs is proposed, where three steps
are carried out for calculation of line losses. Loads
are allocated losses distinctly in the first step and
then DGs in the next step. The residual losses are
redistributed in the third step. Load flow and loss
allocation are performed with and without DGs.
Incentives are provided to the DG units which are
reducing losses in the power system
Fang and Ngan Succinct method The proposed method calculates the loss of each line
[23] as a function of both real and reactive power
injections. The loss allocation technique is not
reliable at certain specific conditions as pointed out
by Carpaneto et al. [5]
Conejo et al. [24] Z-bus method The paper presents a merit-order approach to
allocate transmission losses to DGs and loads based
on the network Z-bus matrix. The load flow result
depends on the non-singularity of the Y-bus
admittance matrix. The innovative approach of this
method is that it exploits the complete set of network
equations without any assumptions. The technique
emphasizes currents instead of injection of power,
but not relevant to network having singular
admittance matrix
Mutale et al. [2] Marginal loss This method is limited to the network having high
coefficients X/R ratio. The solution technique involves
(MLC) and Direct computationally intensive matrices (Hessian and
loss coefficients Jacobian) to compute marginal loss coefficients
(DLC) (MLC) based on Newton–Raphson (NR) load flow.
DLC technique is used to compute the loss
coefficients which are directly related to the nodal
injections with the help of NR approach
Ilic et al. [32] Pro-rata method In this paper, loss allocation is performed by pro-rata
method which first distributes half of the losses to
DGs and other half to the loads, and then assigns
losses based on the power level of both DGs and
loads. In this approach, the position of DG/load is
neglected. The difficulty present in this method is
that, in the presence of only one DG in the system, it
has to pay 50% of the total loss
Loss Allocation Strategies in Active Power … 899

2.3 Quadratic Sharing Scheme of Loss Allocation

The active power loss in a radial distribution network increases quadratically with
power flows. Hence, the “Quadratic Scheme” of LA can be applied to RDNs [25, 33].
Thus,
α(ai ) α(ak )
 2 =  2 (17)
ILR(ai ) ILR(ak )
β(ai ) β(ak )
 2 =  2 (18)
ILI(ai ) ILI(ak )

Following the same steps as in the case of proportional sharing method:

 2
2 · ILR(ai )
α(ai ) =  2  2 (19)
ILR(ai ) + ILR(ak )
 2
2 · ILI(ai )
β(ai ) =  2  2 (20)
ILI(ai ) + ILI(ak )

Based on the quadratic sharing principle, the consumers connected beyond branch-
b (i.e., to ai , for i = FSB to LSB of branch-b) are assigned losses by the expression:
⎡ 2  2 ⎤
ILR(ai ) + ILI(ai )
⎢  2 ⎥
⎢ 
LSB(b)
2 · ILR(ai ) ⎥
⎢ + ILR(ai ) · ILR(ak ) .  ⎥
⎢ 2  2 ⎥
ploss(ai ) = R(b).⎢
⎢ k = FSB(b),k=i ILR(ai ) + ILR(ak ) ⎥ ⎥
⎢  2 ⎥
⎢ 
LSB(b)
2 · ILI(ai ) ⎥
⎣ ⎦
+ ILI(ai ) · ILI(ak ) ·  2  2
k = FSB(b),k=i ILI(ai ) + ILI(ak )
(21)

2.4 Exact Method of Loss Allocation

This method establishes a direct relationship between the nodes injected complex
powers and their respective node voltages. It assigns the exact amount of losses
to the RDN participants by simplifying the influence of cross-term mathematically
from LA formulation. It allocates losses to the network users according to their load
demand and physical locations in the network. Hence, the consumers having same
load demand at different locations can be designed by this method to have a fair loss
allocation [10, 13, 34].
900 A. P. Hota et al.

2.5 Loss Allocation by Network Reconfiguration

Network reconfiguration (NR) using radial topology provides a significant solution

to the identification of fault, separation of faulty location, and restoration of services.
Hence, the objective of NR is to obtain a radial network again as a result of reconfig-
uration. As per literature, the “Branch Exchange” type NR is the technique, where
restoration of radial topology is performed by opening any branch within the loop.
Hence, the network reconfiguration can be performed by two steps using “branch
exchange” method. A closed-loop is formed by closing the tie line (first step) and the
restoration of the radial topology is performed by opening a branch within the closed-
loop (second step). The branch to be opened is known as sectionalizing switch. The
optimal radial distribution network (ORDN) is obtained after several such switching
steps on the initial RDN. The objective of this technique is to decrease the active
power loss a RDN [33].

3 Discussion and Future Scope

Loss allocation by pro-rata method is found to be unsuitable for the consumers

who are electrically connected nearer to the substation, as the cost of loss is equal
for consumers having same load demand but connected far away. The proportional
and quadratic techniques are developed from the branch currents; hence these two
techniques allocate losses to the load points that are located beyond that branch.
Quadratic loss allocation scheme makes the assumption that the loss allocation factors
of a particular consumer are proportional to the square of real/reactive load current
of that consumer whereas the proportional loss allocation scheme assumes that the
loss allocation factors are proportional to real/reactive load current. However, in the
‘Exact Method’ no assumptions are made for LA. Most of the LA methods have been
discussed at a particular load factor and DG capacity. But, the investigation of losses at
different power factors, various DG capacities, and several load levels will aid more
significant results in this area of research with dispatchable and non-dispatchable
DGs.

4 Conclusion

This paper presents a brief overview of loss allocation techniques applied to a radial
distribution network. It can be stated that NR and DG power injection reduces
power loss of an RDN. Hence, loss allocation after reconfiguration with proper DG
installation will provide efficient results for the practical field of application.
Loss Allocation Strategies in Active Power … 901

References

1. Macqueen, C.N., Irving, M.R.: An algorithm for the allocation of distribution system demand
and energy losses. IEEE Trans Power Syst. 11, 338–343 (1996)
2. Mutale, J., Strabac, G., Curcic, S., Jenkins, N.: Allocation of losses in distribution systems with
embedded generation. IEE Proc. Gen. Trans. Distrib. 147, 7–14 (2000)
3. Conejo, A.J., Arroyo, J.M., Alguacil, N., Guijarro, A.L.: Transmission loss allocation: a
comparison of different practical algorithms. IEEE Trans. Power Syst. 17, 571–576 (2002)
4. Costa, P.M., Matos, M.A.: Loss allocation in distribution networks with embedded generation.
IEEE Trans Power Syst. 19, 384–389 (2004)
5. Carpaneto, E., Chicco, G., Akilimali, J.S.: Branch current decomposition method for loss
allocation in radial distribution systems with distributed generation. IEEE Trans Power Syst.
21, 1170–1179 (2006)
6. Carpaneto, E., Chicco, G., Akilimali, J.S.: Characterization of the loss allocation techniques
for radial systems with distributed generation. Electr. Power Syst. Res. 78, 1396–1406 (2008)
7. Carpaneto, E., Chicco, G., Akilimali, J.S.: Loss partitioning and loss allocation in three phase
radial distribution systems with distributed generation. IEEE Trans. Power Syst. 23, 1039–1049
(2008)
8. Savier, J.S., Das, D.: Energy loss allocation in radial distribution systems: a comparison of
practical algorithms. IEEE Trans. Power Deliv. 24, 260–267 (2009)
9. Mishra, S., Das, D., Paul, S.: Active power loss allocation schemes in radial distribution systems.
Cogener. Distrib. Gener. J. 25, 26–43 (2010)
10. Savier, J.S., Das, D.: An exact method for loss allocation in radial distribution systems. Electr.
Power Energy Syst. 36, 100–106 (2012)
11. Atanasovski, M., Taleski, R.: Power summation method for loss allocation in radial distribution
networks with DG. IEEE Trans. Power Syst. 26, 1747–1755 (2011)
12. Atanasovski, M., Taleski, R.: Energy summation method for loss allocation in radial distribution
networks with DG. IEEE Trans. Power Syst. 27, 1433–1440 (2012)
13. Mishra, S., Das, D., Paul, S.: A simple algorithm to implement active power loss allocation
schemes in radial distribution systems. J. Inst. Eng. (India) Ser. B 93, 223–232 (2012)
14. Mishra, S., Das, D., Paul, S.: A simple algorithm for distribution system load flow with
distributed generaion. In: Proceedings of IEEE ICRAIE, pp. 1–5. Jaipur, India (2014)
15. Ghofani-Jahromi, Z., Mahmoodzadeh, Z., Ehsan, M.: Distribution loss allocation for radial
systems including DGs. IEEE Trans. Power Deliv. 29, 72–80 (2014)
16. Jagtap, K.M., Khatod, D.K.: Loss allocation in radial distribution networks with different load
models and distributed generations. IET Gener. Transm. Distrib. 9, 1275–1291 (2015)
17. Jagtap, K.M., Khatod, D.K.: Loss allocation in distribution network with distributed genera-
tions. IET Gener. Transm. Distrib. 9, 1628–1641 (2015)
18. Jagtap, K.M., Khatod, D.K.: Loss allocation in radial distribution network with various
distributed generators and load models. Int. J. Electr. Power Energy Syst. 75, 173–186 (2016)
19. Jagtap, K.M., Khatod, D.K.: Allocation of distribution network losses with different types
of distributed generations. In: Annual IEEE India conference (INDICON), New Delhi, India
(2015)
20. Jagtap, K.M., Khatod, D.K.: Novel approach for loss allocation of distribution networks with
DGs. Electr. Power Syst. Res. 143, 303–311 (2017)
21. Kashyap, S.S., De, M.: A novel loss allocation method for radial distribution system with
distributed generations. In: 1st IEEE International Conference on Power Electronics. Intelligent
Control and Energy Systems (ICPEICES-2016), pp. 1–6. Delhi, India (2016)
22. Kumar, P., Gupta, N., Niazi, K.R., Swarnkar, A.: A circuit theory-based loss allocation method
for active distribution systems. IEEE Trans. Smart Grid 10, 1005–1012 (2017)
23. Fang, W., Ngan, H.: Succinct method for allocation of network losses. IEE Proc. Gener. Transm.
Distrib. 149, 171–174 (2002)
24. Conejo, A.J., Galiana, F.D., Kockar, I.: Z-bus loss allocation. IEEE Trans. Power Syst. 16,
105–110 (2001)
902 A. P. Hota et al.

25. Exposito, A.G., Santos, J.M.R., Garcia, T.G., Velasco, E.A.R.: Fair allocation of transmission
power losses. IEEE Trans. Power Syst. 15, 184–188 (2000)
26. Kumar, P., Gupta, N., Niazi, K.R., Swarnkar, A.: Exact cross-term decomposition method for
loss allocation in contemporary distribution systems. Arab. J. Sci. Eng. 44, 1–12 (2019)
27. Usman, M., Coppo, M., Bignucolo, F., Turri, R., Cerretti, A.: A novel methodology for the
management of distribution network based on neutral losses allocation factors. Int. J. Electr.
Power Energy Syst. 110, 613–622 (2019)
28. Masoud, K., Hassan, M., Aliabadi, M.H.: Loss allocation in distribution network including
distributed energy resources (DERs). Int. Trans. Electr. Energy Syst. 28, e2548 (2018)
29. Kumar, P., Gupta, N., Niazi, K.R., Swarnkar, A.: Branch current decomposition method for loss
allocation in contemporary distribution systems. Int. J. Electr. Power Energy Syst. 99, 134–145
(2018)
30. Sharma, S., Abhyankar, A.R.: Loss allocation of radial distribution system using Shapley value:
A sequential approach. Int. J. Electr. Power Energy Syst. 88, 33–41 (2017)
31. Sharma, S., Abhyankar, A.R.: Loss allocation for weakly meshed distribution system using
analytical formulation of Shapley value. IEEE Trans. Power Syst. 32, 1369–1377 (2017)
32. Ilic, M., Galiana, F., Fink, L.: Power System Restructuring: Engineering and Economics.
Kluwer, Norwell (1998)
33. Savier, J.S., Das, D.: Impact of network reconfiguration on loss allocation of radial distribution
systems. IEEE Trans Power Deliv. 22, 2473–2480 (2007)
34. Mishra, S., Das, D.: A novel active power loss allocation scheme for unbalanced radial power
distribution system. Int. Rev. Electr. Eng. 3, 543–550 (2008)
HSL Color Space Based Skin Lesion
Segmentation Using Fuzzy-Based
Techniques

P. Ganesan, B. S. Sathish, and L. M. I. Leo Joseph

Abstract Skin lesion is the anomalous intensification contrast to the skin just about
it. It is categorized as primary or secondary. The primary lesions are uncharacteristic
skin circumstances existence at birth. The secondary lesions are the result of manip-
ulated primary lesions. There are more than 20 types of skin lesions. Segmentation
is the process of partition of the test image into number of significant clusters. Every
cluster should be unique in terms of any one of the image attributes such as texture,
intensity, or color. The accomplishment of image analysis primarily based on the
upshot of the segmentation process. The proposed approach performs the skin lesion
segmentation using fuzzy c-means clustering (FCM), Possibilistic c-means clustering
(PCM). Possibilistic fuzzy c-means clustering (PFCM) and modified fuzzy c-means
clustering (PFCM). The experimental result reveals the competency of the MFCM
for skin lesion segmentation.

Keywords Segmentation · Color space · Skin lesion · Clustering · Fuzzy c-means

clustering

P. Ganesan (B)
Department of Electronics and Communication Engineering, Vidya Jyothi Institute of
Technology, Aziz Nagar, C.B. Post, Hyderabad, India
e-mail: [email protected]
B. S. Sathish
Department of Electronics and Communication Engineering, Ramachandra College of
Engineering, Eluru, Andhra Pradesh, India
e-mail: [email protected]
L. M. I. Leo Joseph
Department of Electronics and Communication Engineering, S.R. Engineering College,
Warangal, Telangana, India
e-mail: [email protected]

© Springer Nature Singapore Pte Ltd. 2020 903

G. Pradhan et al. (eds.), Advances in Electrical Control and Signal Systems,
Lecture Notes in Electrical Engineering 665,
https://doi.org/10.1007/978-981-15-5262-5_69
904 P. Ganesan et al.

1 Introduction

In segmentation, the complete image is clustered into number of subimages in a

meaningful manner. In a cluster, there should be some uniqueness among all the
image elements (pixels) [1–4]. In image processing and computer vision, the role of
the segmentation process is inevitable. The outcome of the segmentation process is
the basis for image analysis [5]. The segmentation is based on image attributes such as
color and texture [6]. Most of the segmentation techniques are application-oriented,
i.e., it is very hard to get a meaningful result for more than one application using
the same segmentation method [7–9]. This is a major drawback of the segmentation
techniques. There are lots of image segmentation techniques based on threshold,
edge, region, clustering, and so on. Skin lesion is the anomalous intensification
contrast to the skin just about it. It is categorized as primary or secondary. The primary
lesions are uncharacteristic skin circumstances existence at birth. The secondary
lesions are the result of manipulated primary lesions. There are more than 20 types
of skin lesions. In this work, fuzzy cluster-based methods are utilized to segment
the skin lesion images. The efficiency of the segmentation methods is measured in
terms of PSNR and computational cost. Prior to segmentation, the preprocessing to
progress the eminence of the skin lesion image and sharpening is performed on the
preprocessed image to enhance fine details. Most importantly, the enhanced image in
RGB is transformed into user-oriented, HSL space which fairly accurate the human
visual perception.

2 HSL Color Space

This cylindrical color space characterizes the color in more perceptually by three
elements as hue [color type varies from 0° to 360° i.e., red (0 or 360), yellow (60),
green (120), cyan (180), blue (240) and magenta (300)], the saturation which defines
the purity of the color as it varies from 0 to 100% and the lightness in terms of
percentage i.e., 100% of lightness is white and 0% is black [10–13]. The renovation
of an image form RGB to HSL is as follows:

2 (2R
1
− G − B)
H = arccos  (1)
(R − G) − (R − B)(G − B)
2

max(R, G, B) + min(R, G, B)
L= (2)
2

max(R, G, B) − min(R, G, B)
S= for L < 0.5 (3)
max(R, G, B) + min(R, G, B)

max(R, G, B) − min(R, G, B)
S= for L ≥ 0.5 (4)
2 − max(R, G, B) − min(R, G, B)
HSL Color Space Based Skin Lesion Segmentation … 905

3 Fuzzy Clustering for Skin Lesion Segmentation

Fuzzy c-means clustering is the standard algorithm based on cluster centers and
membership to partition a complete dataset into number of clusters [14, 15]. This
algorithm utilizes the Euclidean distance to compute the space between the pixels
(data points) and cluster center [16]. The major function of the algorithm is to lessen
its objective function provided in (5).


c 
n
 
m
Fm (U, V ) = μik xk − vi2  (5)
i=1 k=1

where m = the weighting exponent (fuzziness) parameter. Most cases, m = 2.

vi = cluster centers and μik = membership.
n
k=1 μik X k
m
vi =  n (6)
k=1 μik
m

⎧ ⎫
⎨ c 2/m−1 ⎬−1
x − vi 
μik =  k  (7)
⎩ xk − v j  ⎭
j=1

Possibilistic c-means (PCM) clustering provided the solution for the column sum
constraint of FCM [8, 17]. The objective function of PCM is illustrated in (8)


n 
c 
c 
n
Pm (T, V ; X, γ ) = tikm dki2 + γi (1 − tki )m (8)
i=1 k=1 i=1 k=1

where γ = weighting exponent parameter. When γ = 0 and m = 1, PCM behaves

as Hard C-Means (HCM) CLustering. In general, m = 2 and γ > 0 is the prefer-
able choice. The optimum (minimum) can be achieved if it satisfies the following
conditions.
 
dik 1/m−1
tki = 1/ 1 + , 1 ≤ i ≤ c; 1 ≤ k ≤ n (9)
γi
n
xk tkim
vi = k=1n m (10)
k=1 tki

The noise sensitivity problem of FCM and coincident clusters issue of PCM is
addressed by Possibilistic Fuzzy C-Means (PFCM) clustering [3]. The objective
function of PCM is shown in (11)


n 
c
 m η

c 
n
P Fm (T, V, U ; X, γ ) = aμik + btik dki2 + γi (1 − tki )η (11)
i=1 k=1 i=1 k=1
906 P. Ganesan et al.

where m, η > 1 and a, b, γ > 0. The optimum (minimum) can be achieved if it satisfies
the following conditions.
n  m η
k=1 aμik + btik x k
Vi = n  m η (12)
k=1 aμik + btik
1
tik =  2
1/(m−1) (13)
dik
1+ γi

1
μik =  2/m−1 (14)
c dik
j=1 d jk

The major drawback of standard FCM is that doesn’t provide any spatial informa-
tion which is significant for clustering problems. In the modified FCM (MFCM), this
is taken into consideration and the spatial information is incorporated as a weighted
sum of the membership function [2, 13].

 k∈( X j ) Uik αk2
Si j = Uik αk1 + c  (15)
k∈W ( X j ) t=1 k∈W ( X j ) Utk

The modified membership function is given by (16)

p q
Ui j ∗ Si j
Ui j(new) = c p q (16)
k=1 Uk j ∗ Sk j

It is noted that every pixel has a weight (W ji ) corresponding to clusters.

1
W ji = (17)
− n
 X j −Vi2 
j=1  X j −Vi ( n )
2 c
1+e

The objective function of MFCM is illustrated in (18)


n 
c
 m m  
MF = Uik W ji  X k − Vi2  (18)
k=1 i=1

The proposed approach for skin lesion segmentation is illustrated in Fig. 1.

HSL Color Space Based Skin Lesion Segmentation … 907

Skin lesion image is acquired from image

Test Image
database

Preprocessing Preprocessing to progress the eminence

of the skin lesion image

Image Enhancement Sharpening is performed on the pre

processed image to enhance fine details

Color Space Skin lesion image is transformed to HSV

Transformation space

Fuzzy based Fuzzy based methods applied on HSV

segmentation transformed image

Segmented lesion Outcome is the segmented lesion from

the image

Fig. 1 The proposed approach for skin lesion segmentation

4 Experimental Results and Discussion

Figure 2a illustrates the skin lesion test image to evaluate the efficiency of fuzzy-based
techniques for its segmentation. The original size of the input image is 586 * 561
(95.9 KB). The test image is sharpened to enhance its fine details as illustrated in
Fig. 2b. The HSL version of the image is depicted in Fig. 2c.
Figure 3a illustrated FCM clustering outcome of the HSL based skin lesion image.
This process took 11 iterations to segment the image into three clusters. PSNR is
computed as 45.436. The deviation of the segmented image from the input image is
error image.
Figure 3b depicts PCM clustering outcome of the HSL based skin lesion image.
This process took 15 iterations to segment the image into three clusters. PSNR is
computed as 44.960.
Figure 3c illustrated PFCM clustering outcome of the HSL based skin lesion
image. This process took 14 iterations to segment the image into three clusters.
PSNR is computed as 45.517.
MFCM clustering outcome of the HSL based skin lesion image is portrayed in
Fig. 3d. This process took 15 iterations to segment the image into three clusters.
PSNR is computed as 52.463.
The comparative end result of the proposed approach is demonstrated in Table 1.
908 P. Ganesan et al.

(a) test image (RGB) (b) sharpened image

(c) HSL version of the sharpened image

Fig. 2 Test image to test the efficiency of the proposed approach

MFCM has very PSNR and PFCM has low computational cost as compared to other
methods.

5 Conclusion

The skin lesion segmentation using fuzzy-based clustering methods is highlighted

in the proposed approach. The skin lesion image in RGB is transformed into user-
oriented, HSL which fairly accurate the human visual perception. The experimental
analysis clearly explained that the competency of the MFCM for skin lesion seg-
mentation. Even though the execution time for PFCM is lesser, MFCM has higher
PSNR as compared to other methods.
HSL Color Space Based Skin Lesion Segmentation … 909

(a) FCM result and its error image

(b) PCM result and its error image

(c) PFCM result and its error image

(d) MFCM result and its error image

Fig. 3 Outcome of HSL color space based segmentation

910 P. Ganesan et al.

Table 1 Comparative end result of the proposed approach

Method No. of No.of Computational Cluster centers PSNR
cluster iteration cost (s)
FCM 3 11 8.6083 232.9 155.0 164.7 45.436
13.44 42.84 126.1
14.15 49.10 116.4
MFCM 3 14 6.8608 245.1 222.5 13.66 52.463
253.1 82.34 45.77
253.3 97.34 121.6
PFCM 3 14 2.30138 246.8 254.5 255.6 45.517
235.5 84.46 92.58
13.47 45.47 121.5
PCM 3 15 11.7967 230.0 137.4 149.7 44.960
44.78 53.57 79.07
13.50 45.68 121.9

References

1. Lei, T., Jia, X., Zhang, Y., He, L., Meng, H., Nandi, A.K.: Significantly fast and robust fuzzy C-
means clustering algorithm based on morphological reconstruction and membership filtering.
IEEE Trans. Fuzzy Syst. 26(5), 3027–3041 (2018)
2. Ganesan, P., Sajiv, G.: User oriented color space for satellite image segmentation using fuzzy
based techniques. In: International Conference on Innovations in Information, Embedded and
Communication Systems (ICIIECS), pp. 1–6 (2017)
3. Pal, NR., Pal, K., James Bezdek, A.: A possibilistic fuzzy C means clustering algorithm. IEEE
Trans. Fuzzy Syst. 13(4), 517–530 (2005)
4. Ganesan, P., Palanivel, K., Sathish, B.S., Kalist, V., Shaik, K.B.: Performance of fuzzy based
clustering algorithms for the segmentation of satellite images—A comparative study. In:
IEEE Seventh National Conference on Computing, Communication and Information Systems
(NCCCIS), pp. 23–27 (2015)
5. Krinidis, S., Chatzis, V.: A robust fuzzy local information C-means clustering algorithm. IEEE
Trans. Image Process. 19(5), 1328–1337 (2010)
6. Sathish, B.S., Ganesan, P., Shaik, K.B.: Color Image segmentation based on genetic algorithm
and histogram threshold. Int. J. Appl. Eng. Res. 10(6), 5205–5209 (2015)
7. Chuang, K.S., Tzeng, H.L., Wu, J., Chen, T.J.: Fuzzy c-means clustering with spatial
information for image segmentation. Comput. Med. Image Graph. 30(1), 9–15 (2006)
8. Krishnapuram, R., Keller, J.: The possibilistic c means algorithm: Insights and recommenda-
tions. IEEE Trans. Fuzzy Syst. 1(2), 98–110 (1993)
9. Correa, C.: A comparison of fuzzy clustering algorithms applied to feature extraction on
vineyard. Lecture Notes in Computer Science, pp. 56–65. Springer, Berlin (2012)
10. Ganesan, P., Rajini, V.: Assessment of satellite image segmentation in RGB and HSV color
space using image quality measures. In: International Conference on Advances in Electrical
Engineering (ICAEE), pp. 1–5 (2014)
11. Kalist, V., Ganesan, P., Sathish, B.S., Jenitha, J.M.M.: possibilistic-fuzzy C-means clustering
approach for the segmentation of satellite images in HSL color space. Procedia Comput. Sci.
57, 49–56 (2015)
12. Sajiv, G., Ganesan, P.: Comparative study of possibilistic fuzzy C-means clustering based image
segmentation in RGB and CIELuv color space. Int. J. Pharm. Technol. 8(1), 10899–10909
(2016)
HSL Color Space Based Skin Lesion Segmentation … 911

13. Ganesan, P., Sathish, B.S., Sajiv, G: A comparative approach of identification and segmentation
of forest fire region in high resolution satellite images. In: 2016 World Conference on Futuristic
Trends in Research and Innovation for Social Welfare (Startup Conclave), pp. 1–6 (2016)
14. Ganesan, P., Sajiv, G.: Unsupervised clustering of satellite images in CIELab color space
using spatial information incorporated FCM clustering method. Int. J. Appl. Eng. Res. 10(20),
18774–18780 (2015)
15. Wang, Z., Song, Q., Soh, Y.C., Sim, K.: An adaptive spatial information-theoretic fuzzy clus-
tering algorithm for image segmentation. Comput. Vis. Image Understand. 117, 1412–1420
(2013)
16. Havens, T., Bezdek, J.C., Leckie, C., Hall, L.O., Palaniswami, M.: Fuzzy c-means algorithms
for very large data. IEEE Trans. Fuzzy Syst. 20(6), 1130–1146 (2012)
17. Jia, S., Zhang, C.: Fast and robust image segmentation using an superpixel based FCM
algorithm. In: IEEE International Conference on Image Processing (ICIP), pp. 947–951 (2014)
Industrial IoT-Based Gas Pipe Leakage
Detector Robot

Henil Goswami, Udhav Goyal, and John Sahaya Rani Alex

Abstract Energy resources such as gas often transported through pipes which play
a significant role in cities, industries. Gas leakage in pipes prompts misfortune just
as a danger since they can likewise prompt flame mishaps. The point of this work
is to structure an independent robot for in-pipe investigation. The system utilized
includes a focal pole whereupon a translational component is fitted which thusly is
associated with three edges of connections and wheels. DC motors are joined to the
wheels to accomplish the drive required. The component considers little settlement
in-pipe distances across. Setting sensors on each segment of pipe is in all respects
exorbitant. A pipe crawling robot prototype that can detect the leakage of Methane,
Butane, LPG, Smoke in a pipe by moving horizontally and vertically to prevent
mishaps in industries. The Internet of Things (IoT) module is interfaced with a robot
to give a real-time analysis on the cloud platform and GPS to give the exact location
of the leakage.

Keywords Pipe crawling robot · Oil and gas · Leak detection · IoT

1 Introduction

Vitality assets like fuel, gas, or oil moved through funnels and conveyed to the
spot of utilizations, for example, control age sources like atomic and warm power
plants which have a system of pipelines for different transportation capacities. These
systems square measure in actuality, very enormous for any developed nation. Trans-
portation of gaseous petrol by pipelines is quickly expanding in developing nations
like India. As of now, India has around 15,000 km of gas pipelines. These pipelines,
in any case, have restricted life because of different kinds of static and dynamic
burdens began from both inside and outside the pipe, just as because of normal pro-
cedures of debasement like oxidation and consumption of the pipe surface, and joint
disappointments because of the scraped spot.

H. Goswami · U. Goyal · J. S. R. Alex (B)

School of Electronics Engineering, Vellore Institute of Technology, Chennai 600127, India
e-mail: [email protected]

© Springer Nature Singapore Pte Ltd. 2020 913

G. Pradhan et al. (eds.), Advances in Electrical Control and Signal Systems,
Lecture Notes in Electrical Engineering 665,
https://doi.org/10.1007/978-981-15-5262-5_70
914 H. Goswami et al.

The Internet of things (IoT) is the system of electronic devices, which are identified
with installed frameworks and furthermore different areas through the web. The
framework resembles a moving robot with three motors and it comprises a gas sensor
to distinguish the gas leakage in the pipe. Gas leakage assumes imperative jobs for
urban areas, ventures, and accordingly in developing economies. Along these lines,
gas leakages cause misfortunes likewise as square measure a danger because they
will moreover cause hearth mishaps. Placing sensors on each section of the pipe is
very costly. So here we tend to propose associate degree innovative automaton that
adheres to the outer surface of the gas pipe and moves with the pipe to examine for
leakages. It consists of an MQ2 gas sensor to detect the gas leakages. The robot will
be moving continuously along the metal pipe if there any presence of leakage the
GPS sensor module will transmit the location to the cloud.

2 Literature Survey

A review of existing channel examination robots has been led to survey the pos-
sibility of this work and to figure out what may be the best methodology. It was
discovered that few pipe-examination robots have been developed tentatively; some
are even industrially accessible items. The least complex of those (and thusly the
chief fluctuated inside the market) territory unit stages laying on wheels or tractor
steps underneath. Such robots experience the ill effects of partner powerlessness
to deal with vertical segments of pipe. What’s more, many have restricted guiding
capacities and might be too long to even consider passing through even level elbows.
Envirosight LLC offers a progression of such robots in their line of ROVVER inves-
tigation crawlers; this region unit bound four-wheeled vehicles that convey a spread
of camcorders are accessible in sizes as little as 3 inches. Further developed pipe
reviewers territory unit equipped for going through vertical channels. One of these,
created by RoboProbe Technologies Inc., is equipped for going through complex
pipe systems from 8 to 12 in inward measurement three elastic track units squeezed
against the dividers in symmetrical design push this gadget. It isn’t verbalized, yet is
short enough to arrange generally turns. Various robots, to a great extent examina-
tion comes, are made of this sort, exploitation either tracks or wheels. An alternate
variation of crawlers is proposed in the writing [1–4]. The main objective behind
remote monitoring is to maintain a local (LAN) or remote (Internet) record of the
current status of appliances for managing and controlling home automation devices.
We have built such a system whereby the system administrator can track the status
and also automatically send a warning SMS to a mobile phone authority concerned
via GSM network if the conditions are unusual [5].
Industrial IoT-Based Gas Pipe Leakage Detector Robot 915

3 System Design

In this work, the design is to recognize gas line leakage is under pipeline through
web network and observing it every day. The current framework utilizes an LPG gas
sensor to detect LPG gas when LPG gas leakage happens. An LPG gas sensor module
to identify LPG gas has been utilized. At the point when LPG gas leakage happens,
a HIGH pulse will send to the Arduino board. At the point when the Arduino board
gets a HIGH pulse from a gas sensor, it shows a message on the LCD show and
enacts the signal to produce the blare sound. At the point when an LPG gas sensor
gives a LOW pulse to Arduino board, at that point, the showcase indicates “no gas
spillage” message. In the proposed framework, the robot continues moving along
the metal pipe it continues checking for any gas leakage, on discovery it utilizes an
interface GPS sensor to transmit the location of the leakage distinguished over to the
IoT login framework, here we use IoT to identify and found the location of leakage.
In this way, there is a completely computerized bug like a robot that moves with the
gas pipe and identifies gas spillages in a split second at a low spending plan [5–7].
This work can also be used in industrial applications for detecting pipeline leakages
with the larger size. The IoT plays a major role in this because the location of the
leakage can be seen on the IoT cloud platform. The block diagram for the proposed
work can be seen in Fig. 1.

