Power Systems

Farhad Shahnia
Ali Arefi
Gerard Ledwich Editors

Electric
Distribution
Network
Planning
Power Systems
More information about this series at http://www.springer.com/series/4622
Farhad Shahnia Ali Arefi
•

Gerard Ledwich
Editors

Electric Distribution
Network Planning

123
Editors
Farhad Shahnia Gerard Ledwich
School of Engineering and Information Faculty of Science and Engineering
Technology Queensland University of Technology
Murdoch University Brisbane, QLD
Perth, WA Australia
Australia

Ali Arefi
School of Engineering and Information
Technology
Murdoch University
Perth, WA
Australia

ISSN 1612-1287 ISSN 1860-4676 (electronic)

Power Systems
ISBN 978-981-10-7055-6 ISBN 978-981-10-7056-3 (eBook)
https://doi.org/10.1007/978-981-10-7056-3
Library of Congress Control Number: 2018935954

© Springer Nature Singapore Pte Ltd. 2018

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, express 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.

Printed on acid-free paper

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

Electric distribution networks are critical parts of power delivery systems. In recent
years, many new technologies and distributed energy resources have been inte-
grated into these networks. To provide electricity at the possible lowest cost and at
required quality, long-term planning is essential for these networks. In distribution
planning, optimal location and size of necessary upgrades are determined to satisfy
the demand and the technical requirements of the loads and to tackle uncertainties
associated with load and distributed energy resources. To this aim, an optimization
algorithm is utilized to find the optimal net present cost of augmentation over the
planning period. The distribution network is usually formulated as a mixed-integer
nonlinear programming problem, which is solved using various approaches
including mathematic and heuristic-based algorithms.
Over the last decades, several researches have been carried out around the world
on electric distribution planning, whose results are available as journal articles,
conference papers or technical reports. However, to the best of the editors’
knowledge, no single book has covered the different aspects of distribution net-
works’ planning so far. The interested readers had to search among several hun-
dreds of papers on this topic through various databases in order to build up their
knowledge on the subject. This book is the first one entirely focused on the dis-
tribution networks planning and is an effort to provide a research-oriented and a
coherent book on the subject for postgraduate students and researchers.
This book is benefited from the inputs and comments of a large number of
researchers and experts from the academia and industry. It contains 13 chapters.
The breakdown of the chapters is as follows:
• Chapter 1 reviews the multi-stage expansion planning problem of distribution
networks where investments in the distribution network and distributed gener-
ations are jointly considered;
• Chapter 2 presents static and dynamic models for the planning of distribution
networks;
• Chapter 3 discusses the mathematical formulations of unbalance networks, for
operation optimization analysis to support decision-making processes;

v
vi Preface

• Chapter 4 presents an integrated distributed generation and primary–secondary

expansion planning in the presence of wholesale and retail markets;
• Chapter 5 discusses a new planning tool based on the concept of a multi-agent
system;
• Chapter 6 presents an efficient method for sizing and siting distributed gener-
ations in distribution networks;
• Chapter 7 describes probabilistic and possibilistic-based planning methodolo-
gies of battery energy storage systems in distribution networks;
• Chapter 8 introduces an optimally distributed generation placement problem
towards power and energy loss minimization;
• Chapter 9 presents a hybrid methodology based on a local search algorithm and
a genetic algorithm, for the multi-objective and multi-stage distribution
expansion planning problem;
• Chapter 10 introduces simultaneous optimization concept of distribution net-
work reconfiguration and distributed generation sizing;
• Chapter 11 studies the implementation of optimal incentive plans for plug-in
electric vehicle aggregator to participate in the energy market;
• Chapter 12 presents a survey of optimization techniques used to find the optimal
sizes and locations of compensators; and
• Chapter 13 discusses a methodology for the allocation of automatic reclosers
within electric distribution networks.
As the editors of the book, we would like to thank all the contributors for their
support and hard work. We also would like to thank the reviewers who provided
valuable comments for improving the quality of the book. Also, we are grateful to
the publisher Springer Nature for agreeing to publish this book. Last but not least,
we would like to thank our families—Farhad thanks his parents (Nahideh and Ali)
and his spouse (Negar), Ali thanks his wife and son (Behnaz and Amin), and Gerard
thanks his family for their continuous encouragement and support.

Perth, Australia Farhad Shahnia

Perth, Australia Ali Arefi
Brisbane, Australia Gerard Ledwich
January 2018
Reviewers

Aggelos S. Bouhouras, Aristotle University of Thessaloniki, Greece; University of

Applied Sciences, Greece
Ali Ahmadian, K. N. Toosi University of Technology, Iran
Ali Elkamel, University of Waterloo, Canada
Alexandre M. F. Dias, University of Lisbon and INESC-ID, Portugal
Almoataz Y. Abdelaziz, Ain Shams University, Egypt
Alireza Heidari, University of New South Wales, Australia
Carlos F. M. Almeida, University of Sao Paulo, Brazil
Carlos F. Sabillón, São Paulo State University, Brazil
David Pozo, Skolkovo Institute of Science and Technology, Russia
Dimitris P. Labridis, Aristotle University of Thessaloniki, Greece
Gabriel Quiroga, University of Sao Paulo, Brazil
Gregorio Muñoz-Delgado, Universidad de Castilla-La Mancha, Spain
Hazlie Mokhlis, University of Malaya, Malaysia
Henrique Kagan, University of Sao Paulo, Brazil
Jan Kays, Amprion GmbH, Dortmund, Germany
Javier Contreras, Universidad de Castilla-La Mancha, Spain
John F. Franco, São Paulo State University, Brazil
José M. Arroyo, Universidad de Castilla-La Mancha, Spain
Julio López, University of Cuenca, Cuenca, Ecuador
Karar Mahmoud, Aswan University, Egypt
Kazem Zare, University of Tabriz, Iran
Mehdi Rahmani-Andebili, Sharif University of Technology, Iran
Mahdi Sedghi, K. N. Toosi University of Technology, Iran
Mahmood Reza Haghifam, Tarbiat Modares University, Iran
Mahmud Fotuhi Firuzabad, Sharif University of Technology, Iran
Marcos J. Rider, University of Campinas, Brazil
Masoud Aliakbar Golkar, K. N. Toosi University of Technology, Iran
Mehrdad Setayesh Nazar, Shahid Beheshti University, Iran
Michael Fowler, University of Waterloo, Canada
Moein Moeini-Aghtaie, Sharif University of Technology, Iran

vii
viii Reviewers

Mohamed Ebeed, Aswan University, Egypt

Nelson Kagan, University of Sao Paulo, Brazil
Paschalis A. Gkaidatzis, Aristotle University of Thessaloniki, Greece
Pedro M. S. Carvalho, University of Lisbon and INESC-ID, Portugal
Rubén Romero, São Paulo State University, Brazil
Salah Kamel, Aswan University, Egypt
Shady H. E. Abdel Aleem, Higher Institute of Engineering, Cairo, Egypt
Wardiah Mohd Dahalan, Universiti Kuala Lumpur, Malaysia
Yorino Naoto, Hiroshima University, Japan
Contents

1 Distribution System Expansion Planning . . . . . . . . . . . . . . . . . . . . 1

Gregorio Muñoz-Delgado, Javier Contreras and José M. Arroyo
2 Static and Dynamic Convex Distribution Network
Expansion Planning . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
Julio López, David Pozo and Javier Contreras
3 Mathematical Optimization of Unbalanced Networks
with Smart Grid Devices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
Carlos F. Sabillón, John F. Franco, Marcos J. Rider
and Rubén Romero
4 Multi-stage Primary-Secondary Planning Considering
Wholesale-Retail Markets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115
Mehrdad Setayesh Nazar, Alireza Heidari
and Mahmood Reza Haghifam
5 Multi-agent Based Planning Considering the Behavior
of Individual End-Users . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 143
Jan Kays
6 Optimal Siting and Sizing of Distributed Generations . . . . . . . . . . . 167
Karar Mahmoud and Yorino Naoto
7 Battery Energy Storage Planning . . . . . . . . . . . . . . . . . . . . . . . . . . 185
Mahdi Sedghi, Ali Ahmadian, Ali Elkamel, Masoud Aliakbar Golkar
and Michael Fowler
8 Optimal Distributed Generation Placement Problem
for Power and Energy Loss Minimization . . . . . . . . . . . . . . . . . . . . 215
Aggelos S. Bouhouras, Paschalis A. Gkaidatzis
and Dimitris P. Labridis

ix
x Contents

9 Optimal Planning of Grid Reinforcement with Demand

Response Control . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 253
Alexandre M. F. Dias and Pedro M. S. Carvalho
10 Simultaneous Network Reconfiguration and Sizing
of Distributed Generation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 279
Wardiah Mohd Dahalan and Hazlie Mokhlis
11 Optimal Incentive Plans for Plug-in Electric Vehicles . . . . . . . . . . . 299
Mehdi Rahmani-Andebili, Mahmud Fotuhi Firuzabad
and Moein Moeini-Aghtaie
12 Optimal Allocation of Compensators . . . . . . . . . . . . . . . . . . . . . . . 321
Mohamed Ebeed, Salah Kamel, Shady H. E. Abdel Aleem
and Almoataz Y. Abdelaziz
13 Optimal Allocation of Automatic Reclosers . . . . . . . . . . . . . . . . . . . 355
Carlos Frederico Meschini Almeida, Gabriel Albieri Quiroga,
Henrique Kagan and Nelson Kagan
About the Editors

Farhad Shahnia received his Ph.D. in Electrical Engineering from Queensland

University of Technology, Brisbane, Australia, in 2011. Currently, he is a Senior
Lecturer of Power Engineering at Murdoch University, Perth, Australia. His pro-
fessional experience includes three years at Research Office-Eastern Azarbayjan
Electric Power Distribution Company, Tabriz, Iran. Prior to joining Curtin
University, he was a Research Fellow at the Queensland University of Technology,
Brisbane, Australia, and a Lecturer at Curtin University, Perth, Australia. He has
authored 1 book, 9 book chapters and over 120 research articles as well as editing 4
books.

Ali Arefi received his Ph.D. in Electrical Engineering in 2011. He is currently a

Senior Lecturer of Power Engineering at Murdoch University, Perth, Australia.
Prior to that, he was a Lecturer and Research Fellow at the Queensland University
of Technology. He also has six years’ experience with electric distribution industry
and has been the supervisor for five industry-funded research projects. His research
interests are in the areas of electric delivery planning, state estimation, power
quality and energy efficiency. He has published more than 80 research articles.

Gerard Ledwich received his Ph.D. in Electrical Engineering from University of

Newcastle, Newcastle, Australia, in 1976. He has been the Chair Professor in Power
Engineering at the Queensland University of Technology, Brisbane, Australia, since
1998. Previously, he was with the University of Queensland, Brisbane, Australia,
from 1976 to 1994. His research interests are in the areas of power system operation
and control. He is the supervisor of more than 35 Ph.D. graduates and has published
1 book, 3 chapters and over 350 research articles.

xi
Abbreviations

3P3W Three-phase three-wire

3P4W Three-phase four-wire
ABC Artificial bee colony
AI Artificial intelligence
ALO Ant lion optimization
AR Automatic recloser
BES Battery energy storage
BFOA Bacterial foraging optimization algorithm
CAMG Customers’ active microgrid
CDF Cumulative distribution function
CDMG Customer’s dispatchable microgrid
CF Capacity factor
CNDMG Customer’s non-dispatchable microgrid
CS Cuckoo search
DA Distribution automation
D-FACTS Distributed flexible AC transmission system
DFIG Doubly fed induction generator
DG Distributed generation
DisCo Distribution company
DL Dispatchable load
DNUDG Dispatchable non-utility distributed generation
DOD Depth of discharge
DR Demand response
DSM Demand side management
DSO Distribution system operator
DSTATCOM Distributed static compensator
D-SVC Distribution static VAR compensator
DVS Decision variable set
EDNEP Distribution network expansion planning
ELF Exhaustive load flow

xiii
xiv Abbreviations

EM Energy management
ENS Energy not supplied
EP Evolutionary programming
EPSO Evolutionary particle swarm optimization
ESS Energy storage system
EV Electric vehicle
EVCC Electric vehicle charging coordination
FCB Fixed capacitor bank
FLF Fuzzy load flow
G2V Grid-to-vehicle
GA Genetic algorithm
GAP Genetic algorithm population
GEV Generalized extreme value
GIS Georeferenced information system
GOA Grasshopper optimization algorithm
GPSO Global particle swarm optimization
GRASP Greedy randomized adaptive search procedure
GS Gaussian search
GUPFC Generalized unified power flow controller
GWO Grey wolf optimization
HIC Hours of interrupted customer
HP Hydro-plant
HS Harmony search
HV High voltage
IA Improved analytical
IC Interrupted customers
ICT Information and communications technology
IDGNEP Integrated distributed generation and primary–secondary network
expansion planning
IG Induction generator
IGBT Insulated gate bipolar transistor
IPFC Interline power flow controller
LP Linear programming
LPSO Local particle swarm optimization
LS Load smoothing
LSF Loss sensitivity factor
LV Low voltage
MAS Multi-agent system
MCS Monte Carlo simulation
MGA Modified genetic algorithm
MI Merit index
MICQP Mixed-integer conic quadratic programming
MILP Mixed-integer linear programming
MINLC Mixed-integer-nonlinear constrained
MINLP Mixed-integer nonlinear programming
Abbreviations xv

MPSO Modified particle swarm optimization

MU Monetary unit
MV Medium voltage
NC Normally closed
NC-AR Normally closed automatic recloser
NDL Non-dispatchable load
NDNUDG Non-dispatchable non-utility distributed generation
NLP Nonlinear programming
NO Normally opened
NO-AR Normally opened automatic recloser
NR Network reconfiguration
NUDG Non-utility distributed generation
NURMP Non-utility retail market participants
O&M Operation and maintenance
ODGP Optimal distributed generation placement
OESSP Optimal energy storage system placement
OLTC On-load tap changer
OMS Outage management system
OPF Optimal power flow
ORESP Optimal renewable energy sources placement
PCC Point of coupling connection
PDF Probability density function
PEM Point estimate method
PEV Plug-in electric vehicle
PLF Probabilistic/possibilistic load flow
POPF Probabilistic/possibilistic optimal power flow
PRF Price factor
PSO Particle swarm optimization
PV Photovoltaic
PWM Pulse-width modulation
RES Renewable energy source
RPF Reverse power flow
SABC Satisfied artificial bee colony
SCA Sine cosine algorithm
SCB Switched capacitor bank
SiG Synchronous generator
SoC State of charge
SSSC Static synchronous series compensator
SVC Static VAR compensator
TCSC Thyristor controlled series capacitor
UDG Utility distributed generation
UPQC Unified power quality conditioner
UPS Uninterruptible power supply
UPSO Unified particle swarm optimization
V2G Vehicle-to-grid
xvi Abbreviations

VR Voltage regulator
VSC Voltage source converter
VSS Value of the stochastic solution
VVC Volt-VAR control
WRI Weighted reliability index
WT Wind turbine
Chapter 1
Distribution System Expansion Planning

Gregorio Muñoz-Delgado, Javier Contreras and José M. Arroyo

Abstract The widespread growth of distributed generation (DG), mainly due to its
numerous operational and planning benefits and to the penetration of renewable
energy, inevitably requires the inclusion of this kind of generation in distribution
planning models. This chapter addresses the multistage expansion planning prob-
lem of a distribution system where investments in the distribution network and in
DG are jointly considered. The optimal expansion plan identifies the best alterna-
tive, location, and installation time for the candidate assets. The incorporation of
DG in distribution system expansion planning drastically increases the complexity
of the optimization process. In order to shed light on the modeling difficulties
associated with the co-optimized planning problem, a deterministic model is pre-
sented first. The model is driven by the minimization of the net present value of the
total cost including the costs related to investment, maintenance, production, losses,
and unserved energy. As a relevant feature, radiality conditions are specifically
tailored to accommodate the presence of DG in order to avoid the islanding of
distributed generators and the issues associated with transfer nodes. Since a large
portion of DG relies on non-dispatchable renewable-based technologies, the
uncertainty associated with the high variability of the corresponding energy sources
needs to be properly characterized in the planning models. Based on the previous
deterministic model, uncertainty is incorporated using a stochastic programming
framework. Within such a context, the uncertainty featured by renewable-based
generation and demand is characterized through a set of scenarios that explicitly

G. Muñoz-Delgado (&)
J. Contreras
J. M. Arroyo

E.T.S. de Ingenieros Industriales,
Universidad de Castilla-La Mancha, Ciudad Real, Spain
e-mail: [email protected]
J. Contreras
e-mail: [email protected]
J. M. Arroyo
e-mail: [email protected]

© Springer Nature Singapore Pte Ltd. 2018 1

F. Shahnia et al. (eds.), Electric Distribution Network Planning,
Power Systems, https://doi.org/10.1007/978-981-10-7056-3_1
2 G. Muñoz-Delgado et al.

capture the correlation between uncertainty sources. The resulting stochastic pro-
gram is driven by the minimization of the total expected cost. Both deterministic
and stochastic optimization problems are formulated as mixed-integer linear pro-
grams for which finite convergence to optimality is guaranteed and efficient
off-the-shelf software is available. Numerical results illustrate the effective perfor-
mance of the approaches presented in this chapter.

Keywords Distributed generation Distribution system planning



Multistage Network expansion Stochastic programming Uncertainty

1.1 Introduction

Power systems are used to produce, transport, and distribute electric power from
generation centers to consumers. A power system usually consists of generation
units, transmission and subtransmission networks, distribution networks, con-
sumption centers, system protection devices, and control equipment [1].
Distribution networks are an important part of the electric system since they supply
energy from distribution substations to end users. The main components of the
distribution network are the substations and the feeder branches. Distribution
substations are fed by one or several subtransmission networks, although sometimes
they can be directly connected to the transmission network. Distribution substations
reduce the voltage level from high voltage to medium voltage ranging between 4.16
and 34.5 kV through the use of transformers. Moreover, from distribution substa-
tions energy is injected into the distribution grid by means of one or several primary
feeders [2].
Regardless of their topologies, which may be either meshed or radial, most
distribution networks are operated in a radial way since it is the cheapest and
simplest method from the viewpoints of planning, design, and system protection.
Traditionally, these networks have been designed with wide operating ranges,
which allows them to be passively operated, thereby resulting in more economical
management. However, investment in distribution networks is several times more
costly than investment in transmission grids [1], which reveals the economic
importance of distribution system planning. Moreover, a viable investment plan
must not only be economically driven, but it must also satisfy several criteria and
guidelines related to design, components, layout, or performance [3].
From a centralized standpoint, distribution companies are responsible for the
operation and planning of distribution networks so that the growing demand is
continuously satisfied with quality standards and in a secure fashion. Therefore,
planning models are used to obtain an optimal investment plan at minimum cost
while meeting the security and quality requirements. Traditionally, these planning
models determine the optimal expansion decisions related to the reinforcement and
installation of branches, substations, and transformers [3, 4]. However, the wide-
spread growth of distributed generation (DG), mainly due to its numerous
1 Distribution System Expansion Planning 3

operational and planning benefits [5] and to the impetus of renewable energy,
inevitably requires the inclusion of this kind of generation in distribution planning
models [4, 6]. This new context where DG comes into play calls for changes in the
way distribution systems are operated and planned [3].
DG comprises small-scale power units located close to consumption centers.
Manifold technologies are currently used for DG including wind turbines, photo-
voltaic (PV) plants, mini hydro plants, fuel cells, cogeneration plants, micro gas
turbines, internal combustion engines, and energy storage devices such as batteries
[5, 7]. The use of DG has numerous advantages related to system planning and
operation [8], such as
• Reduction in energy losses
• Control of the voltage profile
• Improvement of power quality
• Increase in system reliability
• Reduction or deferral of the network expansion
• Decrease in the emissions of CO2
• Short lead time
• Low investment risk
• Modularity
• Reduced physical size
• Availability of a wide range of DG technologies.
The presence of DG may have a significant impact on power flows, voltage
profiles, system efficiency, and protection devices. As a consequence, the tradi-
tionally passive distribution networks are transitioning to a new paradigm where an
active role is played. The operational impact of DG depends on many factors such
as the type, size, and location of generation units; the types of control equipment;
and the characteristics of branches and loads, among others.
The increasing penetration of non-dispatchable renewable-based technologies
for DG, such as wind and photovoltaic energy, requires the consideration of the
uncertainty associated with the high variability of these energy sources.
Furthermore, load demand is another source of uncertainty with a huge impact on
generation. Although many tools have been successfully developed to forecast
demand as well as wind and PV energy production, the incorporation of such
uncertainty sources in planning models is still challenging.
In the technical literature, many works have addressed the joint expansion
planning of distribution network assets and DG [4]. However, under a dynamic or
multistage framework, only few works have considered a complete expansion
model accounting for topological changes due to the connection of new load nodes
through the installation of new branches. Relevant works addressing this problem
can be classified in two groups, namely those disregarding uncertainty [9–13], and
those considering uncertainty [14–19].
In [9], a genetic algorithm combined with an optimal power flow is applied to
solve the deterministic co-optimized expansion planning problem driven by the
4 G. Muñoz-Delgado et al.