4 Design Approach

One of the principle concerns influencing all structures and ventures is the required
support. This suggests squandering of time and cash important to keep from future
harms and to fix those as of now occurred. Specifically, extraordinary issues are
related to the support of channels conducts.

4.1 Design Calculations

The pipe crawling robot is a kind of inspection device, which can inspect inside the
pipe, by the way of electromechanical setup arrangements. The electromechanical
arrangements are helping to inspect inside the pipe. This setup arrangements having,
three motor wheel arrangements, which will be arranged in an equilateral triangular
shape. This is helping the movement of the robot inside the pipe. The 12 V battery,
Arduino system will give the control signal to the motor wheels and sensors. At
contraction position, is the initial position of the robot. At this position the all motors
and sensors are at rest position. Now, any other movement was happening in this
position.
916 H. Goswami et al.

ULTRASONIC
SENSOR (1)

LCD
GPS Module

MQ2 Gas Sensor (1)

Arduino Mega
Motor driver

BLUETOOTH
HC05 (1)
DC MOTOR (2)
ROBOT CHASE

IOT Module

POWER SUPPLY

1. 12V BATTERY (1)

Fig. 1 Block diagram of GAS pipe leakage detector robot

At extraction, in this position, the robot movement happened. So that is called

movement position. The movement was starting at, three motor wheel arrangements
are touching the internal surface of the pipe. After touching the wheel inside the
surface of the pipe, the ultrasonic sensor was sensing the surface, and given to the
Arduino system. This system could provide the control action for the central actuation
motor, and this motor stops the rotational motion to the main shaft, this impact could
stop the extraction movement. After touching the wheel, three motor wheels are
started by giving the electrical power input. So, the robot was moved forward to the
linear axis. The extraction position is 23 cm from the central axis [8–11].

5 Constraints and Trade-off

As per the requirements of the pins in this work use of Arduino mega is done. The
design parameter calculation is shown in Table 1. Table 2 indicates the pin details of
all the embedded boards available in the market. Table 3 indicates the pin details for
Industrial IoT-Based Gas Pipe Leakage Detector Robot 917

Table 1 Design parameters calculation of the robot

Calculation Result
For extraction
Height from the top 13 cm
For contraction
Height from the top 20 cm
Contraction distance from the extraction position 20–13 cm 7 cm
No of transferring threads 35
Contraction
√
The radius of the circumscribed circle (R) a/ 3 19.1 cm
Diameter of the pipe 2*R 38.1 cm
For extraction
Triangle base 36 cm
√
Radius of the circumscribed circle (R) a/ 3 20.78 cm
Diameter of the pipe (D) 2*R 41.5 cm
At contraction
The radius of wheel contact 17.8 cm
At extraction
The radius of the wheel contact 23 cm

Table 2 Comparison of
Pins. Raspberry pi 3 ESp8266 Arduino mega
different controllers based on
(NodeMCU)
the available pins
GPIO’s 40 (13 pin) 17 54
UART 1+1 2 4
(pinmuxed)
Cost (Rs) 4000 350 800

Table 3 Pin requirement for

Requirement Total PINs Components Pins
the components used
GPIO’s 18 LCD 7
Ultrasonic 2
sensor
motor pins 8
Gas sensor 1
UART 3UART GPS 1UART (2
(6pins) pins)
Bluetooth 1UART (2
pins)
IoT Module 1UART (2
pins)
918 H. Goswami et al.

Table 4 Placement of robot

Placing robot inside pipe Placing robot outside pipe
over the pipe pros and cons
Not required to go over Cannot go over clamps that
clamps join two pipes
Cannot be stolen from inside Can be stolen from outside
Will detect the change in the Can detect gases from the
density of gas from inside. atmosphere from nearby
sources
Manufacturing cost of the Manufacturing cost of the
robot is less robot is higher
Cannot be affected by the Can be affected by the
atmospheric gases atmosphere gases

the components used in this work. The placement of the robot whether to inside the
pipe or outside the pipe are explored and reasons are listed in Table 4.
Another constraint was whether to place the robot inside or outside the pipe. For
the listed reasons in Table 4, we decided to place it inside the pipe. These are some
of the issues that need to be addressed. When two pipes are joined, there are clamps
that are used to join the pipes. The robot designed cannot climb over the clamps,
which is one of the loopholes in making the pipe crawl outside the pipe. Integrating
the pipe outside the pipe will also increase the chances of theft [12]. To compare the
performance of methods used to detect gas, seven key attributes are defined: leak
sensitivity, location estimate capability, operational change, availability, false alarm
rate, maintenance requirement and cost [13].

6 Results and Discussions

A pipe crawling robot was structured as proposed and another controller board was
assembled to control the functioning of the robot. The control board contains a
number of modules. They are the IoT module for sending data back to the IoT
cloud platform. A GPS module is also added to give the real-time coordinates of the
robot with respect to the leakage. An HC-05 Bluetooth module is connected to the
mobile for giving commands to the robot via the Arduino Bluetooth application. An
ultrasonic sensor is placed to detect any obstacle inside the pipe, and finally, an MQ2
sensor to detect the leakage of LPG gas. Some of the necessary components required
are the motor circuit for connecting the robot to the controller and also the power
circuit to convert the 12 V into the required voltages for the specific components.
The robot structure contains a flat disc, threaded iron rod, joining fixed rods, L-
shape aluminum sheet, DC motor, and wheeled motor which is shown in Figs. 2, 3
and 4.
Industrial IoT-Based Gas Pipe Leakage Detector Robot 919

Fig. 2 L-shape rod

Fig. 3 Twisted rod

Fig. 4 The prototype of the

crawling robot

The robot structure can be seen in the next picture. As visible, the robot has three
tires that are used to crawl inside the pipe and the wheeled motors attached with the
tires to rotate them accordingly.
920 H. Goswami et al.

The controller board unit controls all the operations in this structure. The LCD
displays the value of the gas density and also the output of the ultrasonic sensor.
When the gas density is greater than 170, the LCD displays gas detected is shown in
Fig. 5.
The output of this unit is sent to the IoT cloud platform which is shown in Fig. 6.
The output is the coordinates of the robot at that particular time instance, as the robot
will only stop when it detects gas leakage.

Fig. 5 Electronic control

system for pipe crawler

Fig. 6 Gas leakage monitoring over cloud

Industrial IoT-Based Gas Pipe Leakage Detector Robot 921

7 Conclusion and Future Work

The exact location of the gas leak is located with the help of the robot. The location
of the gas leak is available on the cloud platform so the operation can be done from
any part of the world. The information is password protected so only an authorized
person can access the information. False Alarm rate is reduced.
In the future, for this robot to be implemented in the industrial sector, an ATEX
Certification is a must. This certification covers “equipment and protective systems
intended for use in potentially explosive atmospheres”. Integrating the current system
with this robot could cover any small loopholes. This will reduce the search time and
power consumption of the robot. It will also reduce the false alarm rate to almost
nil. Also, to make the robot fit for fixing the harms straightforwardly from inside,
sparing an immense measure of time and money.

References

1. A Compact and Compliant External Pipe Crawling Robot. IEEE Trans. Robot. 29(1) (2014)
2. Haas, D.K.: Small diameter pipe crawler. Savannah River National Laboratory Tech briefs,
U.S. patent 6,427,602 (2002)
3. Choi, C., Jung, S., Kim, S.: Feeder pipe inspection robot with inch worm mechanism using
pneumatic actuators. Int. J. Control Autom. Syst. 4(1), 87–95 (2006)
4. Schempf, H., Mutschler, E., Chemel, B., Boehmke, S.: BOA II and PipeTaz: Robotic pipe-
asbestos insulation abatement systems. In: Proceedings of IEEE International Conference on
Robotics and Automation, pp. 52–59. Albuquerque, NM (1997)
5. Sahani, M., Rout, S.K., Mandal, A.: Remote monitoring in home automation using low cost
microcontroller. In: 2014 International Conference on Communication and Signal Processing,
pp. 925–929. Melmaruvathur (2014)
6. IoT based smart emergency response system for fire hazards. In: 2017 3rd International Con-
ference on Applied and Theoretical Computing and Communication Technology (iCATccT)
(2017)
7. IoT based smart home automation system using sensor node. In: 2018 4th International
Conference on Recent Advances in Information Technology (RAIT)
8. US20060135065A1, [2004-12-17] [2006-06-22] Samsung Electronics Co., Ltd. Bluetooth
device and method for providing service determined according to bluetooth pin
9. Non destructive inspection of pipe using pipe inspection robot. In: IOSR Journal of Mechanical
and Civil Engineering (IOSR-JMCE) vol. 13, Issue 3, Ver. I e-ISSN: 2278-1684, p-ISSN:
2320-334X (2016)
10. IOSR Journal of Mechanical and Civil Engineering (IOSR-JMCE), 13(3), Ver. I (2016)
11. Zhang, Y., Yan, G.: In-pipe inspection robot with active pipe-diameter adaptability and
automatic tractive force adjusting. Mech. MachTheory 42(12), 1618–1631 (2007)
12. Roh, S, Choi, H.: Strategy for navigation inside pipelines with differential-drive inpipe robot.
In: Proceedings of the 2002 IEEE International Conference on Robotics and Automation (Cat.
No. 02CH37292), vol. 3, pp. 2575 (2002)
13. Designing a Cost Effective and Reliable Pipeline Leak Detection System by Dr Jun Zhang
REL Instrumentation Limited, Manchester, UK
Damping Control of a Multi—Area
Thermal System Using SSO Optimized
2DOF-PID-UPFC Based Controller

Puja Dash and Lalit Chandra Saikia

Abstract In this paper, a study is made to mitigate the damping and better dynamics
of an unequal multi-area thermal system by using UPFC based 2DOF-PID controller.
Controller’s Gains and other parameters are optimized concurrently via SSO tech-
nique. The dynamic responses are performed with and without the presence of UPFC
in between Area 1 & 2 and found a better response with UPFC in terms of various
stability parameters. Further sensitivity analysis has been performed for the pro-
posed controller with UPFC to verify the controller flexibility with variable changes
in system conditions with respect to the base case.

Keywords AGC · SSO technique · GRC · UPFC · Sensitivity analysis · Second

degree of freedom controller

1 Introduction

Application of FACTS Devices in AGC plays the key complementary role to uphold
system frequency and tie line power between the interconnected areas at their respec-
tive nominal values. The base interconnected system thermal system has been opted
from [1–3], which are revealing the detailed concepts of modeling three-area system
in presence of second degree of freedom controllers. As the FACTS controllers con-
trol the power flow along with power system stability improvement, many researchers
have incorporated FACTS Devices in AGC [2, 4–9]. In [2], the comparative stud-
ies between various FACTS controllers as SSSC, TCSC, TCPS, IPFC have been
performed in three-area thermal systems, wherein [10] it has been performed in
coordination with SMES in an interconnected two-area multiple-units hydro–hydro

P. Dash (B)
Department of EEE, Gayatri Vidya Parishad College of Engineering (A), Visakhapatnam, India
e-mail: [email protected]
L. C. Saikia
Department of EE, National Institute of Technology, Silchar, India
e-mail: [email protected]

© Springer Nature Singapore Pte Ltd. 2020 923

G. Pradhan et al. (eds.), Advances in Electrical Control and Signal Systems,
Lecture Notes in Electrical Engineering 665,
https://doi.org/10.1007/978-981-15-5262-5_71
924 P. Dash and L. C. Saikia

system. In [11, 12], the authors have proposed the redox flow batteries (RFB) coor-
dinate with IPFC for the improvement of the system stability in presence of load
disturbance of multi-area system in both conventional and deregulated environment.
Like IPFC, UPFC has also the versatility as it consists of both series and shunt con-
verter connected by a common dc-link capacitor [10], [7–9]. It has the capability for
simultaneous control of active and reactive power flow control [4–9]. In [6] and [8],
the performance of UPFC has been studied for two-area systems with PID, fuzzy PID
& PI2 controllers, respectively, in the conventional and deregulated environment. In
[9] the authors have been studied the single-machine system with a modified model
of UPFC. From the literature, it has been clearly seen that all the studies have been
done for mostly two areas and single-machine systems. Thus, the research scope of
system dynamic improvement with UPFC in combination to 2DOF-PID controller
[18] is there for three-area interconnected systems.
Numerous approaches as: DE [3], PSO [13], CS technique [1, 2], ABC [14], FA
[15, 8] have been adapted for the optimization of variables of various controllers
in AGC. A recent swarm optimization technique called Social spider optimization
(SSO) technique introduced by Cuevas et al. [10] has been applied by Attia et al.
[11] for the optimization of frequency controllers for autonomous two-area hybrid
microgrid system. Yet its effectiveness of the technique has not been verified in a
three unequal area thermal system with second degree of freedom controller in the
presence of any FACTS controllers. Hence, this requires further studies.
In view of the above, following are the main objectives of the present work.
(a) Employment of SSO optimized 2DOF-PID controller for the automatic control
of an unequal three-area thermal system.
(b) Employment of UPFC based 2DOF-PID controller for the improvement of
system stability.
(c) Analysis of the ability of the proposed Controlling Strategy to a wide change in
system conditions.

2 Investigated System

Three unequal areas interconnected thermal systems have been considered with
capacity ratio of 1:3:4. The thermal systems have single reheat turbine and 3% per
minute GRC.
The system parameters are taken from [2] and [16]. UPFC has been incorporated
in tie line 1–2.1% step load disturbance is considered in area 1. The transfer model of
the system is shown in Fig. 1. The objective function considered here is an Integral
squared error (ISE) and followed with the minimization of the ACE [1].
Damping Control of a Multi—Area Thermal System … 925

Fig. 1 UPFC connected between area 1 & 2 in the interconnected thermal systems

3 Mathematical Models of UPFC in AGC

The Unified Power Flow Controller (UPFC) is placed in the Tie line 1–2 shown
in Fig. 1 [7]. By controlling the parameters of a UPFC, the power flow through
the transmission lines/tie lines can be regulated [6, 7]. The solved transfer function
equation of the UPFC Controller is [6]

1
PUPFC (s) = F(s) (1)
1 + sTUPFC

where T UPFC is the Time Constant of UPFC.

4 Social Spider Optimization (SSO)

Social Spider Optimization (SSO) has been developed by Erik et al. [13] which
is based on the supportive manners of the social-spiders. The communal web of
the spiders is assumed as the search space. Each solution is represented by a spider
position and weight of each spider assigned based on the fitness value. Higher weight
value represents the maximum and lowest presents the minimum. The SSO algorithm
is referred to [13, 17]. The basic steps of the algorithm are shown in Fig. 2. For this
proposed system, the objective is the minimization of the Area Controller Error
(ACE). The best-tuned parameters are obtained for the particular problem are N =
250, N f = 69%.
926 P. Dash and L. C. Saikia

Fig. 2 Algorithm for SSO technique

Damping Control of a Multi—Area Thermal System … 927

5 Performance Analysis

A. Performance analysis of the system with: the 2DOF-PID and 2DOF-PID

based UPFC.

The dynamic responses of the system are compared with the 2DOF-PID controller
and 2DOF-PID controller-UPFC and revealed in Fig. 3.
From this comparison, it is clearly observed that the performance of the 2DOF-
PID based UPFC controller is better than that of the 2DOF-PID controller in the
improvement of system stability, i.e., rapid settling after the transient behavior, lower
damping, and lower peak deviations. The respective stability parameters are shown
in Table 1.
B. Sensibility Analysis

Fig. 3 Comparison of dynamic responses at nominal conditions

Table 1 Values of setting time, peak overshoot, and peak undershoot of Fig. 3
Controllers T S (s) MP MU
2DOF 2DOF 2DOF 2DOF 2DOF 2DOF
PID PID-UPFC PID PID-UPFC PID (−ve) PID-UPFC (−ve)
Figure 3a 35.01 31.1 0.01 0.0023 0.0244 0.0167
Figure 3b 39.56 28.01 0.0064 0.0019 0.0082 0.0109
Figure 3c 63.44 55.94 0.0005 0 0.0024 0.0023
928 P. Dash and L. C. Saikia

Sensibility analysis has been performed to observe the heftiness of the optimal gains
and parameters of 2DOF-PID based UPFC controller obtained at base loading condi-
tions to varied changes in loading conditions and SLP. For the analysis, optimization
of KPi , KIi , KDi , bi , ci , and N i , is for ±25% changed system loading from nominal,
placement of UPFC in Area 1 & 2 (in Tie line 1–2) at the changing magnitude of
SLP at 2 and 3% using SSO. The dynamic responses based on 2% SLP case & 25%
changed loading case with the corresponding values of KPi *, KIi *, KDi *, bi *, ci *,
N i * with respect to the base case and changed condition are compared. The com-
parisons of the responses are shown in Figs. 4 and 5. The settling time and the peak
deviations are noted from the Figs shown in Table 2 and the optimum values are
shown in Table 3.
From the dynamic performance curves, it has been clearly observed that as in
both the cases, the responses are approximately the same; therefore the parameters

Fig. 4 Comparison of dynamic responses at 2% SLP in area 1

Fig. 5 Comparison of Dynamic responses at 25%

Table 2 Setting time, peak

Optimized T S (s) MP MU
overshoot and peak
undershoot from Figs. 4 and Figure 4a 37.47 0.0053 −0.0243
5 Figure 4b 68.08 0 −0.0191
Figure 5a 34.25 0.0010 −0.0084
Figure 5b 68.01 0 −0.0093
Damping Control of a Multi—Area Thermal System … 929

Table 3 Optimum values of 2DOF-PID controller at different system

Gains Loading Size of SLP in area 1
25% 75% 1% 2% 3%
KP1 * 0.0515 0.0303 0.0003 0.6490 0.2440
KP2 * 0.0183 0.0491 0.0797 0.0417 0.0340
KP3 * 0.0013 0.0747 0.0024 0.0137 0.0388
KI1 * 0.0519 0.0190 0.0890 0.0710 0.0894
KI2 * 0.0929 0.0732 0.0362 0.0597 0.0199
KI3 * 0.7109 0.5000 0.3884 0.2222 0.2049
KD1 * 0.0120 0.0731 0.0511 0.0177 0.0055
KD2 * 0.3507 0.3087 0.4251 0.2335 0.2888
KD3 * 0.3100 0.3835 0.0431 0.3078 0.0513
b1 * 0.0179 0.0431 0.0198 0.2500 0.3800
b2 * 0.0498 0.0339 0.0438 0.0388 0.1387
b3 * 0.0317 0.0333 0.4657 0.0751 0.1880
c1 * 0.0292 0.0473 0.0695 0.2470 0.3391
c2 * 0.0435 0.1393 0.1213 0.1631 0.2176
c3 * 0.1214 0.4224 0.1719 0.4491 0.4033
N 1* 86.78 111.03 94.314 89.6336 77.751
N 2* 112.57 99.103 114.40 64.442 150.00
N 3* 122.517 75.300 55.071 96.1373 58.532

obtained at nominal conditions need not be returned to an extensive range of changes

in system loading conditions, and size of SLP.

6 Conclusion

The dynamic performance of the 2DOF-PID controller with UPFC in an intercon-

nected thermal system has been executed. The SSO technique is employed for simul-
taneous optimization of 2DOF-PID gains and parameters in the presence of UPFC in
Tie line 1–2. The following conclusions have been drawn from the vigorous studies
as:
1. In the presence of 2DOF-PID with UPFC, the system performed better in terms of
various stability factors as faster settling time, lower overshoots, and mitigation
of damping.
2. SSO is a very simple optimization technique in terms of application and faster
convergence time.
3. SSO optimized 2DOF-PID with UPFC is robust for wide changes in the system
loading or size of SLP.
930 P. Dash and L. C. Saikia

Appendix

System Data
T gi = 0.08 s; T ti = 0.3 s; T ri = 10 s; K ri = 0.5; K pi = 120 Hz/pu MW; T pi = 20 s;
T 12 = 0.086 pu MW/rad; T UPFC = 0.01 s; Population of spiders (N) = 250; Choice
of female spider (N f ) = 0.69 (69%)

References

1. Dash, Puja, Saikia, L.C., Sinha, N.: Comparison of performances of several cuckoo search
algorithm based 2DOF controllers in AGC of multi-area thermal system. Int. J. Electr. Power
Energy Syst. 55, 429–436 (2014)
2. Dash, Puja, Saikia, L.C., Sinha, N.: Comparison of performances of several FACTS devices
using cuckoo search algorithm optimized 2DOF controllers in multi-area AGC. Int. J. Electr.
Power Energy Syst. 65, 316–324 (2015)
3. Sahu, R.K., Panda, S., Rout, U.K.: DE optimized parallel 2-DOF PID controller for load
frequency control of power system with governor dead-band nonlinearity. Electr. Power Energy
Syst. 49, 19–33 (2013)
4. Huang, Z., Ni, Y., Shen, C.M., Wu, F.F., Chen, S., Zhang, B.: Application of unified power
flow controller in interconnected power systems modeling, interface, control strategy and case
study. IEEE Power Eng. Soc. Summer Meet. (1999)
5. Kazemi, A., Shadmesgaran, M.R: Extended supplementary controller of UPFC to improve
damping inter-area oscillations considering inertia coefficient. Int. J. Energy 2(1) (2008)
6. Sahu, R.K., Gorripotu, T.S, Panda, S: A hybrid DE–PS algorithm for load frequency control
under deregulated power system with UPFC and RFB. Ain Shams Eng. J. 6, 893–911 (2015)
7. Thirunavukarasu, R., Chidambaram, I.A.: PI2 controller based coordinated control with redox
flow battery and unified power flow controller for improved restoration indices in a deregulated
power system. Ain Shams Eng. J. 7, 1011–1027 (2016)
8. Pradhan, P.C., Sahu, R.K., Panda, S: Firefly algorithm optimized fuzzy PID controller for AGC
of multi-area multi-source power systems with UPFC and SMES. Eng. Sci. Technol. Int. J. 19,
338–354 (2016)
9. Nahak, N., Nahak, R.K.: Damping of power system oscillations by a novel DE-GWO optimized
dual UPFC controller. Eng. Sci. Technol. Int. J. 20, 1275–1284 (2017)
10. Bhatt, P., Roy, R., Ghoshal, S.P.: Comparative performance evaluation of SMES–SMES, TCPS–
SMES and SSSC–SMES controllers in automatic generation control for a two-area hydro–
hydro system. Electr. Power Energy Syst. 33, 1585–1597 (2011)
11. Chidambaram, I.A., Paramasivam, B.: Control performance standards based load-frequency
controller considering redox flow batteries coordinate with interline power flow controller. J.
Power Sources 219, 292–304 (2012)
12. Chidambaram, I.A., Paramasivam, B.: Optimized load-frequency simulation in restructured
power system with redox flow batteries and interline power flow controller. Electr. Power
Energy Syst. 50, 9–24 (2013)
13. Cuevas, E., Cienfuegos, M., Zaldivar, D., Cisneros. Perez M.: A swarm optimization algorithm
inspired in the behavior of the social-spider. Expert. Syst. Appl. 40(16), 6374–6384 (2013)
14. Gozde, H., Taplamacioglu, M.C., Kocaarslan, I.: Comparative performance analysis of artificial
Bee Colony algorithm in automatic generation control for interconnected reheat thermal power
system. Int. J. Electr. Power Energy Syst. 42, 167–178 (2012)
15. Saikia, L.C., Sahu, S.K.: Automatic generation control of a combined cycle gas turbine plant
with classical controllers using firefly algorithm. Int. J. Electr. Power Energy Syst. 53, 27–33
(2013)
Damping Control of a Multi—Area Thermal System … 931

16. Sancheza, J., Visioli, A., Dormido, S.: A two-degree-of-freedom PI controller based on events.
J. Process Control 21, 639–51 (2011)
17. Attia, A.E.F., Mohammed, A.E.H.: Efficient frequency controllers for autonomous two-area
hybrid microgrid system using social-spider optimizer. IET Gener. Transm. Distrib. 11(3),
637–648 (2017)
18. Araki, Mituhiko, Taguchi, Hidefumi: Two-degree-of-freedom PID controllers. Int. J. Control
Autom. Syst. 1(4), 401–411 (2003)
Integration of Electric Vehicles
in Distribution Grid for Maximizing
Profits Using MOGOA

Kumari Kasturi, Abhimanyu Behera, and Manas Ranjan Nayak

Abstract Integration of electric vehicles (EVs) in the existing grid creates an unex-
pected increase in load demand which may cause instability in the power system. In
this paper, the effects of integration of EV in 33-bus IEEE radial distribution system
(RDS) are analyzed. Here, grid to vehicle (G2V) and vehicle to grid (V2G) mode
of operations of EVs are considered. A Multi-Objective grasshopper optimization
algorithm (MOGOA) is used for finding optimal allocation of EVs in RDS while
minimizing cost of power loss, cost of power drawn from utility grid, and cost borne
by EV owner. The result shows the performance of EV operated in G2V and V2G
mode as compared to EV operated in the only G2V mode, so that both EV owner
and utility grid will be benefited.

Keywords Electric vehicles · Time of use pricing · Grid to vehicles · Vehicles to

grid · Multi-objective grasshopper optimization algorithm (MOGOA)

1 Introduction

Nowadays the world is facing some major problems like climate change and the
scarcity of natural oil. The petroleum-oil fed vehicles contribute more than half of
the carbon monoxide, nitrogen oxides, and almost a quarter of the hydrocarbons to
the environment. So the use of electric vehicles is becoming one of the key solutions
for this problem. Compared to gasoline-powered vehicles, electric vehicles are con-
sidered much clearer and cost-effective [1]. Therefore, adoption of electric vehicles
can reduce our dependence on fossil fuel-powered vehicles. EVs have advantages
like no carbon emission, less noisy, and highly efficient. But the integration of EV has

K. Kasturi (B) · A. Behera

Department of Electrical Engineering, Siksha ‘O’ Anusandhan University (Deemed to be
University), Bhubaneswar 751030, India
e-mail: [email protected]
M. R. Nayak
Department of Electrical Engineering, Biju Patnaik University of Technology,
Rourkela 769015, India

© Springer Nature Singapore Pte Ltd. 2020 933

G. Pradhan et al. (eds.), Advances in Electrical Control and Signal Systems,
Lecture Notes in Electrical Engineering 665,
https://doi.org/10.1007/978-981-15-5262-5_72
934 K. Kasturi et al.

some social, economical, and technical challenges [2]. While the EVs are integrated
with grid, it may draw energy from the service main in G2V mode of operation, or
it may deliver energy to the grid in V2G mode of operation [3]. While drawing the
energy in G2V mode it causes loading effects on the distribution transformer. In [4],
a stochastic modeling and simulation technique for analyzing the impacts of electric
vehicles charging on distribution network is presented by considering all operation
security risk information aspects. Three-phase distribution load flow is used for the
calculation of voltage and current. Result shows, the only impacts of electric vehicle
charging on the distribution grid.
In [5], behavioral characteristics and modeling requirements of PV, plug-in elec-
tric vehicles (PEV), and energy storage devices are discussed. Distributed resource
adoption coupled with smart grid implementations is expected to significantly alter
the nature of the distribution system which can be noticed form results. To solve
these problems we need to observe both the load demands of households and EVs.
Utilization of prepaid Watt meters can give us the different costs of power according
to peak and off-peak hours; this concept is known as time of use (TOU) [6]. By fol-
lowing the TOU concept, if EV owners are charging the EVs in the off-peak period
and discharging the EVs during the peak period they can generate some revenue for
them and the loading effect of the transformer can be minimized.
The main contributions of this paper are as follows:
(a) Impacts of EV on 33 bus radial distribution system (RDS) are analyzed.
(b) Comparison is done between EV operated in only G2V mode and EV operated
in both G2V and V2G mode.
(c) A novel Multi-objective grasshopper optimization algorithm (MOGOA) is used
to solve the proposed multi-objective problem.
In Sect. 2 the system modeling is described. In Sect. 3 problem formulation is
defined, whereas Sect. 4 describes MOGOA. Section 5 shows the results and analysis.
Finally, Sect. 6 gives the conclusion.

2 System Modeling

Here the electric vehicles are connected at the load bus. For charging, EV uses the
same bus where household utilities are connected. During charging hour of EV, it
takes power from grid and during the discharging period, EV delivers the power
to the grid. A smart controller is installed to control the power flow in an efficient
manner.
Integration of Electric Vehicles in Distribution Grid … 935

Fig. 1 Charging/discharging schedules of the EVs

2.1 Electric Vehicle Model

2.1.1 Electric Vehicle Model

In this paper, 33-bus IEEE radial distribution is considered as the test system [7]. EVs
can be charged/discharged during parking periods only, which is between 1.00 am
to 6.00 am and again between 5.00 pm and 12.00 am. The discharging of EV should
be such that, the EV should have a sufficient amount of power available in it for the
upcoming travels. The average energy consumed by the electric vehicle is 3 kWh.
The charging and discharging process depends on the peak or off-peak hour of the
power system. From 7.00 am to 1.00 pm and 4.00 pm to 10.00 pm is considered
as peak hours of the day, with an electricity price of 4.80 INR/kWh. Rest of the
hours is considered as off-peak hours, with an electricity price of 2.20 INR/kWh.
The charging or discharging schedules of the electric vehicles are shown in Fig. 1.

2.1.2 EV Charging/Discharging Model

The EV gets charged when it is parked during off-peak hours. Rest of hours during
day time, it gets discharge with a state of charge (SOC) of 40% to reserve power for
the need of traveling during upcoming hours.
The output power at the nth parking point is calculated as:

βn (t) · powmax
ch/dch with EVs connected
powPL,n (t) = (1)
0 otherwise

max maximum power received or supplied by the EVs, β (t) reflects the
where powch/dch n
variation of the EVs. The βn is considered as (i) −1 ≤ βn < 0 for (Grid to Vehicle),
936 K. Kasturi et al.

(ii) βn (t) = 0, when EVs are idle, and (iii) 0 < βn (t) ≤ 1 for (Vehicle to Grid). The
power of EVs for all parking stations are calculated as


NPL
powV2G/G2V (t) = powPL,n (t) (2)
n=1

NPL is the number of EVs are connected. The charging or discharging rate of qth
EV connected to nth parking point is shown as:

powPL,n (t) tn,q
s
≤ t ≤ tn,q
e
n,q (t) =
powEV (3)
0 otherwise
s e
where tn,q and tn,q are the initial and final instant of charging or discharging.
The state of charge (SOC) of EV battery for upcoming time is calculated as


T
SOC(t + 1) = SOCin + SOC(t) (4)
t=1

where, SOCin is the initial of EV battery.

2.2 Load Model

Here, the load demand of 33-bus IEEE RDS is modeled with the help of load factor.
The IEEE reliability test system load profile [7] is taken to formulate the 24-h profile
of load factors which is shown in Fig. 2. The load at bus n at the required time t can
be found as follows:

Fig. 2 Hourly weight factors

Integration of Electric Vehicles in Distribution Grid … 937

powLoad,n (t) = wh (t) × pn (5)

where wh (t) is the hourly weight factor and pn is the maximum load at bus.

3 Problem Formulation

Here we have to find the optimal allocation for electric vehicles, so that the
transmission loss, the load demand, and expenditure of EV buyers can be reduced.

3.1 Objective Function

The objective function is

Min( f 1obct , f 2obct ) (6)

f 1obct = cPL + cg (7)

where cPL and cg are the annual cost due to power loss in the branch and the annual
cost of power drawn from the main grid, respectively.cPL is calculated as


nt
cPL = [(powEV
L (t) − pow L (t)) × ce (t)]
B
(8)
t=1

where powBL and powEV L are the power loss before and after the installation of EVs
at the RDS, nt is the total time slot, i.e., 8760 h for one year.
At npth branch, we can calculate the power loss by the following equation:


nt
pow(t) L = 2
Rnp Inp (t) (9)
t=1

2
where, Rnp and Inp are the resistance and current of npth branch, respectively.
The amount of power that is drawn from the bus bar can be calculated as:


nt
∗
pow(t)sub = real(Vsub (t) × Isub (t)) (10)
t=1

where Vsub (t) is the voltage at the grid at time t and Isub (t) is the current drawn from
the grid at time t.
938 K. Kasturi et al.

The cost of the power drawn from national grid per year is:


nt
Cg = [(powEV
g (t) − powg (t)) × ce (t)]
B
(11)
t=1

where pow(t)EV B
g and pow(t)g are power drawn from substation after and before
integration of EV at time t, respectively.
Now, if the cost of the power that is drawn from the national grid and the self-
discharge of EV battery due to degradation can be minimized then the profit of the
EV buyers can be increased. This is the next aim of the multi-objective function. It
is calculated as


nt
f 2obct = [(powch (t) − powdch (t)) × ce (t)] + BDC (12)
t=1

where pow(t)ch and pow(t)dch are the power required for the battery for charge and
the discharge power supplied to the grid at time t, respectively.
The degradation of battery (BDC) is calculated as


Ndch
BDC = k
Cdch (DODin , DODfn ) (13)
k=1

where DODin is the initial depth of discharge. DODfn shows the final depth of dis-
charge in the kth iteration. The degradation cost for a discharge cycle is calculated
as
 
DODfn DODin
Cdch (DODin , DODfn ) = Cbat × E bat − (14)
a · DODbfn a · DODbin

where Cbat and E bat are storage cost of battery and battery capacity, respectively.
a and b are Wohler curve parameters vary from one kind of battery to another
kind.

3.2 System Operational Constraints

We have to operate the system within the limits of some equality and inequality
constraints which are given below.
EV
wch/dch ≤ wParking
EV
(15)
Integration of Electric Vehicles in Distribution Grid … 939


NEV
EV
wdemand = [(SOCmax − SOCinitial ) × wbat ] (16)
n=1

SOCmin ≤ SOC(t) ≤ SOCmax (17)

powsub (t) = powLoad (t) + pow L (t) + powEV (t) (18)

vimin ≤ vi (t) ≤ vimax (19)

max
Inp (t) ≤ Inp (20)

EV EV
where wch/dch and wParking energy absorbed and supplied by EV battery and maximum
EV
energy that can be supplied or absorbed by the EVs, respectively, wdemand is the EV
demand, SOCmax is the maximum SOC and SOCinitial initial SOC, vimin and vimax
are minimum voltage and maximum voltage at ith bus respectively, powsub (t) is the
max is maximum current in ijth branch. As per Electricity
substation power at time t,Inp
ACT, the voltage regulation allowed in a limit of 10% of the rated value (1 p.u.), i.e.,
vmin = 0.9 p.u. and vmax = 1.1 p.u. Maximum allowed capacity of the conductors
(I max ) used in the power system are described in reference [8].

4 Multi-objective Grasshopper Optimization Algorithm

(MOGOA)

The multi-objective optimization technique, can be realized by

Minimize f (x) = { f 1 (x), f 2 (x), . . . , f 0 (x)} (21)

Subject to g n (x) ≥ 0, n = 1, 2, 3, . . . , i, h n (x) = 0, n = 1, 2, …, p, and

l ≤ x n ≤ u n , i = 1, 2, . . . , m.
n

In a multi-objective problem, the best values from the set of solutions are selected
as the solution. For this Pareto optimal dominance is used in the multi-objective
problem (MOP), which is expressed bellow [9]:

∀n ∈ {1, 2, . . . , p} : f n (x) ≤ f n (y) ∧ ∃n ∈ {1, 2, . . . , p} : f n (x) < f n (y)

(22)

The solutions found by the Pareto dominance are known Pareto optimal solution.
This can be realized by

ps = {x, y ∈ s|∃y < x} (23)

940 K. Kasturi et al.

4.1 Gross Hopper Optimization Algorithm

The swarming behavior of grasshopper is simulated by the help of the Grasshopper

optimization algorithm. The position of grasshopper provides the solution to the
optimization problem. X n indicates the of grasshopper at nth location.

X n = Sn + G n + An (24)

where Sn : social interaction, Gn : gravity forces on nth grasshopper, and An : wind

advection.
The social interaction component of grasshopper can be visualized as


N
 
Sn = s dnp d̂np (25)
j=1
j =n

where, d np : distance of nth and pth grasshopper. It can be calculated as dnp =

 
x p − xn , s: strength of social forces, and d̂np = x p −xn is the unit normal vector
dnp
from nth grasshopper to the pth grasshopper. Here the x p and xn are constants.
The function s is the social force, which can be calculated as
−r
S(r ) = f e l − e−r (26)

where f shows the intensity of attraction and l is the attractive length scale. For large
distances the function s, is unable to produce strong forces between the grasshoppers.
So to solve this problem the distance need to be mapped or normalized to.
Now consider the ‘G’ in Eq. (24)

G n = −g êg (27)

where g is a constant for gravity and êw is the unit vector in the direction to the center
of earth.
Further, the A component in (24) can be realized as:

An = −u êw (28)

where u indicates the drift and êw is a unit vector in the direction of wind.
Integration of Electric Vehicles in Distribution Grid … 941

Finally, we can write Eq. (24) with all components as

N
  x p − x n
xi = s x p − xn  − g êg + u êw (29)
j=1
dnp
j =n

Here, N denotes the number of grasshoppers.

A stochastic algorithm needs to be performed for exploration and exploita-
tion effectively to find an accurate approximation of the global optimum. Special
parameters are used further to show the exploration and exploitation of optimization.
⎛ ⎞
⎜ 
N
ubd − lbd  i  x p − x n ⎟ 

xid = c⎜
⎝ c s x p − xnd  ⎟ + Td (30)
j=1
2 dnp ⎠
j =n

where ubd is the maximum limit in the dth dimension, lbd indicates the minimum

−r
limit in the dth direction. S(r ) = f e l − e−r , T d shows dth dimension in the target
and ‘c’ is a decaying coefficient which minimizes the Comfort area, repulsion area,
and attraction area. Here the gravity is not considered and for wind direction, it is


assumed to be towards the target, i.e., T d . The decrease in the attraction or repulsion
among the grasshoppers is formulated by the inner ‘c’. It is proportional to the number
of cycles. The decrease in the search area is realized by the outer ‘c’. We can update
the ‘c’ by the help of the equation:

cmx − cmn
c = cmx − l (31)
L
here, cmx denotes the maximum value, cmn denotes the minimum value, l denotes the
current iteration, and L indicates the number of iterations.

4.2 Multi-objective Grasshopper Optimization Algorithm

(MOGOA)

The multi-objective algorithm has two aims. Pareto optimal dominance is used to
compare the results in MOGOA. We have to choose the target in designing MOGOA
based on GOA. The target takes the search to the promising part of search area. The
updating process of the target is the main difference in MOGOA. The probability of
getting the target from the archive can be expressed as:

1
Pn = (32)
Nn
942 K. Kasturi et al.

where N n indicates the number of solutions near to the nth solution.

A roulette wheel distributes the solution properly, in the less distributed part of the
search area. In the case of early convergence, the crowded neighborhood is selected
as target. The problem here is that the archive should have minimum number of
solutions, which will reduce the computational cost of MOGOA. If a solution has
crowded neighborhood then we need to remove it. To do so we use inverse function
of Pi and roulette wheel. The solutions outside the archive are compared with the
solutions inside the archive with the help of MOGOA. If one of the members inside
the archive dominates the external solution then the solution is neglected with the
addition of the non-dominated solution. The main step in designing the MOGOA
[10] is the initialization, selection, generation, and replacement.

5 Result

In this proposed method, the power flow of 33-Bus IEEE is done by the help of
forward sweep and backward sweep algorithm. Time of use (TOU) concept is con-
sidered here. The buying and selling cost of the power is considered as same. The
MOGOA optimization technique is used to solve the proposed problem with popu-
lation of 50 and maximum iteration of 100. This simulation is done for one year. The
simulated results are shown in Table 1. The negative sign indicates the total revenue
generated by the EV owners due to the operation of their vehicles in G2V and V2G
mode.
In Fig. 3, the hourly regulation of voltage at bus number 18 is shown as bus 18 is
the weakest bus among all other buses of the proposed test system. From the figure,
it is observed that during peak hours if EV is operated as G2V mode the voltage
magnitude is just the same to the base case whereas, if EV is operated in G2V &
V2G mode the voltage magnitude is improved as EV discharges its power to the
grid. Maximum voltage during peak hours is 0.9281 p.u. if EV in G2V mode but it
is increased to 0.9354 p.u. if EV in G2V and V2G mode of operation.
Figure 4 shows the hourly change in load at bus number 18. During peak load
demand EVs shave off the peak load, making the system economically profitable
as EV is operated as G2V/V2G mode. The load on components of RDS is reduced

Table 1 Best solution for 33-bus radial distribution systems (RDS)

Location of Number of Power loss f 1,obj in INR f 2,obj in INR
bus EVs (KW)
Base case – – 802374 – –
EV operated 16,33 10,14 820673 30191.09 238251.72
in G2V mode
EV operated 16,33 10,14 823105 −19132.26 −116629.62
in G2V/V2G
Integration of Electric Vehicles in Distribution Grid … 943

Fig. 3 Hourly variation of voltage at bus number 18

Fig. 4 Hourly variation of load at bus number 18

during peak hours and is also proved to be cost-effective to EV buyers. But if EV is

operated as only G2V mode, the load on distribution grid is increased during charging
period of EVs. But in both G2V/V2G mode when EV is discharging power to the
grid, the demand is below the rated generation during peak hours, this will generate
some revenue for the EV owners.
SOC of EVs is shown in Fig. 5. As shown in figure initially SOC is at 90% and
it is decreasing during traveling hours only if EV is operated in G2V mode only.
During this operation mode, EV battery is charging in off-peak hours to ensure its
traveling.
If EV is operated as G2V and V2G mode, EV starts discharging during peak
hours and charging in off-peak hours to get benefit but at the same time, it also has
sufficient charge for traveling.
944 K. Kasturi et al.

Fig. 5 SOC of EV batteries

6 Conclusion

This paper shows the effectiveness of EV operated in G2V and V2G mode as com-
pared to EV operated in the only G2V mode so that both EV owner and utility grid
will be benefited. The MOGOA optimization technique is designed to reduce the
cost of both utility grid and EV owners. A significant improvement in power quality
can be achieved, if EVs will operate in both the V2G and G2V mode.

References

1. Bevis, K., Smyth, A., Walsh, S.: Plugging the gap—can planned infrastructure address resis-
tance to adoption of electric vehicles? In: European Transport Conference, Frankfurt, Germany,
pp. 1–21 (2013)
2. Cazzola, P., Gorner, M., Schuitmaker, R., Maroney, E.: Global EV Outlook 2016. International
Energy Agency, France (2016)
3. Kasturi, K., Nayak, C.K., Nayak, M.R.: Electric vehicles management enabling G2V and V2G
in smart distribution system for maximizing profits using MOMVO. Int. Trans. Electr. Energy
Syst. e12013 (2019)
4. Leou, R.C., Su, C.L., Lu, C.N.: Stochastic analyses of electric vehicle charging impacts on
distribution network. IEEE Trans. Power Syst. 29(3), 1055–1063 (2013)
5. Taylor, J., Smith, J.W., Dugan, R.: Distribution modeling requirements for integration of PV,
PEV, and storage in a smart grid environment. In: 2011 IEEE Power and Energy Society General
Meeting, pp. 1–6. IEEE (2011)
6. Dubey, A., Santoso, S., Cloud, M.P., Waclawiak, M.: Determining time-of-use schedules for
electric vehicle loads: a practical perspective. IEEE Power Energy Technol. Syst. J. 2(1), 12–20
(2015)
7. Grigg, C., Wong, P., Albrecht, P., Allan, R., Bhavaraju, M., Billinton, R., Chen, Q., Fong, C.,
Haddad, S., Kuruganty, S., Li, W., Mukerji, R., Patton, D., Rau, N., Reppen, D., Schneider, A.,
Shahidehpour, M., Singh, C.: The IEEE reliability test system-1996. A report prepared by the
reliability test system task force of the application of probability methods subcommittee. IEEE
Trans. Power Syst. 14(3), 1010–1020 (1999)
8. Baran, M.E., Wu, F.F.: Network reconfiguration in distribution systems for loss reduction and
load balancing. IEEE Trans. Power Deliv. 4(2), 1401–1407 (1989)
9. Censor, Y.: Pareto optimality in multiobjective problems. Appl. Math. Optim. 4, 41–59 (1977)
10. Holland, J.H., Reitman, J.S.: Cognitive systems based on adaptive algorithms. In: Pattern-
Directed Inference Systems, pp. 313–329. Academic Press, London (1978)
Video Object-Tracking Using Particle
Filtering and Feature Fusion

Jyotiranjan Panda and Pradipta Kumar Nanda

Abstract In this paper, a novel video tracking scheme is proposed using the notion
of particle filtering. For each pixel of the frame, two features namely Local Binary
Pattern (LBP) and the RGB are fused to generate a new feature. Fusion is carried
out in the probabilistic framework and the fusion coefficients are determined based
on trial and error. Particle filter based modeling is used to track the object in the
feature plane. The proposed scheme has been tested on different frames of different
benchmarked data sets and the performance of the proposed scheme is found to be
superior than the existing method.

Keywords Particle filtering · Color distribution · LBP · Bhattacharyya

coefficient · Fusion

1 Introduction

Tracking of a moving target object is an important research problem in the field of

computer vision. The problem becomes challenging due to the presence of occlusion,
clutter, intensity variation, camouflage, dynamic background, deformation of the tar-
get, etc. in the scene. As the object is moving of its own interest and the distribution
of object state is non-Gaussian, it makes the problem nonlinear and non-Gaussian
type which is more challenging. To handle such complexity and uncertainty, many
methods have been proposed. Out of all such methods, particle filter (PF) [1] is
more focused. Particle filter algorithm is based on optimal Bayesian estimation and
Monte Carlo model [2]. To represent the non-Gaussian posterior probability density
function of the state variables in the state-space it utilizes a set of weighted random
samples (known as particles) [3]. To overcome the nonlinearity it uses the Bayesian

J. Panda (B) · P. K. Nanda

Image and Video Analysis Lab, Department of ECE, Siksha ‘O’Anusandhan (Deemed to be
University), Odisha 751030, India
e-mail: [email protected]
P. K. Nanda
e-mail: [email protected]

© Springer Nature Singapore Pte Ltd. 2020 945

G. Pradhan et al. (eds.), Advances in Electrical Control and Signal Systems,
Lecture Notes in Electrical Engineering 665,
https://doi.org/10.1007/978-981-15-5262-5_73
946 J. Panda and P. K. Nanda

estimation method iteratively to have newly updated particles, which can reconstruct
the posterior probability density function of the next state of the system. The particle
weight is determined on the basis of features similarity between target object model
[4] and the observed image. Therefore, the key of the tracking algorithm based on
particle filtering is to find a reliable and robust target object feature. In the tradi-
tional object-tracking method based on particle filtering, a color histogram feature
is extracted and used for object-tracking [5]. The color distribution of the target is
by and large stable. Hence the color distribution is not sensitive to partial occlusion
and deformation. But in the case of occlusion, intensity variation in the scene, or
camouflage, this global color feature fails to describe the object accurately. In such
a situation, the performance of this particle filter object-tracking method using only
color feature deteriorates.
Besides color features, motion and edge features are also the preferred features for
tracking. But typically in outdoor scenarios, the scene has many textural attributes.
Hence texture [6, 7] is considered as another important appealing feature to represent
the target object in outdoor scene. When the color of the object is similar to the
background scene, there may be variation in the textural properties. Ding et al. [8]
have proposed a linear weighted fusion method to fuse the color and texture features
of the object and established a reference model for it in particle filtering framework.
The method has achieved superior robustness and stability while tracking in the
presence of illumination variation and occlusion. Lin et al. [9] have proposed an
improved patch-based appearance model for object, considering color and motion
vector feature in particle filtering framework. They have used speeded-up robust
features to recover the error and successfully track an object with partial occlusion
and large change in the appearance of object. Ruohong et al. [10] have proposed
a new anti-occlusion particle filter object-tracking method based on fusion of LBP
and color feature with a deterministic coefficient. Their proposed method effectively
describes an object which results improve in tracking stability and robustness under
the occlusion conditions.
In this paper, we have proposed a particle filter based object tracker using fusion
of two different features with a view to characterizing the object efficiency. We have
in this regard, proposed a block-based method to enhance the robustness of color-
based tracking in outdoor scene by fusing texture features as well as color features.
As Local Binary Pattern (LBP) is one of the efficient methods to model the texture,
for each pixel of the frame, two features namely Local Binary Pattern (LBP) and the
RGB are fused to generate a new feature. Fusion is carried out in the probabilistic
framework and the fusion coefficients are determined based on trial and error. The
particle filter based modeling is used to track the object in the new feature plane.
This paper is organized as follows. Basics of particle filtering is explained in
Sect. 2, extraction of feature is described in Sect. 3, the feature fusion strategy is
presented in Sect. 4, particle filter based modeling is explained in Sect. 5, results
and their analysis is described in Sect. 6 and Sect. 7 presents conclusions and future
research directions.
Video Object-Tracking Using Particle Filtering and Feature Fusion 947

2 Particle Filtering

Particle filtering [1, 2] is initially used to track the target objects in clutter. The
state of the target object is described by the state vector X t and all the observations
{z 1 , z 2 , . . . , z t } up to time step t are denoted by the vector Z t . When the posterior den-
sity p(X t |Z t ) and the observation density p(z t |X t ) are non-Gaussian, particle filters
are preferred to use for state-space representation of both the density function. The
key idea of particle filtering is to approximate the probability distribution of the state
variables by recursively reconstructing  the posterior
 probability distribution using a
weighted random sample set S = s (n) , π (n) |n = 1, 2 . . . , N [11, 12]. Each sam-
ple ‘s’ represents one hypothetical state of the object, with a corresponding N discrete
sampling probability π known as the weight of the particles, where n=1 π (n) = 1.
The evolution of the state variables with time step are described by propagating each
sample according to a system model. Each element of the  sample set is then weighted
in terms of the observations with probability π = p z t |X t = st(n) . At each time
(n)
N
step the mean state of an object is estimated by E[S] = n=1 π (n) · s (n) . Hence
particle filtering provides a robust tracking framework to track the distribution of
state variables, as it models the uncertainty.

3 Feature Extraction

In the case of moving object-tracking based on particle filtering, each particle weight
depends on the similarity between the template feature and the object feature. Higher
similarity measure leads to larger weights of the particles. In this paper, the nature
and the characteristics of the target object [13] are described by color feature and
texture feature. Both of the two characteristics are brought together to fuse by the
histogram method.

3.1 Color Feature Extraction

A color histogram [8] uses the proportion of number of pixels with particular color
values with respect to total number of pixels describe that image in that image frame.
The similarity of two images is compared by the distance on the histogram which
can be determined by using the Bhattacharyya coefficient [14]. The shorter will be
the distance between two histograms the more similar will be the pixel colors.
The weighted color distribution model of the target area pu (y) = { pu (y)}u=1,2,...,m
is as follows:
948 J. Panda and P. K. Nanda

N
||y − xi ||
pu (y) = f k δ[h(xi ) − u] (1)
i=1
a

1 − r 2, r < 1
k(r ) = (2)
0, r ≥1

where r is the radial distance from the selected region center y · k() is the weighting
function to the color value with respect to the location of observed pixel from bound-
ary. xi is the point in the target location. δ() is the Kronecker Delta function. h(xi )
assigns color to xi . The parameter a represents the size of the target location. The
degree of similarity between target templates q(y0 ) and the candidate region p(y) is
evaluated by Bhattacharrya distance dc [4, 14] which is expressed as,

dc = 1 − ρ( p, q) (3)

where,
m
ρ( p, q) = pu (y)qu (y0 ) (4)
u=1

3.2 Texture Feature Extraction

Texture gives information regarding the local irregularity of the image [6, 7]. It
describes the relationship of one pixel with its surrounding pixels. As it is insensitive
to changes in light intensity, it can overcome the deficiency of the color histogram
method whose performance degrades by the change of intensity and background
color. Local binary patterns (LBP) is an operator used to describe the local texture
features of an image. LBP is defined as follows:

p−1
 
LBP P,R (xc , yc ) = s g p , gc 2 p (5)
p=0

1, x ≥ T
s(x) = (6)
0, x < T

where P represents the neighborhood pixels. R represents the distance between the
center pixel xi and the neighborhood pixels. gc is the gray level information of the
center point yc while g p represents the gray level information of the Nth equal points
in the circle taking center as yc and radius R. T is a threshold that is used to model
the uniform regions effectively. LBP value of each pixel is evaluated corresponding
to the surrounding pixels to obtain the LBP histogram of the object.
Video Object-Tracking Using Particle Filtering and Feature Fusion 949

4 Feature Fusion

The ability to describe objects in a complex outdoor environment is poor for a single
feature. Hence fusion of multiple features can be an effective method to increase
tracking accuracy and robustness. In our tracking method, the computed weights of
the particles are found to be more accurate in case of fusion of feature than that of
independent features, thus improving the tracking ability. The process details are
shown in Fig. 1. The target object is allowed to be represented by a rectangular
template, and this template is characterized by jointly the color and LBP histogram.
In order to capture the local properties by LBP features, the tracking window is
partitioned into four equal sub-windows. In each block (sub-widows), we have found
the mean RGB color histogram and LBP histogram. The two histograms of a given
partition are fused probabilistically. In this case, each bin of the histogram is weighted
and added to obtain the fused bin. Fusion of each bin of the histogram results in
the distribution of the fused features. Generally, in the case of additive fusion, the
observed probability is defined as,

p(z1, z2|x) = ηp(z1|x) + (1 − η) p(z2|x) (7)

where p(z1 | x) and p(z2 | x) are the probabilities of observations corresponding to

the color and LBP features, and η is the coefficient of fusion that is chosen on trial
and error basis. Additive mean value of the new feature probability of all four blocks
produces a new feature probability of that corresponding rectangular object region
known as particle. The weight of each particle w in the particle filter tracking algo-
rithm is calculated by using the similarity measure of targets and tracking rectangular
regions. It gives us a more accurate estimated mean state of the object of interest as
compared to the simple color-based tracking.

Fig. 1 Feature fusion strategy

950 J. Panda and P. K. Nanda

5 Particle Filter Based Tracking

Salient steps of the algorithm are enumerated below.

Step 1: Initialization

(a) Each sample of the prior distribution represents

 a rectangle, which
 is con-
sidered to be a particle and given as, s = x, y, vx , v y , Hx , Hy , where x, y
specifies the location of rectangle, vx , v y represents the motion vector and
Hx , Hy represents the height and width of the rectangle. Let we have a sample
set St−1 .
(b) We have manually selected a specified rectangle over the target object in a ref-
erence frame. By applying our proposed fusion strategy we have formulated
our target model q which is represented as follows.

1
4
  1 4
 
q= p j (z1, z2|x) = ηp j (z1|x) + (1 − η) p j (z2|x) (8)
4 j=1
4 j=1

where j represents each block.

Step 2: Selection
We have selected N = 100 samples (particles) from the sample set St−1 with
(n)
corresponding probabilities πt−1 , n = 1, 2, …, 100.

(a) The normalized cumulative probabilities ct−1 is computed as follows.

(0)
ct−1 =0
(n) (n−1) (n)
ct−1 = ct−1 + πt−1
(n)

(n) ct−1
ct−1 = (N )
(9)
ct−1

(b) Generate a uniformly distributed random number γ ∈ [0, 1] and

( j)
(c) Apply the binary search to find out the smallest j for which ct−1 ≥
γ , for j = 1, . . . , 100. Thus
(n) ( j)
(d) We have set all 100 samples as st−1 = st−1 .

Step 3: Propagation
As the particle filter tracker has the capability to handle the multiple hypothesis
(n)
simultaneously [1], each sample from sample set st−1 (called a particle) is prop-
agated through an application of a dynamic model to the time step t. The model
can be represented by linear stochastic differential equation as,
(n)
st(n) = Ast−1 (n)
+ wt−1 (10)
Video Object-Tracking Using Particle Filtering and Feature Fusion 951

(n)
where n = 1, 2 . . . , 100, and wt−1 is a multivariate Gaussian random variable.
“A” is the state transition matrix that defines the deterministic components of the
state for the state transition model.
Step 4: Observation

(a) For each of the hypothetical rectangular region, specified by the sample set
st(n) , we have applied our fusion strategy described by (8) and found out the
distribution of the new feature Pst(n) at each of the hypothetical region.
(b) On comparing the distribution of the target histogram q, the Bhattacharyya
coefficient is computed for each of the samples representing the hypothetical
region as,

  m 
ρ Pst(n) , q = p u(n) · q u (11)
st
u=1

where u is the bins of the histogram.

The Bhattacharyya distance “d” will be calculated as,
  
d= 1 − ρ Pst(n) , q (12)

(c) Weights πt(n) of each particle of the set st(n) are calculated by using the
Bhattacharyya distance “d” and a Gaussian distribution of variance σ 2 as,
 
1 d2
− 2σ
πt(n) = √ exp 2
(13)
2π σ

Step 5: Estimation
With the help, the normalized weight calculated on the above step all the samples
are moved towards their mean state as,

N
E[St ] = πt(n) st(n) (14)
n=1

Step 6: Iterations
Repeat steps 2–5 for a fixed number of iterations.
952 J. Panda and P. K. Nanda

6 Results and Discussions

In our experiments, LASIESTA and VISUALTRACKER benchmark video databases

are used to validate our proposed method. Testing video 1 is from the VISUAL-
TRACKER and Testing video 2 is from the LASIESTA video database. Video 1
has outdoor sequences of the resolution (240 × 320)p × 2 with frame rate 25 fps.
These frames include the challenges of clutter background, partial occlusion, moder-
ate shadows, out of phase rotation due to camera motion, and dynamic background.
Video 2 has outdoor sequences of the resolution (288 × 352)p × 2 with frame rate
25 fps. It includes a dynamic background, camouflage with hard shadows. The per-
formance of the traditional particle filter with color histogram is compared with that
of the proposed method based on feature fusion. We set the number of particles
N = 100 and one of the best possible fusion coefficient η = 0.4, and we have used
the Bhattacharyya distance as the similarity measure.
Test video 1 is an outdoor environment having a pedestrian video sequence, where
the target is occluded by a pole while walking at side of the road. Figures 2 and 3
show the experimental results of both algorithms. It is seen that traditional tracking
algorithm in particle filtering framework based on single-color histogram feature
results in wrong tracking just after the occurrence of occlusion. Whereas the proposed
method in this paper achieves more accurate tracking even if occlusion occurred. Due
to occlusion, the color feature distribution has been changed by the intensity of color
of the occluded body. Hence the tracker is unable to come back from the deviated
state and leads to the loss of tracking. As the local texture feature is invariant to
intensity change, our proposed method keeps the tracker intact in the neighborhood.

Fig. 2 Tracking results of different frames for test video 1 in traditional particle filter method.
a Frame no. 14, b 18, c 22, d 26, e 30, f 34
Video Object-Tracking Using Particle Filtering and Feature Fusion 953

Fig. 3 Tracking results of different frames for test video 1 in the proposed particle filter method.
a Frame no. 14, b 18, c 22, d 26, e 30, f 34

Test video 2 is an outdoor environment having a pedestrian video sequence, where

the target is walking just after the rain. Camouflage with hard shadows is the challenge
in this scene. Figures 4 and 5 show the experimental results of both algorithms. It
is seen that, due to hard shadows, as intensity variation occurred drastically in the

Fig. 4 Tracking results of different frames for test video 2 in traditional particle filter method.
a Frame no. 129, b 132, c 135, d 138, e 141
954 J. Panda and P. K. Nanda

Fig. 5 Tracking results of different frames for test video 2 in proposed particle filter method.
a Frame no. 129, b 132, c 135, d 138, e 141

scene, the color distribution of the object changes a lot. Hence the tracker deviates by
the influence of the camouflage and intensity variation. But the presence of LBP in
our proposed algorithm, which is invariant to the intensity change, keeps the tracker
intact with the object. Thus it proves that the improved method of this paper is clearly
superior to the traditional tracking algorithm.
The ability of moving object-tracking in feature fusion based particle filter method
is clearly superior to the particle filter method based on the single-color feature. The
tracking rectangular region of the proposed feature fusion method shows more accu-
rate tracking of the target person as compared to the single-feature-based tracking.
To verify the accuracy and effectiveness further, the errors in tracking are compared
and analyzed for both the methods. Tables 1 and 2 present the center coordinate of
the actual object center and tracking object center of video 1 and video 2. Tables 3
and 4 represents the tracking errors of both the method for video 1 and video 2 which

Table 1 Center coordinates

Sequence Actual Traditional Proposed
of actual object and tracked
number of coordinates method method with
object obtained by both the
frame η = 0.4
methods for test video 1
14 (118, 72) (121, 87) (111, 75)
18 (118, 80) (115, 113) (104, 91)
22 (116, 88) (108, 138) (97, 106)
26 (114, 94) (102, 164) (90, 122)
30 (114, 101) (110, 151) (83, 137)
34 (112, 107) (139, 158) (76, 148)
Video Object-Tracking Using Particle Filtering and Feature Fusion 955

Table 2 Center coordinates

Sequence Actual Traditional Proposed
of actual object and tracked
number of coordinates method method with
object obtained by both the
frame η = 0.4
methods for test video 2
129 (165, 55) (165, 49) (169, 55)
132 (170, 64) (166, 56) (172, 67)
135 (172, 73) (166, 62) (174, 78)
138 (172, 82) (167, 69) (177, 90)
141 (174, 90) (167, 75) (179, 101)

Table 3 Tracking errors

Sequence number Traditional Proposed method
obtained for object center by
of frame method with η = 0.4
both the methods for test
video 1 14 15 9.4
18 33 17
22 50 26
26 71 36
30 50.4 47
34 57 54

Table 4 Errors obtained for

Sequence number Traditional Proposed method
tracking the object center by
of frame method with η = 0.4
both the methods for test
video 2 129 6 5
132 8.9 3.6
135 12.5 5.4
138 14 9.4
141 16.5 12

is also graphically represented in Fig. 6. The error is the Euclidean distance between
the centroid of the tracked rectangle frame and center of the real object (manual
acquisition). The error is defined as

e= (x  − x)2 + (y  − y)2 . (15)

On the basis of the results, we can easily analyze that, the proposed color and
texture features fusion method has better tracking accuracy than the traditional color-
based object-tracking method in particle filtering framework.
956 J. Panda and P. K. Nanda

Fig. 6 Tracking errors obtained for object centers in both the test videos a video 1, b video 2

7 Conclusions

A feature fusion based particle filter for tracking video objects is proposed. It is
found that the color feature based tracking algorithm in particle filtering framework
is highly sensitive to variation of illumination and also easily affected by background.
The proposed method can overcome the effect of illumination change and camouflage
more effectively. When the influence of the environment is serious over the target,
the numbers of particles can be varied to achieve the tracking accuracy. The results
from the experiments of the proposed method can achieve the target of performance
accuracy and tracking stability, in case of different challenges. The focus of the
future work is to study the fusion of features adaptively with scene dependant fusion
coefficients and achieve more tracking accuracy in complex scenes for single and
multiple objects.