minimization of the overall cost. Reference [10] uses a heuristic method based on
particle swarm optimization and shuffled frog leaping to solve a multiobjective
version of the optimization problem where two objective functions are minimized,
namely cost and reliability. Reference [11] applies a modified particle swarm
optimization algorithm to solve this combinatorial optimization problem. Reference
[12] presents a multiobjective reliability-based distribution expansion planning
model. A hybrid self-adaptive global-based harmony-search algorithm and an
optimal power flow are used whereas a fuzzy satisfying method is applied in order
to obtain the best solution. Reference [13] presents a mixed-integer linear pro-
gramming model to solve the joint multistage expansion planning problem.
Reference [14] proposes a genetic algorithm to solve the distribution system
expansion planning problem under uncertainty using a multiobjective optimization
framework. Reference [15] applies a genetic algorithm combined with an optimal
power flow. The uncertainty of demand, electricity prices, and wind is represented
through scenarios generated on the basis of their corresponding probability density
functions. Reference [16] uses particle swarm optimization to address the planning
problem considering load and price uncertainties under an electricity market
environment. These uncertainties are modeled through probability density functions
and Monte Carlo simulation. In [17], a multiobjective model is formulated to
represent the different objectives of a distribution company and private DG
investors while considering the uncertainty of demand. The concept of system of
systems is proposed to model the expansion of DG owned by private developers.
Particle swarm optimization is applied to solve the proposed model. Finally, in [18,
19], a multistage and stochastic mixed-integer linear programming model is
developed to support the decision-making process of distribution system planners.
This chapter is focused on multistage expansion planning of distribution network
assets and DG [20, 21]. In this problem, the optimal location, alternative, and
installation time for each candidate asset is provided, thereby constituting a
dynamic approach. Moreover, the connection of new load nodes is considered.
Thus, topological changes due to the installation of new branches to supply the
demand at those new load nodes are explicitly accounted for. First, a deterministic
model for co-optimized expansion planning is described. This model is driven by
the minimization of the present value of the total cost, which comprises costs
related to investment, maintenance, production, energy losses, and unserved energy.
As a consequence of this co-optimization, the radial operation of the system needs
to be specifically imposed in order to avoid the islanding of distributed generators
and the issues associated with transfer nodes. Second, in order to represent the
uncertainty associated with demand and renewable-based generation, a stochastic
programming model relying on the previous deterministic model is presented.
Uncertainty is modeled through different scenarios that explicitly capture the cor-
relation between uncertainty sources. This model is driven by the minimization of
the present value of the total expected cost. The resulting optimization problems are
formulated as mixed-integer nonlinear programs, which, using some well-known
linearization schemes, are recast as mixed-integer linear programs.
1 Distribution System Expansion Planning 5

We recognize that the use of the models described in this chapter leads to results
that may be optimistic and that a complete study would require the consideration of
a more sophisticated operational model. This consideration would, however, render
the problem essentially intractable through optimization and would have to be
solved by heuristics or repeated simulations. These modeling limitations notwith-
standing, the use of our models is acceptable for distribution planning purposes and
provides the planner with a first estimate of a cost-effective expansion plan.
Additionally, it is worth mentioning that the use of mixed-integer linear pro-
gramming features three advantages of utmost importance from a practical per-
spective, namely (1) the guaranteed finite convergence to optimality, (2) the
availability of a measure of the distance to the global optimum along the solution
process, and (3) the ready availability of efficient off-the-shelf software.
Finally, note that regardless of the ownership of DG assets, the models described
hereinafter provide valuable information about the best investment plan in terms of
economics. In the case of DG owned by independent producers, such information
may be used to devise appropriate incentive strategies, which are beyond the scope
of this chapter.

1.2 Deterministic Model

In this section, a deterministic optimization model for the joint expansion planning
of distribution network assets and DG is presented. Built on the distribution net-
work expansion planning models described in [7, 22, 23], our model considers a
centralized framework. Thus, the planner is responsible for expanding the existing
distribution network in order to meet the growing demand at minimum cost over a
planning horizon comprising several stages. To that end, the planner has the pos-
sibility of installing new branches, transformers, substations, and generators for
which several investment alternatives exist. For each stage, demand, wind speed,
and solar irradiation profiles are divided into several time blocks. Figure 1.1 shows
an example with a 4-time-block discretization. To that end, first, historical hourly
data of system demand, wind speed, and solar irradiation throughout a year are
expressed as per-unit factors by dividing each data by the corresponding maximum
level. Hence, each set of factors represents the per-unit demand, wind speed, and
solar irradiation profile. Second, triplets of hourly factors for demand, wind speed,
and solar irradiation are sorted by demand in descending order. Next, the factor
curves are discretized into nB time blocks characterized by a pre-specified number
of hours. For each time block, the average values of the ordered demand, wind
speed, and solar irradiation factors are calculated.
Moreover, as done in [22, 23], radial operation of the system is explicitly
imposed and a lossless approximate network model is used. In addition, as done in
[23], the costs of losses are included in the objective function. Next, the deter-
ministic model is described in detail.
6 G. Muñoz-Delgado et al.

1.00

Demand factor (pu)

Time blocks

0.65

0.30
0 8760
Time (h)

1.00
Wind speed factor (pu)

0.50

0.00
0 8760
Time (h)

1.00
Solar irradiation factor (pu)

0.50

0.00
0 8760
Time (h)

Fig. 1.1 Time block discretization of demand, wind speed, and solar irradiation factors for the
deterministic model

1.2.1 Objective Function and Cost-Related Terms

The goal of the distribution planner is to address the estimated demand growth by
installing the required system components in an economic and secure fashion. To
that end, an optimization problem is presented wherein the sum of investment and
operating costs is minimized along the planning horizon. The investment cost is
related to the purchase and installation of distributed generation and distribution
network assets such as branches, transformers, and substations. The operating cost
consists of four terms related to maintenance, production, energy losses, and
unserved energy. The maintenance cost represents the cost associated with keeping
all system components in good condition through regular inspection and eventual
1 Distribution System Expansion Planning 7

repair if needed. The production cost is related to the energy purchased at sub-
stations and the energy produced by generators. The cost of energy losses com-
prises the energy lost as heat in branches and transformers due to the Joule effect.
Finally, the unserved energy cost is a penalty for the demand not supplied under
normal operation.
The objective function (1.1) to be minimized represents the present value of the
total cost.

X ð1 þ IÞt X  
cTPV ¼ cIt þ ð1 þ IÞt cM
t þ ct þ ct þ ct
E R U
I
t2T t2T ð1:1Þ
ð1 þ IÞ nT  
þ nT þ cnT þ cnT þ cnT :
cM E R U
I

As formulated in [23], the total cost consists of three terms. In the first term, the
present worth value of the investment cost is represented under the assumption of a
perpetual or infinite planning horizon [24]. The second term characterizes the
present value of the sum of the operating costs including maintenance, production,
energy losses, and unserved energy costs. Finally, in the third term, the present
value of the sum of the operating costs incurred after the last time stage is modeled.
Note that such a term depends on the values of those costs at the last time stage
while also assuming a perpetual planning horizon. The cost terms in (1.1) are
formulated as
X X X X X X
cIt ¼ RRl CkI;l ‘ij xlijkt þ RRSS it þ RR
CiI;SS xSS NT
CkI;NT xNT
ikt
l2fNRB;NABg k2K l ði;jÞ2! l
i2WSS k2K NT i2W SS
X XX p
þ RR p
CkI;p pf Gk xpikt ; 8t 2 T
p2P k2K p i2Wp

ð1:2Þ
XX X   X X X
t ¼
cM CkM;l ylijkt þ yljikt þ CkM;tr ytrikt
l2L k2K l ði;jÞ2!l tr2TR k2K tr i2WSS
XXX ð1:3Þ
þ CkM;p ypikt ; 8t 2 T
p2P k2K p i2Wp

!
X X X X XXX
cEt ¼ Db pf CbSS gtriktb þ CkE;p gpiktb ; 8t 2 T ð1:4Þ
b2B tr2TR k2K tr i2WSS p2P k2K p i2Wp
8 G. Muñoz-Delgado et al.

X X X X  2
cRt ¼ Db CbSS pf Zktr gtriktb
b2B tr2TR k2K tr i2WSS
1 ð1:5Þ
XX X  2
þ Zkl ‘ij fijktb
l
þ fjiktb
l A; 8t 2 T
l2L k2K l ði;jÞ2!l

X X
t ¼
cU Db CU pf ditb ; 8t 2 T ð1:6Þ
U

b2B i2WLN
t

l
Ið1 þ IÞg
where capital recovery rates are computed as RRl ¼ gl
, 8l 2 fNRB; NABg;
ð1 þ IÞ 1
NT p SS
Ið1 þ IÞg Ið1 þ IÞg Ið1 þ IÞg
RRNT ¼ gNT
; RRp ¼ p , 8p 2 P; and RRSS ¼ gSS
.
ð1 þ IÞ 1 ð1 þ IÞg 1 ð1 þ IÞ 1
It is worth pointing out that, for each time stage, one set of binary variables per
branch, xlijkt , is used to model the associated investment decisions. In contrast, two
sets of binary variables, ylijkt and yljikt , as well as two sets of continuous variables,
l l
fijktb and fjiktb , are associated with each branch in order to respectively model the
direction and magnitude of the corresponding current flow. Note that fijktb l
is positive
and equal to the branch current flow between nodes i and j measured at node i only
when the current flows from i to j, being 0 otherwise.
Expressions (1.2) represent the amortized investment cost at each stage, which is
formulated as the sum of the costs associated with the replacement and addition of
branches, the reinforcement and construction of substations, the installation of new
transformers, and the installation of DG. Expressions (1.3) model the maintenance
costs of existing and newly added branches, transformers, and generators. In (1.4),
the production costs associated with substations and generators are characterized.
Expressions (1.5) represent the cost of energy losses in branches and transformers,
which, as done in [23], are modeled as quadratic terms. Such nonlinearities can be
accurately approximated by a set of tangent lines, as explained in Sect. 1.2.5. This
approximation yields piecewise linear functions, which, for practical purposes, are
indistinguishable from nonlinear models if enough segments are used. Finally,
expressions (1.6) correspond to the penalty cost imposed for the unserved energy.

1.2.2 Kirchhoff’s Laws and Operational Limits

System operation is an important factor in expansion planning models since

investment and operating decisions have a great impact on each other. The con-
straints associated with system operation are formulated as
1 Distribution System Expansion Planning 9

V  vitb  V; 8i 2 WN ; 8t 2 T ; 8b 2 B ð1:7Þ
l
0  fijktb
l
 ylijkt F k ; 8l 2 L; 8i 2 Wlj ; 8j 2 WN ; 8k 2 K l ; 8t 2 T ; 8b 2 B ð1:8Þ
tr
0  gtriktb  ytrikt Gk ; 8tr 2 TR; 8i 2 WSS ; 8k 2 K tr ; 8t 2 T ; 8b 2 B ð1:9Þ

0  ditb
U
 lD
b Dit ; 8i 2 WLN
t ; 8t 2 T ; 8b 2 B ð1:10Þ
C
0  gCiktb  yCikt Gk ; 8i 2 WC ; 8k 2 K C ; 8t 2 T ; 8b 2 B ð1:11Þ

^ p ; 8p 2 fW; Hg; 8i 2 Wp ; 8k 2 K p ; 8t 2 T ; 8b 2 B
0  gpiktb  ypikt G ð1:12Þ
ikb
XXX p X
giktb  n b Dit ; 8t 2 T ; 8b 2 B
lD ð1:13Þ
p2P k2K p i2Wp i2WLN
t

XX X h i
l
fijktb  fjiktb
l

l2L k2K l ði;jÞ2!l

X X XX
¼ gtriktb þ gpiktb  lD
b Dit þ ditb ;
U
8i 2 WN ; 8t 2 T ; 8b 2 B
tr2TR k2K tr p2P k2K p

ð1:14Þ
h  i
ylijkt Zkl ‘ij fijktb
l
 vitb  vjtb ¼ 0;
ð1:15Þ
8l 2 L; 8i 2 Wlj ; 8j 2 WN ; 8k 2 K l ; 8t 2 T ; 8b 2 B:

Constraints (1.7) limit nodal voltage magnitudes by setting upper and lower
bounds. Similarly, constraints (1.8) set the bounds on branch current flows. Note
that, if a branch is not used, i.e., ylijkt ¼ 0, the corresponding current flow is 0.
Analogously, current injection levels of transformers are restricted in constraints
(1.9), whereby, if a transformer is not used, i.e., ytrikt ¼ 0, the corresponding current
injection is 0. Constraints (1.10) model the variables associated with nodal unserved
energy as continuous and nonnegative, with maximum levels equal to the corre-
sponding level of nodal demand. Production limits for conventional and
renewable-based generation are formulated in constraints (1.11) and (1.12),
respectively. Note that, as done for branches and transformers, the maximum level
of generation is imposed through the use of binary utilization variables. Moreover,
the upper bound for the production of each conventional generator is the corre-
sponding rated generation capacity whereas the upper generation bound for each
renewable generator is the available power associated with the corresponding
generation technology. Constraints (1.13) limit the level of penetration of DG to a
fraction, n, of the demand.
Finally, the effect of the distribution network is characterized through expres-
sions (1.14) and (1.15), which represent Kirchhoff’s laws using the linearized
10 G. Muñoz-Delgado et al.

network model proposed in [22]. As described in [22], the linearized network model
is an adapted version of the per-unit dc model used for the transmission network
that is based on three assumptions: (1) nodal voltages lie in a narrow range around
the rated value used as base voltage, (2) all per-unit branch current flows and nodal
power injections have the same power factor, and (3) the per-unit voltage drop
across a branch is equal to the difference between the per-unit magnitudes of the
nodal voltages at both ends of the branch.
As per assumption (1), per-unit values of nodal apparent power injections and
nodal current injections are identical. Assumption (2) allows representing per-unit
branch current flows and nodal power injections, which are complex numbers,
through their magnitudes. Hence, Kirchhoff’s current law can be expressed as a set
of linear scalar equalities in terms of per-unit magnitudes of branch current flows
and nodal power injections, giving rise to nodal balance equations (1.14). In
addition, assumption (3) allows formulating Kirchhoff’s voltage law for each
branch in use as a linear expression relating the per-unit magnitudes of branch
current flows, nodal voltages, and branch impedances. Constraints (1.15) extend
this result to account for the utilization state of all branches. Note that constraints
(1.15) include nonlinearities involving the products of binary variables and con-
tinuous variables, for which a linear equivalent is formulated in Sect. 1.2.5.

1.2.3 Investment and Utilization Constraints

The investment and utilization decisions are constrained according to

xlijkt 2 f0; 1g; 8l 2 fNRB; NABg; 8ði; jÞ 2 !l ; 8k 2 K l ; 8t 2 T ð1:16Þ

it 2 f0; 1g;
xSS 8i 2 WSS ; 8t 2 T ð1:17Þ

ikt 2 f0; 1g;

xNT 8i 2 WSS ; 8k 2 K NT ; 8t 2 T ð1:18Þ

xpikt 2 f0; 1g; 8p 2 P; 8i 2 Wp ; 8k 2 K p ; 8t 2 T ð1:19Þ

ylijkt 2 f0; 1g; 8l 2 L; 8i 2 Wlj ; 8j 2 WN ; 8k 2 K l ; 8t 2 T ð1:20Þ

ytrikt 2 f0; 1g; 8tr 2 TR; 8i 2 WSS ; 8k 2 K tr ; 8t 2 T ð1:21Þ

ypikt 2 f0; 1g; 8p 2 P; 8i 2 Wp ; 8k 2 K p ; 8t 2 T ð1:22Þ

XX
xlijkt  1; 8l 2 fNRB; NABg; 8ði; jÞ 2 !l ð1:23Þ
t2T k2K l
1 Distribution System Expansion Planning 11

X
it  1;
xSS 8i 2 WSS ð1:24Þ
t2T
X X
ikt  1;
xNT 8i 2 WSS ð1:25Þ
t2T k2K NT
XX
xpikt  1; 8p 2 P; 8i 2 Wp ð1:26Þ
t2T k2K p

X
t

ikt 
xNT is ;
xSS 8i 2 WSS ; 8k 2 K NT ; 8t 2 T ð1:27Þ
s¼1

ijkt þ yjikt  1;
yEFB 8ði; jÞ 2 !SW;EFB ; 8k 2 K EFB ; 8t 2 T ð1:28Þ
EFB

X
t
ylijkt þ yljikt  xlijks ; 8l 2 fNRB; NABg; 8ði; jÞ 2 !SW;l ; 8k 2 K l ; 8t 2 T
s¼1
ð1:29Þ

X
t X
ijkt þ yjikt  1 
yERB ijjs ; 8ði; jÞ 2 !SW;ERB ; 8k 2 K ERB ; 8t 2 T ð1:30Þ
ERB
xNRB
s¼1 j2K NRB

 
ijkt þ yjikt ¼ 1;
yEFB 8ði; jÞ 2 !EFB n!SW;EFB ; 8k 2 K EFB ; 8t 2 T ð1:31Þ
EFB

X
t  
ylijkt þ yljikt ¼ xlijks ; 8l 2 fNRB; NABg; 8ði; jÞ 2 !l n!SW;l ; 8k 2 K l ; 8t 2 T
s¼1
ð1:32Þ

X
t X  
ijkt þ yjikt ¼ 1 
yERB ijjs ; 8ði; jÞ 2 !ERB n!SW;ERB ; 8k 2 K ERB ; 8t 2 T
ERB
xNRB
s¼1 j2K NRB

ð1:33Þ

X
t

ikt 
yNT iks ;
xNT 8i 2 WSS ; 8k 2 K NT ; 8t 2 T ð1:34Þ
s¼1

X
t
ypikt  xpiks ; 8p 2 P; 8i 2 Wp ; 8k 2 K p ; 8t 2 T ð1:35Þ
s¼1
12 G. Muñoz-Delgado et al.

X X X
CkI;l ‘ij xlijkt
l2fNRB;NABg k2K l ði;jÞ2!l
X X X XXX
þ CiI;SS xSS  p xp  IBt ; 8t 2 T :
CkI;p pf G
it þ CkI;NT xNT
ikt þ k ikt
i2WSS k2K NT i2WSS p2P k2K p i2Wp

ð1:36Þ

Expressions (1.16)–(1.22) set the binary nature of investment and utilization

variables. Constraints (1.23)–(1.27) are related to investment decisions. As per
constraints (1.23), a maximum of one investment can be made in each branch along
the planning horizon. Constraints (1.24) impose that investment at substation nodes
can only be made once along the planning horizon. One new transformer at most
can be installed at each substation node along the time span, as modeled in con-
straints (1.25). The installation of generators at each candidate node throughout the
planning horizon is limited to one by constraints (1.26). Constraints (1.27) guar-
antee that new transformers can only be added in substations that have been pre-
viously expanded or built.
Expressions (1.28)–(1.35) are related to the utilization of existing and newly
added components. Note that binary utilization variables are related to investment
variables so that a system component cannot be used if it has not been previously
installed. Constraints (1.28)–(1.30) model the utilization of switchable branches
while explicitly characterizing the direction of current flows. Switching of those
branches is considered under normal operation, thereby allowing for network
reconfiguration. Analogously, constraints (1.31)–(1.33) are associated with the
utilization of branches that are not reconfigurable under normal operation.
Constraints (1.34) and (1.35) model the utilization of new transformers and dis-
tributed generators, respectively. Finally, constraints (1.36) set the budget for
investments at each stage.

1.2.4 Radiality Constraints

The radial operation is modeled by

XXX
ylijkt ¼ 1; 8j 2 WLN
t ; 8t 2 T ð1:37Þ
l2L i2Wlj k2K l

XXX
ylijkt  1; 8j 62 WLN
t ; 8t 2 T ð1:38Þ
l2L i2Wlj k2K l

XXX 
~f l  ~f l ¼ ~gSS  D
~ it ; 8i 2 WN ; 8t 2 T ð1:39Þ
ijkt jikt it
l2L k2K l j2Wli
1 Distribution System Expansion Planning 13

0  ~fijkt
EFB
 nDG ; 8i 2 WEFB
j ; 8j 2 WN ; 8k 2 K EFB ; 8t 2 T ð1:40Þ
!
X
t X
0  ~fijkt
ERB
 nDG 1  xNRB
ijjs ; 8ði; jÞ 2 !ERB ; 8k 2 K ERB ; 8t 2 T
s¼1 j2K NRB

ð1:41Þ
!
X
t X
0  ~fjikt
ERB
 nDG 1  xNRB
ijjs ; 8ði; jÞ 2 !ERB ; 8k 2 K ERB ; 8t 2 T
s¼1 j2K NRB
ð1:42Þ

X
t
0  ~fijkt
l
 nDG xlijjs ; 8l 2 fNRB; NABg; 8ði; jÞ 2 !l ; 8k 2 K l ; 8t 2 T ð1:43Þ
s¼1

X
t
0  ~fjikt
l
 nDG xlijjs ; 8l 2 fNRB; NABg; 8ði; jÞ 2 !l ; 8k 2 K l ; 8t 2 T ð1:44Þ
s¼1

0  ~gSS
it  nDG ; 8i 2 WSS ; 8t 2 T ð1:45Þ

where
(   
~ it ¼ 1; 8i 2 WC [ WW [ WH \ WLN ; 8t 2 T
D  C H
 t  ð1:46Þ
0; 8i 62 W [ W [ W \ Wt ; 8t 2 T :
W LN

Radiality is modeled through traditional constraints (1.37) and (1.38) [23, 25] in
conjunction with constraints (1.39)–(1.45), which were recently presented in [26].
Constraints (1.37) impose load nodes to have a single incoming flow while
expressions (1.38) set a maximum of one incoming flow for the remaining nodes.
The new radiality constraints (1.39)–(1.45) avoid the issues with transfer nodes and
islanded DG that would arise should traditional radiality constraints be solely used
in the co-optimized expansion planning problem. The idea behind these new
radiality constraints is to set a fictitious demand in those load nodes that could be
islanded due to the installation of DG units. Fictitious nodal demands can only be
supplied by fictitious substations located at the original substation nodes, where
fictitious energy flowing through the branches of the system is injected. As a
consequence, the islanding of areas with load demand is prevented under normal
operation. Constraints (1.39) represent the nodal fictitious current balance equa-
tions. Constraints (1.40)–(1.44) bound the fictitious flows through branches.
Finally, constraints (1.45) set the limits for the fictitious currents injected by the
fictitious substations.
14 G. Muñoz-Delgado et al.