References

1. Gordon, N.J., Salmond, D.J., Smith, A.F.M.: Novel approach to nonlinear/non-Gaussian

Bayesian state estimation. IEE Proc. F 140(2), 107–113 (1993)
2. Doucet, A., Freitas, N.D., Gordon, N.: Sequential Monte Carlo methods in practice. Springer,
New York (2001)
3. Arulampalam, M.S., MasKell, S., Gordon, N., Clapp, T.: A tutorial on particle filters for online
nonlinear/non-Gaussian Bayesian tracking. IEEE Trans. Signal Process. 50(2), 174–188 (2002)
4. Nummiaro, K., Koller-Meier, E., Van Gool, L.: An adaptive color based particle filter. Image
Vis. Comput. 21(1), 99–110 (2003)
5. Nummiaro, K., Koller-Meier, E. Van Gool, L.: A color-based particle filter. In: First Interna-
tional Workshop on Generative-Model-Based Vision in conjunction with ECCV’02, pp. 53–60
(2002)
6. Marko, H., Matti, P.: A texture-based method for modelling the background and detecting
moving objects. IEEE Trans. Pattern Anal. Mach. Intell. 28(4), 657–662 (2006)
Video Object-Tracking Using Particle Filtering and Feature Fusion 957

7. Ojala, T., Pietikainen, M., Maenpaa, T.: Multiresolution gray-scale and rotation invariant texture
classification with local binary patterns. IEEE Trans. Pattern Anal. Mach. Intell. 24(7), 971–987
(2002)
8. Ding, D., Jiang, Z., Liu, C.: Object tracking algorithm based on particle filter with color and
texture feature. In: Proceedings of the 35th Chinese Control Conference, China, pp. 4031–4036
(2016)
9. Lin, S.D., Lin, J.J., Chuang, C.Y.: Particle filter with occlusion handling for visual tracking.
IET Image Process. 9(11), 959–968 (2015)
10. Ruohong, H., Shenglin, B., et al.: Anti-occlusion filter object-tracking method based on feature
fusion. IET Image Process. 12(9), 1519–1530 (2018)
11. Isard, M., Blake, A.: Contour tracking by stochastic propagation of conditional density. In:
European Conference on Computer Vision, pp. 343–356 (1996)
12. Isard, M., Blake, A.: CONDENSATION—conditional density propagation for visual tracking.
Int. J. Comput. Vis. 1(29), 5–28 (1998)
13. Pérez, P., Hue, C., Vermaak, J., et al.: Color-based probabilistic tracking. In: Proceedings of
the European Conference on Computer Vision (ECCV’02), pp. 661–675 (2002)
14. Aherne, F., Thacker, N., Rockett, P.: The Bhattacharyya metric as an absolute similarity measure
for frequency coded data. Kybernetika 32(4), 1–7 (1997)
Voltage and Frequency Stabilization
of a Hybrid Autonomous System Using
SMC-Based STATCOM Equipped
with BESS

Abhijeet Choudhury, Soumya Mohanty, Swagat Pati, Amar Bijay Nanda,

Amiya Kumar Naik, and Sanjeeb Kumar Kar

Abstract This paperwork focuses on a hybrid microgrid framework, which consists

of various energy sources like a micro-hydro-based SEIG (22 kW) and a photovoltaic
system (18 kW). Both the energy sources are interfaced by a bidirectional converter
(BDC) which acts as a STATCOM. At the DC side of STATCOM, along with PV,
a battery energy storage system is connected. During heavy/light loading condition,
the power (reactive and real) generated by SEIG is insufficient, which further tends
to fluctuation in system voltage and frequency. Since the STATCOM has capability
of transferring power in bidirectional way, hence, the STACOM stores the excess real
power in the BESS and supplies it when required by the system, and one of the major
advantages of STACOM is it enhances the quality of power throughout the system.
The entire hybrid system is subjected to three loads, i.e., dynamic, nonlinear, and
linear loads. The entire system is simulated using Simulink/MATLAB environment,
and the results are studied.

Keywords Static synchronous compensator (STATCOM) · Self-excited induction

generator · Battery energy storage system (BESS) · Photovoltaics (PV) ·
Independent control

A. Choudhury (B) · S. Mohanty · S. Pati · A. B. Nanda · A. K. Naik · S. K. Kar

Department of EE, Institute of Technical Education and Research, Siksha ‘O’ Anusandhan
(Deemed to be University), Bhubaneswar, Odisha 751030, India
e-mail: [email protected]
S. Mohanty
e-mail: [email protected]
S. Pati
e-mail: [email protected]
A. B. Nanda
e-mail: [email protected]
A. K. Naik
e-mail: [email protected]
S. K. Kar
e-mail: [email protected]

© Springer Nature Singapore Pte Ltd. 2020 959

G. Pradhan et al. (eds.), Advances in Electrical Control and Signal Systems,
Lecture Notes in Electrical Engineering 665,
https://doi.org/10.1007/978-981-15-5262-5_74
960 A. Choudhury et al.

1 Introduction

This paper depicts a hybrid autonomous system, in accordance with time-varying

loads. Though it is an isolated system, the power deficit/excess is taken care by
STATCOM [1]. Among the accessibility of generators, self-excited induction gener-
ators (SEIGs) are truly useful for inexhaustible generation of power at remote areas
[4]. SEIGs are robust; brushless, less upkeep arrangement and requirement of any
external source for excitation is negligible which makes these generators generally
satisfactory at remote areas.
In order to maintain the voltage at the SEIG terminal to be constant, a capacitor
bank is used. Though SEIG is sufficient for delivering power to various loads in
repeated interval of time, due to load fluctuations, the system voltage and frequency
get disturbed. Apart from other generators working in knee region, the self-excited
induction generators always work in the saturation region. Broader research in this
area has been done to ensure that the voltage across the SEIG terminal remains con-
stant; for this reason, various FACTS controllers are used like saturable core reactors,
SVCs [2], and short shunt connection of capacitors [3]. Regardless of whether all
these previously mentioned schemes can intently fulfill the prerequisite and to some
degree advantageous, however, it additionally injects harmonics to the generating
system. In any case, steady-state improvement in the region of solid-state semicon-
ductors gives another face to a VR called STATCOM; it has come into the limelight
which trades consistently adjusting reactive power capacity to/from the system to
keep constant voltage across the SEIG terminal. From the last two decenniums, not
only solar [6–10] but also wind energy has evolved into a wide range of awareness
which proves to generate huge amount of clean energy. But the challenge with all
kinds of renewable energy sources is to provide uninterrupted power in stand-alone
mode at remote locations. All these facts bring attention to hybridization of renewable
energy sources (RES) and storage systems [5].
In this work, the researchers have inspected a hybrid isolated system which con-
sists of a micro-hydro and PV independent system alongside various time-changing
loads, in particular, dynamic, nonlinear, and linear loads. A STATCOM is used to
maintain the reactive and real power of the entire system, and a SEIG is imple-
mented for generation of power throughout the system. A photovoltaic system and
BESS are connected across the DC link of STATCOM. Here, the PV system partakes
in real power compensation and the surplus power is put away in the BESS which
can control up the system at deficiency and evening time. For any point of time, the
STATCOM completely controls the flow of power. The entire system is planned and
designed utilizing Simulink/MATLAB, and the yields are given in result outcome.
Voltage and Frequency Stabilisation of a Hybrid Autonomous … 961

Fig. 1 System diagram of hybrid autonomous system

2 System Description

This work comprises of an isolated hybrid system, including distinctive environ-

mentally friendly power vitality sources which are incorporated to afford constant
power to the variable load. A self-excited induction generator-based small-scale
hydro framework contributes power to time-varying loads. At whatever point load
increments as SEIG is bolstered from a steady power source it cannot supply more
power, around then the 3φ 6 pulses IGBT-based STATCOM becomes an integral
factor. STATCOM gives the extra power required by the load. The DC part of the
STATCOM is associated with a PV array alongside a BESS. At light loading con-
dition, the surplus power produced by the SEIG and PV framework is redirected by
bidirectional converter-based STATCOM to be put away in battery. During substan-
tial loading, necessary real power is provided by the STATCOM which is given by
the BESS and the reactive power required is provided by the STATCOM. To direct
the power stream of STATCOM, a control structure utilizing sliding mode controller
has been utilized. The hybrid small-scale hydro PV framework is shown in Fig. 1.

3 Modeling and Control of STATCOM

This hybrid autonomous system consists of a SEIG which is capable of delivering

power to time-varying load. At the point heavy loading, the STATCOM enhances the
necessary power. The frequency and voltage of the entire framework are improved
by regulating real and reactive power outflow of the STATCOM. With the ability
of producing variable shunt impedance by STATCOM, the required compensation
power for the microgrid framework can be taken care by adjusting it in a continuous
962 A. Choudhury et al.

manner. The DC side of the STATCOM related to BESS stores the overabundance
real power intensity of the system. In this way, by this procedure, the STATCOM
can control and repay both reactive and real power. Consequently, the STATCOM
acts like a static synchronous generator. In order to control the ‘Q’, the compensat-
ing currents quadrature quantity needs to be controlled. V t can be constrained by
limiting the STATCOM’s reactive power outflow. The 3φ voltages (V ax , V bx , V cx ) of
self-excited induction generator and the frequency of the microgrid framework are
detected together. This actual frequency is contrasted with the reference value; the
errors found are given to SMC, to obtain the active part of the current (Im p ) ampli-
tude. SMCs used in bidirectional converter control algorithm that attempts to limit
the error (F and V ) by giving references of current. The amplitude can be determined
from the 3φ sinusoidal voltage which is given in Eq (1).

Vt = {(2/3) ∗ (Vax
2
+ Vbx
2
+ Vcx2 )}1/2 (1)

In order to obtain reactive current (Im Q ), the actual voltage amplitude is compared
with its respective reference value; the errors obtained from both the amplitudes are
fed to another SMC. Both the parts, i.e., in-phase parts and quadrature elements of
the SEIG’s voltage, have been determined, and Im p and Im Q components of the
circulating currents are inferred. The in-phase unit vectors are as follows:

Vax
ua = (2)
Vt
Vbx
ub = (3)
Vt
Vcx
uc = (4)
Vt

The above three equations give rise to the quadrature unit vectors which is given
as (Fig. 2)
⎡ ⎤
⎡ ⎤ 0 − √13 √13 ⎡ ⎤
wa ⎢ 3√ ⎥ ua
⎣ wb ⎦ = ⎢ √1
− √1 ⎥⎣
ub ⎦ (5)
⎣ 2√ 2 3 2 3⎦
wc − 23 2√1 3 − 2√1 3 uc

The product of phase component (u a , u b , u c ) with the active current component

(Im P ) determines the 3φ active currents. So

i a P = Im P ∗ u a (6)

i b P = Im P ∗ u b (7)
Voltage and Frequency Stabilisation of a Hybrid Autonomous … 963

Fig. 2 Control algorithm of bidirectional converter

i c p = Im P ∗ u c (8)

The 3φ reactive current component is the product of Im Q and wa , wb , wc

i a Q = Im Q ∗ wa (9)

i b Q = Im Q ∗ wb (10)

i c Q = Im Q ∗ wc (11)

The sum of active and reactive current gives rise to the compensating current of
the STATCOM which is given by

Ica = Ia P + Ia Q (12)

Icb = Ib P + Ib Q (13)

Icc = Ic P + Ic Q (14)
964 A. Choudhury et al.

These Ica , Icb ,Icc tend the actual value of compensating current which is compared
with the reference values, i.e., Ic∗a ,Ic∗v , Ic∗c , and the current errors are compared with
a triangular signal of value 12.5 kHz in order to obtain the switching pulses for the
bidirectional converter.

4 Structure of Sliding Mode Controller

As contrasted with different controllers, sliding mode controller (SMC) has numer-
ous points of interest like insensitivity toward parameter varieties, better strength,
quick reaction, and powerful control. The SMCs used for the control of voltage and
frequency are of first order. The sliding surface used for the controllers that are struc-
tured is underneath. Here, two sliding mode controllers are utilized specifically to
control the real and reactive power by comparing the frequency and voltage individ-
ually. The autonomous control of real power is finished by contrasting the frequency
errors as given in Eq. (15). The equivalent is accomplished for reactive power control
by looking at the voltage errors.

S e, e∗ = αe∗ + βe = 0 (15)

where ‘α’ and ‘β’ are (+ve) constants. The ‘error-e’ and ‘change in error-ė’,
respectively. The error is defined as follows:

e = x − x∗ (16)

where ‘x’ is the state to be controlled Viz, ‘V dc ’ and ‘V t ’ and ‘x*’ is the reference
value. The general control law for the SMC is given as follows:

u = −k · sign(s) (17)

The above control law causes chattering at steady state. So the modified control
law used in this work for reduced chattering is

u = −k · sign(λ) (18)

where ‘λ’ is defined by the following equations.

⎧ ⎫
⎪
⎨ 1, s > φ ⎪
⎬
λ = −1, s < −φ (19)
⎪
⎩ s , otherwise. ⎪
⎭
φ

The tuning of parameters α, β, φ, and k is done by trial and error process to

give desired results. The parameters of the sliding mode controller used in terminal
Voltage and Frequency Stabilisation of a Hybrid Autonomous … 965

voltage control are as follows: α = 1, β = 2500, φ = 40,000, and k = 600. The

parameters of the sliding mode controller used for DC-link voltage control are as
follows: α = 2, β = 1350, φ = 30,000, and k = 1500.

5 Result Analysis and Discussion

Case I. Linear RL Load The smaller-scale microgrid framework delineated in the

work is first presented to a period of fluctuating resistive-inductive-type load. With
variations at step intervals at 3.2 s, there is a stride increment in load which regains
its past rate at 3.4 s, and over at 3.3 s, a stride decrement in load appears till 3.4 s.
Load current IL is shown in Fig. 3. Notwithstanding the deviations in IL, VG and F
remain unaltering; due to this, the stage and abundance IG must be taken care. IC, for
instance, is shown in Fig. 3, due to which the VG and IG are kept up at a relentless
value as shown in Fig. 3. IPV, I B , and current I dc are shown in Fig. 3. At whatever
point the load is diminished the leftover real power fabricates whatever makes the
extension in DC-link voltage as a result of which IB develops and the BESS stores
the abundance bulk of power. The extension in voltage of DC link makes IPV decline
on account of the compensation of real power, and F is kept up at a really steady
value as shown in Fig. 3.

Case II. Nonlinear Load In this loading condition, the microgrid framework delivers
power to the resistive-inductive load. During time t = 3 s, a NL load is turned ON.
The NL-type load is a 3φ diode bridge full-wave rectifier which reinforces power to
R load. Because of the nonlinear loading, current earns perverted and twisted wave
shape, which injects harmonics to the system. The shunt active filters conduct is seen
in the STATCOM to decrease the infusion of harmonics in IG; similarly in the act of
controlling P and Q, it has the ability to keep the F and V consistent. Because of the
expansion in the IL, the I dc interface voltage lessens that moreover gives a decrease
in IB and at the same time expands the IPV. Because of the accurate P and Q, the

Fig. 3 Parameter responses for linear RL load

966 A. Choudhury et al.

Fig. 4 Parameter response for NL load

Fig. 5 Parameter response of dynamic load

dynamic operation of the BDC (STATCOM) helps keeping up V and F levels steady
throughout the framework and it aids in decreasing the harmonics contented in IG
also (Fig. 4).
Case III. Dynamic Load In dynamic loading condition, which was at first working
with linear load, at 3 s, a dynamic (induction motor) load is switched ON. A 7.5 kW
IM which is dynamic in nature is begun legitimately online at 3 s. The abrupt neces-
sity of P and Q is satisfied by the BDC-based STATCOM. The various currents for
this loading condition are shown in Fig. 5. Faster request of power matters an unex-
pected increment in the remunerating current and consequently IB and I dc also. The
V dc lessens by around 22.93(V) yet recovers its incentive after transient response.
Regardless of the unexpected variety of the IL, the V and F are kept up at a genuinely
consistent incentive as depicted in Fig. 5.

6 Conclusions

Here, the system performance of various parameters is analyzed. This system was
subjected to various types of time-varying loads, where the advantages of STACOM
Voltage and Frequency Stabilisation of a Hybrid Autonomous … 967

came into the picture. During linear load switching, it was clearly seen that the
compensation current supplied reactive power to the system when the load was
increased. Overall STATCOM has the capability to control both the power flows;
during heavy loading, it supplies active power from BESS, and during surplus power,
it stores the power in the BESS. In terms, the system is quite adaptive in nature, and
accordingly, independent power control strategies using sliding mode controllers
have been designed so as to respond faster in any load fluctuations.

References

1. Pati, S., Mohanty, K.B., Kar, S.K., Panda, D.: Voltage and frequency stabilization of a micro
hydro-PV based hybrid micro grid using STATCOM equipped with Battery Energy Storage
System. In: 2016 IEEE International Conference on Power Electronics, Drives and Energy
Systems (PEDES), December, pp. 1–5. IEEE (2016)
2. Jena, B., Choudhury, A.: Voltage and frequency stabilisation in a micro-hydro-PV based hybrid
microgrid using FLC based STATCOM equipped with BESS. In: 2017 International Conference
on Circuit, Power and Computing Technologies (ICCPCT), April, pp. 1–7. IEEE (2017)
3. Pati, S., Mohanty, K. B., Choudhury, A., Kar, S.: Integration and power control of a micro-
hydro-PV-wind based hybrid microgrid. In: 2017 International Conference on Circuit, Power
and Computing Technologies (ICCPCT), April, pp. 1–6. IEEE (2017)
4. Bhim Singh, S., Murthy, S., Gupta, S.: STATCOM-based voltage regulator for self-excited
induction generator feeding nonlinear loads. IEEE Trans. Ind. Electron. 53(5) (2006)
5. Zhan, C.-J., Wu, X.G., Kromlidis, S., Ramachandaramurthy, V.K., Barnes, M., Jenkins, N.,
Ruddell, A.J.: Two electrical models of the lead-acid battery used in a dynamic voltage restorer.
IEE Proc. Gener. Transm. Distrib. 150(2) (2003)
6. Rezkallah, M., Hamadi, A., Chandra, A., Singh, B.: Real-time HIL implementation of sliding
mode control for standalone system based on pv array without using dumpload. IEEE Trans.
Sustain. Energy 6(4) (2015)
7. Dhanapal, S., Anita, R.: Voltage and frequency control of stand alone self-excited induction gen-
erator using photovoltaic system based STATCOM. J. Circuits Syst. Comput. 25(04), 1650031
(2016)
8. Tamilselvan, K., Anita, R.: Performance evaluation of PV-supported STATCOM for voltage
and frequency regulation of standalone SEIG system. J. Test. Eval. 46(1), 266–282 (2017)
9. Pingping, G., Ziguang, L., Zhuo, L., Di, W.: PI-PSO algorithm based voltage controller of
STATCOM for self-excited induction generator. In: 2015 34th Chinese Control Conference
(CCC), July, pp. 4349–4354. IEEE (2015)
10. Tandekar, J.K., Ojha, A., Das, S., Swarnkar, P., Jain, S.: SEIG-based renewable power gener-
ation and compensation in MVDC ship power system. Int. Trans. Electr. Energy Syst. 29(4),
e2785 (2019)
Characterization of Power Quality
Disturbances and Their Efficient
Classification
Laxmipriya Samal, Hemanta Kumar Palo, Badri Narayan Sahu,
and Debashisa Samal

Abstract Characterization and classification of power quality (PQ) disturbances are

an essential component in the field of power engineering to meet consumer demands.
Accurate analysis of supply power, its processing, and distribution requires identi-
fication of the noise and disturbances associated during power generation, trans-
mission, and distribution. In general, pure-tone power signals are non-stationary
with both time- and frequency-varying statistical parameters. Thus, the use of either
the time domain or frequency domain analysis cannot characterize or classify the
PQ signal adequately. This motivates the authors to approach the problem domain
employing time–frequency (TF) characterization using a spectrogram initially. TF
distribution is one of the best application tools for PQ analysis and is emphasized
using the short-time Fourier transform (STFT) and wavelet-based features in this
paper. Finally, the TF-based features extracted from the normal and different PQ
disturbances are applied to an efficient probabilistic neural network (PNN) model
for classification. We have shown that PNN with TF-based wavelet features provides
an efficient classification result as compared to other chosen techniques.

Keywords Characterization · Power quality · Disturbances · Accuracy ·

Probabilistic neural network · Classification

L. Samal · H. K. Palo (B) · B. N. Sahu · D. Samal

Institute of Technical Education and Research, Siksha ‘O’ Anusandhan (Deemed to be
University), Bhubaneswar, Odisha 751030, India
e-mail: [email protected]
L. Samal
e-mail: [email protected]
B. N. Sahu
e-mail: [email protected]
D. Samal
e-mail: [email protected]

© Springer Nature Singapore Pte Ltd. 2020 969

G. Pradhan et al. (eds.), Advances in Electrical Control and Signal Systems,
Lecture Notes in Electrical Engineering 665,
https://doi.org/10.1007/978-981-15-5262-5_75
970 L. Samal et al.

1 Introduction

The customer has the interest to obtain a clean power without fluctuations, or PQ
disturbances. However, the involvement of different PQ disturbances such as tran-
sients, harmonics, spikes, flickers, sags, and swells makes it practically impossible.
Arguably, an efficient characterization and detection of these PQ disturbances remain
a challenge to emerging researchers endeavoring to supply clean power [1, 2]. The
PQ issues are growing day-by-day that drive towards new research. There is an ever
increase in demand to tackle the PQ issues both in scientific and industrial commu-
nities because of three major reasons. These are (i) the deregulation of the electricity
market (ii) electronic and power electronic equipment, and (iii) sensitivity of the
existing nonlinear devices to PQ disturbances that pose difficulties to consumers [3].
Improvement in supply power requires the identification and elimination of poten-
tial sources that degrade PQ. An intelligent approach to meet this requirement is to
involve detection mechanism in or around the electronic or electric to detect, charac-
terize, capture, and classify the PQ disturbances. With the identification of the type
of disturbances and their effect on the load, it can be possible to find means for their
elimination [4]. Due to the influence of both the time and frequency component of the
PQ disturbances on the supply power, the approaches that focus on either the time or
frequency component of a signal remain inadequate. On the contrary, the approaches
that apply both the time and frequency resolution of the designated signal remain
more appealing. This has made the authors explore the capability of the spectrogram
in characterizing and identifying the PQ disturbances of a signal. Nevertheless, the
classification of these identified disturbances demands the extraction of reliable and
robust features and an equally efficient machine learning algorithm [5, 6].
A literature survey points out the availability of many efficient signal process-
ing mechanisms to tackle PQ disturbances. However, the non-stationary PQ distur-
bances need to be represented by features that suitably and adequately reflect these
disturbances. Among popular feature extraction techniques, the statistical parame-
ters (mean, variance, maximum, etc., of signal amplitudes or frequency), the Fourier
transform (FT), STFT, short-time energy, and wavelet transform (WT) have been
widely employed in this field [7, 8]. While the conventional FT provides the fre-
quency representation of a signal, the STFT proves to be more efficient as a time–
frequency tool. It remains an effective method to analyze stationary signals although
cannot be suitably applied to non-stationary signals due to the use of fixed windows.
This has made the WT a state-of-the-art time–frequency tool to monitor and control
PQ issues [3]. The multi-resolution capability of WT and its ability to accommodate
a varying window create the desired motivation for this study. The features extracted
in the wavelet domain from PQ disturbances have been compared with that of STFT
and FT for a possible enhancement in recognition accuracy.
There have been many learning algorithms found which are mentioned in the lit-
erature for the detection and recognition of PQ disturbances for their effectiveness.
The classification accuracy varies with feature extraction techniques chosen, limita-
tions, and complexity of these machine learning algorithms [7, 9]. The choice of the
Characterization of Power Quality Disturbances … 971

classifier depends mostly on the application domain or task in hand. In this regard,
the neural network (NN) remains a widely suggested technique applied in the field of
non-stationary signal classification like speech or image. Simplicity, better process-
ing quality, and the ability to discriminate patterns make the NNs popular [10–14].
Among NNs, the learning procedure in much-debated deep NN spreads across many
numbers of hidden layers; hence, it is difficult to comprehend the learning attributes.
It requires a large memory space and is slower due to an equally large number of
hidden neurons in a hugely large number of hidden layers. There is no theoretical
background in the training methods or model or topology or, flavor, which makes
the algorithm a black-box approach. On the contrary, the PNN has the least con-
straint in the choice of performance-dependent coefficients in precision. It requires
the adjustment of only one parameter to simulate the feature sets and hence is faster
as compared to either the multilayer perceptron or the radial basis function network.
To train or run the algorithm, the PNN utilizes matrix manipulation and can easily
approach the Bayes optimal solution when specific conditions are fulfilled. This has
made the network versatile and a natural choice for the proposed work.
The paper is summarized as follows: Different PQ disturbances that distort the
power signal and different signal processing techniques have been briefed in Sect. 2.
The PNN classification scheme is elaborated in Sect. 3. The simulation and clas-
sification results are provided in Sect. 4. Section 5 concludes the work with future
directions.

2 Characterization of Power Quality Disturbances

The FT has been extensively employed to analyze the stationary signal in many
signal processing applications. In this, an arbitrary signal can be described in terms
of weighted sinusoids. However, the technique provides only frequency information
of a signal and fails to analyze non-stationary signal mostly encountered in real life.
Thus, aperiodic signals such as speech, music, and power signals whose frequency
varies with time cannot be adequately represented using FT [11–14]. Further, it is not
advisable to represent the spectral characteristics of any non-stationary signals by
only the frequency argument. The illustration of the signal can be more informative
with a joint time–frequency characterization. Such analysis can fetch information
on the behavior of the spectral content concerning time and hence suits for studying
non-stationary signals. This has motivated the authors to represent PQ disturbances
using the joint time–frequency transform such as the spectrogram so that the exact
behavior of the time-varying frequency content of the signal can be extracted and
analyzed.
972 L. Samal et al.

2.1 PQ Disturbances

The work considers the detection and characterization of the following PQ distur-
bances.

2.1.1 Voltage Sag

It is the decrease in RMS voltage at the power frequency between 0.1 and 0.9 pu for
a duration in the range of 0.5 cycles to 1 min.

2.1.2 Voltage Interruption

When the load current of the supply voltage reduces to less than 0.1 pu below 1 min,
the signal is considered interrupted [15]. Voltage interruption can occur when the
supply voltage or load current decreases due to a faulty power system or control mal-
functions. However, the interruption may be repetitive for a short time or momentary.
These are highly unpredictable due to random occurrence and can cause loss of con-
trol/computer memory, hardware damage, loss of product, shutdown, or failure of
equipment.

2.1.3 Signal Swell

It is the increase in the RMS value of a PQ signal at the power frequency with
durations in the range of 0.5 cycles to 1 min. The magnitude of voltage swell lies in
the range of 1.1 and 1.8 pu.

2.1.4 Signal Harmonics

Harmonics in a signal are sinusoidal with frequencies as integer multiples of the

fundamental frequency of the signal under consideration. It distorts the waveform of
the original signal when added with it.

2.1.5 Signal Transients

The term transient in a signal signifies the presence of non-dying impulsive or oscilla-
tory variations for a certain time. The transient phenomena in a signal can be studied
accurately using spectral information. The impulsive transient is described by the rise
and decay times of a signal. On the contrary, the oscillatory transient is either high
frequency (>500 Hz) or low frequency (<500) and has a duration in microseconds.
Characterization of Power Quality Disturbances … 973

2.1.6 Signal Flickers

The signal flicker indicates a series of random changes in voltage or current or power
and thus is undesirable. It is defined by the RMS magnitude of a signal quantity
with a range between 0.9 and 1.1 pu. It is expressed as a percentage of the original
signal magnitude and varies with time and duration of measurement of a signal. A
few important sources that produce flicker in a signal are arc furnaces, rolling mills,
and variable load industrial motors.

2.2 Signal Representation

2.2.1 Fourier Transform (FT)

The FT of a PQ signal x(t) with time t and frequency ω can be represented as

∞
X (ω) = x(t)eiωt (1)
−∞

It provides the spectrum of the analyzed signal and comprises the components
at the entire frequency range over which it is considered nonzero. It provides the
frequency information of a signal but not the time information. However, the PQ
signal involving bulge, surge, drift, transients, trends, etc., is non-stationary. An
adequate representation and characterization of such events require due emphasizes
on both time and frequency contents of the signal. It makes the FT not suitable to
analyze such signals. To alleviate the issue, the signal analysis must be carried out
over a short segment or frame in which the signal can be assumed to be stationary
[8]. This has led to the concept of short-time Fourier transform (STFT).

2.2.2 Short-Time Fourier Transform (STFT)

The FT is customized to deal only with the stationary signal whereas real-world
signals are mostly non-stationary. On the contrary, the STFT is adaptive in a two-
dimensional space corresponding to the time and frequency. It gives information on
the time and the frequencies of a non-stationary signal event [11].
In STFT, the analyzed signal under consideration is framed into small segments
to make it look stationary. Each of the framed signals is passed through a window
function such as rectangular, Hamming, and Hanning of fixed to validate the sta-
tionary signal. Care must be taken to choose a window narrow enough to assume a
signal to be stationary within the designated window. Due to the short duration of the
window, the FT of the signal to window product reflects the local frequency charac-
teristics of the analyzed signal. The derivation of STFT of the windowed signal can
974 L. Samal et al.

be formulated as

x(t) = x(τ )h(t − τ ) (2)

where t and τ denote the fixed and running time, respectively.


x(τ ) for τ near t
x(t) = (3)
0 for τ away from t

The window FT of a signal can be expressed as


1
X (ω) = √ eiωτ x(τ )dτ (4)
2π

The windowed FT of STFT with a window function w(t) is represented as


1
X STFT (t, ω) = √ e−iωτ w(τ − t)dτ (5)
2π

The size and shape of the window often determine the precision of the information
extraction using STFT. Thus, it limits the applicability in the representation of the sig-
nal adequately in a two-dimensional space. Nevertheless, the use of a fixed window
puts a constraint on the analysis of non-stationary signal analysis that can be repre-
sented better using window functions with varying width. This has led researchers
to explore other flexible time and frequency tools as the wavelet transform (WT).

2.2.3 Wavelet Transform (WT)

The WT has the multi-resolution capability in describing the signal under consid-
eration and hence remains an efficient time–frequency tool in the analysis of non-
stationary signals such as speech, music, and power signals. It decomposes the signal
into low- and high-frequency components designated as approximation ax (i, τ ) and
detailed parameters dx (i, τ ) as shown below


i=I  
I 
x(t) = ai (t) + di (t) = ax (i, τ ) ϕi,k (t) + dx (i, τ ) ϑi,k (t) (6)
i=1 k i=1 k

where ϕi,k ϑ0 represent the scaling function and the mother wavelet, respectively, in
the nested space. The coefficients ai (t) consider approximating the signal coarser
than the coefficients di (t). Thus, there is very likely a loss of information during the
transition of the signal between the approximation coefficients. This has made the
authors consider the detailed coefficients for classifying the power signal in place of
the coefficients ax (i, τ ).
Characterization of Power Quality Disturbances … 975

di (t) = ai−1 (t) − ai (t) (7)

As the extracted detailed coefficients provide differential approximated informa-

tion, we can assume these as wavelets or small waves for further analysis in this
work.
The Daubechies (DB) wavelet has been opted in this paper. The wavelet consists
of vanishing or zero moments half of the number of wavelet coefficients. Thus, DB2,
DB4, DB6, etc., have one, two, and three zero moments, respectively. The number
of zero moment decides the polynomial behavior of a signal and puts a limit on
this. For example, DB2 can encode polynomials of a constant signal component or
polynomials with a single coefficient. Similarly, DB4 encodes the constant or linear
or polynomial signals corresponding to two coefficients, whereas DB6 can encode
these signals corresponding to three coefficients. Due to this reason, DB6 wavelet
has been chosen in this work.