1.2.5 Mixed-Integer Linear Formulation

The deterministic model for the joint expansion planning of DG and distribution
network assets is a mixed-integer nonlinear program for which no exact method is
currently available. This issue motivates us to recast the original problem as an
instance of mixed-integer linear programming by replacing nonlinear expressions
(1.5) and (1.15) with linear terms. As a consequence, finite convergence to the
optimum is guaranteed while providing a measure of the distance to optimality
along the solution process. Moreover, efficient off-the-shelf software based on the
state-of-the-art branch-and-cut algorithm is available.
Based on [27], a piecewise linear approximation is used for the quadratic terms
in (1.5). Thus, expressions (1.5) are replaced with

X X X XX
nH
cRt ¼ Db CbSS pf tr tr
Mkh diktbh
b2B tr2TR k2K tr i2WSS h¼1
1 ð1:47Þ
XX X X
nH  
þ l
Mkh ‘ij dlijktbh þ dljiktbh A; 8t 2 T
l2L k2K l ði;jÞ2! h¼1
l

X
nH
gtriktb ¼ dtriktbh ; 8tr 2 TR; 8i 2 WSS ; 8k 2 K tr ; 8t 2 T ; 8b 2 B ð1:48Þ
h¼1

0  dtriktbh  Atrkh ; 8h ¼ 1. . .nH ; 8tr 2 TR; 8i 2 WSS ; 8k 2 K tr ; 8t 2 T ; 8b 2 B

ð1:49Þ

X
nH
l
fijktb ¼ dlijktbh ; 8l 2 L; 8i 2 Wlj ; 8j 2 WN ; 8k 2 K l ; 8t 2 T ; 8b 2 B ð1:50Þ
h¼1

0  dlijktbh  Alkh ; 8h ¼ 1. . .nH ;

ð1:51Þ
8l 2 L; 8i 2 Wlj ; 8j 2 WN ; 8k 2 K l ; 8t 2 T ; 8b 2 B

where expressions (1.47) are the linearized costs of energy losses, while (1.48)–
(1.49) and (1.50)–(1.51) are related to the linearization of energy losses in trans-
formers and branches, respectively.
In addition, using the disjunctive-constraint-based transformation described in
[28], nonlinear expressions (1.15) have the linear equivalent of
     
J 1  ylijkt  Zkl ‘ij fijktb
l
 vitb  vjtb  J 1  ylijkt ;
ð1:52Þ
8l 2 L; 8i 2 Wlj ; 8j 2 WN ; 8k 2 K l ; 8t 2 T ; 8b 2 B:
1 Distribution System Expansion Planning 15

If ylijkt is equal to 1, the corresponding constraint (1.52) becomes

 
0  Zkl ‘ij fijktb
l
 vitb  vjtb  0, which is identical to the condition Zkl ‘ij fijktb
l

 
vitb  vjtb ¼ 0 resulting from (1.15). Conversely, if yijkt is equal to 0, the corre-
l
   
sponding constraint (1.52) yields J   vitb  vjtb  J, i.e., vitb  vjtb   J,
l
since fijktb is equal to 0 as per (1.8). Thus, for a sufficiently large positive value for
parameter J, no relation between nodal voltage magnitudes vitb and vjtb is imposed,
as modeled in (1.15) for ylijkt equal to 0. Since nodal voltage magnitudes are
 
bounded by V and V in (1.7), the largest possible value for vitb  vjtb  is V  V,
which is thus the minimum value for J.
Thus, the resulting mixed-integer linear program is formulated as

X ð1 þ I Þt X  
Minimize cTPV ¼ cIt þ ð1 þ I Þt cM
t þ ct þ ct þ ct
E R U
N
DT
t2T
I t2T
ð1 þ I ÞnT  M 
þ cnT þ cEnT þ cRnT þ cU
nT
I
ð1:53Þ

subject to
Constraints (1.2)–(1.4), (1.6)–(1.14), (1.16)–(1.45), and (1.47)–(1.52) (1.54)
where
n
NDT ¼ cEt ; cIt ; cM t ; ct ; ct ; c
R U TPV
; ditb
U
; fijktb
l
; ~fijkt
l
; gpiktb ; gtriktb ; ~ it ; vitb ;
gSS
o
p p
xlijkt ; xNT
ikt ; xikt ; xit ; yijkt ; yikt ; yikt ; dijktbh ; diktbh
SS l tr l tr

1.3 Stochastic Programming Model

In this section, based on the previous deterministic model, the uncertainty related to
renewable-based generation and demand is incorporated using a scenario-based
stochastic programming framework [29]. To that end, uncertainty is characterized
through a set of scenarios that explicitly capture the correlation between uncertainty
sources.
Next, the procedure for generating scenarios of demand, wind speed, and solar
irradiation is presented. The objective function and constraints of the stochastic
model are subsequently described in detail. This section concludes with the for-
mulation of the resulting mixed-integer linear program.
16 G. Muñoz-Delgado et al.

1.3.1 Uncertainty Modeling

For investment planning, the increasing penetration of stochastic resources in

electric distribution networks requires the accurate modeling of the associated
uncertainty. Based on the methodology described in [30], a set of scenarios rep-
resenting the uncertainty associated with demand, wind speed, and solar irradiation
is generated using historical data. The scenario-generation procedure comprises six
steps that are described as follows:
Step (1) Historical hourly data of system demand, wind speed, and solar irradiation
throughout a year are expressed as per-unit factors by dividing each data by the
corresponding maximum level. Hence, each set of factors represents the per-unit
demand, wind speed, and solar irradiation profile.
Step (2) Triplets of hourly factors for demand, wind speed, and solar irradiation are
sorted by demand factor in descending order. Figure 1.2 shows an ordered demand
factor curve and the corresponding profiles of wind speed and solar irradiation
factors.
Step (3) The factor curves resulting from step (2) are discretized into nB time blocks.
In order to accurately model the peak demand, which usually has a big influence on
investment decisions, a relatively small time block related to such peak demand is
defined. For each time block, the corresponding wind speed and solar irradiation
factors are sorted in descending order. An example with a 4-time-block dis-
cretization is depicted in Fig. 1.3.
Step (4) For each time block, the cumulative distribution functions (cdf) of the
ordered demand, wind speed, and solar irradiation factors are built. As an example,
Fig. 1.4 the cdf corresponding to the curves depicted in Fig. 1.3.
Step (5) The cdfs are divided into segments with their corresponding probabilities.
H
The pre-specified numbers of segments are denoted by nD W
S , nS , and nS for demand,
wind speed, and solar irradiation factor curves, respectively. In addition to its
pre-specified probability, each segment s is characterized by an average factor equal
to the average value of the factors within such a segment. As a consequence, pairs
probability-average factor are generated for demand, namely pD sb lsb , for wind
D
H H
speed, namely pW sb lsb , and for solar irradiation, namely psb lsb . In Fig. 1.5, the
W

cdf of the first time block for the demand factor curve is divided into three segments
with probabilities, pD D D
11 , p21 , and p31 , equal to 0.4, 0.5, and 0.1, respectively. The
demand factors associated with those segments respectively lie in the ranges [0.00,
0.70], (0.70, 0.86], and (0.86, 1.00]. As can be seen in Fig. 1.5, the corresponding
average demand factors, lD D D
11 , l21 , and l31 , are equal to 0.67, 0.76, and 0.90,
respectively.
1 Distribution System Expansion Planning 17

1.00

Demand factor (pu)

0.65

0.30
0 8760
Time (h)

1.00
Wind speed factor (pu)

0.50

0.00
0 8760
Time (h)
Solar irradiation factor (pu)

1.00

0.50

0.00
0 8760
Time (h)

Fig. 1.2 Ordered demand factor curve and the corresponding profiles of wind speed and solar
irradiation factors

Step (6) Scenarios, indexed by x, for each time block result from combining all
H H
pairs pDsb lsb , psb lsb , and psb lsb . Hence, nodal demands in each scenario are
D W W

equal to the product of the forecasted values and the corresponding factor lD b ðxÞ.
H
For each scenario, average factors lW sb and l sb are converted to wind speed and solar
irradiation levels so that the maximum levels of wind and photovoltaic power
generation, G ^ W ðxÞ and G ^ H ðxÞ, are determined. Thus, for each time block b,
ikb ikb
18 G. Muñoz-Delgado et al.

1.00

Demand factor (pu)

0.65

Time blocks

0.30
0 8760
Time (h)

1.00
Wind speed factor (pu)

0.50

0.00
0 8760
Time (h)
Solar irradiation factor (pu)

1.00

0.50

0.00
0 8760
Time (h)

Fig. 1.3 Time block discretization of demand, wind speed, and solar irradiation factors

scenario x comprises an average demand factor, lD b ðxÞ, a vector of maximum

levels of wind power generation, G ^ W ðxÞ, and a vector of maximum levels of
ikb
^ H ðxÞ. Mathematically, the set of scenarios Xb is
photovoltaic power generation, G ikb
formulated as

n o
Xb ¼ lD ^W ^ H ðxÞ
b ðxÞ; Gikb ðxÞ 8i2W W
;8k2K W
; G ikb 8i2W H
;8k2K H
; 8b 2 B:
8x¼1...nX
ð1:55Þ
1 Distribution System Expansion Planning 19

Time block 1 Time block 2 Time block 3 Time block 4

Cumulative probability

1 1 1 1

0 0 0 0
0 1 0 1 0 1 0 1
Demand factor (pu)
Cumulative probability

1 1 1 1

0 0 0 0
0 1 0 1 0 1 0 1
Wind speed factor (pu)
Cumulative probability

1 1 1 1

0 0 0 0
0 1 0 1 0 1 0 1
Solar irradiation factor (pu)

Fig. 1.4 Cumulative distribution functions of demand, wind speed, and solar irradiation factors

1.00
Cumulative probability

0.90

0.40

0.00
0.00 0.70 0.86 1.00
Demand factor (pu)

Fig. 1.5 Cumulative distribution function for the first block of the demand factor curve

In addition, the probability of each scenario, pb ðxÞ, is equal to the corresponding

W H D W H
product pD sb psb psb . The number of scenarios per time block, nX , is equal to nS nS nS ,
whereas the number of scenarios or operating conditions is equal to nB nX .
It should be noted that if data for more than one year were available, the
scenario-generation procedure could be applied for each year in order to create
different operating conditions at each stage.
20 G. Muñoz-Delgado et al.

1.3.2 Objective Function and Cost-Related Terms

The objective function to be minimized is the net present value of the expected total
cost, which can be also represented through expression (1.1) described for the
deterministic model. Expected investment and maintenance costs are modeled as in
(1.2) and (1.3), respectively. The other cost terms, i.e., the expected costs of pro-
duction, energy losses, and unserved energy, are modeled as

XX
nX X X X
cEt ¼ pb ðxÞDb pf CbSS gtriktb ðxÞ
b2B x¼1 tr2TR k2K tr i2WSS
! ð1:56Þ
XXX
þ CkE;p gpiktb ðxÞ ; 8t 2 T
p2P k2K p i2Wp

XX
nX X X X  2
cRt ¼ pb ðxÞDb CbSS pf Zktr gtriktb ðxÞ
b2B x¼1 tr2TR k2K tr i2WSS
1 ð1:57Þ
XX X  2
þ Zkl ‘ij fijktb
l
ðxÞ þ fjiktb
l
ð x Þ A; 8t 2 T
l2L k2K l ði;jÞ2!l

XX
nX X

t ¼
cU pb ðxÞDb C U pf ditb ðxÞ; 8t 2 T : ð1:58Þ
U

b2B x¼1 i2WLN

t

Similar to the deterministic model, it is worth pointing out that, for each time
stage, one set of binary variables per branch, xlijkt , is used to model the associated
investment decisions. In contrast, two sets of binary variables, ylijkt and yljikt , as well
l
as two sets of continuous variables, fijktb ðxÞ and fjiktb
l
ðxÞ, are associated with each
branch in order to respectively model the direction and magnitude of the corre-
sponding current flow. Note that fijktb
l
ðxÞ is positive and equal to the branch current
flow between nodes i and j measured at node i only when the current flows from i to
j, being 0 otherwise.
The three expected costs modeled in (1.56)–(1.58) respectively are the stochastic
counterparts of the costs formulated in (1.4)–(1.6) for the deterministic model.
Expressions (1.56)–(1.58) mainly differ from (1.4)–(1.6) in the consideration of
scenarios and their probabilities in order to model uncertainty.
1 Distribution System Expansion Planning 21

1.3.3 Kirchhoff’s Laws and Operational Limits

For the stochastic model, the constraints associated with system operation are
formulated as

V  vitb ðxÞ  V; 8i 2 WN ; 8t 2 T ; 8b 2 B; 8x ¼ 1. . .nX ð1:59Þ

0  fijktb
l kl ;
ðxÞ  ylijkt F
ð1:60Þ
8l 2 L; 8i 2 Wlj ; 8j 2 WN ; 8k 2 K l ; 8t 2 T ; 8b 2 B; 8x ¼ 1. . .nX

 tr ;
0  gtriktb ðxÞ  ytrikt G k
ð1:61Þ
8tr 2 TR; 8i 2 WSS ; 8k 2 K tr ; 8t 2 T ; 8b 2 B; 8x ¼ 1. . .nX

0  ditb
U
ð x Þ  lD
b ðxÞDit ; 8i 2 WLN
t ; 8t 2 T ; 8b 2 B; 8x ¼ 1. . .nX ð1:62Þ
C
0  gCiktb ðxÞ  yCikt Gk ; 8i 2 WC ; 8k 2 K C ; 8t 2 T ; 8b 2 B; 8x ¼ 1. . .nX ð1:63Þ

^ p ðxÞ;
0  gpiktb ðxÞ  ypikt G ikb
ð1:64Þ
8p 2 fW; Hg; 8i 2 Wp ; 8k 2 K p ; 8t 2 T ; 8b 2 B; 8x ¼ 1. . .nX
XXX X
gpiktb ðxÞ  n b ðxÞDit ; 8t 2 T ; 8b 2 B; 8x ¼ 1. . .nX ð1:65Þ
lD
p
p2P k2K p i2W i2WLN
t

XX X h i
l
fijktb ðxÞ  fjiktb
l
ðxÞ
l2L k2K l ði;jÞ2!l
X X XX
¼ gtriktb ðxÞ þ gpiktb ðxÞ ð1:66Þ
tr2TR k2K tr p2P k2K p

b ðxÞDit þ ditb ðxÞ; 8i 2 W ; 8t 2 T ; 8b 2 B; 8x ¼ 1. . .nX

 lD U N

h  i
ylijkt Zkl ‘ij fijktb
l
ðxÞ  vitb ðxÞ  vjtb ðxÞ ¼ 0;
ð1:67Þ
8l 2 L; 8i 2 Wlj ; 8j 2 WN ; 8k 2 K l ; 8t 2 T ; 8b 2 B; 8x ¼ 1. . .nX :

It is worth mentioning that expressions (1.59)–(1.67) correspond to the adap-

tation of expressions (1.7)–(1.15) in the deterministic model, respectively, to the
scenario-based setting. Thus, for the description for (1.59)–(1.67), the interested
reader is referred to Sect. 1.2.2.
22 G. Muñoz-Delgado et al.

1.3.4 Investment and Utilization Constraints

For the stochastic model, investment- and utilization-related decisions for distri-
bution network assets and distributed generators along with the budgetary limit for
investments at each stage are modeled as in (1.16)–(1.36), which are described in
Sect. 1.2.3.

1.3.5 Radiality Constraints

Radiality constraints for the stochastic model are identical to those described for the
deterministic model in Sect. 1.2.4, namely expressions (1.37)–(1.46).

1.3.6 Mixed-Integer Linear Formulation

The stochastic model is formulated as a mixed-integer nonlinear program, where

nonlinearities are related to the quadratic energy losses in (1.57) and to the bilinear
terms in (1.67) involving the product of a binary variable and a continuous variable.
As done for the deterministic model, the quadratic energy losses in (1.57) are
recast as piecewise linear expressions as [27]

XX
nX X X XX
nH
cRt ¼ pb ðxÞDb CbSS pf diktbh ðxÞ
tr tr
Mkh
b2B x¼1 tr2TR k2K tr i2WSS h¼1
1 ð1:68Þ
XX X X
nH  
þ l
Mkh ‘ij dlijktbh ðxÞ þ dljiktbh ðxÞ A; 8t 2 T
l2L k2K l ði;jÞ2!l h¼1

X
nH
gtriktb ðxÞ ¼ dtriktbh ðxÞ;
h¼1 ð1:69Þ
8tr 2 TR; 8i 2 W ; 8k 2 K ; 8t 2 T ; 8b 2 B; 8x ¼ 1. . .nX
SS tr

0  dtriktbh ðxÞ  Atrkh ;

8h ¼ 1. . .nH ; 8tr 2 TR; 8i 2 WSS ; 8k 2 K tr ; 8t 2 T ; 8b 2 B; 8x ¼ 1. . .nX
ð1:70Þ

X
nH
l
fijktb ðx Þ ¼ dlijktbh ðxÞ;
h¼1 ð1:71Þ
8l 2 L; 8i 2 Wlj ; 8j 2 WN ; 8k 2 K l ; 8t 2 T ; 8b 2 B; 8x ¼ 1. . .nX
1 Distribution System Expansion Planning 23

0  dlijktbh ðxÞ  Alkh ;

8h ¼ 1. . .nH ; 8l 2 L; 8i 2 Wlj ; 8j 2 WN ; 8k 2 K l ; 8t 2 T ; 8b 2 B; 8x ¼ 1. . .nX :
ð1:72Þ

Using the disjunctive-constraint-based transformation described in [28], non-

linear expressions (1.67) have the linear equivalent of
     
 J 1  ylijkt  Zkl ‘ij fijktb
l
ðxÞ  vitb ðxÞ  vjtb ðxÞ  J 1  ylijkt ;
ð1:73Þ
8l 2 L; 8i 2 Wlj ; 8j 2 WN ; 8k 2 K l ; 8t 2 T ; 8b 2 B; 8x ¼ 1. . .nX :

Note that expressions (1.68)–(1.73) respectively represent the stochastic coun-

terparts of expressions (1.47)–(1.52) in the deterministic model.
Thus, the deterministic equivalent associated with the stochastic model can be
formulated as a mixed-integer linear program suitable for commercially available
software as

X ð1 þ I Þt X  
Minimize cTPV ¼ cIt þ ð1 þ I Þt cM
t þ ct þ ct þ ct
E R U
NST t2T
I t2T
ð1 þ I Þ nT  
þ nT þ cnT þ cnT þ cnT
cM E R U
I
ð1:74Þ

subject to:
Constraints (1.2), (1.3), (1.16)–(1.45), (1.56), (1.58)–(1.66), and
(1.68)–(1.73) (1. 75)
where n
NST ¼ cEt ; cIt ; cM
t ; ct ; ct ; c
R U TPV
; ditb
U
ðxÞ; fijktb
l
ðxÞ; ~fijkt
l
; gpiktb ðxÞ;
NT p p
gtriktb ðxÞ; ~gSS
it ; vitb ðxÞ; xijkt ; xikt ; xikt ; xit ; yijkt ; yikt ; yikt ; dijktbh ðxÞ; diktbh ðxÞg:
l SS l tr l tr

1.4 Numerical Results

Both models have been applied to a distribution system based on the benchmark
presented in [31]. As shown in Fig. 1.6, the system comprises 50 load nodes,
represented by circles; 4 substation nodes, depicted as squares; and 63 branches,
indicated by lines. Base power and base voltage are 1 MVA and 13.5 kV,
respectively. Upper and lower bounds for voltages at load nodes are equal to 1.05
and 0.95 pu, respectively. The system power factor pf is set at 0.9 and a three-block
piecewise linearization is used to approximate energy losses.
24 G. Muñoz-Delgado et al.

20
17
51
1 3
19 9
2 7 4
18
5
22 23 24 26
25
21 8 6

27
54 32
10 28
31 33 34
30 35 36
29 37 38 39
53
40
43 41
44 15 16
13 45
42

12
11 14 46 47

48
50
52
49

Fig. 1.6 One-line diagram of the test system

The currency used in the simulations is U.S. dollars, hereinafter denoted by $.

Investment decisions are made over a ten-year planning horizon divided into yearly
stages considering a 10% interest rate and an investment budget equal to
$3.5 million per year. Nodal peak demands are presented in Table 1.1.
The uncertainty characterization of demand, wind speed, and solar irradiation
relies on historical data of year 2012 for an actual site with maximum levels of wind
speed and solar irradiation equal to 17.08 m/s and 1114.21 W/m2, respectively. As
done in [32], the same demand, wind speed, and solar irradiation profiles are used
for all nodes. Moreover, four time blocks are considered, with durations equal to
350, 2650, 3900, and 1860 h/year, respectively. For the deterministic model,
average values for demand, wind speed, and solar irradiation within each time block
are used, which are presented in Table 1.2. For the stochastic model, the cumulative
distribution functions for demand, wind speed, and solar irradiation are divided into
three equiprobable segments. Thus, as reported in Table 1.3, for each time block,
three different conditions of demand, wind speed, and solar irradiation are con-
sidered according to the procedure described in Sect. 1.3.1. As a result, 27
equiprobable scenarios are generated for each time block, thereby totaling 108
scenarios for each stage.
In Fig. 1.6, existing branches not subject to modification are represented by solid
lines, existing branches that can undergo replacement are depicted as solid double
lines, and non-existing branches that are candidates for the installation are drawn as
Table 1.1 Nodal peak demands (MVA)
Node Stage
1 2 3 4 5 6 7 8 9 10
1 2.0500 2.1115 2.1730 2.2345 2.2960 2.3575 2.4190 2.4805 2.5420 2.6035
2 0.7800 0.8034 0.8268 0.8502 0.8736 0.8970 0.9204 0.9438 0.9672 0.9906
3 1.5800 1.6274 1.6748 1.7222 1.7696 1.8170 1.8644 1.9118 1.9592 2.0066
4 0.3200 0.3296 0.3392 0.3488 0.3584 0.3680 0.3776 0.3872 0.3968 0.4064
5 0.2800 0.2884 0.2968 0.3052 0.3136 0.3220 0.3304 0.3388 0.3472 0.3556
6 1.1700 1.2051 1.2402 1.2753 1.3104 1.3455 1.3806 1.4157 1.4508 1.4859
7 1.2400 1.2772 1.3144 1.3516 1.3888 1.4260 1.4632 1.5004 1.5376 1.5748
8 0.7200 0.7416 0.7632 0.7848 0.8064 0.8280 0.8496 0.8712 0.8928 0.9144
9 1.1400 1.1742 1.2084 1.2426 1.2768 1.3110 1.3452 1.3794 1.4136 1.4478
1 Distribution System Expansion Planning

10 1.5600 1.6068 1.6536 1.7004 1.7472 1.7940 1.8408 1.8876 1.9344 1.9812
11 1.9100 1.9673 2.0246 2.0819 2.1392 2.1965 2.2538 2.3111 2.3684 2.4257
12 0.9300 0.9579 0.9858 1.0137 1.0416 1.0695 1.0974 1.1253 1.1532 1.1811
13 1.1500 1.1845 1.2190 1.2535 1.2880 1.3225 1.3570 1.3915 1.4260 1.4605
14 1.3500 1.3905 1.4310 1.4715 1.5120 1.5525 1.5930 1.6335 1.6740 1.7145
15 1.6200 1.6686 1.7172 1.7658 1.8144 1.8630 1.9116 1.9602 2.0088 2.0574
16 2.1600 2.2248 2.2896 2.3544 2.4192 2.4840 2.5488 2.6136 2.6784 2.7432
17 1.4000 1.4420 1.4840 1.5260 1.5680 1.6100 1.6520 1.6940 1.7360 1.7780
18 2.1000 2.1630 2.2260 2.2890 2.3520 2.4150 2.4780 2.5410 2.6040 2.6670
19 1.8100 1.8643 1.9186 1.9729 2.0272 2.0815 2.1358 2.1901 2.2444 2.2987
20 0.0000 1.2900 1.3287 1.3674 1.4061 1.4448 1.4835 1.5222 1.5609 1.5996
21 0.0000 0.1600 0.1648 0.1696 0.1744 0.1792 0.1840 0.1888 0.1936 0.1984
(continued)
25
Table 1.1 (continued)
26

Node Stage
1 2 3 4 5 6 7 8 9 10
22 0.0000 1.6300 1.6789 1.7278 1.7767 1.8256 1.8745 1.9234 1.9723 2.0212
23 0.0000 0.0000 0.3400 0.3502 0.3604 0.3706 0.3808 0.3910 0.4012 0.4114
24 0.0000 0.0000 2.5100 2.5853 2.6606 2.7359 2.8112 2.8865 2.9618 3.0371
25 0.0000 0.0000 1.7200 1.7716 1.8232 1.8748 1.9264 1.9780 2.0296 2.0812
26 0.0000 0.0000 0.0000 1.4300 1.4729 1.5158 1.5587 1.6016 1.6445 1.6874
27 0.0000 0.0000 0.0000 1.6300 1.6789 1.7278 1.7767 1.8256 1.8745 1.9234
28 0.0000 0.0000 0.0000 1.2200 1.2566 1.2932 1.3298 1.3664 1.4030 1.4396
29 0.0000 0.0000 0.0000 0.0000 0.1600 0.1648 0.1696 0.1744 0.1792 0.1840
30 0.0000 0.0000 0.0000 0.0000 1.3500 1.3905 1.4310 1.4715 1.5120 1.5525
31 0.0000 0.0000 0.0000 0.0000 1.7900 1.8437 1.8974 1.9511 2.0048 2.0585
32 0.0000 0.0000 0.0000 0.0000 0.2300 0.2369 0.2438 0.2507 0.2576 0.2645
33 0.0000 0.0000 0.0000 0.0000 0.0000 1.4700 1.5141 1.5582 1.6023 1.6464
34 0.0000 0.0000 0.0000 0.0000 0.0000 1.6700 1.7201 1.7702 1.8203 1.8704
35 0.0000 0.0000 0.0000 0.0000 0.0000 2.0700 2.1321 2.1942 2.2563 2.3184
36 0.0000 0.0000 0.0000 0.0000 0.0000 1.2300 1.2669 1.3038 1.3407 1.3776
37 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.8200 0.8446 0.8692 0.8938
38 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 1.7100 1.7613 1.8126 1.8639
39 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.2400 0.2472 0.2544 0.2616
40 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.9400 0.9682 0.9964
41 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 1.3000 1.3390 1.3780
42 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 1.7400 1.7922 1.8444
(continued)
G. Muñoz-Delgado et al.
Table 1.1 (continued)
Node Stage
1 2 3 4 5 6 7 8 9 10
43 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 1.0900 1.1227 1.1554
44 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.4400 0.4532
45 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 1.1200 1.1536
46 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 1.3000 1.3390
47 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.2000 0.2060
48 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 1.0300
49 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 1.1700
50 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.5500
1 Distribution System Expansion Planning
27
28 G. Muñoz-Delgado et al.

Table 1.2 Operational conditions used for the deterministic model (pu)
Time Average demand Average wind speed Average solar irradiation
block factor factor factor
1 0.7418 0.3345 0.0734
2 0.5248 0.3279 0.2441
3 0.3914 0.3236 0.2216
4 0.3011 0.3177 0.0207

Table 1.3 Operational conditions used for the stochastic model (pu)
Time Average demand Average wind speed Average solar irradiation
block factor factor factor
1 0.83340 0.50535 0.22026
0.72168 0.32588 0.00000
0.67027 0.17218 0.00000
2 0.58940 0.49425 0.58421
0.51504 0.31629 0.14816
0.47014 0.17309 0.00000
3 0.42664 0.51211 0.54507
0.38973 0.31093 0.11964
0.35800 0.14763 0.00000
4 0.32606 0.53766 0.06212
0.30166 0.29549 0.00000
0.27546 0.11993 0.00000

dashed lines. Branch lengths are presented in Table 1.4. Network reconfiguration
under normal operation can be implemented by switching branches 10-31, 13-43,
23-24, 33-39, and 38-44. The capacity and unitary impedance of existing branches
are 6.28 MVA and 0.557 Ω/km, respectively. Table 1.5 presents the data for
candidate conductors in branches subject to replacement and in non-existing
branches. Note that two conductor alternatives are available per branch. Based on
[23], all branches are characterized by a maintenance cost equal to $450/year and a
25-year lifetime.
Existing substations, which are located at nodes 51 and 52 (Fig. 1.6), include a
12-MVA transformer characterized by an impedance equal to 0.16 Ω and a
maintenance cost equal to $2000. Nodes 53 and 54 represent the locations of
candidate substations. Voltages at substation nodes are set at 1.05 pu. Based on the
cost data reported in [33], the costs of the energy supplied by all substations, CbSS ,
are identical and equal to $225.33/MWh, $182.72/MWh, $154.43/MWh, and
$81.62/MWh, for time blocks 1–4, respectively. The cost of unserved energy, CU ,
is $1000/MWh. Investment decisions consist in (i) expanding existing substations
1 Distribution System Expansion Planning 29

Table 1.4 Branch lengths (km)

From To Length From To Length From To Length
1 2 0.66 12 45 1.33 30 43 1.47
1 9 0.86 13 43 1.07 30 54 1.02
1 51 1.11 14 15 1.81 31 37 0.45
3 4 0.90 14 46 1.31 32 39 1.46
3 51 2.06 14 50 2.25 33 34 0.81
4 5 1.45 14 52 2.21 33 39 1.19
4 7 1.24 15 16 0.91 34 35 0.76
5 6 0.81 16 40 1.29 35 36 0.45
6 28 1.55 17 18 1.83 36 53 1.28
7 8 1.00 18 19 0.68 37 43 1.01
8 25 0.79 18 21 0.98 38 39 1.19
8 27 1.60 19 20 0.96 38 44 1.27
8 33 1.92 21 54 0.58 40 41 1.39
9 17 1.61 22 23 1.85 41 42 1.52
9 22 2.08 22 54 1.89 41 53 1.73
9 23 1.36 23 24 0.82 42 47 1.82
10 23 1.89 24 25 0.89 42 48 1.77
10 31 0.92 26 27 0.68 44 45 1.02
11 12 1.42 27 28 1.15 46 47 1.29
11 52 1.50 28 53 1.64 48 49 1.58
12 13 1.70 29 30 1.17 49 50 0.92

Table 1.5 Data for candidate conductors

Type Alternative 1 Alternative 2
l Upper Impedance Investment Upper Impedance Investment
limit (MVA) (Ω/km) cost ($/km) limit (MVA) (Ω/km) cost ($/km)
NRB 9.00 0.478 19140 12.00 0.423 29,870
NAB 6.28 0.557 15020 9.00 0.478 25,030

by adding a new transformer and (ii) building a new substation from scratch.
Expansion costs of substations at nodes 51–54 are $100000, $100000, $150000,
and $150000, respectively. Data for candidate transformers are listed in Table 1.6,
where the same two alternatives are available for each substation. The lifetime of all
candidate transformers is 15 years. Moreover, it is assumed that gSS is considerably
larger than the lifetimes of the other distribution assets. Thus, RRSS ¼ I.
For the sake of simplicity, investment in renewable-based DG is only allowed,
with the penetration limit, n, set at 25%. Candidate nodes for installation of wind
generators are 3, 15, 23, 35, and 42. Candidate nodes for installation of PV gen-
erators are 4, 12, 24, 36, and 43. The economic and technical features of candidate
30 G. Muñoz-Delgado et al.

Table 1.6 Data for candidate transformers

Alternative 1 Alternative 2
Rated Impedance Maintenance Investment Rated Impedance Maintenance Investment
capacity (Ω) cost ($) cost ($) capacity (Ω) cost ($) cost ($)
(MVA) (MVA)
12 0.16 2000 750,000 15 0.13 3000 950,000

Table 1.7 Data for candidate DG units

Alternative k Wind generators Photovoltaic generators
Rated Investment Cost of Rated Investment Cost of
capacity (MVA) cost ($/MW) energy capacity (MVA) cost energy
supplied ($/MW) supplied
($/MWh) ($/MWh)
1 0.91 185,000 0 0.70 172,000 0
2 2.05 184,000 0 1.65 171,000 0

DG units are presented in Table 1.7, where two alternatives are considered for each
p
technology. Maintenance costs for DG units are set so that CkM;p ¼ 0:05CkI;p pf Gk .
Wind generation limits for the deterministic model, G ^ W , and for the stochastic
ikb
model, G ^ W ðxÞ, associated with wind speed levels, are determined using the data
ikb
available in [34] related to wind generators E44 and E82, which correspond to
alternatives 1 and 2, respectively. PV generation limits for the deterministic model,
^ H , and for the stochastic model, G
G ^ H ðxÞ, associated with solar irradiation levels,
ikb ikb
are determined using the data available in [35] related to 5000 solar panels of types
KD100-36 and KD300-80 within F Series, which correspond to alternatives 1 and
2, respectively. A 20-year lifetime is considered for all units.
Both models have been implemented on a Dell PowerEdge R920X64 with four
Intel Xeon E7-4820 processors at 2.00 GHz and 768 GB of RAM using CPLEX
12.6 [36] and GAMS 24.8 [37]. The stopping criterion for the branch-and-cut
algorithm of CPLEX is based on an optimality gap equal to 1%. Under this stop-
ping criterion, computing times were equal to 3.73 min for the deterministic model
and 9.88 h for the stochastic model.
The solutions to both cases are depicted in Fig. 1.7. As can be observed, both
expansion plans differ in both investment decisions and topology. The solution
provided by the stochastic model features a radial topology. In contrast, a meshed
solution was identified by the deterministic model, which is compliant with the
consideration of network reconfiguration under normal operation. Thus, both
solutions topologically differ in the branches connecting new load nodes to the
distribution system. For the solution attained by the deterministic model, the
demands at nodes 22 and 24 are supplied by substation 51 through branches 9-22
and 24-25, respectively. For the solution provided by the stochastic model, those
demands are respectively fed by substation 51 through branch 23-24 and by sub-
station 54 through branch 22-54. Additionally, for the solution to the deterministic
1 Distribution System Expansion Planning 31

(a) 20 TR1-T4
17 R2-T1 51 R1-T1
A1-T2 R2-T2 3
1
19 A1-T1 9 R2-T5
W2-T6 W
A1-T1 A1-T2 2
18 7 4
A1-T3 5
A1-T1
22 23 24 26
25 Θ2-T1
21 W 8 A1-T4
6
A1-T3
A1-T1 W2-T3 Θ2-T5
27 A1-T4
TR1-T1 54 32
A1-T5 A1-T10 10 A1-T6 28
A1-T5 A2-T6 A2-T6
31 33 34 A2-T4
A1-T5 35
30 A1-T7 36 TR2-T3
A1-T5 A1-T6
29 37 38 39 W2-T9 W Θ2-T8 53
A1-T5 A1-T9 40 A2-T8
43 A1-T8
41
A1-T5 A1-T10
44 15 16 A1-T8
13 Θ2-T7 45
W 42
A1-T9 W2-T10 R1-T1 A1-T9
12 A1-T9 A1-T10
R1-T9 11 14 46 47

Θ2-T2 R1-T1 R2-T1 48

A1-T10
50
52
A1-T10 49

(b) 20 TR1-T8
17 R2-T1 51 R2-T1
A1-T2 R2-T3 3
1
19 A1-T1 9 R2-T3
A1-T1 2
18 7 4
A1-T2 5
A2-T1
22 23 24 26
25 Θ2-T2
21 W 8 A1-T4
6
A1-T2 A1-T2
A2-T1 W2-T6 Θ2-T3
27 A1-T4
TR2-T1 54 32
A1-T5 10 A1-T6 28
A1-T5 A1-T6 A2-T6
31 33 34 A2-T4
A1-T5 35
30 36 TR2-T3
A1-T5 A1-T6
29 37 38 39
A1-T5 Θ2-T7 53
A1-T5 A1-T7 40 A2-T8
43 A1-T8
41
A1-T7
44 15 16 A1-T8
13 Θ2-T5 45
W 42
A1-T7 W2-T10 R1-T1 A1-T9
12 A1-T9 A1-T10
R1-T9 11 14 46 47

Θ2-T4 R2-T7 R2-T1 48

A1-T10
50
52
TR1-T9 A1-T10 49

Load node Substation node W Wind generator Photovoltaic generator

Branch Tt Stage t of installation TRk Alternative k for candidate transformer
Rk Alternative k in branch subject to replacement Ak Alternative k in prospective branch
Wk Alternative k for wind generator Θk Alternative k for photovoltaic generator

Fig. 1.7 Solutions: a Deterministic, b Stochastic

32 G. Muñoz-Delgado et al.

model, substation 52 supplies the demands at nodes 31, 37, and 43 at stages 5–9
through branches 13-43, 37-43, and 31-37, while substation 51 supplies those
demands at stage 10 through branches 10-31, 31-37, and 37-43. However, for the
solution to the stochastic model, those demands are supplied by substation 54
through branches 30-43, 37-43, and 31-37 at stages 5–10. Finally, additional dif-
ferences arise between both expansion plans related to the demand supply at load
nodes 32, 38, and 39.
Moreover, the installation of transformers at substation nodes differs in three
aspects. The installation of a transformer at substation node 51 takes place at
different stages. At substation node 52, a transformer is installed for the solution
attained by the stochastic model, while no transformer is installed for the deter-
ministic model. Furthermore, at substation node 54, another difference arises in the
transformer alternative being installed. Regarding DG investment, the expansion
plan obtained by the deterministic model includes the same generator types,
alternatives, and locations as those determined by the stochastic model with the
addition of a wind generator and a photovoltaic generator at nodes 3 and 35,
respectively. The installation times of generators also differ except for that installed
at node 15.
Finally, it is worth mentioning that the number of generators installed for the
solution obtained by the deterministic model is higher than that for the stochastic
solution. On the contrary, the solution provided by the stochastic model includes a
larger number of transformers with bigger capacity than those of the solution
obtained by the deterministic model.
Table 1.8 lists the present values of the different costs associated with the
solutions to both models. As can be seen, the consideration of uncertainty yields an
expected production cost that is greater than the production cost incurred by the
deterministic solution. However, this production cost increase is offset by the
reduction in costs related to investment, maintenance, losses, and unserved energy.
This case study is also useful to substantiate the use of both the approximate
network model and the piecewise linear approximation for quadratic energy losses.
In this regard, load flow results for the high load level at the last time stage provided
by the deterministic approach have been compared with those achieved with a full
ac load flow model. The average errors obtained for branch current flows, injections
at substations, and nodal voltage magnitudes are 0.72, −0.26, and 0.49%, respec-
tively. These results, which are consistent with the experience reported in [22],
validate the suitability of the linearized network model. Moreover, the quality of the
piecewise linearization for losses has been assessed by solving the deterministic
problem with quadratic losses wherein binary investment and utilization variables

Table 1.8 Present values of costs (106 $)

Case Investment Maintenance Production Losses Unserved energy cTPV
Deterministic 21.08 8.88 266.87 9.43 1.45 307.72
Stochastic 17.38 6.80 273.46 8.07 1.13 306.84
1 Distribution System Expansion Planning 33

were fixed to the values attained by the mixed-integer linear program with a
three-piece linearization. For this particular case study, the total cost only differs by
0.87%, thereby corroborating the appropriateness of the piecewise linear approxi-
mation in terms of solution quality.
Additionally, the convenience of using a stochastic solution approach rather than
a simpler deterministic one is analyzed. To that end, a widely-used metric, namely
the value of the stochastic solution (VSS), is calculated [29]. For a minimization
problem, the VSS is defined as the difference between two cost terms. The first
term, denoted by cDP , represents the value of the objective function obtained from
the stochastic model by fixing decision variables not depending on scenarios to the
values resulting from solving the associated deterministic problem. The second
term, denoted by cSP , represents the value of the objective function resulting from
the stochastic model. Thus, the VSS quantifies the potential gain associated with the
stochastic solution.
For the problem under consideration, cDP is equal to the value of cTPV obtained
from solving the stochastic model (1.74) and (1.75) by fixing decision variables
p p
xlijkt , xNT SS l tr
ikt , xikt , xit , yijkt , yikt , and yikt to the values resulting from solving the deter-
ministic model (1.53) and (1.54). For this particular case study, cDP is equal to
$311.49 million, whereas cSP is equal to the value of cTPV provided by the
stochastic model (1.74) and (1.75), which, as can be seen in Table 1.8, is equal to
$306.84 million. Thus, the VSS is equal to $4.56 million, which represents a
potential 1.5% improvement upon the deterministic solution.

Appendix

The symbols used throughout this chapter are listed below:

Indices
b Index for time blocks.
h Index for the blocks used in the piecewise linearization of energy losses.
i,j Indices for nodes.
k; j Indices for available investment alternatives.
l Index for branch types.
p Index for generator types.
s Index for segments of the cumulative distribution functions.
t; s Indices for time stages.
tr Index for transformer types.
x Index for scenarios.
34 G. Muñoz-Delgado et al.

Sets
B Index set of time blocks.
Kl Index set of available alternatives for branches of type l.
Kp Index set of available alternatives for generators of type p.
K tr Index set of available alternatives for transformers of type tr.
L Set of branch types. L ¼ fEFB, ERB, NRB, NABg where EFB, ERB, NRB,
and NAB denote existing fixed branch, existing replaceable branch, new
replacement branch, and newly added branch, respectively.
P Set of generator types. P ¼ fC , W, Hg where C, W, and H stand for
conventional, wind power, and photovoltaic generation, respectively.
T Index set of time stages.
TR Set of transformer types. TR ¼ fET , NTg where ET and NT denote existing
transformer and new transformer, respectively.
NDT Set of variables associated with the deterministic model.
NST Set of variables associated with the stochastic model.
!l Index set of branches of type l.
!SW;l Subset of !l comprising those branches that are switchable under normal
operation.
Wli Index set of nodes connected to node i by a branch of type l.
WLN
t
Index set of load nodes at stage t.
W N Index set of system nodes.
W p
Index set of candidate nodes for the installation of generators of type p.
WSS Index set of substation nodes.
Xb Set of scenarios for time block b.