2.2.4 Statistical Parameters

The advantages of representing a signal using statistical analysis are good robustness
and simplicity. For this work, the maximum amplitude, minimum amplitude, mean
amplitude, and variance of the spectrograms representing the PQ disturbances have
been considered. These parameters are extracted and fed to the PNN classifier for
simulation and classification of the disturbances along with the pure-tone signal.

2.2.5 Spectrogram

The periodogram estimates the power spectrum describing the frequency content of
a signal and hence cannot provide the desired information on its temporal character-
istics. To alleviate this issue, it is essential to represent the signal in both time–fre-
quency domains. For any non-stationary random discrete signal x(n) of length N, the
spectrogram can provide both the time and frequency representation. It is estimated
as
 M−1 2
1   − j xkm


Px [n, k] =  x[m]w[m − n]e M  (8)
M  m=0 

where 0 ≤ n ≤ N − 1, 0 ≤ k ≤ M − 1, M < N . Here, w(n) represents the

windowed signal of length M, and N denotes the number of samples.
Figure 1 provides the framework of the PQ disturbance characterization and
classification.
976 L. Samal et al.

Fig. 1 Proposed
characterization and PQ Signal Feature
classification of PQ Extraction
disturbances
WT FT STFT

Spectrogram Classification
Characterization
Statistical
Parameters

3 The Classification Approach

The PNN machine learning algorithm has been opted-in this work as it is faster in
comparison to other NN models such as MLP, RBFN, and the DNN. It provides a near-
optimal solution and also allows the accommodation of additional training samples
or their removal at ease. It is not demanding to retrain the network extensively as the
network is insensitive to outliers [16]. The use of single processing while training,
single parameter manipulation, tolerance to erroneous data, and adaptable training
using a sparse input sample makes it versatile in the field of pattern recognition.
The PNN model has four layers (input, pattern, summation, and decision). The
features from the input are propagated to the pattern layer to form the desired PQ
disturbance pattern. The summation layer estimates a maximum likelihood corre-
sponding to the specific pattern by summarizing and averaging and forwards it to
the decision layer for the desired PQ disturbance classification. The decision in the
output layer is taken based on the layer which follows the Bayes’ rule that provides
an optimum solution.

4 Results and Discussion

Figure 2 shows the current noise in picoampere (pA) at different periods of time in
seconds. It consists of different segments of signal disturbance such as weak and
sporadic, spike and bursting components. The aim is to characterize the disturbances
of supply current which endanger the electrical and electronic appliances used in
industries and scientific communities.
In Fig. 3, the differentiation between the pure-tone ideal PQ signal and the pure-
tone signal with spikes has been made. The signal amplitude at different periods in
seconds has been plotted for a voltage signal of 0.3-second duration.
Different spectrogram representations of PQ disturbances are graphically shown
in Figs. 4, 5, 6, 7, 8 and 9. The time domain sag disturbance signal and its spectrogram
characterization are shown in Fig. 4 for a signal of 0.2 s duration.
Figure 5 provides the spectrogram representation of voltage swell disturbance
of 0.2 s duration and its time domain representation for characterization. Similarly,
Characterization of Power Quality Disturbances … 977

Fig. 2 Current noise (pA) with disturbances at different times in seconds

Fig. 3 Plot of a synthetic PQ voltage signal at different periods in seconds with pure tone and
spikes

Fig. 4 Simulated voltage sag and its spectrogram

978 L. Samal et al.

Fig. 5 Simulated voltage swell signal and its spectrogram

Fig. 6 Simulated voltage harmonics and its spectrogram

Fig. 7 Simulated voltage oscillatory transient and its spectrogram

Fig. 8 Simulated voltage flicker and its spectrogram

Characterization of Power Quality Disturbances … 979

Fig. 9 The simulated voltage interruption and its spectrogram

the characterization of signal voltage harmonics of 0.2-seconds duration is shown in

Fig. 6 both in the time domain and also using the spectrogram.
The time-domain representation of the oscillatory transients and the flicker of
the PQ voltage disturbances of 0.2 s duration have been provided in Figs. 7 and 8
respectively. Similarly, the voltage interruption in a signal has been represented using
the spectrogram in Fig. 9.
The classification error of the PNN classifier with different feature extraction
techniques such as the FT, STFT, statistical, and WT is compared in Table 1. In this
table, an attempt is made to find out the recognition error of the PNN using the chosen
features extracted from different PQ disturbances analyzed in this work. In particular,
the aim is to compare the applicability and effectiveness of the WT time–frequency
tool with other conventional methods such as the statistical, FT, and STFT. The
recognition error using WT has found to be less followed by the STFT as observed
from our results. The time–frequency representation of WT and the application of
an adaptive window have provided the desired resolution with enhanced accuracy.
As compared to the traditional FT or statistical parameter, the time–frequency STFT
has outperformed in recognizing PQ disturbances as revealed from our results. The
statistical features extracted from the spectrograms of the PQ signal have shown
better performance than the FT as shown in this table.
Figure 10 provides the correlation factors corresponding to voltage sag for dif-
ferent voltage gauge levels as compared to short interruptions. The factor goes on
decreasing as the voltage gauge level in percentage increases.

Table 1 PNN classification error (%) of PQ disturbances with different feature extraction
techniques
Dataset Pure-tone Swell Sag Harmonics Spikes Transients Flickers
signal
FT 11.6 10.3 10.6 10.8 10.9 9.9 9.5
Statistical 10.2 8.9 9.3 9.7 9.8 9.1 8.7
STFT 9.5 8.4 8.8 9.1 9.0 8.3 8.2
Wavelets 8.4 7.2 7.7 7.9 8.1 6.9 6.5
980 L. Samal et al.

Fig. 10 The correlation

factors corresponding to
voltage sag for different
voltage gauge levels

5 Conclusion

The objective of this work is to test the ability of time–frequency tools such as the
STFT and the WT in classifying the PQ disturbances. It has additionally attempted to
characterize the PQ disturbances both in the time domain and using the spectrogram.
Further, the statistical parameters of the spectrograms have been fetched and used to
compare the PNN accuracy with other chosen feature sets. It has used the MATLAB
platform for characterization, feature extraction, and simulation of the PNN network.
The PQ disturbances such as interruption, swell, sag, harmonics, transients, flick-
ers, and the pure-tone ideal signals have been used for characterization and classifi-
cation purposes. The work has been carried out by taking 70 examples per class with
training to the testing ratio of 90 to 10%. The signals are generated for different power
qualities for a 0.2-second duration and random distortion to make the simulation reli-
able as none of these disturbances are predictable in a real power system. Although
the work shows the excellent results with the chosen time–frequency tools, still there
is a scope for improvement. The exploration of other effective time–frequency tools
and efficient classifiers may provide new feature directions.

References

1. Abdelsalam, A.A., Eldesouky, A.A., Sallam, A.A.: Characterization of power quality distur-
bances using hybrid technique of linear Kalman filter and fuzzy-expert system. Electr. Power
Syst. Res. 83(1), 41–50 (2012)
2. Biswal, M., Dash, P.K.: Detection and characterization of multiple power quality disturbances
with a fast S-transform and decision tree based classifier. Digit. Signal Process. 23(4), 1071–
1083 (2013)
3. Dehghani, H., Vahidi, B., Naghizadeh, R.A., Hosseinian, S.H.: Power quality disturbance
classification using a statistical and wavelet-based hidden Markov model with Dempster-Shafer
algorithm. Int. J. Electr. Power Energy Syst. 47, 368–377 (2013)
4. Erişti, H., Yıldırım, Ö., Erişti, B., Demir, Y.: Automatic recognition system of underlying
causes of power quality disturbances based on S-transform and Extreme Learning Machine.
Int. J. Electr. Power Energy Syst. 61, 553–562 (2014)
Characterization of Power Quality Disturbances … 981

5. Huang, N., Xu, D., Liu, X., Lin, L.: Power quality disturbances classification based on S-
transform and probabilistic neural network. Neurocomputing 98, 12–23 (2012)
6. Hooshmand, R., Enshaee, A.: Detection and classification of single and combined power quality
disturbances using fuzzy systems oriented by particle swarm optimization algorithm. Electr.
Power Syst. Res. 80(12), 1552–1561 (2010)
7. De Yong, D., Bhowmik, S., Magnago, F.: An effective power quality classifier using wavelet
transform and support vector machines. Expert Syst. Appl. 42(15–16), 6075–6081 (2015)
8. Shen, Y., Abubakar, M., Liu, H., Hussain, F.: Power quality disturbance monitoring and clas-
sification based on improved PCA and convolution neural network for wind-grid distribution
systems. Energies 12(7), 1280 (2019)
9. Ribeiro, M.V., Szczupak, J., Iravani, M.R., Gu, I.Y., Dash, P.K., Mamishev, A.V.: Emerging
Signal Processing Techniques for Power Quality Applications (2007)
10. Palo, H.K., Mohanty, M.N.: Comparative analysis of neural networks for speech emotion
recognition. Int. J. Eng. Technol. 7, 112–116 (2018)
11. Palo, H.K., Mohanty, M.N., Chandra, M.: Efficient feature combination techniques for
emotional speech classification. Int. J. Speech Technol. 19, 135–150 (2016)
12. Palo, H.K., Sagar, S.: Comparison of neural network models for speech emotion recognition.
In: 2018 2nd International Conference of Data Science and Business Analytics, pp. 127–131.
IEEE (2018)
13. Palo, H.K., Mohanty, M.N.: Wavelet-based feature combination for recognition of emotions.
Ain Shams Eng. J. 9, 1799–1806 (2018)
14. Palo, H.K., Chandra, M., Mohanty, M.N.: Recognition of human speech emotion using variants
of Mel-frequency cepstral coefficients. Advances in Systems, Control and Automation, pp. 491–
498. Springer, Singapore (2018)
15. Abdullah, A.R., Sha’ameri, A.Z., Sidek, A.R.M., Shaari, M.R.: Detection and classification
of power quality disturbances using time-frequency analysis technique. In: 2007 5th Student
Conference on Research and Development, pp. 1–6. IEEE (2007)
16. Palo, H.K., Sagar, S.: Characterization and classification of speech emotion with spectrograms.
In: 2018 IEEE 8th International Advance Computing Conference (IACC), pp. 309–313. IEEE
(2018)
A Novel Control Approach
for Multi-level Inverter-Based Microgrid

Sangram Keshari Routray, Buddhadeva Sahoo, and Sudhansu Sekhar Dash

Abstract Previously, neutral-point-clamped NPC inverters are widely used due to

its flexibility. NPC has more advantages than conventional voltage source inverter
(VSI) by reducing harmonics and providing more voltage levels, but the circuit
of the inverter is much complicated due to the implementation of large number
of electronic switches. This paper focused on the design of seven-level reduced
switch cascaded inverter (RSCI) through minimum number of switches, which is
integrated in between the generating station and grid. To maintain the requested
output voltage, MPPT technique is used and the inverter also operates by using space
vector modulation technique (SVMT). The strength of the paper is that it can produce
the accurate AC voltage under pulsating DC voltage condition. In this manuscript, the
theoretical concept of the proposed control architecture and its design are discussed.
To minimize the inverter output harmonic voltage and reduce the nonlinearity of
the system, LCL filter is used. To show the effectiveness of the proposed approach,
the designed approach is undergone through MATLAB/Simulink software by using
different conditions such as changing irradiance and battery charging or discharging
conditions.

Keywords Battery · Photovoltaic cell (PV) · MPPT technique · Space vector

modulation technique (SVMT) · RSCI · LCL filter

S. K. Routray (B) · B. Sahoo

Department of EEE, Institute of Technical Education and Research, Siksha ‘O’ Anusandhan
(Deemed to be University), Bhubaneswar, Odisha 751030, India
e-mail: [email protected]
B. Sahoo
e-mail: [email protected]
S. S. Dash
Department of DSM, Capital Institute of Management and Science, Bhubaneswar, India
e-mail: [email protected]

© Springer Nature Singapore Pte Ltd. 2020 983

G. Pradhan et al. (eds.), Advances in Electrical Control and Signal Systems,
Lecture Notes in Electrical Engineering 665,
https://doi.org/10.1007/978-981-15-5262-5_76
984 S. K. Routray et al.

1 Introduction

In the power generation sector because of the uncertainty in the environment and
energy crisis, there is a new energy generation sector gaining interest that is known
as renewable energy. The main advantages of the renewable energy sector are its
used natural sources like solar, wind, hydro, etc. In the above sectors, the most
efficient sector and easily available sector is solar and wind [1, 2]. To utilize and
development in the renewable energy sector in an efficient manner, power electronic
devices are required. Generally, most of the power in photovoltaic system comes
from the supply. In this, the most important part to utilize the power with less loss is
by using power electronic devices [3, 4]. There are two types of conversion system
generally implemented to convert energy from photovoltaic cell to utility grid: one-
way and two-way conversion. In a two-way conversion, the first part is used to convert
pulsating DC to actual DC and second part is used to convert actual DC to AC [4–6].
To connect the generation system to grid, there should be a conversion device
used. The conversion device must contain an active and passive device. Due to these
devices, the conversion device produces power loss. The power loss is broadly divided
into two types (1) conduction losses; (2) switching losses. The losses produced by
the switches are the major point of concern because the inverter can be made up
by using large number of switches. A filter is needed to mitigate the harmonics.
In addition to the filter, it is needed that the inverter also produces high voltage
level. In the past decades, different other topology inverters are used to generate
more voltage levels. To generate high voltage level, flying capacitor inverters are
used. But the design of conventional approaches is more complex and contains more
switches [7–9]. Therefore, the paper is focused on the design of a RSCI to convert
the DC voltage to AC and integrate with the grid. By using the seven-level inverter,
the system can generate higher voltage stages with lesser components. By using
the MPPT technique, the efficiency of the solar output is increased [10, 11]. The
efficiency of the system can be controlled by incremental conductance method.
One bigger problem in renewable energy system is that they are uncertain. Due
to this, the generated power is fluctuated. Therefore, to increase the performance,
the system needs a storing device to store the power [12, 13]. By using a battery
cell, we will overcome these problems and rise the efficiency and reliability of the
system. To reduce the switching loss and reduce the harmonics, the system uses fuzzy
logic controller (FLC) [14]. In this proposed approach, the manuscript focuses on
the performance of the MLI by properly regulating the switching sequence.
This paper is divided into five sections. Section 2 analyzes about reduced switch
seven = level inverter, Sect. 3 describes control technique which is used to increase
system efficiency of the system, Sect. 4 analyzes the simulation results at different
conditions, and Sect. 5 concludes the major findings of the study.
A Novel Control Approach for Multi-level Inverter-Based Microgrid 985

2 Seven-Level RSCI

The seven-level RSCI is a cascade connection of full bridge inverter. The inverter
operates for both ‘+’ and ‘−’ conditions. The IGBT and diodes are supposed to be
ideal when the voltage in the capacitor is V dc , V dc /3, or 2V dc /3. The PV output current
is to be controlled. The inverter operation may be explained for four positive half
cycle modes out of total eight modes.
Mode 1: From Fig. 1, when the IGBTs S1 and S2 are turned off, the C1 discharges
through D1. The voltage output is V dc /3. During this period, T1 and T4 are turned
on.
Mode 2: Here, S1 remained as turned off but S2 is turned on, so the C1 discharges
the current through S2 and D2. The S1 and D2 get shorted. Capacitor output
voltage remains as 2V dc /3. As in case of Mode 1, the switches T1 and T4 remain
turned on. Inverter output voltage becomes 2V dc /3.

Fig. 1 Two-phase seven-level inverter

986 S. K. Routray et al.

Table 1 For positive half condition: (three-phase) switching condition

S1 S2 T1 T3 T5 T4 T6 T2 V out (+)
Off Off On Off Off Off On On V dc /3
Off On On Off Off Off On On 2V dc /3
On Off On Off Off Off On On V dc
Off Off Off Off Off Off Off On 0

Table 2 For negative half condition: (three-phase) switching condition

S1 S2 T1 T3 T5 T4 T6 T2 V out (−)
Off Off Off On On On Off Off V dc /3
Off On On On On On Off Off 2V dc /3
On Off On On On Off Off Off V dc
Off Off Off Off On Off Off Off 0

Mode 3: S1 is now turned on, which makes diode D2 get short circuited. S2
cannot oppose the current flow. Now, both the capacitors are discharged in series,
and the output voltage is V dc . Already, the switches T1 and T4 are on.
Mode 4: Here, both S1 and S2 are turned off. Capacitor voltage is V dc /3. Here,
only T4 is on, so the inverter output current passes through the filter inductor.
Feedback diode of the T2 is on. So, the voltage of the inverter is 0.
Figure 1 shows a two-phase seven-level inverter. The voltage steps for positive
half cycles are 0, V dc , V dc /3, 2V dc /3, and for ‘−’ half conditions, the inverter output
voltage is negative. After completing the positive and negative half, the inverter offers
seven output voltage levels like ±Vdc , ± V3dc , ± 2V3dc , 0. Table 1 shows the triggering
sequences for the positive pulses, and Table 2 shows the triggering sequences for
negative pulses.
The three-phase seven-level proposed inverter is presented in Fig. 2. This is the
switching operation for the three-phase inverter as like as two-level inverter. The
switches are conducting by varying the conduction angle.

3 Control Techniques

To generate the accurate DC voltage, three series-connected solar cells are integrated.
The mathematical equation of each PV unit is taken [14]
  V  
pv
−7
Ipv = Isc − 10 e 2574×10 −3
−1 (1)
A Novel Control Approach for Multi-level Inverter-Based Microgrid 987

Fig. 2 Three-phase seven-level inverter

By using MPPT technique, desired voltage (dc) can be generated to achieve max-
imum power. Here, to increase the efficiency of the system, incremental conductance
method is used. This method is based upon change in power with respect to voltage,
which is same as zero [maximum power point (MPP)].
DP > 0: perturbation moved toward MPP.
DP < 0: perturbation moved away from MPP.
The seven-level inverter is operated by space vector modulation techniques, which
is more advantageous than PWM techniques. To generate the appropriate inverter
output current in d q frame id and iq , are necessary to convert as per the active, reactive
and grid conditions.

p = Vsd i d + Vsq i q (2)

q = Vsq i d − Vsd i q (3)

By solving (2) and (3)

p − Vsq i q q + Vsd i q
id = , id =
Vsd Vsq

Equating the above two equations

p × Vsq − Vsq2 i q = q × Vsd + Vsd2 i q (4)

988 S. K. Routray et al.

p × Vsq − q × Vsd
iq = (5)
Vsd2 + Vsq2

Similarly,

p × Vsd + q × Vsq
id = (6)
Vsd2 + Vsq2

The control architecture of the proposed system is indicated in Fig. 3. To control

the inverter output current, here fuzzy controller and decoupling control method are
used. By using the above strategies, the inverter voltage vector is also computed.
Here, to generate the harmonic free switching pulses, fuzzy rule is implemented. To
send the requested power to AC side, the battery plays a very important role. When

Va PLL Vsd
Vb Dq Vsq
Grid
Ipv Vdc Vc convertion Id
Ia PLL
ca Ib Dq
Ic
convertion Iq
Vca Reduced switch
Solar seven-level
PV LCL filter
Lbat cascaded
inverter
cb
Vbattery
Ia
Ib

Ic

Vca
eVca
Vdc* Vcb*
Ipv +_ +_ ×÷ K2

MPPT control Space vector

_
(IC method) pulse width
+
Vdc Vdc modulation

Vca*-Vbattery eVc
Ppv +_ ×÷ b K1

Vca
Id
p* sd − *Vsq q Id* V_ Fuzzy
P* *
I = +
Grid d controller
Network sd + Vs2q V2
Data
supervisory q* *
Battery
Vsd −qp *Vsq +_
Fuzzy
Block I q* = 2 2
controller
condito sd +V sq VIq* Iq
n

Fig. 3 Control topology model simulation

A Novel Control Approach for Multi-level Inverter-Based Microgrid 989

there is surplus PV power available, it stores the energy, and when there is less power
generated or fluctuation of power occurs, then it maintains the requested power.
For capacitor C a and C b , V ca , V cb = Actual voltage capacitor.
Vca∗ , Vcb∗ = Desired voltage at capacitor.

Vca∗ = Vbattery

Vdc − Vca = Vcb

Error in C b ,

Vcb∗ − Vcb
eVcb = (7)
Vcb

Error in C a ,

Vca∗ − Vca
eVca = (8)
Vca

As shown in Eq.7 and Eq. 8, the capacitor voltage errors are used to compute
the short vectors for pulse generation. Capacitor charging or discharging is totally
dependent upon the selection of short vector. So, relative error (E) of capacitor voltage
is computed as below.

E = (K 1 × eVca ) − (K 2 × eVcb ) (9)

K1, K2 = gain which is related to capacitor voltage.

The main function of gain K1 and K2 is to regulate which relative error and
capacitor voltage is much more significant to give a appropriate control. In MLI, tra-
ditionally K1 and K2 must be same value like reference voltage, but here, capacitor
voltages are unequal. From the above equations at K2  K1, the capacitor voltage
is adjusting the battery operation. In every step, symbol of the relative error deter-
mines which short vector is needed to charge or discharge. If E = positive, C a is
charging condition and C b is discharging condition; otherwise, C b is charging and
C a is discharging mode. By using this technique, the system generated accurate the
AC side active power and reactive power. To calculate relative error and adjusting
the selection of vector, the proposed system calculate the charging and discharging
condition of the capacitor.
The input control variable contains seven variables named as high positive (HP),
average positive (AP), low positive (LP), high negative (LN), average negative (AN),
and low negative (LN). However, the fuzzy control architecture is constructed by
using 49 rules as indicated in Table 3.
990 S. K. Routray et al.

Table 3 Fuzzy logic rules

Ve Voltage error (V e )
HN AN LN Z LP AP HP
HN HN HN HN AN LN Z LP
AN HN HN HN AN Z LP AP
LN HN HN AN LN Z AP HP
Z HN AN LN Z LP AP HP
LP HN AN Z LP AP HP HP
AP AN LN Z AP HP HP HP
HP LN Z LP AP HP HP HP

4 Simulation Results

Case-1: System operating at fixed irradiance

By fixing the irradiation, the short-circuit current at 5.6 A, the PV system produces
the voltage and current as 117.3 V dc and 4.75 A, respectively as shown in Fig. 4a. So,
multiplying both current and voltage gives the total power 558 W. But the required
power at the grid is varying in the range of 662–445 W. The power changes at a time
duration t = 40 ms. Thus, our PV output power generates accurate power by which
we get the requested power at the grid side. (b) Due to the presence of inductor and
capacitor, the system generates reactive power from zero to 250 var which is clearly
shown in the simulation. (c) The third graphical curve shows the output DC voltage
which is coming from the PV module. (d) Battery is a more important part here;
when the PV module generates less power, battery discharges, and when the PV
module generates more power with respect to grid, then battery charging condition
is available. As the theory shows here also in the simulation part, the current becomes
1.9 A at a time duration 0–40 ms; it clearly shows the battery discharging condition
of battery. The battery current at −1.9 A during 40–100 ms interval indicates the
charging condition of battery. (e) The fifth graphical curve shows the inverter side
AC current, but it contains a lot of harmonics. This cannot be used to operate at
the grid side. (f) The high contained of harmonic is not tolerable at the grid side.
Therefore, by using an additional LCL filter, the system minimizes the nonlinearity
and the graph shows in figure. The single phase current and voltage output results
are illustrated in Fig. 5(a–e).
Figure 6 shows the FFT analysis of the grid side current. LCL filter reduces the
output distortions of the grid side and calculates the harmonic alterations of the
output current. There are totally eight cycles, but here studying four cycles out of
eight cycles by fixing the frequency at 50 Hz. The THD is coming 0.51%, which is
very less and well fit in the environmental condition and gives very good dynamic
response.
A Novel Control Approach for Multi-level Inverter-Based Microgrid 991

(a)

(b)

(c)

(d)

(e)

(f)

Fig. 4 a Change in real power. b Change in reactive power. c Solar voltage. d Current through
battery energy system. e Seven-level inverter current. f Grid current
992 S. K. Routray et al.

(a)

(b)

(c)

(d)

(e)

Fig. 5 Under constant environmental condition a p-p voltage of seven-level RSCI; b seven-level
inverter p-voltage because of neutral point; c filtered p-voltage of RSCI w.r.t neutral point; d filtered
Von filtered neutral point voltage reference to neutral; e filtered Van-filtered phase voltage reference
to neutral
A Novel Control Approach for Multi-level Inverter-Based Microgrid 993

Fig. 6 FFT results of grid current

Case-2: The system operates at different irradiance

At different irradiance, the output of the proposed approach is illustrated in
Fig. 7. Figure 8 shows the FFT analysis of the grid side current. Using LCL fil-
ter minimizes the output harmonics of the grid side and calculates the total harmonic
distortions of the output current. There are totally eight cycles containing in the
figure, but here, studying four cycles out of eight cycles by fixing the frequency at
50 Hz. The THD is coming 0.18%, which is very less and well fit in the environmental
condition and gives very good dynamic response.

5 Conclusion

This paper shows that reduced switched seven-level inverter is easily incorporated
in the microgrid system. It provides increase in voltage level by which the solar-
battery system achieves decrease in voltage errors. The switching losses are mini-
mized because the inverter only contains eight switches and provides seven voltage
levels. This concept gives better result as compared to the conventional VSI. This
manuscript also illustrates the battery operation according to the generation demand.
By implementing the appropriate MPPT technique and SVPWM technique, the sys-
tem achieves maximum power and also is able to generate the actual switching pulses
for the proposed inverter. Fuzzy logic controller is used to generate the actual current
signal for controlling the switching pulses, and an inductor–capacitor–inductor filter
is additionally used to mitigate the nonlinearity content of the inverter output current
and voltage. By analyzing the above simulation results, FFT analysis report shows
that the proposed system minimizes the harmonics as compared to the conventional
approach.
994 S. K. Routray et al.

(a)

(b)

(c)

(d)

(e)

Fig. 7 a Phase voltage; b solar voltage; c battery condition; d grid voltage; e phase voltage and
current
A Novel Control Approach for Multi-level Inverter-Based Microgrid 995

Fig. 8 FFT results of grid current

References

1. Sahoo, B., Routray, S.K., Rout, P.K.: A new topology with the repetitive controller of a reduced
switch seven-level cascaded inverter for a solar PV-battery based microgrid. Eng. Sci. Technol.
21(4), 639–653 (2018)
2. Bragard, M., Soltau, N., Thomas, S., De Doncker, R.W.: The balance of renewable sources and
user demands in grids: power electronics for modular battery energy storage systems. IEEE
Trans. Power Electron. 25(12), 3049–3056 (2010)
3. Sahoo, B., Routray, S.K., Rout, P.K.: Repetitive control and cascaded multilevel inverter with
integrated hybrid active filter capability for wind energy conversion system. Eng. Sci. Technol.
(2019)
4. Yazdani, A., Di Fazio, A.R., Ghoddami, H., Russo, M., Kazerani, M., Jatskevich, J., Strunz, K.,
Leva, S., Martinez, J.A.: Modeling guidelines and a benchmark for power system simulation
studies of three-phase single-stage photovoltaic systems. IEEE Trans. Power Deliv. 26(2),
1247–1264 (2011)
5. Glassley, W.: Geothermal Energy: Renewable Energy and the Environment, CRC Press, Boca
Raton (2010). Chapman, P.L., Esram, T.: Comparison of photovoltaic array maximum power
point tracking technique
6. Burusteta, S., Pou, J., Ceballos, S., Marino, I., Alzola, J.A.: Capacitor voltage balance limits
in a multilevel-converter-based energy storage system. In: Proceedings of the 14th European
Conference on Power Electronics and Applications, Aug/Sept, pp. 1–9 (2011)
7. Xinchun, L., Gao, S., Li, J., Lei, H., Kang, Y.: A new control strategy to balance neutral-point
voltage in three-level NPC inverter. In: Proceedings of the IEEE 8th International Conference
on Power Electronics, ECCEAsia, May/Jun, pp. 2593–2597 (2011)
8. Rodriguez, J., Bernet, S., Steimer, P.K., Lizama, I.E.: A survey on neutral-point-clamped
inverters. IEEE Trans. Ind. Electron. 57(7), 2219–2230 (2010)
9. Karimi, M., et al.: A developed single-phase cascaded multilevel inverter with reduced number
of circuit devices. Int. J. Ambient Energy 40(3), 254–262 (2019)
10. Arcos-Aviles, D., et al.: Fuzzy logic-based energy management system design for residential
grid-connected microgrids. IEEE Trans. Smart Grid 9(2), 530–543 (2016)
11. Sahoo, B., Routray, S.K., Rout, P.K.: Artificial neural network-based PI-controlled reduced
switch cascaded multilevel inverter operation in wind energy conversion system with solid-state
transformer. Iran. J. Sci. Technol. Trans. Electr. Eng. 1–21 (2019)
12. Nabae, A., Takahashi, I., Akagi, H.: A new neutral-point-clamped PWM inverter. IEEE Trans.
Ind. Appl. IA-17(5), 518–523 (1981)
996 S. K. Routray et al.

13. Geetha, R., Ramaswamy, M.: New PWM strategy for three-phase multilevel inverter. Int. J.
Power Electron. 7(1–2), 86–108 (2015)
14. Siddique, M.D., et al.: Optimal design of a new cascaded multilevel inverter topology with
reduced switch count. IEEE Access 7, 24498–24510 (2019)
An AIA-Based Parameter Extraction
Method for PV System

Swati Sucharita Pradhan and Raseswari Pradhan

Abstract This paper presents parameter extraction using an artificial intelligence

algorithm (AIA) based on Photovoltaic (PV) system. The mathematical model of a
PV system is highly nonlinear in nature. AIA methods are very effective for parameter
extraction of this system. Numerous AIA methods are tested in this field. But, Cuckoo
Search Algorithm (CSA) is a recent AIA method that is found to be very effective in
many problems. But, it is never used in PV system parameter extraction. Therefore,
the proposed AIA method is designed using CSA-based algorithm. The efficacy of
the proposed method has been compared with two other approaches such as Genetic
algorithm Villalva’s Numerical Iterative method. Here, for the study of the above two
techniques, a 200 W PV array called KC200GT has been taken as a case study. The
performance of this studied PV model has been realized with the help of different
simulation results. The parameters are extracted under Standard Test Conditions
(STC). Comparing results obtained from the discussed methods, it is found that
CSA-based method is more accurate with fast convergence property.