Parameters
Alkh Width of block h of the piecewise linear energy losses for alternative k for
branches of type l.
Atrkh Width of block h of the piecewise linear energy losses for alternative k for
transformers of type tr.
CkE;p Cost coefficient for the energy supplied by alternative k for generators of
type p.
CkI;l Investment cost coefficient for alternative k for branches of type l.
CkI;NT Investment cost coefficient for alternative k for new transformers.
CkI;p Investment cost coefficient for alternative k for generators of type p.
CiI;SS Investment cost coefficient for the substation at node i.
CkM;l Maintenance cost coefficient for alternative k for branches of type l.
CkM;p Maintenance cost coefficient for alternative k for generators of type p.
CkM;tr Maintenance cost coefficient for alternative k for transformers of type tr.
1 Distribution System Expansion Planning 35

CbSS Cost coefficient for the energy supplied by substations for time block b.
CU Cost coefficient for unserved energy.
Dit Actual peak demand at node i and stage t.
~ it
D Fictitious peak demand at node i and stage t.
Fk
l Upper limit for the actual current flow through alternative k for branches
of type l.
p
Gk Rated capacity for alternative k for generators of type p.
^
G p Maximum power availability for alternative k for generators of type p at
ikb
node i and time block b.
^ p ðxÞ Maximum power availability for alternative k for generators of type p at
G ikb
node i, time block b, and scenario x.
tr
Gk Rated capacity of alternative k for transformers of type tr.
I Annual interest rate.
IBt Investment budget for stage t.
J Sufficiently large positive constant.
‘ij Length of the branch connecting nodes i and j.
l Slope of block h of the piecewise linear energy losses for alternative k for
Mkh
branches of type l.
tr
Mkh Slope of block h of the piecewise linear energy losses for alternative k for
transformers of type tr.
nB Number of time blocks.
nDG Number of candidate nodes for installation of distributed generation.
nH Number of blocks of the piecewise linear energy losses.
nT Number of time stages.
nX Number of scenarios per time block.
nDS Number of segments for demand factors at each time block.
npS Number of segments for factors for generation of type p at each time
block.
pf System power factor.
RRl Capital recovery rate for investment in branches of type l.
RRNT Capital recovery rate for investment in new transformers.
RRp Capital recovery rate for investment in generators of type p.
RRSS Capital recovery rate for investment in substations.
V Lower bound for nodal voltages.
V Upper bound for nodal voltages.
l Unitary impedance magnitude for alternative k for branches of type l.
Zk
Zktr Impedance magnitude for alternative k for transformers of type tr.
Db Duration of time block b.
gl Lifetime of branches of type l.
gNT Lifetime of new transformers.
gp Lifetime of generators of type p.
gSS Lifetime of substation assets other than transformers.
D Average demand factor of time block b.
lb
36 G. Muñoz-Delgado et al.

lDsb Average factor for demand in segment s of time block b.

lpsb Average factor for generation of type p in segment s of time block b.
lDb ðxÞ Average demand factor of time block b and scenario x.
n Penetration limit for distributed generation.
pDsb Probability of the average factor for demand in segment s of time block b.
ppsb Probability of the average factor for generation of type p in segment s of
time block b.
pb ðxÞ Probability of scenario x of time block b.

Variables
cEt Production cost at stage t.
cIt Amortized investment cost at stage t.
cMt Maintenance cost at stage t.
cRt Energy losses cost at stage t.
cUt Unserved energy cost at stage t.
cTPV Present value of the total cost.
U Unserved energy at node i, stage t, and time block b.
ditb
U
ditb ðx Þ Unserved energy at node i, stage t, time block b, and scenario x.
l
fijktb Actual current flow through alternative k for the branch of type
l connecting nodes i and j at stage t and time block b.
l
fijktb ðx Þ Actual current flow through alternative k for the branch of type
l connecting nodes i and j at stage t, time block b, and scenario x.
~f l Fictitious current flow through alternative k for the branch of type
ijkt
l connecting nodes i and j at stage t.
gpiktb Current injection at node i for alternative k for the generator of type p at
stage t and time block b.
gpiktb ðxÞ Current injection at node i for alternative k for the generator of type p at
stage t, time block b, and scenario x.
gtriktb Actual current injection at substation node i for alternative k for the
transformer of type tr at stage t and time block b.
gtriktb ðxÞ Actual current injection at substation node i for alternative k for the
transformer of type tr at stage t, time block b, and scenario x.
~gSS
it
Fictitious current injection at substation node i and stage t.
vitb Voltage magnitude at node i, stage t, and time block b.
vitb ðxÞ Voltage magnitude at node i, stage t, time block b, and scenario x.
xlijkt Binary investment variable for alternative k for the branch of type
l connecting nodes i and j at stage t.
xNT
ikt Binary investment variable for alternative k for the new transformer at
substation node i and stage t.
xpikt Binary investment variable for alternative k for the generator of type p at
node i and stage t.
1 Distribution System Expansion Planning 37

xSS
it
Binary investment variable for the substation at node i and stage t.
ylijkt Binary utilization variable for alternative k for the branch of type
l connecting nodes i and j at stage t.
ypikt Binary utilization variable for alternative k for the generator of type p at
node i and stage t.
ytrikt Binary utilization variable for alternative k for the transformer of type tr
at substation node i and stage t.
dlijktbh Current in block h of the piecewise linear energy losses for alternative
k for the branch of type l connecting nodes i and j at stage t and time
block b.
dlijktbh ðxÞ Current in block h of the piecewise linear energy losses for alternative
k for the branch of type l connecting nodes i and j at stage t, time block
b, and scenario x.
dtriktbh Current injection in block h of the piecewise linear energy losses for
alternative k for the transformer of type tr at substation node i, stage t,
and time block b.
dtriktbh ðxÞ Current injection in block h of the piecewise linear energy losses for
alternative k for the transformer of type tr at substation node i, stage t,
time block b, and scenario x:

References

1. A. Gómez-Expósito, A.J. Conejo, C. Cañizares, Electric Energy Systems. Analysis and

Operation (CRC Press, Boca Raton, FL, USA, 2009)
2. W.H. Kersting, Distribution System Modeling and Analysis, 3rd edn. (CRC Press, Boca
Raton, FL, USA, 2012)
3. H.L. Willis, Power Distribution Planning Reference Book, 2nd edn. (Marcel Dekker Inc, New
York, NY, USA, 2004)
4. P.S. Georgilakis, N.D. Hatziargyriou, A review of power distribution planning in the modern
power systems era: models, methods and future research. Electr. Power Syst. Res. 121, 89–
100 (2015)
5. N. Jenkins, R. Allan, P. Crossley, D. Kirschen, G. Strbac, Embedded Generation (The
Institution of Engineering and Technology, London, UK, 2000)
6. A. Keane, L.F. Ochoa, C.L.T. Borges, G.W. Ault, A.D. Alarcon-Rodriguez, R.A.F. Currie, F.
Pilo, C. Dent, G.P. Harrison, State-of-the-art techniques and challenges ahead for distributed
generation planning and optimization. IEEE Trans. Power Syst. 28(2), 1493–1502 (2013)
7. W. El-Khattam, Y.G. Hegazy, M.M.A. Salama, An integrated distributed generation
optimization model for distribution system planning. IEEE Trans. Power Syst. 20(2),
1158–1165 (2005)
8. R. Viral, D.K. Khatod, Optimal planning of distributed generation systems in distribution
system: a review. Renew. Sust. Energ. Rev. 16(7), 5146–5165 (2012)
9. H. Falaghi, C. Singh, M.-R. Haghifam, M. Ramezani, DG integrated multistage distribution
system expansion planning. Int. J. Electr. Power Energy Syst. 33(8), 1489–1497 (2011)
10. M. Gitizadeh, A.A. Vahed, J. Aghaei, Multistage distribution system expansion planning
considering distributed generation using hybrid evolutionary algorithms. Appl. Energy 101,
655–666 (2013)
38 G. Muñoz-Delgado et al.

11. M. Sedghi, M. Aliakbar-Golkar, M.-R. Haghifam, Distribution network expansion consid-

ering distributed generation and storage units using modified PSO algorithm. Int. J. Electr.
Power Energy Syst. 52, 221–230 (2013)
12. M. Shivaie, M.T. Ameli, M.S. Sepasian, P.D. Weinsier, V. Vahidinasab, A multistage
framework for reliability-based distribution expansion planning considering distributed
generations by a self-adaptive global-based harmony search algorithm. Reliab. Eng. Syst. Saf.
139, 68–81 (2015)
13. A. Tabares, J.F. Franco, M. Lavorato, M.J. Rider, Multistage long-term expansion planning of
electrical distribution systems considering multiple alternatives. IEEE Trans. Power Syst. 31
(3), 1900–1914 (2016)
14. C.L.T. Borges, V.F. Martins, Multistage expansion planning for active distribution networks
under demand and distributed generation uncertainties. Int. J. Electr. Power Energy Syst. 36
(1), 107–116 (2012)
15. A. Bagheri, H. Monsef, H. Lesani, Renewable power generation employed in an integrated
dynamic distribution network expansion planning. Electr. Power Syst. Res. 127, 280–296
(2015)
16. R. Hemmati, R.-A. Hooshmand, N. Taheri, Distribution network expansion planning and DG
placement in the presence of uncertainties. Int. J. Electr. Power Energy Syst. 73, 665–673
(2015)
17. H. Arasteh, M.S. Sepasian, V. Vahidinasab, P. Siano, SoS-based multiobjective distribution
system expansion planning. Electr. Power Syst. Res. 141, 392–406 (2016)
18. S.F. Santos, D.Z. Fitiwi, M. Shafie-Khah, A.W. Bizuayehu, C.M.P. Cabrita, J.P.S. Catalão,
New multistage and stochastic mathematical model for maximizing RES hosting capacity–
part I: problem formulation. IEEE Trans. Sustain. Energy 8(1), 304–319 (2017)
19. S.F. Santos, D.Z. Fitiwi, M. Shafie-Khah, A.W. Bizuayehu, C.M.P. Cabrita, J.P.S. Catalão,
New multi-stage and stochastic mathematical model for maximizing RES hosting capacity–
part II: numerical results. IEEE Trans. Sustain. Energy 8(1), 320–330 (2017)
20. G. Muñoz-Delgado, J. Contreras, J.M. Arroyo, Joint expansion planning of distributed
generation and distribution networks. IEEE Trans. Power Syst. 30(5), 2579–2590 (2015)
21. G. Muñoz-Delgado, J. Contreras, J.M. Arroyo, Multistage generation and network expansion
planning in distribution systems considering uncertainty and reliability. IEEE Trans. Power
Syst. 31(5), 3715–3728 (2016)
22. S. Haffner, L.F.A. Pereira, L.A. Pereira, L.S. Barreto, Multistage model for distribution
expansion planning with distributed generation–part I: problem formulation. IEEE Trans.
Power Deliv. 23(2), 915–923 (2008)
23. R.C. Lotero, J. Contreras, Distribution system planning with reliability. IEEE Trans. Power
Deliv. 26(4), 2552–2562 (2011)
24. L. Blank, A. Tarquin, Engineering Economy, 7th edn. (McGraw-Hill, New York, NY, USA,
2012)
25. P.C. Paiva, H.M. Khodr, J.A. Domínguez-Navarro, J.M. Yusta, A.J. Urdaneta, Integral
planning of primary-secondary distribution systems using mixed integer linear programming.
IEEE Trans. Power Syst. 20(2), 1134–1143 (2005)
26. M. Lavorato, J.F. Franco, M.J. Rider, R. Romero, Imposing radiality constraints in
distribution system optimization problems. IEEE Trans. Power Syst. 27(1), 172–180 (2012)
27. S.P. Bradley, A.C. Hax, T.L. Magnanti, Applied Mathematical Programming
(Addison-Wesley, Reading, MA, USA, 1977)
28. S. Binato, M.V.F. Pereira, S. Granville, A new Benders decomposition approach to solve
power transmission network design problems. IEEE Trans. Power Syst. 16(2), 235–240
(2001)
29. J.R. Birge, F. Louveaux, Introduction to Stochastic Programming, 2nd edn. (Springer, New
York, NY, USA, 2011)
30. L. Baringo, A.J. Conejo, Wind power investment within a market environment. Appl. Energy
88(9), 3239–3247 (2011)
1 Distribution System Expansion Planning 39

31. V. Miranda, J.V. Ranito, L.M. Proença, Genetic algorithms in optimal multistage distribution
network planning. IEEE Trans. Power Syst. 9(4), 1927–1933 (1994)
32. Y.M. Atwa, E.F. El-Saadany, M.M.A. Salama, R. Seethapathy, Optimal renewable resources
mix for distribution system energy loss minimization. IEEE Trans. Power Syst. 25(1), 360–
370 (2010)
33. Red Eléctrica de España (2017) [Online], Available: https://www.esios.ree.es/en
34. ENERCON, ENERCON wind energy converters: products overview (July 2010) [Online],
Available: http://www.enercon.de
35. KYOCERA SOLAR Europe (2017) [Online], Available: http://www.kyocerasolar.eu
36. IBM ILOG CPLEX (2017) [Online], Available: https://www.ibm.com/analytics/data-science/
prescriptive-analytics/cplex-optimizer
37. GAMS Development Corporation (2017) [Online], Available: http://www.gams.com
Chapter 2
Static and Dynamic Convex Distribution
Network Expansion Planning

Julio López, David Pozo and Javier Contreras

Abstract This chapter presents static and dynamic optimization-based models for
planning the electric distribution network. Based on a branch flow model, two
Mixed-Integer Conic Quadratic Programming (MICQP) convex formulations are
proposed to solve the network expansion planning models including high modeling
fidelity of the intrinsic interaction of the manifold elements of the networks. The
objective of the presented models is to minimize investment and operation costs by
optimally deciding on installing new feeders and/or changing existing ones for
others with larger capacities, installing new substations or expanding existing ones
and, finally, installing capacitor banks and voltage regulators, modifying the net-
work topology. In addition, discrete tap settings of voltage regulators are modeled
as a set of mixed-integer linear equations, which are embedded in an ac optimal
power flow. The presented MICQP models are convex optimization problems.
Therefore globality and convergence are guaranteed. Computational results to
verify the efficiency of the proposed methodology are obtained for a 24-node test
system. Finally, conclusions are duly drawn.

Keywords Capacitor banks Convex optimization Dynamic model



Electric distribution network expansion planning Static models
Voltage regulators

J. López (&)
Electrical Engineering School, University of Cuenca, Cuenca, Ecuador
e-mail: [email protected]
D. Pozo
Center of Energy Systems, Skolkovo Institute of Science and Technology,
Moscow, Russia
e-mail: [email protected]
J. Contreras
Escuela Técnica Superior de Ingenieros Industriales,
Universidad de Castilla – La Mancha, Ciudad Real, Spain
e-mail: [email protected]

© Springer Nature Singapore Pte Ltd. 2018 41

F. Shahnia et al. (eds.), Electric Distribution Network Planning,
Power Systems, https://doi.org/10.1007/978-981-10-7056-3_2
42 J. López et al.

2.1 Introduction

The rapid increase of renewable power generation connected to electric distribution

networks has complicated their operation. Because this operational complexity, it
could lead to a high impact on the economic efficiency of networks due to the
significant investment costs of new control devices that properly guarantee
appropriate levels of security, quality, and reliability at competitive costs. In this
vein, optimization tools to solve the Electric Distribution Network Expansion
Planning (EDNEP) problem has recently attracted more attention representing a
shifting toward feasible-based to optimization-based planning paradigms. It is clear
that the use of optimization tools in EDNEP represent substantial gains or savings
in planning electric distribution networks. However, it is essential to properly
capture the complexity of the non-linear interactions of the manifolds elements and
physic laws with high fidelity. This chapter is devoted to this propose presenting
convex formulations that could be implemented on off-the-shelf solvers with
globality and convergence guarantee.
In its simplest version, the EDNEP problem consists of determining the
investments that guarantee an economical and reliable distribution network oper-
ation. Technical constraints such as maximum current flows through feeders,
maximum power from substation transformers, voltage magnitude limits in nodes
and network radiality must be considered [1, 2]. The EDNEP problem can be
established as follows: an electric distribution network needs to meet the demands
of a fixed number of consumers due to demand growth, hence, it is necessary to
carry out expansion planning consisting of: installing new feeders and/or changing
existing ones for others with larger capacities, installing new substations or
expanding existing ones and finally, installing capacitor banks and voltage regu-
lators. The objective is to minimize the total costs related to those investments and
network operation costs, subject to a set of physical, operating and economic
constraints [3].
The installation of capacitor banks in electric distribution networks is important,
mainly to maintain the voltage magnitude and energy losses within pre-established
limits. Their optimal sizes and locations make these improvements feasible [4].
Thus, optimal capacitor bank placement aims at placing and sizing them, mini-
mizing the costs associated with capacitor banks and energy losses. Within this
context, another important aspect is the location of voltage regulators [5].
Some works about the EDNEP problem have independently addressed feeder
and substation installation [7–13, 17–20], allocation of capacitor banks [5, 21–23],
allocation of voltage regulators [24–26], and joint allocation of capacitor banks and
voltage regulators [27–30]. However, the EDNEP problem should focus not only
on a single technology–or device–based planning. In this context, an integral
EDNEP problem considering a co-optimization of all the above control devices
could foster benefits to the electric distribution network by reducing operation costs
and losses, increasing flexibility and reliability is proposed.
2 Static and Dynamic Convex Distribution … 43

However, this problem becomes a large-scale mixed-integer non-linear pro-

gramming (MINLP) problem. The EDNEP problem has been solved using different
techniques, such as heuristic and meta-heuristic algorithms and classical opti-
mization techniques [6]. Heuristic algorithms have produced solutions with a rel-
atively low computational effort, like branch-exchange in [7, 8] and the constructive
heuristic algorithm in [9]. Meta-heuristic algorithms have also been used, like
evolutionary algorithms in [10], genetic algorithms in [11] and [12], ant colony
algorithms in [13], simulated annealing algorithms in [14, 15] and particle swarm
algorithm in [16]. Although metaheuristics are flexible and achieve good results,
they also present many problems, such as a high computational demand, the need
for adjusting and fine-tuning the parameters and the definition of a stopping cri-
terion. In addition, they cannot guarantee convergence to a global optimum, or
indicate the quality of the final solution, because they do not provide a distance
indicator to the optimal solution.

2.2 Time Framework

An important aspect to be considered in the optimal EDNEP problem is the

decision-making process in the planning horizon. According to [31], the EDNEP
problem can be divided into two periods: short-term planning (1 up to 4 years) and
long-term planning (5 up to 20 years), leading to two types of EDNEP optimization
models, static and dynamic.
In a static model, EDNEP decisions are only made at the beginning of the
planning horizon, i.e., at a single point in time, where the load demand data con-
sidered remains constant until the end of the planning horizon. This modeling type
is known as a single-stage model as well and considers the whole planning horizon
in a single period, which is the target period. Since the EDNEP in mainly condi-
tioned by the load demand in the electric distribution network, which usually
increases over time, the reference period is usually selected as the last year of the
planning horizon.
In a dynamic model, EDNEP decisions are made at different points in time. This
modeling type is known as multi-period as well and represents the real behavior of
the electric distribution network. In this approach, the planning horizon is divided
into different time periods, each one comprising a specific number of years [32, 33].
The advantage of using a static approach for the EDNEP problem is that the
resulting model is relatively simple. One of the disadvantages is that the EDNEP is
solved for the last year of the planning horizon. Another drawback is that, if the
EDNEP problem is solved for a long-term planning horizon, the load demand in the
electric distribution network will probably be much higher than the load demand in
the short-term. Therefore, the EDNEP will probably result in an oversize of the
installed components and higher investments.
44 J. López et al.

2.3 AC Power Flow in Electric Distribution Networks

The analysis of an electric distribution network requires the solution of the power
flow problem to calculate the state of the system represented by voltage magnitudes
in nodes, current flows in feeders, energy losses and other variables of interest.
Therefore, power flow models are widely tools used in the steady-state analysis of
the networks. Most ac power flow models in electric distribution networks are
based on power-mismatch and current-mismatch formulations either in polar or
rectangular formats, mainly using Newton-Raphson algorithms [35]. On the other
hand, radial networks are characterized by a high R/X ratio. This renders the load
flow problem ill-conditioned. Previous research indicates that standard load flow
methods fail to converge in ill-conditioned test systems [36, 37].
In this work, the equations that represent the steady-state of radial networks are
obtained from the branch flow model proposed in [38–40] as
X  X
Pk ¼ Pkm þ Rkm Ikm
2
 Pjk 8k 2 B ð2:1Þ
j2aðkÞ j2aðkÞ

X  X
Qk ¼ Qkm þ Xkm Ikm
2
 Qjk 8k 2 B ð2:2Þ
j2aðkÞ j2aðkÞ

 2
Vk2  Vm2 ¼ 2ðRkm Pkm þ Xkm Qkm Þ  R2km þ Xkm
2
Ikm 8km 2 BR ð2:3Þ

Vm2 Ikm
2
¼ P2km þ Q2km 8km 2 BR ð2:4Þ

where constraints (2.1) and (2.2) are the active and reactive power injections; (2.3)
describes the forward voltage drop in each line and (2.4) defines apparent power
flow injection at the head bus of each line. Equations (2.1)–(2.4) are frequently
used in the power flow sweep method of radial networks and can be used to
formulate the MINLP model for the EDNEP problem.
Without loss generality, the power flow optimization problem can be formulated
using the above steady-state equations of radial networks, including an objective
function that minimizes real power loss [27]. The compact form of the non-linear ac
power flow problem can be expressed as
P
min 2
Rkm Ikm 8km 2 BR
j2aðkÞ
subject to:
constraints (2:1Þ  ð2:4Þ ð2:5Þ
Pk ¼ PSE
k  Pk
D
8km 2 B
Qk ¼ QSEk  Q D
k 8km 2 B

where the two last equations are the nodal active and reactive power balancing
conditions, respectively.
2 Static and Dynamic Convex Distribution … 45

The optimization problem (2.5) is a non-linear and non-convex in the ac power

flow for radial networks due to the quadratic terms in their constraints and objective
function. However, the square terms in (2.1)–(2.4) and the objective function in
(2.5) can be dropped using auxiliary variables that represent those square terms,
dk ¼ Vk2 and lkm ¼ Ikm
2
, and (2.4) can be relaxed to a convex constraint by second
order conic programming (SOCP) [41], where it is relaxed to an inequality con-
straint. The relaxed convex ac power flow problem for radial networks can be
expressed as
X
min Rkm lkm ð2:6Þ
j2aðkÞ

X X
k  Pk ¼
PSE ðPkm þ Rkm lkm Þ  8k 2 B ð2:7Þ
D
Pjk
j2aðkÞ j2aðkÞ

X X
k  Qk ¼
QSE ðQkm þ Xkm lkm Þ  8k 2 B ð2:8Þ
D
Qjk
j2aðkÞ j2aðkÞ

 
dk  dm ¼ 2ðRkm Pkm þ Xkm Qkm Þ  R2km þ Xkm
2
lkm 8km 2 BR ð2:9Þ

dm lkm  P2km þ Q2km 8km 2 BR ð2:10Þ

2.4 Convex Model for the EDNEP Problem

In this section, an optimization model is presented that includes the minimization of

investment costs by installing new feeders and/or changing existing ones for others
with larger capacities, installing new substations or expanding existing ones and
finally, installing capacitor banks and voltage regulators, as well as the timing to
add new assets or expand existing ones in case of dynamic (multi-stage) planning,
including the operating costs associated with energy loss. Equations related to each
device considered are depicted to capture the physical laws that govern them. Then,
a convex formulation is derived to build a MICQP optimization problem that
benefits from the advances of off-the-shelf MICQP solvers. It is worth to clarify that
Mixed-Integer Conic Quadratic Programming (MICQP) problems are non-convex
because of the integer nature of some decision variables. However, MICQP reso-
lution algorithms iteratively solve optimization problems where integer variables
are fixed. In this case, our problem is convex. Additionally, in this case, MICQP
theory guarantees global solution of the optimization problem. For further details
about Conic Programming theory, interested readers are referred to [42].
46 J. López et al.

2.4.1 Capacitor Bank Model

In electric distribution networks, there are two types of capacitor banks: (1) Fixed
Capacitor Banks (FCB) and (2) Switched Capacitor Banks (SCB) [43], as shown in
Fig. 2.1. FCBs are capacitor banks composed of units which, after being installed in
the planning stage, are always connected throughout all load levels; whereas, SCBs
are composed of units which, after being installed at the planning stage, can be fully
or partially connected at every load level. According to Fig. 2.1a, expressions
(2.11)–(2.13) model the allocation and operation of FCBs in the networks.

qfcb fc fc
k ¼ Q nk 8k 2 FCB ð2:11Þ

0  nfck  Nkfcb 8k 2 FCB ð2:12Þ

nfck 2 Z þ 8k 2 FCB ð2:13Þ

Equation (2.11) represents the reactive power produced by the fixed capacitor
banks installed at node k. Constraint (2.12) limits the number of units to be installed

Fig. 2.1 Capacitor bank schemes: a FCB. b SCB

2 Static and Dynamic Convex Distribution … 47

in a fixed capacitor bank and (2.13) establishes the integrality condition for the
number of units installed.
To model SCBs (Fig. 2.1b), it is important to take into account the load varia-
tions in a given period of time, therefore, these equations are modeled considering
the time period and can only be included in the dynamic model. Expressions
(2.14)–(2.16) represent the SCBs installation and operation.

k;t ¼ Q nk;t
qscb 8k 2 SCB; 8t 2 P ð2:14Þ
sc sc

k;t  Nk
0  nsc scb
8k 2 SCB; 8t 2 P ð2:15Þ

þ
k;t 2 Z
nsc 8k 2 SCB; 8t 2 P ð2:16Þ

where (2.14) represents the reactive power produced by the fixed capacitors banks
installed at node k. Constraint (2.15) limits the number of units to be installed in a
fixed capacitor bank and (2.16) establishes the integrality condition for the number
of units installed.

2.4.2 Voltage Regulator Model

To model the voltage regulator, consider an autotransformer with an automatic

changing mechanism of the tap position (number of turns) of the series winding to
maintain a predetermined level of voltage magnitude along an electric distribution
feeder in case of load level variations. Standard voltage regulators contain a
reversing switch that enables a regulating range, which determines the tap step-size
D (+ increases and – decreases the voltage magnitude), taking into account the
reference voltage magnitude and the maximum number of steps, Ns, as shown in
Fig. 2.2, where D and Ns are known parameters, e.g. 0.00625, 32 steps, respec-
tively, a and tp are the tap setting and tap position, respectively, which are con-
sidered variables in the planning and operation of electric distribution networks.

Fig. 2.2 Voltage regulator in operation

48 J. López et al.

Fig. 2.3 Feeder with a voltage regulator model

Consider the voltage regulator located at node m of branch km, where node x is
the non-regulated voltage magnitude node, as shown in Fig. 2.3, where a partial
electric distribution feeder with a voltage regulator is shown. The voltage regulator
in branch km can be divided into branch kx and branch xm, where branch xm only
contains the tap changer and the impedance kx is the same as branch km. With these
considerations, similar to DistFlow [38], the injected active and reactive power
equations of the voltage regulator can be described as
X  X
Pk ¼ Pkm þ Rkm Ikm
2
 Pjk 8k 2 B ð2:17Þ
j2aðkÞ j2aðkÞ

X  X
Qk ¼ Qkm þ Xkm Ikm
2
 Qjk 8k 2 B ð2:18Þ
j2aðkÞ j2aðkÞ

 2
Vk2  Vx2 ¼ 2ðRkm Pkm þ Xkm Qkm Þ  R2km þ Xkm
2
Ikm 8km 2 VR ð2:19Þ

Vm2 Ikm
2
¼ P2km þ Q2km 8km 2 VR ð2:20Þ

Vm2 ¼ a2km Vx2 8km 2 VR ð2:21Þ

where constraints (2.17) and (2.18) are the active and reactive power injections;
(2.19) describes the forward voltage drop in each line, (2.20) defines the apparent
power flow injection at the head bus of each line and (2.21) is the voltage mag-
nitude regulated by the voltage regulator. Equations (2.17)–(2.20) can be convex-
ified in a similar form to the ac power flow in Sect. 2.3.
Equation (2.21) can be transformed into several mixed integer linear expres-
sions. In order to do so, the voltage regulator tap-changer is formulated as

akm ¼ 1 þ Dkm tpkm 8km 2 VR ð2:22Þ

 max 
akm  amin
Dkm ¼ km
8km 2 VR ð2:23Þ
Nskm
2 Static and Dynamic Convex Distribution … 49

Nskm vr Nskm vr
 a  tpkm  a 8km 2 VR ð2:24Þ
2 km 2 km
tpkm 2 Z 8km 2 VR ð2:25Þ

where (2.22) is the expression that determines the tap setting, (2.23) computes the
tap step-size, (2.24) limits the number of tap positions and states if it is necessary to
install a voltage regulator in the electric distribution network and (2.25) establishes
the integrality condition to the number of tap positions.
Integer variable tpkm can be expressed by the binary expansion scheme [44] as

Nskm 
X  
Nskm
tpkm ¼ n akmn 8km 2 VR ð2:26Þ
n¼0
2

X
Nskm
akmn ¼ 1 8km 2 VR ð2:27Þ
n¼0

where akmn 2 f0=1g. Replacing (2.26) in (2.22), yields

Nskm 
X  
Nskm
akm ¼ 1 þ Dkm n akmn km 2 VR ð2:28Þ
n¼0
2

Using the auxiliary variable dk ¼ Vk2 in (2.21), one obtains

dm ¼ a2km dx 8km 2 VR ð2:29Þ

Multiplying both sides of (2.28) by dx , results in

Nskm 
X  
Nskm
akm dx ¼ dx þ Dkm n akmn dx 8km 2 VR ð2:30Þ
n¼0
2

Defining auxiliary variables zkm ¼ akm dx and xkmn ¼ akmn dx , one obtains

Nskm 
X  
Nskm
zkm ¼ dx þ Dkm n xkmn 8km 2 VR ð2:31Þ
n¼0
2

From (2.29), dm ¼ akm zkm , and replacing (2.31) yields

Nskm 
X  
Nskm
dm ¼ zkm þ Dkm n akmn zkm 8km 2 VR ð2:32Þ
n¼0
2

Defining the auxiliary variable ykmn ¼ akmn zkm , one obtains

50 J. López et al.

Nskm 
X  
Nskm
dm ¼ zkm þ Dkm n ykmn 8km 2 VR ð2:33Þ
n¼0
2

Auxiliary variables xkmn and ykmn are the products of binary and continuous
variables, which can be reformulated into an exact linear set of constraints using the
big-M method, resulting in

Makmn  xkmn  Makmn 8km 2 VR; 8n ¼ 0; 1; 2; . . .; Nskm ð2:34Þ

M ð1  akmn Þ  dx  xkmn  M ð1  akmn Þ 8km 2 VR; 8n ¼ 0; 1; 2; . . .; Nskm

ð2:35Þ

Makmn  ykmn  Makmn 8km 2 VR; 8n ¼ 0; 1; 2; . . .; Nskm ð2:36Þ

M ð1  akmn Þ  zkm  ykmn  M ð1  akmn Þ 8km 2 VR; 8n ¼ 0; 1; 2; . . .; Nskm

ð2:37Þ

2.4.3 Static Convex EDNEP Model

As previously explained, the static model considers the whole planning horizon in a
single target period. In addition, the steady state, capacitor banks and voltage
regulators convex formulations previously developed are taken into account. The
static convex EDNEP model can be formulated as
X X X 
min f ¼ KL C
Ckm;c aCkm;c Lkm þ KS CkSE aSE
k þ C k ak
SER SER

km2BR c2C k2SE

X X
þ KC CkFCB nfck þ KVR CkVR avr
km ð2:38Þ
k2FCB km2VR
X  X h 2  SE 2 i
þ 8760Closs k  Pk þ 8760
PSE D OS
Ck PSE
k þ Qk
k2B k2B
X X  X X
k  Pk ¼
PSE Pkm;c þ Rkm;c lkm;c  8k 2 B ð2:39Þ
D
Pjk;c
km2BR c2C jk2BR c2C

fcb
X X  X X
k  Qk þ qk ¼
QSE Qkm;c þ Xkm;c lkm;c  8k 2 B
D
Qjk;c
km2BR c2C jk2BR c2C

ð2:40Þ
2 Static and Dynamic Convex Distribution … 51
Xh    i
dk  M ð1  akm Þ þ dm  2 Rkm;c Pkm;c þ Xkm;c Qkm;c  R2km;c þ Xkm;c
2
lkm;c
c2C
8km 2 BR
ð2:41Þ
Xh    i
dk   M ð1  akm Þ þ dm  2 Rkm;c Pkm;c þ Xkm;c Qkm;c  R2km;c þ Xkm;c
2
lkm;c
c2C
8km 2 BR
ð2:42Þ

dm lkm  P2km þ Q2km 8km 2 BR ð2:43Þ

 min  max
Vk2  dk  Vk2 8km 2 B ð2:44Þ
 max
0  lkm;c  Ikm;c
2
aCkm;c 8km 2 BR, 8c 2 C ð2:45Þ
X
bkm ¼ aCkm;c  1 8km 2 BR ð2:46Þ
c2C
X X
bkm ¼ jBj  jSE0j  aSE
k ð2:47Þ
km2BR k2SE

 2  2  SE0 2 h SE0 SE  SE 2 i SE
PSE
k þ QSE
k  Sk þ 2Sk Sk þ Sk ak 8k 2 SE0 [ SE ð2:48Þ

Nskm 
X  
Nskm vr Nskm Nskm vr
 a  n ykmn  a 8km 2 VR ð2:49Þ
2 km n¼0 2 2 km

Constraints (2:11Þð2:13Þ; ð2:23Þ; ð2:27Þ; ð2:31Þ; ð2:33Þ; ð2:34Þð2:37Þ ð2:50Þ

aCkm;c 2 f0; 1g 8km 2 BR; 8c 2 C ð2:51Þ

k 2 f0; 1g
aSE 8km 2 SE ð2:52Þ

akm 2 f0; 1g 8km 2 BR ð2:53Þ

avr
km 2 f0; 1g 8km 2 VR ð2:54Þ

where objective function (2.38) represents the annualized investment and operation
costs. The first term represents the costs associated with the new feeders installation
and/or changing existing ones, the second term represents the costs of installing or
expanding substations, the third and fourth terms represent the costs associated with
52 J. López et al.

installing fixed capacitor banks and voltage regulators, respectively, and the fifth
and sixth terms represent the costs associated with energy losses and substation
operation, respectively.
Equations (2.39) and (2.40) represent the active and reactive power balance
nodal equations. Equations (2.41) and (2.42) describe the forward voltage drops in
each branch while (2.43) defines the apparent power flow injection at the head bus
of each branch. The limits of voltage and current magnitudes are established in
(2.44) and (2.45). Equation (2.46) accounts for the case when a line with a specific
conductor type is already installed, by setting the corresponding binary variable to
1, or the case when a line is already installed but can have its conductor type
changed by replacing the inequality sign in the summation of (2.46) with an
equality. Equation (2.47), together with power balance constraints (2.39) and
(2.40), ensures that the network has a number of trees equal to the number of
existing and installed substations. Equation (2.48) indicates substation installation
ðSSE0
k ¼ 0 and SSE
k 6¼ 0Þ and substation expansion ðSk
SE0
6¼ 0 and SSE
k 6¼ 0Þ and
binary variable aSE
k is related to the substation capacity limit. Equation (2.49) states
whether it is necessary to install a voltage regulator in the electric distribution
km ¼ 1. Finally, constraints (2.51)–(2.54) represent the binary nature
network, i.e.avr
of the variables. The static model in (2.38)–(2.54) is a convex MICQP formulation
for the EDNEP problem that guarantees global optimum solutions.

2.4.4 Dynamic Convex EDNEP Model

As previously explained, in the EDNEP dynamic model, the decisions are made at
different points in time. In order to do so, the proposed static model in (2.38)–(2.54)
can be extended to a dynamic (multi-period) model by adding the period index and
some constraints, as
X X X XX
min f ¼ KL C
Ckm;c aCkm;c;t Lkm þ KS CkSE aSE
k;t þ Ck ak;t
SER SER

t2P km2BR c2C t2P k2SE

X X X X X X
þ KC CkFCB nfck;t þ KC k;t þ KVR
CkSCB nsc CkVR avr
km;t
t2P k2FCB t2P k2FCB t2P km2VR
XX XX   
2 2
þ Closs k;t  Pk;t þ
PSE D
CkOS PSE
k;t þ QSE
k;t
t2P k2B t2P k2B
ð2:55Þ
X X  X X
k;t  Pk;t ¼
PSE Pkm;c;t þ Rkm;c lkm;c;t  8k 2 B; 8t 2 P
D
Pjk;c;t
km2BR c2C jk2BR c2C

ð2:56Þ
2 Static and Dynamic Convex Distribution … 53
fcb
X X  X X
k;t 
QSE k;t þ qk;t
QD þ qscb
k;t ¼ Qkm;c;t þ Xkm;c lkm;c;t  Qjk;c;t
km2BR c2C jk2BR c2C

8k 2 B; 8t 2 P
ð2:57Þ
  Xh    i
dk;t  M 1  akm;t þ dm;t  2 Rkm;c Pkm;c;t þ Xkm;c Qkm;c;t  R2km;c þ Xkm;c
2
lkm;c;t
c2C
8km 2 BR; 8t 2 P
ð2:58Þ
  Xh    i
dk;t   M 1  akm;t þ dm;t  2 Rkm;c Pkm;c;t þ Xkm;c Qkm;c;t  R2km;c þ Xkm;c
2
lkm;c;t
c2C
8km 2 BR; 8t 2 P
ð2:59Þ

dm;t lkm;t  P2km;t þ Q2km;t 8km 2 BR; 8t 2 P ð2:60Þ

 2  2
Vkmin  dk;t  Vkmax 8km 2 B,8t 2 P ð2:61Þ
 2
0  lkm;c;t  Ikm;c
max
aCkm;c;t 8km 2 BR; 8c 2 C; 8t 2 P ð2:62Þ
X
bkm;t ¼ aCkm;c;t  1 8km 2 BR; 8t 2 P ð2:63Þ
c2C
X X
bkm;t ¼ jBj  jSE0j  aSE
k;t 8t 2 P ð2:64Þ
km2BR k2SE

 2  2  2 h SE0 SE  SE 2 i SE
PSE
k;t þ QSE
k;t  SSE0
k þ 2Sk Sk þ Sk ak;t 8k 2 SE0 [ SE; 8t 2 P
ð2:65Þ

qfcb fc fc
k;t ¼ Q nk;t 8k 2 FCB; 8t 2 P ð2:66Þ

0  nfck;t  Nkfcb 8k 2 FCB; 8t 2 P ð2:67Þ

k;t ¼ Q nk;t
qscb 8k 2 SCB; 8t 2 P ð2:68Þ
sc sc

k;t  Nk
0  nsc scb
8k 2 SCB; 8t 2 P ð2:69Þ
 
km  akm
amax min
Dkm ¼ 8km 2 VR ð2:70Þ
Nskm
54 J. López et al.

Nskm 
X  
Nskm
zkm;t ¼ dx;t þ Dkm n xkmn;t 8km 2 VR,8t 2 P ð2:71Þ
n¼0
2

Nskm 
X  
Nskm
dm;t ¼ zkm;t þ Dkm n ykmn;t 8km 2 VR; 8t 2 P ð2:72Þ
n¼0
2

Makmn;t  xkmn;t  Makmn;t 8km 2 VR; 8t 2 P; 8n ¼ 0; 1; 2; . . .; Nskm ð2:73Þ

   
 M 1  akmn;t  dx;t  xkmn;t  M 1  akmn;t
ð2:74Þ
8km 2 VR; 8t 2 P; 8n ¼ 0; 1; 2; . . .; Nskm

Makmn;t  ykmn;t  Makmn;t 8km 2 VR; 8t 2 P; 8n ¼ 0; 1; 2; . . .; Nskm ð2:75Þ

   
M 1  akmn;t  zkm;t  ykmn;t  M 1  akmn;t
ð2:76Þ
8km 2 VR; 8t 2 P; 8n ¼ 0; 1; 2; . . .; Nskm

X
Nskm
akmn;t ¼ 1 8km 2 VR,8t 2 P ð2:77Þ
n¼0

Nskm 
X  
Nskm vr Nskm Nskm vr
 a  n ykmn;t  a 8km 2 VR; 8t 2 P ð2:78Þ
2 km;t n¼0 2 2 km;t
X
aCkm;c;t ¼ 1 8t 2 P ð2:79Þ
t2P
X
aSE
k;t ¼ 1 8t 2 P ð2:80Þ
t2P
X
avr
km;t ¼ 1 8t 2 P ð2:81Þ
t2P

þ
k;t 2 Z
nsc 8k 2 SCB; 8t 2 P ð2:82Þ

nfck;t 2 Z þ 8k 2 FCB,8t 2 P ð2:83Þ

aCkm;c;t 2 f0; 1g 8km 2 BR,8c 2 C,8t 2 P ð2:84Þ

k;t 2 f0; 1g
aSE 8km 2 SE; 8t 2 P ð2:85Þ

akm;t 2 f0; 1g 8km 2 BR; 8t 2 P ð2:86Þ

2 Static and Dynamic Convex Distribution … 55

avr
km;t 2 f0; 1g 8km 2 VR; 8t 2 P ð2:87Þ

where (2.55)–(2.87) represent the convex dynamic (multi-period) EDNEP problem,

which was extended from a convex static EDNEP problem to a study for all time
periods. In this problem, constraints (2.79)–(2.81) has added, limiting the pieces of
equipment (feeders, substations and voltage regulators) to be installed in a specific
branch or node along the planning horizon.