Keywords PV cell · Parameter extraction · CSA · GA · Villalva’s numerical

iterative method

1 Introduction

Electricity generation using PV has become one of the most propitious sources of
renewable green energy on account of the environment and economy [1]. Again the
PV systems are clean, inexhaustible energy sources and require low maintenance.
Hence PV system is preferred over other renewable energy sources based systems
[2]. But, PV panels have low energy conversion properties [3]. Therefore, different
measures are being taken for increasing this conversion efficiency. One of the vital
measures among them is to select the exact parameters of a studied PV panel in order
to analyze its properties [4]. For these different parameter extraction methods are used

S. S. Pradhan · R. Pradhan (B)

Department of EE, VSSUT, Burla, Odisha, India
e-mail: [email protected]

© Springer Nature Singapore Pte Ltd. 2020 997

G. Pradhan et al. (eds.), Advances in Electrical Control and Signal Systems,
Lecture Notes in Electrical Engineering 665,
https://doi.org/10.1007/978-981-15-5262-5_77
998 S. S. Pradhan and R. Pradhan

to calculate values of unknown parameters by considering data from manufacturer

data sheets [5].
Metaheuristic based parameter extraction techniques can make the extraction pro-
cedure quicker and without hampering its accuracy [6, 7]. Genetic algorithm (GA)
is one of the member of this metaheuristic family. It is inspired by evolution. This
algorithm uses a simple chromosome based data set. GA is often viewed as a fast
function optimizer [8, 9]. However, it is very difficult to choose parameters like gen-
eration size, offspring size in each generation, etc. Another problem, in this case,
is not finding the solution that is most optimum for a certain case [10]. Recently,
another AI-based algorithm known as Cuckoo search algorithm (CSA) is becom-
ing famous due to its simplicity and quick convergence [11]. Since 2009, CSA [12,
13] is widely used metaheuristic based parameter extraction technique. It is an effi-
cient swarm-intelligence-based algorithm. CSA has numerous benefits like its ease
in implementation and efficiency in solving extremely nonlinear optimization prob-
lems as in real-time applications. Hence, we have implemented this method in our
work.
In this paper, the model with five parameters is considered. Techniques based on
CSA [5] are used for the extraction of these five unknown parameters.

2 Parameter Extraction Problem Formulation

An ideal PV panel is nothing but a pure dc current source. Electric power is generated
using the PV cells directly by converting solar energy to electrical energy. In general,
the PV phenomenon consists of one prominent and most used model such as the
single-diode-five-parameter model. The I–V characteristics and the current equations
of the PV panel are described mathematically as follows [14].

I = Ipv,cell − Id (1)

where
  qv  
Id = Io,cell exp −1 (2)
AkT
Therefore,
  qv  
I = Ipv,cell − Io,cell exp −1 (3)
AkT
where I pv,cell , I d , I o,cell are the ideal solar current produced, current-across-diode and
reverse-saturable-current, respectively. Similarly, q, k, T, and A are the unit charge of
electron, the Boltzmann constant, cell-temperature in kelvin, and the diode-ideality-
constant respectively. Again I pv,cell , and I o,cell are computed as follows.
An AIA-Based Parameter Extraction Method for PV System 999

Fig. 1 Equivalent circuit Rs ipv

diagram of single diode
model
Id Ish
Iph
vpv
Rsh
D

Ipv,cell = Ipv,n + K i T (4)

 3   
T q Eg 1 1
Io,cell = Ion exp − (5)
Tn Ak T Tn

The equivalent circuit for single diode model of ideal PV cell is shown in Fig. 1
and is corresponding output current (I) is calculated using Eq. (6).
 
V + I Rs V + I Rs
I = Ipv − Io exp −1 − (6)
Vt A Rp

where V t is the module-thermal-voltage and computed as follows.

Vt = Ns kT /q (7)

Here, N s , Rs and Rp is the no. of series cells connected in series, series resistance
and shunt resistance of PV module.
The solar panel characteristics such as P–V and I–V curves are shown in Fig. 2a
and b, respectively. In these figures, it can be observed that there are three distinct
points such as short-circuit-point, MPP, and open-circuit-point. To have a convenient
PV cell, the value of Rs should be very less than the value of Rp . However, the com-
mercially value of Rp is much larger compared to that of diode-forward-resistance.
The value of V oc , I sc , and MPP are the three important parameters of the I–V curve.
In this paper PV equivalent single diode modeling circuit is taken.
The PV module response be subject to both its internal constituents and environ-
mental influences such as level of solar irradiation and the junction temperature.
The P–V characteristics of the PV cell are shown in Fig. 2b. It depends on V oc , I sc
and MPP. These parameters are provided in the manufacturer’s datasheet. However,
the PV cell mathematical equation consists of some other parameters like Rs , Rp , A,
I o , and I pv , etc. These additional parameters are not mentioned by the manufacturer.
In this study, to obtain the parameter extraction using a CSA-based technique is done.
The layout of the entire procedure is shown in Fig. 3a. Further, the flowchart of the
entire procedure is shown in Fig. 3b.
1000 S. S. Pradhan and R. Pradhan

Fig. 2 a I–V curves and

b P–V curves for PV panel MPP

P (Watts)
V (Volts)
(a)

MPP

I (Amps)

V (Volts)
(b)

3 CSA-Based Parameter Extraction Method

CSA algorithm [3] is one of the recognized artificial intelligent optimization methods
in the recent era. This method can be used for solving both constrained as well
as unconstrained optimization issues. It is widely in applications in technical and
nontechnical areas [8, 11].
The CSA is a stochastic algorithm. It is formulated taking inspiration from nature.
It is capable of giving globally optimum solutions. It follows the following standard
procedures such as:
1. A cuckoo bird randomly dumps its each lays eggs in other bird’s nests. These are
the local solution.
2. From all individual nest, the best nest is chosen that is having the best quality of
eggs. That is the global solution.
3. The best eggs in the best nests have more survival chances. Therefore, that
solution forward to the next generation.
4. The constraint in this problem is a limitation in host nests. Also chances in got
caught by the host bird. In that case also, the egg is not being hatched means the
solution is not being solved.
Therefore, this algorithm can give a solution. In the method of PV parameter esti-
mation, the minimization of the difference of the measured and the calculated data is
done by manipulating the normal electrical parameters provided in the manufacturer’s
datasheet.
An AIA-Based Parameter Extraction Method for PV System 1001

G Analytical
Calculation
T
Error AIA based
Optimization
Calculation
Algorithm
G
Mathematical
Modeling
T

(a)

(b)

Fig. 3 a Layout of proposed AIA-based parameter extraction technique and b flowchart of the
entire procedure
1002 S. S. Pradhan and R. Pradhan

STEP.1-Initialization of ‘n’ number of host nest(population)

STEP.2-Define a stopping criterion

STEP.3-Choosing a egg of cuckoo via levy flights and compute its fitness(Fi)

STEP.4-Randomly choosing an egg from other bird’s nest and calculate its fitness(Fj)

STEP.5-Checking If Fi>Fj, replace jth egg by ith egg

A fraction (pa) of worse nests are demolished and replaced by the new ones

Preserve good nests (best solution)

STEP.6-End program

Fig. 4 Proposed CSA-based parameter extraction algorithm

The cost-function in this problem is calculated by considering the root-mean-

square-of error (RMSE) as follows.

1 
N
RMSE =  ( f (V, I, x))2 (8)
N i=1

where N be the number of experimental data and in case of SDM. Also, x is the
unknown PV module parameters that are to be calculated where x = [I pv , I o , Rs , Rp ].
Here, the errors are calculated as follows.

Error1 = Voc,stc − Voc,calculated (9)

and

Error2 = Isc,stc − Isc,calculated (10)

In other words, Eq. (8) is the homogeneous expression of Eq. (6) [7]. The steps
of the proposed algorithm are shown in Fig. 4.

4 Result Analysis

To test the efficacy of the proposed GA-based parameter extraction system,

KC200GT PV array is considered. The manufacturer datasheet of KC200GT PV
array under the STC conditions is given in Table 1 (Fig. 5).
An AIA-Based Parameter Extraction Method for PV System 1003

Table 1 Manufacturer
S. No. Parameters Value
datasheet of the KC200GT
PV array at STC 1 I mp (A) 7.61
2 V mp (V) 26.3
3 Pmax,e (W) 200.143
4 I sc (A) 8.21
5 V oc (V) 32.9
6 K v (V/K) −0.123
7 K i (A/K) 0.0032
8 Ns 54
9 Np 1

Using the data mentioned in the table and Eqs. (4)–(6) under an iterative process
as mentioned in [1], the values of A, Rs , Rp , I pv , and I o,cell are obtained as 1.3, 0.221 ,
415.405 , 8.214 A and 9.825 × 10−8 A, respectively. For which, the P–V and I–V
characteristics are found as shown in Fig. 4a and b respectively. Again, the same
process of the parameter extraction is applied for the studied PV system using GA-
based algorithm. Here, in the process, the objective function (V, I, x) is minimized
with respect to the electrical parameter range. In theoretical, the value of I pvn is
assumed to be approximately equal to value of I scn . E gn value is in a loose range from
1 to 2 eV, value of K i is provided by the data sheet. The value of I on is usually taken
lesser that of 50 µA. As mentioned in [10], there is a range between 1 and 2 is chosen
for the ideality factor. Most of the manufacturers choose a value lesser than 0.5  for
Rs and for Rp range is chosen (10–500)  [11, 12]. During the parameter estimation
process for the SDM, the applied CSA optimization algorithms are programmed in
m.file using MATLAB code which includes population size, maximum generation
number as 25, and 1000 respectively. In the case of the Genetic algorithm The function
“ga” is used for minimization in MATLAB [13] along with specifying the number
of variables, crossover rate Pc = 0.8, and mutation rate Pm = 0.2, is utilized for the
convergence process test. The results obtained by using CSA is compared with that
of GA and Villalva’s is compared in Table 2. It can be seen in the table that RMSE in
the case of CSA is less compared to both GA and Villalva’s. Therefore, it is giving
more accurate results.
The P–V and I–V characteristic curves are drawn using the parameters obtained
from each optimization techniques like CSA, Villalva’s, and GA. They are compared
in Fig. 6a and b.
As shown in Fig. 6, the green colour curve represents the (P–V ) and (I–V ) char-
acteristics curves obtained using parameters provided in Villalva [1]. Blue curve
represents characteristic curve using the parameter obtained from GA. Similarly, the
red colon curve is for the CSA algorithm. From the above figure, it can be seen that
CSA-based algorithm does the parameter extraction more accurately than that of
Villalva’s and GA.
1004 S. S. Pradhan and R. Pradhan

(a)

(b)

Fig. 5 a I–V curve and P–V curve of studied PV system at STC in case of parameters using
CSA-based algorithm
An AIA-Based Parameter Extraction Method for PV System 1005

Table 2 Values for the

Electrical Villalva et al. GA [6] CSA
extracted parameters for
parameters [5]
different optimization
algorithms I pv (A) 8.21 8.5005 8.3325
I o (A) 9.32 × 10−06 8.02 × 10−04 2.32 × 10−06
Rs () 0.221 0.10 0.1635
Rp () 415.405 451.138 460.4128
A 1.3 1.3 1.3
Time taken 1.412139 0.289045 0.13325
(s)
Electrical Villalva et al. GA CSA
parameters [1]
I pv (A) 8.21 8.5005 8.3325
I o (A) 9.32 × 10−06 8.02 × 10−04 2.32 × 10−06
RMSE 0.0975 0.0191 0.010

The characteristic curves for different solar irradiances are shown in Fig. 7a and
b respectively using the parameters using CSA-based algorithm. From these two
figures, it can be seen that the PV current and PV power are varying with PV voltage
according to the theoretical concept in case of variations in irradiance. When, irra-
diance increases, the MPP power, and MPP current are increasing. Also, V oc point
is increasing.

5 Conclusion

A mathematical model is necessary for analysis of PV system. For that, all parameters
are required to be known. However, values of a few parameters are provided in
the manufacturer datasheet. Whereas few are still remaining unknown. Hence, a
parameter extraction method is required. The mathematical model is nonlinear in
nature. Therefore, metaheuristic methods are the best fit in this problem. A CSA-
based parameter extraction method for PV panel is presented in this paper. For testing
this method, a practical model is considered that is with a maximum 200 W power
conversion capacity. Comparing the performance results of the studied PV panel with
parameters of the proposed method has been done with that of another two methods
such as Villalva’s Iterative technique and GA method. From the results, it is seen that
the proposed CSA technique needs less time of convergence with higher accuracy.
1006 S. S. Pradhan and R. Pradhan

10

CSA
8
GA
Villalva
Current (Amps)

0
0 5 10 15 20 25 30 35 40
Voltage (volts)

(a)
200
CSA
GA
150 Villalva
Power (Watts)

100

50

0
0 5 10 15 20 25 30 35
Voltage (Volts)

(b)
Fig. 6 a Comparison I–V curve and b P–V curve
An AIA-Based Parameter Extraction Method for PV System 1007

9
2
1000 W/m
7.91
2
7 400 W/m
2
Current, I (Amps)

6 800 W/m
5 2
600 W/m
4

0
0 5 10 15 20 25 30 35.55 40
Voltage, V (Volts)

(a)
205
2
600 W/m
2
400 W/m
150
2
Power, P (Watts)

800 W/m
2
1000 W/m
100

50

0
0 5 10 15 20 25 30 35.55 40
Voltage, V (Volts)

(b)
Fig. 7 a I–V curves and b P–V curves for different solar irradiances of PV panel in case of
parameters using CSA-based algorithm

References

1. Díez-Mediavilla, M., Alonso-Tristan, C., Rodríguez-Amigo, M.D., García-Calderón, T.: Imple-

mentation of PV plants in Spain: a case study. Renew. Sustain. Energy Rev. 14(4), 1342–1346
(2010)
2. Ellabban, O., Abu-Rub, H., Blaabjerg, F.: Renewable energy resources: current status, future
prospects and their enabling technology. Renew. Sustain. Energy Rev. 39, 748–764 (2014)
3. Green, M.A., Bremner, S.P.: Energy conversion approaches and materials for high-efficiency
photovoltaics. Nat. Mater. 16(1), 23–34 (2017)
1008 S. S. Pradhan and R. Pradhan

4. Subudhi, B., Pradhan, R.: A comparative study on solar array parameter extraction methods.
Int. J. Renew. Energy Technol. 3(3), 295–315 (2012)
5. Villalva, M.G., Gazoli, J.R., Ruppert Filho, E.: Comprehensive approach to modeling and
simulation of photovoltaic arrays. IEEE Trans. Power Electron. 24(5), 1198–1208 (2009)
6. Gao, X., Cui, Y., Hu, J., Xu, G., Wang, Z., Qu, J., Wang, H.: Parameter extraction of solar cell
models using improved shuffled complex evolution algorithm. Energy Convers. Manag. 157,
460–479 (2018)
7. Chen, Y., Sun, Y., Meng, Z.: An improved explicit double-diode model of solar cells: fitness
verification and parameter extraction. Energy Convers. Manag. 169, 345–358 (2018)
8. Onwubolu, G.C., Mutingi, M.: A genetic algorithm approach to cellular manufacturing systems.
Comput. Ind. Eng. 39(1–2), 125–144 (2001)
9. Mohamed, F.A., Koivo, H.N.: Online management genetic algorithms of microgrid for
residential application. Energy Convers. Manag. 64, 562–568 (2012)
10. Brest, J., Greiner, S., Boskovic, B., Mernik, M., Zumer, V.: Self-adapting control parameters
in differential evolution: a comparative study on numerical benchmark problems. IEEE Trans.
Evol. Comput. 10(6), 646–657 (2006)
11. Shehab, M., Khader, A.T., Al-Betar, M.A.: A survey on applications and variants of the cuckoo
search algorithm. Appl. Soft Comput. 61, 1041–1059 (2017)
12. Brajevic, I., Tuba, M.: Cuckoo search and firefly algorithm applied to multilevel image
thresholding. In: Cuckoo Search and Firefly Algorithm, pp. 115–139. Springer, Cham (2014)
13. Zamani, A.A., Tavakoli, S., Etedali, S.: Fractional order PID control design for semi-active
control of smart base-isolated structures: a multi-objective cuckoo search approach. ISA Trans.
67, 222–232 (2017)
14. Sridhar, R., Jeevananathan, D., ThamizhSelvan, N., Banerjee, S.: Modeling of PV array and
performance enhancement by MPPT algorithm. Int. J. Comput. Appl. 7(5), 0975–8887 (2010)
Smart Solar Electromagnetic Energy
Harvesting Footware

Aneesh R. Nair and Vinit Kotak

Abstract Electronic devices are becoming smarter and miniature which can be
easily integrated into any object and communicated wirelessly. One challenge is to
power them up with available resources. This is the challenge that we face everywhere
including mobile, tablet, laptop and other electronic portable devices. We are trying
to harvest energy using solar, wind or other sources, and here, our main focus is
to harvest the energy that we waste while walking, to charge and power electronic
system inside the shoe. This shoe is designed for personal as well as militarily purpose
where energy conservation is crucial. Soldiers who are on the field needs to power
various devices on their body, and smart shoe can be a useful emergency option. Other
features like tracking, calling, weight measurement, odor exhaust are also provided
for other personal use, and USB out for mobile charging, etc., is also incorporated.
This project focuses on charging a 2000 mAh battery inside the shoe while walking
and using a solar cell mounted on the front of the shoe when the person is resting.

Keywords Energy conservation · Battery · Solar cell · Electromagnet · SIM800 ·

USB

1 Introduction

Nowadays, a lot of gadgets are battery powered and need timely charging, mostly in
military applications where there are a lot of gadgets that needs charging which can be
achieved only by a battery backup. In the field, they do not have any option but go back
to the base camp for charging. The proposed design harvests energy while walking
and through solar cell mounted on the front of the shoe. It also has a USB out which

A. R. Nair (B)
Department of Electronics Engineering, Shah & Anchor Kutchhi Engineering College,
Mumbai 88, India
e-mail: [email protected]
V. Kotak
Shah & Anchor Kutchhi Engineering College, Mumbai 88, India
e-mail: [email protected]

© Springer Nature Singapore Pte Ltd. 2020 1009

G. Pradhan et al. (eds.), Advances in Electrical Control and Signal Systems,
Lecture Notes in Electrical Engineering 665,
https://doi.org/10.1007/978-981-15-5262-5_78
1010 A. R. Nair and V. Kotak

can be connected to the gadget for charging. Different forms of renewable energy
sources are researched to capture and convert into electrical energy which can supply
energy to wearable devices [1]. Using energy where it is generated can solve problem
of storage and distribution. Many attempts have been made to develop generators,
thin solar cells [2], and other technologies like surgically positioned piezoelectric
material in animals [3]. Energy captured while walking or through solar inserted into
shoes is the most efficient form of energy harvesting [4].
Other features like an LCD display for display current time, messages and phone
numbers can be used for personal purpose. An exhaust is mounted on the shoe to
remove foul odor due to sweating, and an LED light for emergency lighting purpose
is also included in the shoe. The system inside the shoe is multipurpose and has
Bluetooth, SD card storage for MP3 playback onto the Bluetooth headset.
The charging is done using electromagnet what moves the flap while walking and
rotates the magnet, this magnet rotates on the coil which produces electricity which is
rectified and stored onto a battery (2000 mAh). Similarly, the solar cell also charges
the battery during bright sunlight using charging circuit onto the same battery, and
this helps the battery to be charged quickly.

2 Features

• Charge while walking

• Charge through solar cell
• 2000 mAh battery backup on USB
• LCD for clock, messaging and other multimedia purpose
• Temperature and humidity indication
• Exhaust fan to remove foul odor
• Vibrator for foot massage
• GPS tracking
• Emergency Light
• Heartbeat measurement.

The system records heartbeat and shows patterns on the screen to track during
running. The proposed system implements all the above features for military and
personal use.

2.1 System Design

Figure 1 shows the block diagram of the generator and solar cell used to charge
the battery, Fig. 2 shows how the peripherals are connected to the controller. The
controller has custom firmware that takes all the data from the sensors and displays
Smart Solar Electromagnetic Energy Harvesting Footware 1011

Fig. 1 Block diagram of power supply

Fig. 2 Block diagram of microcontroller circuit

it on the LCD. All other parameters can be set using the microcontroller like fan
on/off, vibrator, emergency light, temperature and humidity.
1012 A. R. Nair and V. Kotak

Fig. 3 Buck boost for charge stabilization

2.2 Charge Controller Circuit

The charge controller in Fig. 3 is the heart of the circuit, and each and every pulse is
charged to a primary capacitor which then dissipates the charge to the main battery.
Similarly, the solar cell also uses a capacitor which charges when the light is sufficient,
and then, once the capacitor is above a threshold, the charge controller will constantly
charge the Li-ion battery. Once the main battery is above 3.3 V, the microcontroller
switches on and waits till the battery has achieved 4 V, now, all other peripherals can
be activated till the battery drops to 3.3 V, and after which, it deactivates all peripherals
and switches to charging mode. The above circuit shows how MCP73831 and buck
boost circuit are used to charge and stabilize the voltage across the circuit.

2.3 USB Charging

USB charging circuit in Fig. 4 is one of the features that can be used when an
emergency charging of a cell phone is needed. The main feature of the shoe is to
charge the battery while walking and solar cell and then use the charge for emergency
purposes. This is another circuit that gives USB charge output which can be used as
a power bank when urgent charging is needed.

2.4 Sim800H Circuit

Sim800H is an ideal module which has all major features of radio transmission. It
implements online tracking with the help of GSM and GPS, and also, it has Bluetooth
inbuilt as shown in Fig. 5. It has some flash inbuilt which can be used for programming
which puts less load on the controller if sufficient memory is not available.
Smart Solar Electromagnetic Energy Harvesting Footware 1013

Fig. 4 USB charger

Fig. 5 Sim800H circuit

1014 A. R. Nair and V. Kotak

Fig. 6 Microcontroller schematic

2.5 Microcontroller Circuit

In Fig. 6, the microcontroller circuit is the heart of the shoe which monitors the battery
voltages and gives all indication about the status of the battery and the peripherals. It
uses ATXmega8 which interfaces a display LCD and other driver modules, and when
any command is given, it checks the battery status and then executes the command.
The exhaust fan, LCD, GPS tracker, etc., are all controlled by the controller.

2.6 Fan Circuit

The fan circuit shown in Fig. 7 is also controlled by the controller by PWM so
that controlled amount of power is used while directly driving the fan, and also, the
vibrator has similarly circuit which is activated by the controller.
The emergency light circuit shown in Fig. 8 is the only circuit that is driven
directly by the battery and has a small switch to switch on the LED light, and it is an
emergency feature when there is no or low light.

3 Implementation (Energy Harvesting)

The basic idea behind the development of smart Shoe (Fig. 9) was to find an efficient
way to harvest energy while walking and through solar cell. We had lots of challenges
Smart Solar Electromagnetic Energy Harvesting Footware 1015

Fig. 7 Fan circuit

Fig. 8 Emergency LED circuit

Fig. 9 Basic building block

1016 A. R. Nair and V. Kotak

Fig. 10 Generator

like the shoe should not alter the walking style of the person. The energy harvested
should be stored and used like a power bank or to power any electronic devices.
The device also has USB input port for the user to charge the device at home and
use other peripherals and tracking device. They implement the idea we had to design
a generator (Fig. 10) with gears and flap that can convert the mechanical energy
into electrical. We tried with different gear ratios which uses the rotational inertia to
generate power through continuous movement of gear and coil movement.
Different levels of voltages are produced depending on the position of the coil
and magnet [5], more the speed of rotation more is the voltage produced. The turns
of the coil is calculated depending on the speed of flap movement, and the optimum
coil length is chosen depending on the maximum RMS voltage.

4 Result and Gait Analysis

The gait cycle is defined as the systematic study of leg movement using instru-
mentation and human observation. This period is divided into two distinctive
phases:
• Phase of stance, during this phase, the foot is in contact with the terrain. In this
phase, the foot undergoes five movements—initial contact, loading response, mid
stance, terminal stance and toe off. This is the most useful phase which converts
most of the walking energy into electrical with the whole body weight carried on
the foot which is responsible for generator flap movement.
• Phase of swing, in which, the leg moves forward supporting the whole body
support [3], and during this period, the other leg generates energy.
Smart Solar Electromagnetic Energy Harvesting Footware 1017

Fig. 11 Generator waveform

Considering the above points, the generator is designed so that it does not change
the walk pattern and also generate maximum power. Going ahead to test the result,
we first do an analysis of the details of work we did in a step by step manner. Firstly,
the peripherals are tested with the firmware by triggering the drivers manually, and
later, firmware is tested module wise. For example, the board is powered with an
external source, and the peripherals are tested to confirm its working. After which,
the power section is tested separately, and then, the voltages are stabilized. The below
is the screen shot of the output of the generator which is an AC waveform.
The generator screenshot (Fig. 11) shows 6–10 AC when a single walk or flap
movement is done. The voltage is then rectified using a full bridge rectifier and
stored on a capacitor (10–14 VDC) which then is connected to the charge controller
which supplies the voltage required for the microcontroller and other peripherals.
The results are shown below.
Generator: 6–10 VAC
Rectifier: 8–14 VDC
Battery: 2000 mAh
Total charge send to the battery: 80–100 mA.
The solar cell shown in Fig. 12 adds up charge when proper sunlight is present
Solar cell output: 6 V/500 mW
Output capacitor: 100 uF/10 V
Full charge voltage: 4 V.
Once the battery is fully charged, the microcontroller allows all other peripherals
to be activated and used till the battery drops down to 3.3 V where only emergency
light works.
1018 A. R. Nair and V. Kotak

Fig. 12 Solar cell

5 Conclusion

Initially, the proposed model was just to harvest the energy while walking and charge
the battery using solar cell. Later, we thought of implementing other features that
can be of use for military as well as personal use especially when a person is driving
or riding. These features can be life saving or can be used in case of an emergency.
Currently, there are other designs that use energy harvesting, but they are limited
to their design, speed of walking and time. Our design mainly focuses on both type
of charging, i.e., even when the person is sitting idle, the solar charges the battery.
The current design can be scaled with other useful features, and the whole shoe can
be made of paper solar cell which can be more efficient. The future scope of the
product is dependent on the application and usability.
Finally, Fig. 13 shows the final model of the smart shoe works as per specification
with all other peripherals working seamlessly.
Smart Solar Electromagnetic Energy Harvesting Footware 1019

Fig. 13 Smart shoe model

1020 A. R. Nair and V. Kotak

Acknowledgements I sincerely feel that the credit of this thesis could not be narrowed down to
only one individual. It gives me great pleasure to express my deep sense of gratitude to my project
guide Dr. V. C. Kotak for his resourceful and able guidance which lead to timely completion. It
was really his insight and obsession for innovative ideas that motivated me to consider the idea
seriously.
I am grateful to Prof. Subha Subramaniam (H.O.D. of Electronics Department) whose profound
cooperation has been a pillar in making this project work possible. I would also like to express my
deep regards and gratitude to the principal Dr. Bhavesh Patel. I would also like to thank the entire
teaching and non-teaching staff of Electronics department, who extended their kind cooperation.
Thank you all and I wish you all health and happiness.

References

1. Khaligh, A., Zeng, P., Zheng, C.: Kinetic energy harvesting using piezoelectric and electromag-
netic technologies state of the art. IEEE Trans. Ind. Electron. 57(3), 850–860 (2010)
2. Chapman, P., Raju, M.: Designing power systems to meet energy harvesting needs. TechOnline
India 8(42) (2008)
3. Frontoni, E., Mancini, A., Zingaretti, P.: Energy harvesting applied to smart shoes. IJMET 9(8),
412–421 (2018)
4. Chalasani, S., Conrad, J.M.: A survey of energy harvesting sources for embedded systems. In:
Southeastcon, 3–6 Apr 2008, pp. 442–447. IEEE (2008)
5. Duffy, M., Carroll, D.: Electromagnetic generators for power harvesting. In: Proceedings of
Power Electronics Specialists Conference 2004. PESC 04. 2004 IEEE 35th Annual, 20–26 Jun,
vol. 3, pp. 2075–2081 (2004)
A Novel Low Contrast Image
Enhancement Using Adaptive
Multi-Resolution Technique and SVD

Lalit Mohan Satapathy and Pranati Das

Abstract An image is called visually rich when its brightness and contrast are
properly adjusted. But it was reported in different applications that the contrast of an
image was affected by the external interferences. Due to the above phenomena, the
image quality degrades. To improve the quality of the image, contrast enhancement
method has been used. This technique adjusts the intensity distribution of the low-
quality image to solve the problem. But the old techniques were not suitable for all
kinds of images. To overcome the limitations, this paper presents a novel variational
mode decomposition (VMD)-based enhancement technique associated with singular
value decomposition (SVD). Initially, the reference images and the processed image
were decomposed into various modes using variational mode decomposition (VMD).
Then in the second step, the maximum value of the singular matrix was calculated
for the selected mode of each image. An image-dependent correction factor was
calculated using the singular value matrix (SVM). Finally, the image is reconstructed
by simply adding the β-corrected mode image with the unprocessed modes of the
reference image. To avoid the over-enhancement, a new weighted factor was applied
to the original as well as to the enhanced image. To validate the algorithm, the
proposed method was tested on different types of publicly available low contrast
images. The experimental result shows that the proposed algorithm gives significant
enhancement in terms of contrast over other state-of-the-art methods.

Keywords Enhancement · DWT · SVD · VMD

L. M. Satapathy (B)
Institute of Technical Education and Research, Siksha ‘O’ Anusandhan (Deemed to be
University), Bhubaneswar, Odisha 751030, India
e-mail: [email protected]
P. Das
Indira Gandhi Institute of Technology, Sarang, India

© Springer Nature Singapore Pte Ltd. 2020 1021

G. Pradhan et al. (eds.), Advances in Electrical Control and Signal Systems,
Lecture Notes in Electrical Engineering 665,
https://doi.org/10.1007/978-981-15-5262-5_79
1022 L. M. Satapathy and P. Das

1 Introduction

The visual appearance and perception of a digital image were generally characterized
by its brightness and contrast. Brightness preservation and contrast enhancement
were still one of the major challenges to the researchers working in the digital image
domain. Furthermore, contrast enhancement was considered as an empirical method
for both still and moving images. In this enhancement approach, the visual quality
of the image was improved to an appreciable level.
In the last decade, many contrast enhancement techniques have been proposed
by different researchers. Those techniques were coming under either the spatial
domain or transfer domain method. The histogram equalization (HE) is coming
under the spatial domain. HE-based enhancement methods were popularly used for
all types of images due to its simplicity. Concerning the application of enhance-
ment on an image, HE was categorized as global histogram equalization (GHE)
and local histogram equalization (LHE) [1]. The above methods were stretching the
histogram; as a result, a significant loss of contrast in the background is observed.
Over-enhancement, undesirable artifacts were also the drawback of this method. To
overcome these drawbacks, many researchers have proposed many methods [2–7].
Partition-based histogram equalization methods were one of a good example. In
this method, before equalization, the histogram range is divided into parts and then
equalization is applied to separate parts. Some methods like brightness preserving
dynamic histogram equalization (BPDHE) partition the histogram based on its local
maxima [3]. Although these methods give better results as compared to HE, they
fail when the histogram has spikes. Another method proposed was to apply a weight
factor after equalization to limit the enhancement [8, 9].