2.5 Numerical Results

2.5.1 Data Specifications

Two cases are used for the solution of the EDNEP problem using the 24-node
electric distribution network [45]: (1) a static test case and (2) a dynamic test case.
The system considered has 24 nodes, 4 substations, 20 load demand nodes and 34
branches, operating at 13.8 kV. For the static model, the planning horizon con-
sidered is 20 years in only one stage. For the dynamic model, the planning horizon
considered is 20 years divided into 4 periods of 5 years each. These models have
been solved using the CPLEX optimization solver [46] in AMPL [47], in a Dell PC
server with 256 GB of RAM memory and 2.27 GHz.
The base topology of the 24-node network is shown in Fig. 2.4. Investment
alternatives for substations and conductors, which were adopted from [20], are
shown in Tables 2.1 and 2.2. Load demand data for each period is shown in
Table 2.3. The costs and data related to the capacitor banks and voltage regulators
are adapted from [27]. The location cost of capacitor banks is US$1500, the cost of
each module is equal to US$1000, the capacity of each module is equal to
300 kVar. For other hand, the location cost of voltage regulators is equal to US
$8400 and 32 steps. The interest rate is set at 13%. The price of energy is equal to
$0.25 /kWh, and finally the minimum and maximum voltage magnitude limits are
0.95 and 1.05 pu, respectively.

2.5.2 Static Test Case

This case is solved in 6 min 45 s, the objective function found is US$118,321,152.

The EDNEP problem is solved taking into account the load demand in the last stage
(year 20), but the decisions are made in the first year. The solutions obtained are as
follows: substations 23 and 24 are built, feeders 2–12, 4–9, 10–16 and 7–19 are
built with conductor type 1, feeders 4–15, 15–17, 1–14, 5–24, 7–23, 10–23, 11–23,
17–22, 18 – 24 and 13–20 are built with conductor type 2 and feeders 1–21 and
6–22 are changed with conductor type 1. In addition, FCB with 1800 kVar are
56 J. López et al.

3
21 2
23

12

10 11
1 16 7

19

9 4 8
15 17
14

5
6 22

18

24 20 13

Existing Substation
New candidate substation
Node
Existing feeder
New candidate feeder

Fig. 2.4 Base topology of the 24-node electric distribution network

Table 2.1 Conductor’s data C Rkm;c ðX=kmÞ Xkm;c ðX=kmÞ C

Ckm;c ð$=kmÞ
1 0.3655 0.2520 20  103
2 0.2921 0.2466 30  103
3 0.2359 0.2402 40  103

Table 2.2 Substations data

k ðkVAÞ
SSE k ðkVAÞ
SSER CkSE ð$Þ CkSER ð$Þ
SE
21 10,000 8000 0.00 1  106
22 15,000 12,000 0.00 1  106
23 20,000 0.00 5  106 0.00
24 25,000 0.00 8  106 0.00
2 Static and Dynamic Convex Distribution … 57

Table 2.3 Load data (kVA)

Bus t=1 t=2 t=3 t=4 Bus t=1 t=2 t=3 t=4
1 4050 4658 5356 6160 13 0 1350 1553 1785
2 780 897 1032 1186 14 0 3280 3772 4338
3 2580 2967 3412 3924 15 0 1460 1679 1931
4 320 368 423 487 16 0 0 1530 1760
5 280 322 370 426 17 0 2330 2680 3081
6 1170 1346 1547 1779 18 0 0 2310 2657
7 4040 4646 5343 6144 19 0 0 1750 2013
8 720 828 952 1095 20 0 0 4020 4623
9 1140 1311 1508 1734 21 0 0 0 0
10 1560 1794 2063 2373 22 0 0 0 0
11 0 2000 2300 2645 23 0 0 0 0
12 0 850 978 1124 24 0 0 0 0

Fig. 2.5 Solution topology 3

21 2
of the 24-node electric
23
distribution network in static
case
12

10 11
1 16 7

19

9 4 8
15 17
14

5
6 22

18

24 20 13

located at nodes 1, 3, 7, 9, 13, 14 and one voltage regulator is located in circuit 4–

15 with a tap position of 6. The topology obtained from the optimization process is
shown in Fig. 2.5.
58 J. López et al.

2.5.3 Dynamic Test Case

This case is solved in 18 min 23 s, the objective function found is US$55,654,205.

In period 1, feeders 4–9, 17–22, 4–16 are built with conductor type 1, while feeders
10–16, 4–15 and 17–22 are built with conductor type 2 and feeder 8–22 is changed
with conductor type 1. FCB with 1200 kVar is located at node 1 and FCB with
900 kVar is located at nodes 3, 7, 9 and 5.
In period 2, substation 23 is built and feeders 2–12, 1–14 and 11–23 are built
with conductor type 1, while feeders 6–13 and 7–23 are built with conductor type 2
and feeder 1–21 is changed with conductor type 1. FCB with a capacity of
1200 kVar is installed at node 14. In addition, one voltage regulator is installed in
branch 4–15 with a tap position of 5.
In period 3, feeder 10–23 is built with conductor type 1 and feeders 7–19, 14–18,
and 13–20 are built with conductor type 2. In addition, the voltage regulator
installed in branch 4–15 in period 2 changes the tap position to 2.
In period 4, substation 24 is built and feeders 24–18, 24–5 and 24–20 are built
with conductor type 1 and feeder 1–9 is built with conductor type 2. In addition, the
voltage regulator installed in branch 4–15 in period 2 changes the tap position to 1.
The topologies obtained from the optimization process in each period are shown in
Fig. 2.6.
The results found for the static and dynamic test cases show that the solutions
obtained using the dynamic model have lower costs than those generated using the
static model. This can be explained by the appropriate execution of investments in
the dynamic model. Note that the cost difference between the static and the dynamic
test cases is US$62,666,947.87, which represents a 47.03% reduction in total
investment cost.

Appendix

The notations used throughout this chapter are listed below:

Sets
B Set of nodes.
BR Set of branches.
C Set of conductor types.
FCB Set of fixed capacitor banks.
P Set of periods.
SE0 Set of existing substations.
2 Static and Dynamic Convex Distribution … 59

(a) 3
(b)
21 2 3
21 2
23
23

12
12
10 11
1 16 7 10
1 16 7 11
19
19
9 4 8
9 4 8
15 17
14 15 17
14

5
6 22 5
6 22
18
18

24 20 13
24 20 13

(c) 3
(d) 3
21 2 21 2
23 23

12 12

10 11 10 11
1 16 7 1 16 7

19 19

9 4 8 9 4 8
15 17 15 17
14 14

5 5
6 22 6 22

18 18

24 20 13 24 20 13

Fig. 2.6 Solution topologies of the 24-node electric distribution network in dynamic case: a In
period 1. b In period 2. c In period 3. d In period 4

SE Set of new installed substations.

SCB Set of switched capacitor banks.
VR Set of voltage regulators.
Zþ Set of positive integer numbers.

Parameters
km ; akm
amax min Upper and lower tap settings of voltage regulator in branch km.
C Installation cost of branch km using conductor type c (US$).
Ckm;c
CkSE Installation cost of new substation k (US$).
CkSER Expanding cost of substation k (US$).
60 J. López et al.

CkFCB Installation cost of fixed capacitor bank in node k (US$).

CkSCB Installation cost of switched capacitor bank in node k (US$).
VR Installation cost of voltage regulator in branch km (US$).
Ckm
Closs Real power loss cost (US$).
CkOS Operating cost of substation k (US$).
max
Ikm;c Upper current limit from conductor type c in branch km (pu).
KL Interest rate for branches.
KS Interest rate for substations.
KC Interest rate for capacitor banks.
KVR Interest rate for voltage regulators.
Lkm Length of branch km (km).
M Large positive number.
Nkfcb Maximum number of fixed capacitors in installed bank k (pu).
Nkscb Maximum number of switched capacitors in installed bank k (pu).
Nskm Total number of steps of voltage regulator in branch km.
PDk;t Active power demand at node k in period t (pu).
QD k;t Reactive power demand at node k in period t (pu).
Qfc Rating of fixed capacitors (pu).
Qsc Rating of switched capacitors (pu).
Rkm;c Resistance per length of conductor type c installed in branch km (pu).
SSE0
k
Maximum apparent power capacity of existing substation k (pu).
SSE
k
Maximum apparent power capacity of newly installed substation
k (pu).
Vkmin ; Vkmax Lower and upper voltage magnitude limits in substation k (pu).
Xkm;c Reactance per length of conductor type c installed in branch km (pu).
Dkm Tap step-size of voltage regulator in km.

Decision Variables
dk;t 2
Auxiliary variable containing Vk;t .
lkm;c;t 2
Auxiliary variable containing Ikm;c;t .
nfck;t Integer variable associated with the number of fixed capacitors installed in a
bank in node k in period t.
nsc
k;t Integer variable associated with the number of switched capacitors installed
in a bank in node k in period t.
PSE
k;t
Active power from substation k in period t (pu).
Pkm;c;t Active power flow by branch km conductor type c in period t (pu).
QSE
k;t
Reactive power from substation k in period t (pu).
Qkm;c;t Reactive power flow by branch km conductor type c in period t (pu).
2 Static and Dynamic Convex Distribution … 61

qfcb
k;t
Reactive power injection from fixed capacitor bank in node k in period
t (pu).
qscb
k;t
Reactive power injection from switched capacitor bank in node k in period
t (pu).
aCkm;c;t Binary variable associated with installing and/or changing of branch km
using conductor type c in period t.
aSE
k;t
Binary variable associated with installing of a new substation k in period t.
aSER
k;t
Binary variable associated with expansion of substation k in period t.
avr
km;t Binary variable associated with installing of voltage regulator in branch km
in period t.
akm;t Variable used in the calculation of the voltage magnitude drop of branch
km in period t.
bkm;t Binary variable associated with radiality in branch km in period t.

References

1. I.J. Ramirez-Rosado, J.A. Dominguez-Navarro, Possibilistic model based on fuzzy sets for the
multiobjective optimal planning of electric power distribution networks. IEEE Trans. Power
Syst. 19(4), 1801–1810 (2004)
2. I.J. Ramirez-Rosado, J.A. Dominguez-Navarro, New multiobjective tabu search algorithm for
fuzzy optimal planning of power distribution systems. IEEE Trans. Power Syst. 21(1), 224–
233 (2006)
3. T. Gönen, Electric Power Distribution Systems Engineering (McGraw-Hill, New York, 1986)
4. S.F. Mekhamer, M.E. El-Hawary, S.A. Soliman, M.A. Moustafa, M.M. Mansour, New
heuristic strategies for reactive power compensation of radial distribution feeders. IEEE
Trans. Power Delivery 17(4), 1128–1135 (2002)
5. J.Y. Park, J.M. Sohn, J.K. Park, Optimal capacitor allocation in a distribution system
considering operation costs. IEEE Trans. Power Syst. 24(1), 462–468 (2009)
6. N.C. Sahoo, S. Ganguly, D. Das, Recent advances on power distribution system planning: a
stage-of-the-art survey. Energy Syst. 4(2), 165–193 (2013)
7. K. Nara, T. Satoh, H. Kuwabara, K. Aoki, M. Kitagawa, T. Ishihara, Distribution systems
expansion planning by multi-stage branch exchange. IEEE Trans. Power Syst. 7(1), 208–214
(1992)
8. E. Míguez, J. Cidrás, E. Díaz-Dorado, J. García-Dornelas, An improved branch-exchange
algorithm for large-scale distribution network planning. IEEE Trans. Power Syst. 17(4), 931–
936 (2002)
9. M. Lavorato, M. Rider, A.V. Garcia, R. Romero, A constructive heuristic algorithm for
distribution system planning. IEEE Trans. Power Syst. 25(3), 1734–1742 (2010)
10. G. Yang, Z. Dong, K. Wong, A modified differential evolution algorithm with fitness sharing
for power system planning. IEEE Trans. Power Syst. 23(2), 514–522 (2008)
11. V. Miranda, J.V. Ranito, L.M. Proença, Genetic algorithm in optimal multistage distribution
network planning. IEEE Trans. Power Syst. 9(4), 1927–1933 (1994)
12. I. Ramirez-Rosado, J. Bernal-Agustín, Genetic algorithms applied to the design of large
power distribution systems. IEEE Trans. Power Syst. 13(2), 696–703 (1998)
13. J. Gómez, H. Khodr, P. Oliveira, L. Ocque, J. Yusta, R. Villasana, A. Urdaneta, Ant colony
system algorithm for the planning of primary distribution circuits. IEEE Trans. Power Syst. 19
(2), 996–1004 (2004)
62 J. López et al.

14. V. Parada, J. Ferland, M. Arias, K. Daniels, Optimization of electrical distribution feeders

using simulated annealing. IEEE Trans. Power Del. 19(3), 1135–1141 (2004)
15. J.M. Nahman, D.M. Peric, Optimal planning of radial distribution networks by simulated
annealing technique. IEEE Trans. Power Syst. 23(2), 790–795 (2008)
16. S. Ganguly, N. Sahoo, D. Das, Mono- and multi-objective planning of electrical distribution
networks using particle swarm optimization. Appl. Soft Comput. 11(2), 2391–2405 (2011)
17. R. Lotero, J. Contreras, Distribution system planning with reliability. IEEE Trans. Power Del.
26(4), 2552–2562 (2011)
18. S. Ganguly, N. Sahoo, D. Das, Multi-objective planning of electrical distribution systems
using dynamic programming. Int. J. Electr. Power Energy Syst. 46, 65–78 (2013)
19. S.N. Ravadanegh, R.G. Roshanagh, On optimal multistage electric power distribution
networks expansion planning. Electr. Power Energy Syst. 54, 487–497 (2014)
20. R.A. Jabr, Polyhedral formulations and loop elimination constraints for distribution network
expansion planning. IEEE Trans. Power Syst. 28, 1888–1897 (2013)
21. R. Gallego, J. Monticelli, R. Romero, Optimal capacitor placement in radial distribution
networks. IEEE Trans. Power Syst. 16(4), 630–637 (2001)
22. D.F. Pires, A.G. Martins, C.H. Antunes, A multiobjective model for VAR planning in radial
distribution networks based on tabu search. IEEE Trans. Power Syst. 20(2), 1089–1094
(2005)
23. I.C. Silva Junior, S. Carneiro Junior, E.J. Oliveira, J.S. Costa, J.L.R. Pereira, P.A.N. Garcia, A
heuristic constructive algorithm for capacitor placement on distribution system. IEEE Trans.
Power Syst. 23(4), 1619–1626 (2008)
24. C.A.N. Pereira, C.A. Castro, Optimal placement of voltage regulators in distribution systems,
in Proceedings of IEEE Bucharest Power Tech (Bucharest, Romania, 2009), pp. 1–5
25. A.S. Safigianni, G.J. Salis, Optimum voltage regulator placement in a radial power
distribution network. IEEE Trans. Power Syst. 15(2), 879–886 (2000)
26. J. Mendoza, D. Morales, R. López, J. Vannier, C. Coello, Multiobjetive location of automatic
voltage regulators in radial distribution network using a micro genetic algorithm. IEEE Trans.
Power Syst. 22(1), 404–412 (2007)
27. J.F. Franco, M.J. Rider, M. Lavorato, R.A. Romero, A mixed integer LP model for the
optimal allocation of voltage regulators and capacitors in radial distribution systems. Electr.
Power Energy Syst. 48, 123–130 (2013)
28. E.P. Madruga, L.N. Canha, Allocation and integrated configuration of capacitor banks and
voltage regulators considering multi-objective variables in smart grid distribution system, in
Proceedings of International Conference on Industry Applications (São Paulo, Brazil, Nov.
2010), pp. 1–6
29. B.A. de Souza, A.M.F. de Almeida, Multiobjective optimization and fuzzy logic applied to
planning of the volt/var problem in distributions systems. IEEE Trans. Power Syst. 25(3),
1274–1281 (2010)
30. J. Sugimoto, R. Yokoyama, Y. Fukuyama, V.V.R. Silva, H. Sasaki, Coordinated allocation
and control of voltage regulators based on reactive tabu search, in IEEE Russian Power Tech
(St. Petersburg, Russia, 27–30 June 2005), pp. 1–6
31. H. Fletcher, K. Strunz, Optimal distribution system horizon planning-part I: formulation.
IEEE Trans. Power Syst. 22(2), 791–799 (2007)
32. S. Haffner, L.F.A. Pereira, L.A. Pereira, L.S. Barreto, Multistage model for distribution
expansion planning with distributed generation—part I: problem formulation. IEEE Trans.
Power Delivery 23(2), 915–923 (2008)
33. S. Haffner, L.F.A. Pereira, L.A. Pereira, L.S. Barreto, Multistage model for distribution
expansion planning with distributed generation—part II: numerical results. IEEE Trans.
Power Delivery 23(2), 924–929 (2008)
34. A. Sorokin, S. Rebennack, P. Pardalos, N. Iliadis, M. Pereira, Handbook of Networks in
Power Systems I. Energy Systems (Springer, Berlin, 2012)
35. J. Tate, T. Overbye, A comparison of the optimal multiplier in polar and rectangular
coordinates. IEEE Trans. Power Syst. 20(4), 1667–1674 (2005)
2 Static and Dynamic Convex Distribution … 63

36. S.C. Tripathy, G.D. Prasad, O.P. Malik, G. S. Hope, Load-flow solutions for ill-conditioned
power systems by a Newton-like method. IEEE Trans. Power App. Syst. PAS-101(10), 3648–
3657 (1982)
37. R.A. Jabr, Radial distribution load flow using conic programming. IEEE Trans. Power Syst.
21, 1458–1459 (2006)
38. M.E. Baran, F.F. Wu, Network reconfiguration in distribution systems for loss reduction and
load balancing. IEEE Trans. Power Delivery 4, 1401–1407 (1989)
39. M.E. Baran, F.F. Wu, Optimal capacitor placement on radial distribution systems. IEEE
Trans. Power Delivery 4, 725–734 (1989)
40. M.E. Baran, F.F. Wu, Optimal sizing of capacitors placed on a radial distribution system.
IEEE Trans. Power Delivery 4, 735–743 (1989)
41. M. Farivar, S.H. Low, Branch flow model: relaxations and convexification—part I. IEEE
Trans. Power Syst. 28(3), 2554–2564 (2013)
42. D. Luenberger, Y. Ye, Linear and Nonlinear Programming (Springer, 2008)
43. R.A. Jabr, Optimal placement of capacitors in a radial network using conic and mixed integer
linear programming. Electr. Power Syst. Res. 78, 941–948 (2008)
44. J. López, D. Pozo, J. Contreras, J.R.S. Mantovani, A multiobjective minimax regret robust
VAr planning model. IEEE Trans. Power Syst. 32, 1761–1771 (2017)
45. I. Gönen, I. Ramirez-Rosado, Review of distribution system planning models: a model for
optimal multi-stage planning. IEE Proc. Gen. Trans. Dist. 133(7), 397–408 (1986)
46. IBM, IBM ILOG CPLEX V12.1. User’s Manual for CPLEX (2009)
47. R. Fourer, D.M. Gay, B.W. Kernighan, AMPL: A Modeling Language for Mathematical
Programming (Duxbury Press, 2002)
Chapter 3
Mathematical Optimization
of Unbalanced Networks with Smart
Grid Devices

Carlos F. Sabillón, John F. Franco, Marcos J. Rider

and Rubén Romero

Abstract Electric distribution networks should be prepared to provide an eco-

nomic and reliable service to all customers, as well as to integrate technologies
related to distributed generation, energy storage, and plug-in electric vehicles.
A proper representation of the electric distribution network operation, taking into
account smart grid technologies, is key to accomplish these goals. This chapter
presents mathematical formulations for the steady-state operation of electric dis-
tribution networks, which consider the unbalance of three-phase grids.
Mathematical models of the operation of smart grid related devices present in
networks are discussed (e.g., volt-var control devices, energy storage systems, and
plug-in electric vehicles). Furthermore, features related to the voltage dependency
of loads, distributed generation, and voltage and thermal limits are also included.
These formulations constitute a mathematical framework for optimization analysis
of the network operation, which makes it possible to model decision-making pro-
cesses. Different objectives related to technical and/or economic aspects can be
pursued within the framework; in addition, the extension to multi-period and
multi-scenario optimization is discussed. The presented models are built based on
mixed integer linear programming formulations, avoiding the use of conventional
mixed integer nonlinear formulations. The application of the presented framework

C. F. Sabillón (&)
R. Romero

Department of Electrical Engineering, São Paulo State University (UNESP),
Ilha Solteira, Brazil
e-mail: [email protected]
R. Romero
e-mail: [email protected]
J. F. Franco
São Paulo State University (UNESP), Rosana, Brazil
e-mail: [email protected]
M. J. Rider
School of Electrical and Computer Engineering,
University of Campinas, Campinas, Brazil
e-mail: [email protected]

© Springer Nature Singapore Pte Ltd. 2018 65

F. Shahnia et al. (eds.), Electric Distribution Network Planning,
Power Systems, https://doi.org/10.1007/978-981-10-7056-3_3
66 C. F. Sabillón et al.

is illustrated throughout control approaches for the voltage control and the plug-in
electric vehicle charging coordination problems.