G(x, y) = α F(x, y) + β (1)

where α and β are called the gain and bias parameters. In Eq. 1, multiplication and
addition are applied to the equalized image as gain and bias parameters. And (x, y)
indicates the pixel location. The range of α and β is depending on image quality and
fixed either manually or automatically.
The above contrast enhancement algorithm performs well on some of the selected
images, but min while facing contrast overstretching or noise amplification prob-
lems. To overcome the above issue, nowadays image decomposition-based contrast
enhancement techniques are becoming popular in the frequency domain [10–13].
The low contrast images were decomposed into sub-bands from low frequency to
high frequency by using a discrete wavelet transform (DWT). The low-frequency
and high-frequency details of a digital image contain the approximation and detail
information. Due to high-frequency components containing the noise information,
only low-frequency components were enhanced using CLAHE [11]. On the other
hand, to enhance the low contrast image (medical and satellite images) by beta adjust-
ment, singular value decomposition (SVD) methods have been proposed. SVD has no
domain limitations. It can perform well in the pixel domain as well as in the frequency
A Novel Low Contrast Image Enhancement … 1023

domain. The singular value-based image enhancement technique was working on the
fundamental of the singular value matrix of the image pixels which contains the geo-
metric information of the image obtained by SVD [14, 15]. Demirel et al. have
proposed a new technique, which contains the DWT and SVD. The low contrast
image is decomposed using DWT, and the low-low sub-band is updated using SVD.
The enhanced image is reconstructed by applying inverse discrete wavelet trans-
form (IDWT). The above-mentioned technique is known as DWT-SVD [16]. Some
weighted technique has been applied to modify the above method [17]. Furthermore,
in this direction, a combination of discrete cosine transform and SVD is proposed
for enhancing the low contrast images [18]. Although the DWT-SVD and DCT-SVD
enhance the low contrast images by using a singular value matrix and deciding the
scaling factor, they were failing to produce the satisfying result when the factor is
approaching one. With all the advantages the constraint is that the wavelet-based
image decompositions were not fully imaging dependent as these techniques require
a basis function to decompose the image. The main limitation of the DWT is to
choose the suitable mother wavelet and its computational complexity. It motivated
us toward an image-dependent decomposition system.
In recent years, Dragomiretskiy et al. [19] have developed the variational mode
decomposition (VMD) method to compute variational modes from signals or images.
The modes are concurrently extracted from the original image based on the solution
of the optimization algorithm using Fourier isometry and the assignment of center
frequencies [19, 20]. The VMD has been successfully applied for image enhancement
applications [21]. The VMD has advantages such as it is fully image-specific and it
does not require any basis function as wavelet transform [21].
In this paper, the low contrast images are taken into consideration. The images are
decomposed using VMD. Singular value decomposition was applied for calculating
the multiplication factor. A weighted average control is applied for controlling the
over-enhancement. The proposed paper is structured as follows: Sect. 2 gives a global
presentation of VMD, SVD, local HE, and a weighted method for enhancement.
Section 3 presents the experimental results and discussion. The conclusions of the
present study are mentioned in Sect. 4.

2 Methodology

The low contrast grayscale images are considered for the present study. The images
are collected from public domain. The dimension of the images is set as 512 × 512
pixels as a preprocessing step.
1024 L. M. Satapathy and P. Das

2.1 CLAHE

Contrast limited adaptive histogram equalization (CLAHE) is used for medical

images, especially where local information plays a vital rule. The method is well
described as a window-based system where the total image is segregated into several
non-overlapping blocks. The individual blocks are treated using histogram equal-
ization. Each pixel is customized based on pixels near the selected one and redis-
tributed the brightness. Nowadays, CLAHE is popularly adopted as it has improved
the over-amplification of noise problems present in the HE. The proposed method
comprises three parts: (i) variational mode decomposition [21], (ii) singular value
decomposition [16], and (iii) weighting operation [12].

2.2 Variational Mode Decomposition

Two-dimensional variational mode decomposition (2D VMD) is an entirely non-

recursive variational mode decomposition model, where the modes are extracted
concurrently. The model looks for k number of modes and their respective center
frequencies in spectral domain, such that the band-limited modes reproduce the input
image exactly or in least squares sense [19]. During the decomposition process, each
mode is mostly compact around a center pulsation ωk [19]. The optimized problem
for evaluation of modes for an original image (M) is given as [20]
 

K
   2
min αk ∇ X AS,K (n)e− jωk n  (2)
{Mk }{ωk } 2
k=1

Subject to


K
MAS,k (n) = Y (3)
k=1

where M AS,k (ω) is the Fourier transform of the 2D analytical signal MAS,k (n).
This 2D analytical
 signal is evaluatedfrom the original image which is given by
MAS,k (n) = 1 + sgn(ω, ωk ) M k (ω) , where M k (ω) corresponds to the Fourier
transform of the kth mode image. The modes are evaluated by solving Eq. 3 using
the augmented Lagrangian [19]. The VMD technique uses different parameters such
as the bandwidth constant (α), number of modes (K), Lagrangian multipliers dual
ascent time step (τ ), tolerance (tol), and the initialization of center frequencies as
prior to decompose an image into modes [19]. In this study, K is simply set to four.
In this study, the number of modes such as K = 4 is considered. Similarly, the other
parameters are used as α = 1000 and tol = 10−5 , respectively [21]. The modes
of original image (as in Fig. 2a) estimated using VMD technique are depicted in
A Novel Low Contrast Image Enhancement … 1025

Fig. 2b–e. As evident from this figure, the original image information is segregated
into different modes based on the frequency content that is from low frequency to
high frequency.

2.3 SVD

The singular value decomposition (SVD) is a factorization process used for real or
complex matrix. The singular value decomposition of a digital image M with size x
× y is factorized as shown in Eq. 4.

M =U VT (4)

where U is a unitary matrix with size x × x, and diagonal values of are known as
singular value. The is a diagonal matrix with size x × y, and V is a unitary matrix
with size y × y. Such type of refactoring allows the user to represent an image with
a comparatively smaller set of data, and the lower set of data is preserving the useful
features of the original image. The T is indicating the transpose operator. The two
matrixes U and V are representing as the left and right singular vectors of image M.
The above discussed singular vectors are indicating the structure as well as geometry
property of an image, whereas the singular values (diagonal of ) are defining the
luminance property of an image. This concludes that the manipulation of singular
value will bias the intensity information of an image without affecting the structure.
The proposed approach for brightness preserving and contrast enhancement is
applicable for low contrast image. The overall procedure of the proposed technique
is well demonstrated in Figs. 1, 2, 3 and 4. Initially, the original image is decomposed
into four modes as Mode 0, Mode 1, Mode 2, and Mode 3 as shown in Figs. 1 and
2. For an easy approach, the Mode 0 is denoted as A. In the next step, CLAHE is
applied on A for generating an equalized image B. To generate the values of U, ,
and V, the SVD is applied on both A and B as shown in Eqs. 5 and 6.

A = UA VA (5)
A

Fig. 1 Decomposition of Original Image (M)

original image using VMD

VMD

Mode0 Mode1 Mode2 Mode3

1026 L. M. Satapathy and P. Das

Fig. 2 a Original image, the VMD decomposition images as b Mode 0, c Mode 1, d Mode 2,
e Mode 3

Mode0

Mode0
CLAHE

Calculate UB ∑B VB Calculate UA ∑A VA of
Of enhanced image of Mode0 using SVD
Mode0 using SVD

Calculation of ξ using equation (7)

Calculation of NEW Mode0

using equation (11)

New Mode0

Fig. 3 Flow chart for generating new Mode 0


B = UB VB (6)
B

max( B)
ξ= (7)
max( A)

The maximum singular values A and B are considered for computing the
scaling factor. The scaling factor (ξ ) is used for upgrading the Mode 0. The new A
is presented in Eq. 8.
A Novel Low Contrast Image Enhancement … 1027

Fig. 4 Flow chart for NEW Mode1 Mode2 Mode3

generation of enhanced Mode0
image

IVMD (N)

Original Image(M)

Weighted average

Contrast
Enhanced Image(P)


New A = U A ξ VA (8)
A

By practical approach, it was found that the above scaling approach fails to produce
contrast enhancement up to satisfactory level, especially in those images which are
falling in mid-range of brightness. In the above cases, the value of ξ is closer to 1.
As a result, the input low contrast image does not change remarkably. To overcome
this problem, a new weighted averaging method is presented which takes A and
B into consideration

1 1
α= ξ+ (9)
4 ξ
  1
New =α ξ + (10)
A A
ξ B

where ξ is calculated from Eq. 7 and α is calculated from Eq. 9. The new equalized
Mode 0 is given by (Fig. 3)


NEW Mode 0 = U A New VA (11)
A

The modified image can be generated by simple adding NEW Mode 0, Mode 1,
Mode 2, and Mode 3. This process is known as inverse VMD (IVMD) as shown in
Fig. 4.
1028 L. M. Satapathy and P. Das

2.4 Weighting Operation

To attenuate the over-enhancement, the weighted average of the images M and N is

processed as follows [12]:

P = M. ∗ Q + β ∗ N . ∗ (unit matrix − Q) (12)

where M and N are the original and constructed images. The * represents the bit-wise
multiplication. β is the brightness compensation factor and it lies between 1 and 2
(1 < β < 2). Q and (unit matrix—Q) are the weighting coefficients of M and N.
⎡ ⎤
Q(1, 1) · · · Q(1, n)
⎢ .. .. .. ⎥
Q=⎣ . . . ⎦
Q(m, 1) · · · Q(m, n)
⎡ ⎤
f (M(1, 1))γ · · · f (M(1, n))γ
⎢ .. .. .. ⎥
=⎣ . . . ⎦ (13)
γ γ
f (M(m, 1)) · · · f (M(m, n))

The sizes of both M and N are m × n. M(i, j) is the grayscale of pixel (i, j) in
the original image M. For the proper enhancement of image, the γ range is fixed
between [0, 5]. In this approach, γ is fixed at 1.5.
The function f is described as follows:

M(i, j) − min(M)
f (M(i, j)) = (14)
max(M) − min(M)

where max(M) and min(M) are denoted as the maximum and minimum gray values
of image M.

3 Experimental Results

In this paper, the proposed VMD-based and SVD-based image enhancement algo-
rithm was applied over several low contrast images such as brain image, satellite
image, and tissue image. In order to verify the performance of the proposed method,
the results are compared with CLAHE, DCT-SVD, and DWT-SVD [11]. Both
qualitative and quantitative results are evaluated based on visual and performance
measure.
A Novel Low Contrast Image Enhancement … 1029

3.1 Qualitative Assessment

For qualitative assessment, Figs. 5, 6 and 7 are demonstrated for validation of the
effectiveness of the algorithm. The visual quality of the images with various enhance-
ments is depicted in the figure. The DCT-SVD sufferers from over-saturation of var-
ious pixels, as a result, fail to produce the details of the information. The CLAHE
and DWT-SVD enhance the image but fail to describe the details of the image from
a quality point of view. This can be observed that the proposed method describes
significantly better contrast and detail information and avoids the over-saturation as
per the presented evidence.

Fig. 5 a Original low contrast image 1, b CLAHE, c DCT-SVD, d DWT-SVD, and e proposed
method

Fig. 6 a Original low contrast image 2, b CLAHE, c DCT-SVD, d DWT-SVD, and e proposed
method

Fig. 7 a Original low contrast image 3, b CLAHE, c DCT-SVD, d DWT-SVD, and e proposed
method
1030 L. M. Satapathy and P. Das

3.2 Quantitative Assessment

Measuring Parameters
The measuring parameters used for quantitative assessment are entropy ratio (ER),
measure of enhancement (EME), structural similarity index measure (SSIM) [11],
and absolute mean brightness error (AMBE).
Discrete entropy indicates the degree of detail enhancement in a digital image.


L−1
I = p(i) log2 p(i) (15)
i=0

where p(i) denotes the probability intensity of the processed image.

The average contrast of an image (EME) is calculated using the following
equation.

1 
m n
Imax (k, l)
EME = 20 log (16)
mn l=1 k=1 Imin (k, l)

Structural information preservation is assessed by computing structural similarity

(SSIM). The SSIM formula is based on three comparison measures such as luminance
(L), contrast (C), and structure (S).

2μ M μ P + C1
Luminance L(M, P) = (17)
μ2M + μ2P + C1
2σ M σ P + C2
Contrast C(M, P) = (18)
σ M2 + σ P2 + C2
σx y + C 3
Structure S(M, P) = (19)
σx σ y + C 3

The multiplication of Eqs. 17, 18, and 19 will produce the following SSIM:

2(μ M μ P + c1 )(2σMP + c2 )
SSIM(M, P) =  2   (20)
μ M + μ2P + c1 σ M2 + σ P2 + c2

where μM and μP are average, σ M2 and σ P2 are variance, and σ MP is covariance of

image and enhanced image, respectively. C 1 = (k 1 L)2 , C 2 = (k 2 L)2 , and C 3 = C 2 /2
are variables to stabilize the denominator, where L is the dynamic range of the pixel
and K 1 = 0.01 and K 2 = 0.03.
The brightness preservation of an image is described by the AMBE. The AMBE
is defined by the difference between the mean brightness of the input image (μM )
and the processed image (μP ).
A Novel Low Contrast Image Enhancement … 1031

AMBE = |μ P − μ M | (21)

4 Results and Discussion

A sample of five images is collected for evaluation (Fig. 8). The entropy, EME, SSIM,
and AMBE of the images from Fig. 8a–e are calculated for different methods and
tabulated for comparison. A comparison of the entropy value for various methods
is presented in Table 1. It reveals that the proposed method outperforms over the
other enhancement technique except image 3, where DWT-SVD performs better
(Table 1). Table 2 demonstrates the EME value; the higher the value of EME, the better
its performance. In some of the cases, CLAHE performs better than the proposed
method.
The structural similarity index validates the geometrical structure of an image.
The range of the SSIM is from 0 to 1. The closer the value of SSIM of an image
toward 1 indicates the structural preservation (Table 3). It indicates that during the
enhancement process, the structure of the image has not deviated. As a result, the
quality of the image is significantly better.
The mean brightness error is the evidence that the proposed method is preserving
the brightness and enhances the contrast (Table 4). The result shows that the error of
the proposed method is less.

Fig. 8 Low contrast sample images considered for objective evaluation

Table 1 Comparison of entropy over various contrast enhancement methods

Test image CLAHE DCT-SVD DWT-SVD Proposed method
Image (a) 5.33 6.86 6.96 7.02
Image (b) 5.84 6.76 6.84 6.89
Image (c) 5.76 7.09 7.32 7.11
Image (d) 4.98 6.02 6.33 6.8
Image (e) 5.35 6.41 6.73 7.1
Mean entropy 5.45 6.62 6.83 6.98
1032 L. M. Satapathy and P. Das

Table 2 EME comparison of the proposed method over various contrast enhancement methods
Test image CLAHE DCT-SVD DWT-SVD Proposed method
Image (a) 07.70 0.91 01.22 03.58
Image (b) 11.14 0.31 04.43 03.50
Image (c) 29.55 5.66 21.55 20.44
Image (d) 25.41 10.22 09.95 17.55
Image (e) 21.91 9.24 14.67 20.55
Mean EME 19.142 5.268 10.364 13.124

Table 3 SSIM comparison of the proposed method over various contrast enhancement methods
Test image CLAHE DCT-SVD DWT-SVD Proposed method
Image (a) 0.91 0.88 0.85 0.92
Image (b) 0.71 0.66 0.72 0.76
Image (c) 0.88 0.83 0.82 0.85
Image (d) 0.70 0.81 0.86 0.86
Image (e) 0.88 0.93 0.92 0.96
Mean SSIM 0.816 0.822 0.834 0.87

Table 4 AMBE comparison of the proposed method over various contrast enhancement methods
Test image CLAHE DCT-SVD DWT-SVD Proposed method
Image (a) 28.33 15.22 16.19 10.55
Image (b) 38.22 24.19 18.55 15.24
Image (c) 21.55 17.94 24.55 17.22
Image (d) 16.89 14.25 12.42 9.21
Image (e) 44.5 26.99 24.21 21.22
Mean SSIM 29.898 19.718 19.184 14.688

5 Conclusion

The paper presents a new multi-resolution enhancement method that is a combi-

nation of VMD, local histogram equalization, and SVD. The dynamic application
of this technique is centered on the low contrasted image. The proposed method
preserves the brightness and enhances the contrast. To enhance the image informa-
tion, the low-frequency components and high-frequency components of an image are
addressed properly to avoid the over-enhancement by applying the weighted factor.
Experimental results are evidence of the performance of the proposed method.
A Novel Low Contrast Image Enhancement … 1033

References

1. Gonzalez, R., Woods, E.: Digital Image Processing, 3rd edn. Prentice-Hall, Upper Saddle River,
NJ (2008)
2. Kim, Y.-T.: Contrast enhancement using brightness preserving bi-histogram equalization. IEEE
Trans. Consum. Electron. 43(1), 1–8 (1997)
3. Ibrahim, H., Pik Kong, N.S.: Brightness preserving dynamic histogram equalization for image
contrast enhancement. IEEE Trans. Consum. Electron. 53(4), 1752–1758 (2007)
4. Sundaram, M., Ramar, K., Arumugam, N., Prabin, G.: Histogram-modified local contrast
enhancement for mammogram images. Int. J. Biomed. Eng. Technol. 9(1), 60–71 (2012)
5. Chen, S.D., Ramli, A.R.: Preserving brightness in histogram equalization based contrast
enhancement techniques. Digit. Signal Process. 14, 413–428 (2004)
6. Sun, C.C., Ruan, S.J., Shie, M.C., Pai, T.W.: Dynamic contrast enhancement based on histogram
specification. IEEE Trans. Consum. Electron. 51(4), 1300–1305 (2005)
7. Arici, T., Dikbas, S., Altunbasak, Y.: A histogram modification framework and its application
for image contrast enhancement. IEEE Trans. Image Process. 18(9), 1921–1935 (2009)
8. Huang, S.-C., Cheng, F.-C., Chiu, Y.-S.: Efficient contrast enhancement using adaptive gamma
correction with weighting distribution. IEEE Trans. Image Process. 22(3), 1032–1041 (2013)
9. Kaur, N., Singh, E.H.: Enhancement of medical images using histogram based hybrid technique.
Int. J. Adv. Eng. Manage. Sci. 2(9), 1425–1432 (2016)
10. Satapathy, L.M., Dalai, A., Satapathy, S., Jena, A.: Satellite image enhancement based on multi-
technology fusion. In: 2018 Second International Conference on Inventive Communication and
Computational Technologies (ICICCT), Coimbatore, pp. 1677–1680 (2018)
11. Atta, R., Ghanbari, M.: Low-contrast satellite images enhancement using discrete cosine
transform pyramid and singular value decomposition. IET Image Process. 7(5), 472–483 (2013)
12. Lidong, H., Wei, Z., Jun, W., Zebin, S.: Combination of contrast limited adaptive histogram
equalisation and discrete wavelet transform for image enhancement. IET Image Process 9(10),
908–915 (2015)
13. Yang, Y., Su, Z., Sun, L.: Medical image enhancement algorithm based on wavelet transform.
Electron. Lett. 46(2), 120–121 (2010)
14. Demirel, H., Anbarjafari, G., Jahromi, M.N.S.: Image equalization based on singular value
decomposition. In: Proceedings of 23rd International Symposium on Computer and Information
Sciences, pp. 1–5 (2008)
15. Atta, R., Abdel-Kader, R.F.: Brightness preserving based on singular value decomposition for
image contrast enhancement. Opt. Int. J. Light Electron Opt. 126(7–8), 799–803 (2015)
16. Demirel, H., Ozcinar, C., Anbarjafari, G.: Satellite image contrast enhancement using discrete
wavelet transform and singular value decomposition. IEEE Geosci. Remote Sens. Lett. 7(2),
333–337 (2010)
17. Kallel, F., Ben Hamida, A.: A new adaptive gamma correction based algorithm using DWT-SVD
for non-contrast CT image enhancement. IEEE Trans. NanoBiosci. 16(8), 666–675 (2017)
18. Bhandari, A.K., Kumar, A., Padhy, P.K.: Enhancement of low contrast satellite images using
discrete cosine transform and singular value decomposition. World Acad. Sci. Eng. Technol.
55, 35–41 (2011)
19. Dragomiretskiy, K., Zosso, D.: Variational mode decomposition. IEEE Trans. Signal Process.
62, 531–544 (2014)
20. Dragomiretskiy, K., Zosso, D.: Two-dimensional variational mode decomposition. In: Tai,
X.C., Bae, E., Chan, T.F., Lysaker, M. (eds.) Energy Minimization Methods in Computer
Vision and Pattern Recognition (EMMCVPR 2015). Lecture Notes in Computer Science, vol.
8932. Springer, Cham (2015)
21. Satapathy, L.M., Tripathy, R.K., Das, P.: A combination of variational mode decomposition
and histogram equalization for image enhancement. Natl. Acad. Sci. Lett. 42, 333 (2019)
Power Factor Corrected Cuk Converter
with PI and Fuzzy Logic Controller

Alok Kumar Mishra, Akshaya Kumar Patra, Ramachandra Agrawal,

Narayan Nahak, Amaresh Gantayet, Jnana Ranjan Swain,
and Samarjeet Satapathy

Abstract This paper presents a comparative analysis between the two controllers, PI
and fuzzy logic controller used for power factor correction using DC–DC CUK con-
verter. MATLAB/SIMULINK models of CUK converter are developed to improve
the power factor and output voltage regulation. Average current control technique is
used for input power factor correction, and to get a regulated voltage at the output PI
or fuzzy logic controller (FLC) is adopted. The system is tested at both steady states,
transient condition and its performance are then estimated and compared in terms of
various parameters like total harmonic distortion (THD), input power factor, output
voltage ripple for PI and FLC.

Keywords PID · FLC · CUK converter · Average current control · Power factor
correction circuits

A. K. Mishra (B) · A. K. Patra · R. Agrawal · A. Gantayet · J. R. Swain · S. Satapathy

EEE Department, Institute of Technical Education and Research, Siksha ‘O’ Anusandhan
(Deemed to be University), Bhubaneswar, Odisha 751030, India
e-mail: [email protected]
A. K. Patra
e-mail: [email protected]
R. Agrawal
e-mail: [email protected]
A. Gantayet
e-mail: [email protected]
J. R. Swain
e-mail: [email protected]
S. Satapathy
e-mail: [email protected]
N. Nahak
EE Department, Institute of Technical Education and Research, Siksha ‘O’ Anusandhan (Deemed
to be University), Bhubaneswar, Odisha 751030, India
e-mail: [email protected]

© Springer Nature Singapore Pte Ltd. 2020 1035

G. Pradhan et al. (eds.), Advances in Electrical Control and Signal Systems,
Lecture Notes in Electrical Engineering 665,
https://doi.org/10.1007/978-981-15-5262-5_80
1036 A. K. Mishra et al.

1 Introduction

According to the international standards and for the best power transfer and utiliza-
tion, power factor correction has become a necessity. To get a DC output voltage, we
use a rectifier and a parallel capacitor [1, 2] as shown in Fig. 1a. The input voltage
and current are shown in Fig. 1b. Figure 1b depicts a pulse-shaped input current with
much harmonics content causing a very poor power factor. The IEEE and IEC are
some international entities to define or standardize the permissible limits of the har-
monic content in line current such as IEEE 519 and IEC 61000-3-2 [3]. The design
of an AC–DC power converter to overcome these power quality issues like obtaining
UPF at the AC input mains and a close regulation of the DC output voltage has been
discussed in [4–9]. So, for small power uses in single phase supplies, DC–DC con-
verters are used which are switch-based and controlled accordingly to ensure HPF
at the mains side. The main motive is to emulate a resistive circuit when seen from
mains side and improve power factor. Theoretically, there exist three families of non-
isolated power factor correction topologies which are buck, boost and buck-boost
topology [1, 10–13]. The buck topology is generally used when low output voltage is
required but has high frequency commuted current at input. Due to commuted nature,
it exhibits a discontinuous nature and hence should be connected with high speed
recovery circuit. The major drawback faced by buck topology is the inclusion of a
filter of high frequency range within the diode bridge and the power source [14]. The
filtering inductor causes a soft variable current in the input end in the topology of a
boost converter. This topology provides high output voltage and leads to over-voltage
stress at the switches [5, 15]. The third family is buck-boost topology and finds its
applications in power factor correction circuits. It includes CUK and single-ended
primary inductance converter (SEPIC) converter [16, 17]. The output voltage polarity
makes these two converters different from each other. In case of CUK converter, the
position of free-wheeling diode and the inductor is reversed from that of the SEPIC
converter to obtain reversed output voltage polarity [18, 19]. In this paper, we have
used CUK type PFC circuit to improve the input side power factor of the diode bridge
rectifier, lower the THD factor with decreased ripple contained voltage at output [1,
20, 21]. The entire modelling and the design scheme of pf rectified AC–DC power
converters are carried out in the MATLAB/SIMULINK environment.

is
is ,V s
L
AC O
Vs A
Source Co
D

Diode Bridge
(a) (b)

Fig. 1 a AC–DC rectifier b source voltage and current waveform

Power Factor Corrected Cuk Converter … 1037

2 Circuit Configuration and Design Equation

To improve the power factor at the input and to get a regulated voltage at the output,
we have to interface CUK converter in between the bridge rectifier and the load as
shown in Fig. 2.
Like the conventional buck-boost type converter, CUK also provides a reverse
polarity of voltage at the output, but this CUK topology, for a low voltage DC-
bus, provides a low ripple current at its input, which is rare in case of conventional
converters of buck-boost type. With adequate coupling coefficient design in a CUK
converter, the same magnetic core can accommodate the output and input inductors.
The basic circuit diagram of DC–DC CUK converter is shown in Fig. 3. When switch
is ON, inductor L 1 and L 2 stores energy, inductor current increases linearly, and
output capacitor supplies power to the load, and diode is reverse biased as depicted
in Fig. 4.
When switch is ON applying KVL and KCL, the equation is given as follows:

di 1
L1 = vin (1)
dt
di 1
L2 = vC1 + vCo (2)
dt

is L1 C1
Io
L2

Vs SW D Co RL Vo

CUK Converter
Diode Bridge

Fig. 2 CUK converter topology for power factor correction

L1 C1
L2

Vin SW Co RL
D

Fig. 3 Basic circuit diagram of DC–DC CUK converter

1038 A. K. Mishra et al.

L1 C1
L2

Vin Co RL

Fig. 4 CUK converter when switch is ON

L1 C1
L2

Vin Co RL

Fig. 5 CUK converter when switch is OFF

dvC1
C1 = −i L 2 (3)
dt
dvCo vCo
Co = i L2 − (4)
dt R
When switch is OFF as depicted in Fig. 5 applying KVL and KCL, the equation
is given as follows:

di 1
L1 = vin − vC1 (5)
dt
di 2
L2 = −vCo (6)
dt
dvC1
C1 = i L1 (7)
dt
dvCo vcCo
Co = i L2 − (8)
dt R
Using the above equations, we can find the values of circuit components. Where
vs is the rms source voltage, d is the duty cycle, i L represents inductor current
ripple, vc , vo are the voltage ripple of the output voltage and capacitor voltage,
respectively. f s is the switching frequency. P0 is the output power, Vo is the output
Power Factor Corrected Cuk Converter … 1039

voltage, and I0 is the output current.

vs ∗ d
L1 = (9)
i L 1 ∗ f s
vs ∗ d
L2 = (10)
i L 2 ∗ f s
I0 ∗ (1 − d)
C1 = (11)
vc1 ∗ f s
vs ∗ d
Co = (12)
8 ∗ f s2 ∗ L 2 ∗ vo

3 Different Methods of Power Factor Correction

There are two objective of power factor correction: (1) To get a regulated voltage at
the output. (2) The wave shape of current at the input must be a sine wave. To achieve
the first objective, we have to use a feedback loop at the output. There are two methods
to achieve the second objective. First method is called “multiplier approach”, and
the second is called “voltage follower approach”. In multiplier approach, a feedback
loop of input current is used to govern the DC–DC power converter to function as an
input voltage programmed current sink as shown in Fig. 6. In this paper, multiplier
approach is used. Multiplier approach control is further sub-divided into four different
methods for generating the gate pulse for the CUK converter, such as
1. Hysteresis current control
2. Peak current control
3. Average current control
4. Borderline current control.
In this paper, average current control method is considered which allows a better
input current waveform, which is depicted in Fig. 6. In this technique, the sensed
inductor current is to be filtered with the help of a current error amplifier, and the
output of the same drives the PWM modulator. Hence, the error between input current
Ig and its reference gets minimized with the help of inner current loop. The reference
current is generated by a voltage error amplifier (PI or fuzzy logic controller). Due
to the PWM modulator, average current control method gives a constant switching
frequency, and because of current filtering commutation, noises get eliminated. This
method does not need any compensation ramp. Inductor current must have to be
sensed which is the demerits of this method.
1040 A. K. Mishra et al.

is Iactual Io
DC-DC
Converter

Vs Vin RL Vo

SW

Diode Bridge
1
K +
Gate
Sinusoidal -
Pulse
Reference
PWM
- Modulator
+

X Iref
Current error Amplifier
PI Controller
-
Multiplier +
Vo,ref

Voltage error Amplifier

PI or Fuzzy Logic Controller

Fig. 6 Multiplier approach with average current control technique

4 PID Controller for Output Voltage Regulation

The name itself Proportional-Integral-Derivative controller is a type of control loop

mechanism adopted in continuous modulated operation of control. A PID controller
repeatedly calculates the error in between a set point (SP) value and measured pro-
cess value (PV) and gets a correction accordingly in PID base. In real-time oper-
ation, it accordingly provides an accurate and respective change of correction to a
Power Factor Corrected Cuk Converter … 1041

control function. For example, furnace temperature control applies a derivative term
to effectively correct the error despite of a huge change.

5 Fuzzy Logic Controller for Output Voltage Regulation

The basic block diagram of the proposed fuzzy logic controller for output voltage
regulation of CUK converter is shown in Fig. 7. There are two separate meanings of
fuzzy logic. Generally, fuzzy logic is the further application of multi valued logic and
is also known as logic system. We can also say fuzzy logic is same as fuzzy sets theory
that relates to collection of objects with unsharp boundaries and the membership is
a point of degree. There is a specific object in which degree of membership in a
given set can vary between the range 0–1 in fuzzy set theory. Fuzzy logic is based
on sound quantitative and also deals with imprecise information and data. Error in
voltage of a power converter can be defined in linguistic variables like Negative
Big (NB), Negative Medium (NM), Negative Small (NS), Zero (ZE), Positive Small
(PS), Positive Medium (PM), Positive Big (PB), and each variable can be defined by
varying triangular membership function. Seven fuzzy levels were chosen and were
defined by fuzzy set library in which value of the error is e and change in error is de.
The larger the number of fuzzy levels, the higher the input resolution. We know that
a rule is n dimensional and n is the number of variable included in the rule. The sum
of rules is known as rule R. FIS editor edits the input and output variables, which are
e, de and output. After editing, we have to design the membership function for each
variable. The final step involves writing rules in rule editor using the rule given in
Table 1.

Vo - LPF
+
Rule Base
Integrator Imax,ref
Vo,ref
e(n) ce(n)

Fuzzification Rule Evaluator Defuzzification

Data Base

Fig. 7 Basic block diagram of FLC

1042 A. K. Mishra et al.

Table 1 Fuzzy rules for closed loop control of CUK converter

e\de NB NM NS ZE PS PM PB
NB NB NB NB NB NM NS ZE
NM NB NB NB NM NS ZE PS
NS NB NB NM NS ZE PS PM
ZE NB NM NS ZE PS PM PB
PS NM NS ZE PS PM PB PB
PM NS ZE PS PM PB PB PB
PB ZE PS PM PM PB PB PB

6 Results and Discussion

To investigate the performance of the proposed system, a simulink model of a CUK

converter used for power factor correction is developed as shown in Fig. 8. The
designed parameter used in simulation is given in Table 2. The system performance
is investigated in four steps. Step: 1 when CUK converter is not connected to the
system. Step: 2 when CUK converter is connected to the system with inner current
loop only and without the output voltage feedback loop. Step: 3 when CUK converter
is connected to the system with PI controller in the feedback loop. Step: 4 when CUK
converter is connected to the system with fuzzy logic controller in the feedback loop.
To convert AC–DC, we generally connect a bridge rectifier and filter capacitor across
the load as shown in Fig. 1, and simulation result of the same along with its harmonic
spectrum is shown in Fig. 9. To improve the performance of the input current and
output voltage, CUK converter is used as discussed in Fig. 2. Various waveforms at
steady state in open loop are shown in Fig. 10. From Fig. 10d, it is clear that when

Fig. 8 Simulink model of CUK converter

Power Factor Corrected Cuk Converter … 1043

Table 2 Designed parameter

Parameters Values
of CUK converter
Input voltage RMS value 120 V
Supply frequency 50 Hz
L 1, L 2 6 mH, 10 mH
C1 , Co 10 mF, 10 mF
Load power 1 kW
Reference output voltage 100 V
Switching frequency 40 kHz
Voltage and current ripple 0.05
Duty cycle 0.45

Fig. 9 Simulation results of the proposed system without CUK converter

load is increased at t = 1 s, load voltage is not a regulated one. To get a regulated

output voltage and improved input current wave PI or Fuzzy Logic Controller is used
in the output feedback loop.
The dynamic response of the system is shown in Figs. 11 and 12, respectively, for
PI and fuzzy logic controller in the outer loop. To test the validity of the PI and fuzzy
1044 A. K. Mishra et al.

Fig. 10 Simulation results of the proposed system in open loop

logic controller, a load change and reference voltage change are created at t = 1 s
and at t = 2 s, respectively. With this dynamic change, the regulated output voltage
is obtained which can be clearly seen from Figs. 11e and 12e, respectively, for PI
and fuzzy logic controller. Various performance parameter obtained in simulation is
given in Table 3 for comparison.

7 Conclusion

The design, modelling and simulation of CUK converter for power factor correction
purposes are considered in MATLAB/Simulink environment. The outcomes of the
simulation work provided low THD of the supply current with upgraded AC mains
pf and reduced output voltage ripple. Compared with PI and fuzzy logic controller
used in the output feedback loop, fuzzy logic controller gives better result in terms
of THD of supply current, input power factor and output voltage ripple which are
Power Factor Corrected Cuk Converter … 1045

Fig. 11 Simulation results of PI controlled CUK converter

given in Table 3. From Fig. 12e, it is clear that fuzzy logic controller gives a well-
regulated output voltage with reduced ripple when load disturbance and reference
change occur. The prototype of the proposed CUK converter can be developed that
would be attempted as a future work.
1046 A. K. Mishra et al.

Fig. 12 Simulation results of fuzzy logic controlled CUK converter

Table 3 Performance
Performance THD % Power factor Output voltage
parameters of the system
parameter ripple %
under different condition
Different
condition
Without CUK 191.36 0.4479 9.2
converter
Open loop CUK 4.70 0.9986 2.2
converter
PI controlled 3.13 0.9990 2.1
CUK converter
Fuzzy logic 2.97 0.9991 1.8
controlled
CUK converter
Power Factor Corrected Cuk Converter … 1047

References

1. Mishra, A.K., Pathak, M.K., Das, S.: Isolated converter topologies for power factor correction—
a comparison. In: IEEE International Conference on Energy, Automation and Signal, 28 Dec
2011, pp. 1–6
2. Kuiyuan, W.: The comparison and choice of several power factor correction methods. In: IEEE
Vehicle Power and Propulsion Conference, pp. 1–5. IEEE (Sept, 2006)
3. IEEE 519–2014: IEEE recommended practice and requirements for harmonic control in electric
power systems, pp. 1–29, 11 June 2014
4. Pressman, I.: Switching Power Supply Design. McGraw-Hill, New York (1991)
5. Patra, A.K., Rout, P.K.: Adaptive continuous-time model predictive controller for implantable
insulin delivery system in type I diabetic patient. Optimal. Control Appl. Methods 38, 184–204
(2017)
6. Patra, A.K., Rout, P.K.: Optimal H-infinity insulin injection control for blood glucose regulation
in IDDM patient using physiological model. Int. J. Autom. Control 8, 309–322 (2014)
7. Singh, B., Singh, B.N., Chandra, A., Al-Haddad, K., Pandey, A., Kothari, D.P.: A review of
single-phase improved power quality AC-DC converters. IEEE Trans. Ind. Electron. 50(5),
962–981 (2003)
8. Batarseh, I.: Power Electronics Circuits, 3rd ed. Wiley (2006)
9. Richard, R.: The fundamentals of power factor correction. Int. J. Elect. Eng. Edu. 31, 213–229
(1994)
10. Dah, D., Ki, S.K: Light-load efficiency improvement in buck-derived single-stage single-switch
PFC converters. IEEE Trans. Power Electron. 28(5), 2105–2110 (2013)
11. Patra, A.K., Mishra, A.K., Rout, P.K.: Backstepping Model predictive controller for blood
glucose regulation in type-I diabetes patient. IETE J. Res. https://doi.org/10.1080/03772063.
2018.1493404
12. Patra, A.K., Rout, P.K.: Adaptive sliding mode gaussian controller for artificial pancreas in
TIDM patient. J. Process Control 58, 23–27 (2017)
13. Liu, X., Xu, J., Chen, Z., et al.: Single-inductor dual-output buck–boost power factor correction
converter. IEEE Trans. Ind. Electron. 62(2), 943–952 (2015)
14. Kanaan, H.Y., Al-Haddad, K., Fnaiech, F.: Switching-function-based modeling and control
of a SEPIC power factor correction circuit operating in continuous and discontinuous current
modes. IEEE Int. Conf. Ind. Technol. 1(8), 431–437 (2004)
15. Umamaheswari, M.G., Uma, G.: Analysis and design of reduced order linear quadratic regulator
control for three phase power factor correction using Cuk rectifiers. Electr. Power Syst. Res.
1(96), 1–8 (2013)
16. Umamaheswari, M.G., Uma, G., Vijitha, S.R.: Comparison of hysteresis control and reduced
order linear quadratic regulator control for power factor correction using DC–DC Cuk
converters. J. Circuits Syst. Comput. 21(1), 1250002 (2012)
17. Umamaheswari, M.G., Uma, G., Isabella, L.A.: Analysis and design of digital predictive
controller for PFC Cuk converter. J. Comput. Electron. 13(1), 142–154 (2014)
18. Poorali, B., Adib, E.: Analysis of the integrated SEPIC-flyback converter as a single-stage
single-switch power-factor-correction LED driver. IEEE Trans. Ind. Electron. 63(6), 3562–
3569 (2016)
19. Melo, P.F., Gules, R., Ribeiro, E.F., et al.: A modified SEPIC converter for high-power-factor
rectifier and universal input voltage applications. IEEE Trans. Power Electron. 25(2), 310–321
(2010)
20. Sudhakarababu, C., Veerachary, M.: Zeta converter for power factor correction and voltage
regulation. In: 2004 IEEE Region 10 Conference TENCON 2004, vol. 500, pp. 61–64. IEEE,
24 Nov 2004
21. Patra, A.K., Rout, P.K.: Backstepping sliding mode Gaussian insulin injection control for blood
glucose regulation in TIDM patient. J. Dyn. Sys. Meas. Control 140(9), 091006-091006-15
(2018)
A Robust Competitive Optimization
Algorithm Based Energy Management
Control Strategy in a Battery
and Ultracapacitor Based Hybrid Energy
Storage System

Subhashree Choudhury, Nikhil Khandelwal, and Anshuman Satpathy

Abstract Energy Storage Systems (ESSs) plays an important role in microgrid oper-
ation in terms of power quality enhancement, regulation of voltage and frequency,
regularizing the intermittency of the Renewable Energy Sources (RESs) and main-
taining a balance between generation and demand. Among all ESSs, Battery Energy
Storage System (BESS) is found to be more promising. However, BESS alone cannot
fulfil the desire of robustness, active response, life cycle and potential of the system
due to its power delivering limitations. So, the integration of Hybrid Energy Storage
Systems (HESSs) is an emerging solution to the above issues. Basically, battery and
Ultracapacitor (UC) based energy storage systems (UCESS) have compatible per-
formances which makes them appealing in forming a HESS. As a result, the energy
supplying capability of battery and storage capacity of UC is improvised. This paper
attempts to bring out the advantages of the HESS as well as proposes a novel and
robust Competitive Optimization Algorithm (COA) tuned PID based control strat-
egy for improved power efficiency and for enhancing the life span of the battery.
Further, the system configuration considered for study with the proposed controller
is developed in the Matlab/Simulink environment and is led to variation in load. The
results are indicative of the fact that an efficient control strategy is developed using
the HESS. In addition to this, the THD calculation is done through FFT analysis to
justify the enhancement in system stability through the control technique proposed.

Keywords Hybrid energy storage system (HESS) · Battery · Ultracapacitor (UC) ·

Bidirectional buck/boost converter (B-BBC) · Competitive optimization algorithm
(COA) · Total harmonic distortion (THD)

S. Choudhury (B) · N. Khandelwal · A. Satpathy

Department of Electrical and Electronics Engineering, Institute of Technical Education and
Research, Siksha ‘O’ Anusandhan (Deemed to be University), Bhubaneswar, Odisha 751030,
India
e-mail: [email protected]

© Springer Nature Singapore Pte Ltd. 2020 1049

G. Pradhan et al. (eds.), Advances in Electrical Control and Signal Systems,
Lecture Notes in Electrical Engineering 665,
https://doi.org/10.1007/978-981-15-5262-5_81
1050 S. Choudhury et al.

1 Introduction

In today’s era, the major global problems of an instant rise in carbon-dioxide level
in the atmosphere, drastic change in surface temperature and climate, environmental
pollution, the greenhouse effect and high energy utilization must be quickly addressed
and resolved [1]. The population has started to realize the effect of these consequences
and are trying to fulfil the energy demand and reduce pollution. In recent years, the
solution is better described by using RESs [2]. Still, the management in terms of
storing and distribution of energy is difficult without interruption. Thus, ESSs have
been evolved in the recent past that can improve the management by diminishing
fluctuations, filtering the frequency and boosting the power factor of the system.
Battery, as an ESS outperforms as compared to other ESSs due to its stability and
high energy storage capacity. Batteries contain cells that transform the electrical
energy into chemical energy and the inverse of it. They provide a steady voltage
and an efficient lift power level. During variation in frequency, they can adjust the
frequency by absorbing or injecting power into the load [3]. However, due to their low
power density, the system gets slower and a large amount of batteries are required.
Also, the generated heat by the batteries increases their internal resistance and leads
in decreasing of its life cycle. Nickel Cadmium (Ni-Cd), Nickel-Metal Hydride (Ni-
MH), Lead-Acid, Lithium ion (Li-ion) are the most usable batteries as given in the
literature [4]. But Li-ion battery is more used because of its ability of charging and
discharging up to cent percent and having a low self-discharging capability.
Dissimilarly, the devices having high power density such as ultracapacitors are
also frequently used alone as an ESS as they reduce the lifetime of the system and
lack the assets provided by the batteries. Typically, an ESS must cover the demand of
having a high energy and power density. This could be brought about by hybridizing
two or more compatible ESSs. Accordingly, a HESS has been designed in this paper
combining a Battery and a UC. UCs are the electrochemical capacitors that store
energy differently unlike normal capacitors [5]. Rather than using a dielectric between
a pair of electrodes they use an electrolytic solution between two conductors. They
have a faster rate of charging and discharging, thus making it a viable solution for
most of the power system applications. Overall, a HESS can establish the system’s
durability, longer life cycle, lower maintenance, peak power and operation at variant
temperatures [6].
Under varying conditions of load, the management in power can only be performed
by using a controller. In the literature, many authors have proposed numerous con-
trollers such as fuzzy-based PID controller [7], evolutionary-based PID controller
[8], conventional PID controller [9], etc. Conventional PID controller has the advan-
tages of easy implementation and being simple in structure as compared to other
controllers. The controller sets the boundary parameters so that a better response
can be achieved. Nevertheless, the PID controller has a shortcoming that it cannot
respond dynamically to any occurrence of non-linearity in the system parameters.
Therefore, Optimization Techniques (OTs) are used to tune the parameters of the PID
controller dynamically without any physical change. Many nature-based OTs have
A Robust Competitive Optimization Algorithm … 1051

Fig. 1 Single line diagram of the system under study

been proposed by various authors in the literature such as; Bee Colony Optimization
(BCO) [10], Cuckoo Search (CS) [11], Flower Pollination Algorithm (FPA) [12],
Gravitational Search Algorithm (GSA) [13], Harmony Search (HS) [14], Weighted
Superposition Attraction Algorithm (WSA) [15], Competitive Optimization Algo-
rithm (CSO) [16]. However, the COA-based OT is best as compared to other OTs
because of competitiveness among all other OTs, better speed of converging towards
an optimal solution and faster execution [16].
The rest part of this manuscript has been elaborated in the following sections.
Detailed mathematical modelling of the system is described in Sect. 2. In Sect. 3,
the classical and the proposed control approach has been discussed. The simula-
tion model and the result analysis are demonstrated in Sect. 4. Ultimately, Sect. 5
addresses the conclusion extracted from the entire study (Fig. 1).

2 Mathematical Modelling of the System

2.1 Modelling of a Battery

An electrically connected simplified single RC model of a battery is demonstrated

in Fig. 2 and assembled using a dependent voltage source ‘VOC (z)’ where z is the

Fig. 2 A simplified model

of a battery [17]
1052 S. Choudhury et al.

abbreviation for State of Charge (SoC), a capacitor ‘C’ and resistances ‘R0 ’ and ‘R1 ’.
Here, ‘V1 ’ is the voltage drop across the parallel combination, ‘V0 ’ is the terminal
voltage of the battery and ‘I’ is the current flowing through the circuit [17].
Equation (1) is obtained by applying Kirchhoff’s Current Law (KCL) to the circuit:

1 1
V1 (t) = − V1 (t) + I (t) (1)
R1 C 1 C1

On applying Kirchhoff’s Voltage Law (KVL) to the circuit, we get Eq. (2):

V0 (t) = VOC (z(t)) − V1 (t) − R0 I (t) (2)

According to the SoC definition, which states the ratio of the residual capacity to
the normal capacity, the mathematical relation can be written as shown in Eq. (3):

i
n i I (τ )
z(t) = z(0) + (3)
Cn
0

where ‘τ ’ is the present time, ‘Cn ’ is the nominal capacity, and ‘ηi ’ is the Coulombic
efficiency approximately valued to be 1.

2.2 Modelling of an Ultracapacitor

A simple model of a UC is designed with an ideal capacitor ‘C’, equivalent series

resistance ‘R S ’ and parallel resistance ‘R p ’ as shown in Fig. 3. The resistance ‘R S ’
reflects the total internal series resistance of the UC and it causes energy loss during
charging and discharging period of the UC. Resistance, ‘R p ’ is also known as leakage
resistance and it depicts the total leakage current of the UC. In the circuit, ‘V0 ’ is the
UC output voltage, ‘VC ’ is the voltage across the capacitor, ‘i’ is the current flowing
through the UC, and ‘i 1 ’ and ‘i 2 ’ are the currents flowing through the capacitor and
the parallel resistance, respectively [18].
Output voltage across the UC can be written as shown in Eq. (4):

Fig. 3 Dynamic model of a

UC [18]
A Robust Competitive Optimization Algorithm … 1053

V0 = R S i + VC (4)

Current across the capacitor and the parallel resistance can be formulated as in
Eqs. (5) and (6):

dVC
i1 = C (5)
dt

i 2 = VC R p (6)

Total current flowing through the UC is given by Eq. (7):

dVC
i = i1 + i2 = C + VC R p . (7)
dt

2.3 Mathematical Modelling of a Bidirectional Buck/Boost

Converter (B-BBC)

A B-BBC having the features of low weight, small volume and high efficiency are
used in the HESS for the purpose of transferring energy in two directions. It can
operate as both buck and Boost converter in opposite directions depending upon
the working of two switches which are controlled by a PWM signal [19]. Both the
switches operate at the same time, alternatively. Figure 4 represents a B-BBC model
constructed using filtering capacitors ‘C1 ’ and ‘C2 ’, an inductor ‘L’, and switches
‘S1 ’ and ‘S2 ’. Here, ‘i L ’ is the inductor current and ‘U1 ’ and ‘U2 ’ are the output
voltages of the Boost and buck side, respectively [20].
During Buck mode, when ‘S1 ’ turns on and ‘S2 ’ turns off then the inductor current
increases and flows from left to right but, when ‘S1 ’ is turned off and ‘S2 ’ is turned on
then the inductor current decreases and flow of current in the ‘U2 ’ side is continued.
However, in Boost mode when ‘S2 ’ is switched on and ‘S1 ’ is switched off then the

Fig. 4 Bidirectional Buck/Boost converter [20]

1054 S. Choudhury et al.

inductor gets charged up but, when ‘S2 ’ turns off and ‘S1 ’ turns on then the energy
stored in the inductor gets discharged together with the source ‘U2 ’ and supplies
power to the ‘U1 ’ side.

3 Control Approach

3.1 Conventional PID Controller

The different modes of operation of the PID controller can be defined by its name,
proportional, integral and derivative. It is a linear type of controller which uses feed-
back to reduce the errors. Such three parameters that play a vital role in controlling
the system are ‘K p ’, ‘K i ’ and ‘K d ’. The proportional constant ‘K p ’, helps to expand
the rise time, the integral constant ‘K i ’, which decreases the steady-state error and
the derivative constant ‘K d ’, which helps in decreasing settling time and peak over-
shoot. All these parameters work together to carry out a perfect controlled signal.
However, the controller has the limitations of not being able to respond to non-
linearities in the system. So, for precise and proper tuning controlling the parameters
has to be executed offline which reduces the system efficacy. Equation (8) represents
the governing mathematical equation of the PID controller:

d
u(t) = K p e(t) + K i e(t) + K d e(t) (8)
dt

3.2 Competitive Optimization Algorithm (COA)

Concept of COA Living creatures such as birds, lions, cuckoos, cat, ants, etc. have
their own behaviour of living, different from others. Many such methods are there in
the literature that are inspired by these creatures. Each method can be used to solve an
optimization problem. Nevertheless, it is a long-time process to select an applicable
method which should be the best. So, this paper deals with such an OT which is based
on the competition of different creatures in nature to survive. As in real life, animals
consume other ones to survive and those who are being attacked may extinct. By
applying some logic and OT, a new Competitive Optimization Algorithm (COA) has
been implemented in this paper.

COA uses various OTs such as Particle Swarm Optimization Algorithm (PSO)
[21], Ant Colony Optimization Algorithm (ACO) [22], Cat Swarm Optimization
Algorithm (CSO) [23], Artificial Bee Colony Algorithm (ABC) [24] of the creatures
namely birds, ants, cats, and bees respectively as a prototype in groups. Imperialist
Competitive Algorithms (ICA) [25] is also used to compete between creatures of all
A Robust Competitive Optimization Algorithm … 1055

groups and at the end of each COA iteration, the weakest creature will be identified
and devoted to the other creatures of a group decided by the Roulette Wheel. This
operation occurs at each iteration end to strengthen the other creatures and increase
their population. After completion of few iterations, only a group will be left which
ensures the best results from one of the OTs. There is an issue with these OTs that
after reaching a stage near to the optimal solution may not show a better performance
which heads toward stagnation. All groups may extinct except one and due to being
four times in population than initially, its optimization method may lead to stagnancy
rather than having a better performance. Thus, to counteract this problem the final
population is again divided into all four groups randomly with a new position and
the operation of the algorithm is repeated. This must result in a better solution which
means after a few iterations, the problem space will become smaller and new strate-
gies can be adopted to find the best solution. Figure 5 demonstrates a flowchart that
describes the basic working of COA.

Application of COA for finding the optimal parameters of the PID Controller
For acquiring optimal operation of the battery and Ultracapacitor based hybrid energy
system, a COA-based optimization
 technique
 is proposed in this work for optimal tun-
ing of the ID parameters K p , K i and K d . The Integral Time Absolute Error (ITAE)
performance index is adopted here for error minimization of the objective functions.
The objective functions for Battery (J1 ) and Ultracapacitor (J2 ) are formulated as
follows:

t
J1 = (et1 (t))2 dt (9)
0

t
J2 = (et2 (t))2 dt (10)
0

where

et1 (t) = V PBAT − PINV

et2 (t) = SoC = SoCinitial − SoCfinal .

4 Simulation Models and Results

In this paper, an islanded hybrid energy storage system comprising of Battery and
Ultracapacitor has been designed using Matlab/Simulink environment. A single-
phase Bidirectional Buck/Boost Converter is assembled for operation in two oppo-
site directions being controlled by a PWM signal. The HESS is fed to a variable
1056 S. Choudhury et al.

Fig. 5 Flowchart of competitive optimization algorithm (COA)

load. For testing the efficiency of the proposed controller an unsymmetrical fault is
switched on from time t = 0.2 to t = 0.4 s. In addition to this, a detailed comparison
between the proposed technique and traditional technique has been made by con-
sidering various system parameters such as Power, Voltage and SoC of the Battery
and Ultracapacitor. The characteristics of the Voltage and Power of the load being
subjected to an unsymmetrical fault has also been demonstrated taking Battery with
A Robust Competitive Optimization Algorithm … 1057

PID, Battery and Ultracapacitor with PID and proposed COA tuned PID with Battery
and Ultracapacitor into account (Fig. 6).
Figures 7, 8 and 9 represents the transient performance of conventional PID and
COA tuned PID for Power, Voltage, and SoC of Battery respectively. It can be
observed that the COA tuned PID controller surpasses the classical PID controller
with reference to enhance stability, faster response, rise time, peak time, and overall
system operation.
Figures 10, 11 and 12 shows the characteristics of Power, Voltage, and SoC of
Ultracapacitor with conventional PID and COA tuned PID controller. From the fig-
ures obtained it can be concluded that the COA tuned PID responds more dynamically
and quickly as compared to conventional PID. Also, the COA tuned PID restores
system stability more efficiently with better control of system dynamics.
Figures 13 and 14 illustrate the Load Voltage and Load Power of the hybrid energy
storage system when subjected to an unsymmetrical fault for time t = 0.2 s to t =

Fig. 6 Matlab/Simulink model of islanded battery and ultracapacitor based hybrid energy storage
system

Fig. 7 Battery power

1058 S. Choudhury et al.

Fig. 8 Battery voltage

Fig. 9 Battery SoC

0.4 s. It can be observed that the system responds more quickly and has a better
dynamic response and enhanced transient stability in case of Battery and Ultraca-
pacitor with proposed COA tuned PID than that of Battery and Ultracapacitor with
conventional PID and Battery with only conventional PID. So, the results obtained
clearly indicate that the synchronized use of Battery and Ultracapacitor with proposed
COA tuned PID in hybrid energy storage system subjected to an unsymmetrical fault
yields better result in terms of rise time, peak time, peak overshoot, settling time and
faster response as compared to Battery and Ultracapacitor with conventional PID
and Battery with conventional PID controller.
Figures 15, 16 and 17 justifies the system stability by calculating the THD values of
Load Voltage for various cases such as the proposed COA tuned PID with Battery and
Ultracapacitor, conventional PID with Battery and Ultracapacitor and conventional
A Robust Competitive Optimization Algorithm … 1059

Fig. 10 Ultracapacitor
power

Fig. 11 Ultracapacitor
voltage

PID with only Battery. It can be seen that the THD values of Load Voltage for
COA tuned PID with Battery and Ultracapacitor is 0.67%, for conventional PID
with Battery and Ultracapacitor is 5.35 and 8.97% for Battery with conventional
PID. Hence, the proposed COA tuned PID is best in terms of ensuring better voltage
stability by decreasing the harmonics distortions.
1060 S. Choudhury et al.

Fig. 12 Ultracapacitor SoC

Fig. 13 Load voltage

Fig. 14 Load power

A Robust Competitive Optimization Algorithm … 1061

Fig. 15 THD of load

voltage for Battery-
Ultracapacitor-COA-PID

Fig. 16 THD of load voltage

for ultracapacitor-PID

Fig. 17 THD of voltage for

Battery-PID
1062 S. Choudhury et al.

Fig. 18 THD of load power

for Battery-Ultracapacitor-
COA-PID

Fig. 19 THD of load power

for
Battery-Ultracapacitor-PID

Figures 18, 19 and 20 reveals the THD values of Load Voltage for the proposed
COA tuned PID with Battery and Ultracapacitor, conventional PID with Battery and
Ultracapacitor and conventional PID with Battery. The values found out to be 0.41%,
2.51% and 7.15% respectively. This shows the proposed COA tuned PID technique
outperforms the conventional PID based Battery and Ultracapacitor and conventional
PID based Battery in reducing the THD value to a great extent.
Table 1 gives a tabular comparison of the THD values obtained for Load Voltage
and Power with the various
 case of analysis.
 Table 2 enumerates a comparative study
for the PID parameters K p , K i , K d for the proposed COA tuned PID controller
and classical PID controller.
A Robust Competitive Optimization Algorithm … 1063

Fig. 20 THD of load power

for Battery-PID

Table 1 Comparison of THD values for load voltage and load current between conventional battery
based PID, battery and ultracapacitor based PID and proposed battery and ultracapacitor based COA
tuned PID
ESSs with associated controllers Load parameters
THD of load voltage (in %) THD of load power (in %)
Battery-Ultracapacitor-COA-PID 0.69 0.41
Battery-Ultracapacitor-PID 5.35 2.51
Battery-PID 8.97 7.15

5 Conclusion

This paper investigates the optimization of the energy management for Battery and
Ultracapacitor based HESS for an autonomous microgrid system. For enhancing
the Battery life span and efficient operation of the HESS, a novel COA-based opti-
mization technique has been proposed which dynamically tunes the PID parameters,
thus bringing about the enhancement of system dynamic response and stability. To
examine the effectiveness of the technique proposed, the system has been tested by
inclusion of an unsymmetrical fault. Lastly, the stability analysis has been carried
out through THD calculation to verify the robustness of the proposed technique.
 1064

Table 2 Comparison of PID parameters between conventional battery based PID, Battery and Ultracapacitor based PID and proposed Battery and Ultracapacitor
based COA tuned PID
Load parameters Controllers
Battery-Ultracapacitor-COA-PID Battery-Ultracapacitor-PID Battery-PID
Control parameters
Kp Ki Kd Kp Ki Kd Kp Ki Kd
V 0.25 0.37 0.58 1.34 2.11 1.3 4.5 4.2 5.1
P 0.13 0.29 0.38 1.16 2.09 2.10 4.3 4.3 4.56
S. Choudhury et al.
A Robust Competitive Optimization Algorithm … 1065

Appendix

Values of parameters used in the system model

Parameters Values
Battery Battery type-Lithium ion, Battery nominal voltage V 0 = 48 V,
Battery capacity = 1000 Ah, Battery internal resistance R0 =
0.005 
Ultracapacitor Ultracapacitor nominal voltage V 0 = 35 V, Ultracapacitor internal
resistance = 0.015 , Rs = 0.025 , Rp = 0.0145 ,
Ultracapacitor rated capacitance C = 50 F, Number of series
capacitors = 4, Number of parallel capacitors = 1
Bidirectional Buck/Boost L = 1 mH, C 1 = C 2 = 10 µF, F s = 20 kHz, U1 = 48, U2 = 500
Load Nominal frequency—50 Hz, Active power—3000 W, Reactive
power—1600
Line inductance Line inductance H/km—0.8737e−3 for R, Y and B (Normal
operation)
Line inductance H/km—0.5027e−3 during fault for Y and B.
SoCinitial 40
SoCfinal 100

References

1. Bocklisch, T.: Hybrid energy storage systems for renewable energy applications. Energy
Procedia 73, 103–111 (2015)
2. Guo, Y., Zhao, Z., Huang, L.: SoC estimation of lithium battery based on AEKF algorithm.
Energy Procedia 105, 4146–4152 (2017)
3. Oudalov, A., Chartouni, D., Ohler, C.: Optimizing a battery energy storage system for primary
frequency control. IEEE Trans. Power Syst. 22(3), 1259–1266 (2007)
4. Divya, K.C., Østergaard, J.: Battery energy storage technology for power systems—an
overview. Electr. Power Syst. Res. 79(4) 511–520 (2009)
5. Cao, J., Emadi, A.: A new battery/ultracapacitor hybrid energy storage system for electric,
hybrid, and plug-in hybrid electric vehicles. IEEE Trans. Power Electron. 27(1), 122–132
(2011)
6. Khaligh, A., Li, Z.: Battery, ultracapacitor, fuel cell, and hybrid energy storage systems for
electric, hybrid electric, fuel cell, and plug-in hybrid electric vehicles: state of the art. IEEE
Trans. Veh. Technol. 59(6), 2806–2814 (2010)
7. Arcos-Aviles, D., et al.: Fuzzy logic-based energy management system design for residential
grid-connected microgrids. IEEE Trans. Smart Grid 9(2), 530–543 (2016)
8. Dogruer, T., Tan, N.: Design of PI controller using optimization method in fractional order
control systems. IFAC-Papers OnLine 51(4), 841–846 (2018)
9. Merai, M., et al.: An adaptive PI controller design for DC-link voltage control of single-phase
grid-connected converters. IEEE Trans. Ind. Electron. 66(8), 6241-6249 (2018)
1066 S. Choudhury et al.

10. Chiang, H.-S., et al.: A novel artificial bee colony optimization algorithm with SVM for bio-
inspired software-defined networking. Int. J. Parallel Prog. 1–19 (2018)
11. Mulani, K., et al.: Performance analysis and feasibility study of ant colony optimization, particle
swarm optimization and cuckoo search algorithms for inverse heat transfer problems. Int. J.
Heat Mass Transfer 89, 359–378 (2015)
12. Kayabekir, A.E., et al.: A comprehensive review of the flower pollination algorithm for solving
engineering problems. In: Nature-Inspired Algorithms and Applied Optimization, 171–188.
Springer, Cham (2018)
13. Mondal, S., Bhattacharya, A., nee Dey, S.H.: Multi-objective economic emission load dispatch
solution using gravitational search algorithm and considering wind power penetration. Int. J.
Electr. Power Energy Syst. 44(1), 282–292 (2013)
14. Sheng, W., et al. Reactive power coordinated optimisation method with renewable distributed
generation based on improved harmony search. IET Gener. Transm. Distrib. 10(13), 3152–3162
(2016)
15. Baykasoğlu, A., Akpinar, Ş.: Weighted superposition attraction (WSA): a swarm intelligence
algorithm for optimization problems—part 1: unconstrained optimization. Appl. Soft Comput.
56, 520–540 (2017)
16. Sharafi, Y., Khanesar, M.A., Teshnehlab, M.: COOA: competitive optimization algorithm.
Swarm Evol. Comput. 30, 39–63 (2016)
17. Xu, Jun, Li, Shiying, Cao, Binggang: A novel current disturbance estimation method for battery
management systems in electric vehicle. Energy Procedia 105, 2837–2842 (2017)
18. Jun, R., Kai, W., Liwei, L.: Characteristics analysis of ultracapacitor-battery hybrid energy
storage system. In: 2017 Chinese Automation Congress (CAC). IEEE (2017)
19. Inoue, S., Akagi, H.: A bidirectional DC–DC converter for an energy storage system with
galvanic isolation. IEEE Trans. Power Electron. 22(6), 2299–2306 (2007)
20. Wang, C., et al.: Efficiency analysis of a bidirectional DC/DC converter in a hybrid energy
storage system for plug-in hybrid electric vehicles. Appl. Energy 183, 612–622 (2016)
21. Shi, Y.: Particle swarm optimization: developments, applications and resources. In: Proceedings
of the 2001 Congress on Evolutionary Computation (IEEE Cat. No. 01TH8546), vol. 1. IEEE
(2001)
22. Dorigo, M., Stützle, T.: The ant colony optimization metaheuristic: algorithms, applications,
and advances. In: Handbook of Metaheuristics, 250–285. Springer, Boston (2003)
23. Chu, S.-C., Tsai, P.-W., Pan, J.-S.: Cat swarm optimization. In: Pacific Rim International
Conference on Artificial Intelligence. Springer, Berlin (2006)
24. Karaboga, D., Akay, B.: A comparative study of artificial bee colony algorithm. Appl. Math.
Comput. 214(1), 108–132 (2009)
25. Atashpaz-Gargari, E., Lucas, C.: Imperialist competitive algorithm: an algorithm for opti-
mization inspired by imperialistic competition. In: 2007 IEEE Congress on Evolutionary
Computation. IEEE (2007)