Keywords Distribution network operation Mathematical optimization



Mixed integer linear programming Smart grids devices Steady-state operation
point

3.1 Introduction

At the present time, high levels of reliability are demanded to power systems, as
electricity is required, among others, in industry, communication, lightning, heating
and entertainment. Smart grids emerged from earlier attempts of power utilities to use
the improvements on electronic technologies to bring a reliable supply of good
quality electricity to their customers. Nowadays, smart grids stand out as the current
responses in order to cope with the challenges brought by a rising electrical demand.
The electric distribution network is the final stage in the transfer of power to
individual consumers. It routes power from small energy sources nearby or power
substation fed by transmission lines, to residential, industrial, and commercial
customers, through power lines, switches, and transformers [1]. Nowadays, utilities
are in charge of operating the distribution network, maintaining a reliable supply of
electric power to all costumers connected into the grid. Traditionally, the electrical
distribution network has been designed to carry the power from the sources
downstream to the consumers, but lately this one-way electricity delivery model has
been changing.
With the evolution of the smart grids, the distributed generation (DG) growth,
and the introduction of renewable sources and energy storage systems (ESS), the
classic distribution network model is evolving. All of these factors impact directly
over the planning, engineering, construction, operation, and maintenance of the
network. As smart grids technologies continue to strengthen, the current electric
distribution network will become a more intelligent and real-time optimized grid; in
consequence, the complexity of the planning, operation, and maintenance of the
network will increase, as new technologies and distribution practices, which offer
greater efficiency, sustainability, and cost savings, are provided. Thereupon, the
network must evolve in order to engage all network elements and participants
including consumer, generators, and those that do both, in an active management
seeking to fulfill technical, economic, and environmental objectives [2].
Since mid-1960s optimization concepts and techniques have been part of the
power system planning and operation. The development of strong optimization
methods and algorithms and their proper application to the power system depends
on a suitable representation of the electric distribution network behavior under
smart grid schemes. Hence, mathematical modeling is crucial to achieve an
enhanced representation of the network operation, endorsing the decision-making
of optimization algorithms.
3 Mathematical Optimization of Unbalanced Networks … 67

Once a problem has been properly represented, it is up to the planners/operators

to choose the most appropriate method in order to solve it. Heuristic and meta-
heuristic techniques, as well as mathematical optimization have foregrounded
among those techniques to become the most commonly applied methods to problem
solving in electric distribution networks. In the latest years, the accelerated advent
of efficient commercial solvers based on mathematical optimization has increased
the interest of researchers in the development of complex and realistic mathematical
models for optimization problems. Therefore, once the mathematical model is
properly defined, the commercial solver finds the best solution; i.e., the planner/
operator does not need to take care of the development of the solution method.
According to the nature of the adopted formulation for the optimization problem,
the corresponding mathematical model may be classified as
• Linear Programming (LP), where the term ‘linear’ indicates that all constraints,
as well as the objective function, are barren of any nonlinearity.
• Nonlinear Programming (NLP), which aims to deal with problems involving
nonlinear constraints and/or objective function.
• Mixed Integer Linear Programming (MILP), which are a special type of LP,
where all or some of the decision variables are confined to only integer values.
• Mixed Integer Nonlinear Programming (MINLP), a special type of NLP (ana-
logue to MILP).
For each type of mathematical model there are several well-known optimization
techniques. For example, LP problems can be solved by using simplex or interior
point algorithms. For NLP problem one can use several traditional optimization
techniques (gradient-based techniques, Lagrangian relaxation, Newton’s method,
successive linear programming, etc.) or an interior point algorithm. To solve MILP
problems, a branch and bound algorithm, improved versions of branch and bound
such as branch and price or branch and cut, Benders’ decomposition, Gomory’s
cutting planes, among others, might be used. Finally, solving a MINLP problem is a
very complicated task and there is few theory related to classical optimization in
this regard. Thus, commercial solvers are assumed to solve this type of problems
based on branch and bound algorithms, sensitivity and barrier methods, and interior
point methods.
In the decade of 2000, commercial solvers based on classical optimization
techniques excelled to become extremely efficient, taking advantage of the
improvement on resolution techniques. Since then, solvers that target LP and MILP
problems such as CPLEX [3], MOSEK [4], GUROBI [5], and similar, had become
extremely efficient compared to prior versions. In counterpart, the development of
specialized solvers on NLP and MINLP problems is still in progress.
Due to the aforementioned, the interest in the development of mathematical
models to represent the operation and planning of electric distribution networks has
grown among the researchers. Thereby, commercial solvers such as CPLEX have
been used when the problem is represented by a LP or MILP mathematical model.
On the other hand, for those cases in which the problems are represented by NLP or
68 C. F. Sabillón et al.

MINLP formulations, these formulations have been transformed into equivalent or

approximated LP problems, if possible, or have been solved using solvers for NLP
or MINLP problems such as KNITRO [6] or BONMIN [7], even though NLP
specialized software are not equally efficient (i.e., they cannot guarantee the global
optimum of the problem and they usually demand high computational efforts).
In electric distribution networks, mathematical modeling has become an
important tool, as it is widely used in operation and expansion planning problems,
especially those including mixed integer programming. This is due to the fact that
commonly planners and operators of electric distribution networks have to meet
specific goals with limited sources. Hence, a great share of the optimization
problems can be classified as ‘yes’ or ‘no’ problems, which may be represented by
binary variables. Decisions such as
• schedule/not schedule, e.g., electric vehicle charging;
• build/not build, e.g., the construction of a new distribution line;
• K out of N, e.g., the number of capacitors operating in a bank; and
• N-possible values, e.g., the tap-position of a voltage regulator;
can be optimized using mathematical models [1].
Moreover, the inclusion of smart grid devices and technologies into the electric
distribution network urge improvements in mathematical formulations previously
employed to model the grid. The flexibility offered by smart grid devices demand
higher levels of accuracy and resolution in the problem formulations, leading to
more realistic but also more complex models. In this regard, the utilization of
three-phase representation has become crucial in the solution of problems related to
network operation. A three-phase representation, in despite of the commonly used
single-phase, takes into account the imbalance in the network and allows the
inclusion of mutual coupling effects, conveying to a more accurate determination of
the steady-state operation point.
This chapter presents two mathematical formulations to represent the
steady-state operation of unbalanced electric distribution networks. The network is
initially represented by NLP models; hereupon, LP models are developed
throughout linearization techniques and approximations, which are implemented in
order to avoid the complexity associated with the solution of the NLP problems.
These formulations constitute a framework that can be used by planners and
operators as a tool inside optimization methods and algorithms aiming to optimize
specific goals, guaranteeing by these means, the feasibility of the solutions found.
Besides, constraints related to load conditions, DG features, and voltage and
thermal limits are also included in both formulations.
3 Mathematical Optimization of Unbalanced Networks … 69

3.2 Mathematical Representation of Unbalanced Electric

Distribution Networks

Traditionally, the evaluation of the electric distribution network state has been
determined by solving a power flow. The objective of the power flow is to deter-
mine, given a set of specific values, the steady-state operation point of the network,
i.e., obtain the voltage magnitudes, phase angles in all nodes, and derived quantities
(e.g., active and reactive power flows, current magnitudes in the circuits, and power
loss). The power flow is a useful tool for the analysis of networks in steady-state,
being widely utilized in real time operation, as well as in the planning of expansion
and operation. This problem is typically modeled as a system of nonlinear equa-
tions, solved through iterative methods [8, 9].
In this section, two approaches are presented aiming to model the operation of an
unbalanced electric distribution network. In contradistinction to the iterative trait
proper of commonly used power flow methods, these formulations can be solved
using mathematical optimization; and they can be extended in order to be used as
tools in optimization analysis in order to mathematically formalize decision-making
regarding different objectives related to technical and/or economic constraints.
Thereupon, in order to determine the steady-state operation point of a network
using mathematical optimization, the operation of the grid must be modeled as a
conventional mathematical programming problem (3.1). These problems have as a
common feature the involvement of optimization. A goal is established and defined
as an objective function ð f Þ which has to be maximized or minimized by the setting
of a set of control variables ð xÞ and subject to a set of constraints.

max=min f ðxÞ
subject to: gðxÞ  0; ð3:1Þ
hðxÞ ¼ 0;

Although the behavior of an electric distribution network follows a set of

nonlinear constraints, it is desired to reach LP representations, which avoid the
complexity related to the solution of NLP problems. For this, the nonlinear set of
equations is initially presented; later, linearization and approximation techniques
are applied to reach an LP model, in each approach. Moreover, in this section all
loads are considered as constant power loads.

3.2.1 Current-Based Mathematical Formulation

This formulation is based on the real and imaginary parts of currents through
circuits and node voltages in the network. A single branch of an unbalanced net-
work is depicted in Fig. 3.1 Each vector of voltages and currents represent the sum
!
of the corresponding real and imaginary parts, e.g., I ¼ I re þ jI im . Hence, the set of
70 C. F. Sabillón et al.

Vm Vn
Zmn,a Imn,a
a

Zmn,a,b
Zmn,b Imn,b
b

Zmn,b,c Zmn,c,a
Zmn,c Imn,c
c

1 1
2 Bmn 2 Bmn

m n
ImD ImG InG InD

Fig. 3.1 Current-based representation of a single branch of an unbalanced electric distribution

network

nonlinear mathematical relationships that represent the steady-state operation of an

unbalanced network are written in terms of the current and voltages real and
imaginary parts, the active and reactive power demanded by loads, and the circuit
impedance, resistance, and reactance.
In the following formulation, consider the sets F; L, and N representing phases,
re=im
circuits, and nodes, respectively. Furthermore, Vn;f are the real/imaginary parts of
Gre=im Dre=im
the voltage at node n and phase f, while In;f and In;f are the real/imaginary
G=D G=D
parts of the generated and demanded currents, respectively. Pn;f and Qn;f are the
re=im
generated/demanded active and reactive powers. In addition, Imn;f represents the
real/imaginary parts of the current through the circuit connecting nodes m and n,
while Bmn;f represents its shunt susceptance for phase f. Finally, Rmn;f ;h and Xmn;f ;h
are the resistance and reactance for circuit mn between phases f and h, respectively.
From Fig. 3.1, the voltage drop from node m to node n can be derived as
3 Mathematical Optimization of Unbalanced Networks … 71

2 3
D~ 2 32 3
Vmn;a Zmn;a Zmn;a;b Zmn;a;c ~Imn;a
6 7
6 ~ 7 6 76 7
6 DVmn;b 7 ¼ 4 Zmn;a;b Zmn;b Zmn;b;c 54~ Imn;b 5
4 5
Zmn;a;c Zmn;b;c Zmn;c ~
D~
Vmn;c Imn;c
2 3 ð3:2Þ
Zmn;a~
Imn;a þ Zmn;a;b~Imn;b þ Zmn;a;c~ Imn;c
6 7
6 7
¼ 6 Zmn;a;b~
Imn;a þ Zmn;b~Imn;b þ Zmn;b;c~ Imn;c 7
4 5
~ ~ ~
Zmn;a;c Imn;a þ Zmn;b;c Imn;b þ Zmn;c Imn;c

Analyzing the individual case for phase a, (3.3) is formulated as

D~
Vmn;a ¼ Zmn;a~
Imn;a þ Zmn;a;b~
Imn;b þ Zmn;a;c~
Imn;c ð3:3Þ

Extending (3.3), considering that Zmn;f ;h ¼ Rmn;f ;h þ jXmn;f ;h and separating cur-
rents and voltages in their real and imaginary parts, (3.4) and (3.5) are obtained.
re
Vm;a  Vn;a
re
¼ Rmn;a Imn;a
re
þ Rmn;a;b Imn;b
re
þ Rmn;a;c Imn;c
re
 Xmn;a Imn;a
im
ð3:4Þ
 Xmn;a;b Imn;b
im
 Xmn;a;c Imn;c
im

im
Vm;a  Vn;a
im
¼ Rmn;a Imn;a
im
þ Rmn;a;b Imn;b
im
þ Rmn;a;c Imn;c
im
þ Xmn;a Imn;a
re
ð3:5Þ
þ Xmn;a;b Imn;b
re
þ Xmn;a;c Imn;c
re

Generalizing these expressions, the voltage drop of an unbalanced network,

written in terms of the real and imaginary parts of the voltages and currents, can be
expressed by (3.6) and (3.7).
X
re
Vm;f  Vn;f
re
¼ ðRmn;f ;h Imn;h
re
 Xmn;f ;h Imn;h
im
Þ 8mn 2 L; f 2 F ð3:6Þ
h2F
X
im
Vm;f  Vn;f
im
¼ ðXmn;f ;h Imn;h
re
þ Rmn;f ;h Imn;h
im
Þ 8mn 2 L; f 2 F ð3:7Þ
h2F

Furthermore, to model the complete steady-state operation of an unbalanced

network, the first Kirchhoff’s Law for the real and imaginary parts of the currents in
each node is applied, as shown in (3.8) and (3.9). Hereupon, (3.10) and (3.11)
establish the relationship between power, voltage and current for the loads.
!
X X X X im
Vm;f
Gre
Im;f þ re
Ikm;f  re
Imn  Bkm;f þ Bmn;f
km2L mn2L km2L mn2L
2 ð3:8Þ
¼ Dre
Im;f 8m 2 N; f 2 F
72 C. F. Sabillón et al.

!
X X X X re
Vm;f
Gim
Im;f þ im
Ikm;f  im
Imn  Bkm;f þ Bmn;f
km2L mn2L km2L mn2L
2 ð3:9Þ
¼ Dim
Im;f 8m 2 N; f 2 F

n;f ¼ Vn;f In;f þ Vn;f In;f

PD 8n 2 N; f 2 F ð3:10Þ
re Dre im Dim

n;f ¼ Vn;f In;f þ Vn;f In;f

QD 8n 2 N; f 2 F ð3:11Þ
re Dim im Dre

Equation (3.12) presents the complete NLP formulation developed to determine

the steady-state operation point of an unbalanced network.

min a
ð3:12Þ
subject to : ð3:6Þð3:11Þ

where a is the objective function of the NLP model, and can be designed to
minimize or maximize the network operator best interests (e.g., power loss, voltage
deviation, or reliability).
Aiming to achieve a LP model based on (3.12), linearization techniques and
approximations must be applied to the nonlinearities shown in (3.10) and (3.11). In
this regard, [10] proposed the application of Taylor’s approximation around an
re im
estimated point ðVn;f ; Vn;f Þ. Hence, (3.10) and (3.11) are rewritten, expressing the
real and imaginary currents in terms of the power and voltages, as shown in (3.13)
and (3.14). Those equations represent the nonlinear expressions for the real and
imaginary demanded currents as the functions gðPD n;f ; Qn;f ; Vn;f ; Vn;f Þ and
D re im

hðPDn;f ; Qn;f ; Vn;f ; Vn;f Þ, respectively.

D re im

n;f Vn;f þ Qn;f Vn;f

PD re D im
Dre
In;f ¼ Vn;f þ Vn;f
re2 im2 ¼ gðPD
n;f ; Qn;f ; Vn;f ; Vn;f Þ 8n 2 N; f 2 F
D re im
ð3:13Þ

n;f Vn;f Qn;f Vn;f

PD im D re
Dim
In;f ¼ Vn;f þ Vn;f
re2 im2 ¼ hðPD
n;f ;t ; Qn;f ;t ; Vn;f ;t ; Vn;f ;t Þ
D re im
8n 2 N; f 2 F ð3:14Þ

Hereupon, taking advantage of the relatively small and limited variation range of
the voltage magnitude in a distribution network, (3.15) and (3.16) present the
Taylor’s approximation used to linearize (3.10) and (3.11).

@g 
 re  im
Dre
In;f ¼ g þ @V re
re
ðVn;f  Vn;f Þ þ @V@gim  ðVn;f im
 Vn;f Þ 8n 2 N; f 2 F ð3:15Þ

@h 
 re  im
Dim
In;f ¼ h þ @V re
re
ðVn;f  Vn;f Þ þ @V@him  ðVn;f im
 Vn;f Þ 8n 2 N; f 2 F ð3:16Þ

Therefore, the LP model that represents the steady-state operation of an unbal-

anced network is shown as
3 Mathematical Optimization of Unbalanced Networks … 73

min a
ð3:17Þ
subject to : ð3:6Þð3:9Þ; ð3:15Þ; and ð3:16Þ

3.2.2 Power-Based Mathematical Formulation

An additional representation to determine the steady-state operation point of an

unbalanced network using mathematical optimization is presented. This formula-
tion is based on the active and reactive power flow through the circuits and the
voltage magnitudes in the network. Analogue to Figs. 3.1 and 3.2 shows a single
branch of an unbalanced network, depicting the power flows through the circuits.
Thus, each vector of complex power represents the sum of the active and reactive
!
powers, i.e., S ¼ P þ jQ.
For the following formulation let the sets F; L, and N, represent phases, circuits,
and nodes, respectively. Besides, ~Vn;f is the voltage vector at node n and phase f,
qdr
with magnitude Vn;f and angle hn;f ; while, Vn;f is the squared voltage magnitude at
qdr
node n and phase f. Furthermore, Imn;f and Imn;f are the current magnitude and the
squared current magnitude through circuit mn and phase f, respectively; while Pmn;f
and Qmn;f are the active and reactive power flows arriving at node n, respectively.

Vm Vn
Zmn,a Smn,a
a

Zmn,a,b
Zmn,b Smn,b
b

Zmn,b,c Zmn,c,a
Zmn,c Smn,c
c

m n
SmD SmG SnG SnD
Fig. 3.2 Power-based representation of a single branch of an unbalanced electric distribution
network
74 C. F. Sabillón et al.

Vm Vn
Vm’ Vn’
Pmn,a + jQmn,a

Rmn,a,b + jXmn,a,b Rmn,a,c + jXmn,a,c Rmn,a + jXmn,a

m' n'

Pm,a
D m Pm,a
G
Pn,a
G n
Pn,a
D

Qm,a
D
Qm,a
G
Qn,a
G
Qn,a
D

Fig. 3.3 Single branch equivalent for phase a

G=D G=D G=D

Sn;f ; Pn;f and Qn;f are the generated/demanded apparent, active, and reactive
powers, respectively. Finally, Zmn;f ;h ; Rmn;f ;h , and Xmn;f ;h are the impedance, resis-
tance, and reactance for circuit mn between phases f and h, respectively.
Hereby, ~
Imn in (3.2) must be expressed in terms of the active and reactive powers
as shown in (3.18).
! ! !
~ S f ;mn Pf ;mn þ jQf ;mn
If ;mn ¼ ¼ ð3:18Þ
~
Vf ;n ~
Vf ;n

Analyzing only phase A from (3.2), Fig. 3.3 presents a simplified equivalent
which divides the voltage drop considering two fictitious nodes m′ and n′.
Therefore, the mathematical equations for each branch-segment are shown in
(3.19).

~
Vm;a  ~ Vm0 ;a ¼ Zmn;a;b~Imn;b
~ ~ ~
Vm0 ;a  Vn0 ;a ¼ Zmn;a;c Imn;c ð3:19Þ
~ Vn0 ;a ¼ Zmn;a~
Vn0 ;a  ~ Imn;a

In order to reach a general expression for the voltage drop, each term of (3.19) is
analysed separately. Initially, the first term can be written as shown in (3.20) where
the term ~Vm 0 ;a =~
Vm 0 ;a is added to enable some algebraic manipulations aiming to
reach a approximated linear expression for the voltage drop.
!
    Pmn;b  jQmn;b ~
Vm 0 ;a
~Vm;a  ~
Vm0 ;a ¼ Rmn;a;b þ jXmn;a;b ð3:20Þ
~ 
Vn;b ~
V 0
m ;a

Applying the simplification shown in (3.21), and replacing ðRmn;a;b and Xmn;a;b Þ
as presented in (3.22), constraint (3.23) is reached.
3 Mathematical Optimization of Unbalanced Networks … 75

~
Vm 0 ;a Vm0 ;a   Vm0 ;a 
 \ h  h  \  120 ð3:21Þ
~ 
Vn;b Vn;b
m;b m;a
Vn;b
    
~ mn;a;b þ jX
R ~mn;a;b ¼ Rmn;a;b þ jXmn;a;b 1\  120 ð3:22Þ

~ Vm0 ;a  ~  
Vm;a ~
Vm 0 ;a  Vm2 0 ;a ¼ ~mn;a;b Pmn;b  jQmn;b
Rmn;a;b þ jX ð3:23Þ
Vn;b

By separating (3.23) into its real and imaginary parts, (3.24) and (3.25) are
obtained.

  Vm0 ;a  
Vm;a Vm0 ;a cos hm;a  hm0 ;a ¼ ~ mn;a;b Pmn;b þ X
R ~mn;a;b Qmn;b þ Vm2 0 ;a ð3:24Þ
Vn;b
  Vm0 ;a  
Vm;a Vm0 ;a sin hm;a  hm0 ;a ¼ ~mn;a;b Pmn;b  Qmn;b R
X ~ mn;a;b ð3:25Þ
Vn;b

Later, by adding the square power of (3.24) and (3.25), the expression for the
voltage drop corresponding to the first segment of Fig. 3.3 is shown in (3.26).

Vm0 ;a  ~ 
~mn;a;b Qmn;b þ Z~mn:a:b
2
Vm;a  Vm2 0 ;a ¼ 2 Rmn;a;b Pmn;b þ X 2 2
Imn;b ð3:26Þ
Vn;b

Therefore, in order to avoid the nonlinearities shown in (3.26), assume

Vm0 ;a =Vn;a  1, and the variables V 2 and I 2 are replaced by V sqr and I sqr , respec-
tively. Hence,
 
sqr
Vm;a  Vmsqr0 ;a ¼ 2 R~ mn;a;b Pmn;b þ X
~mn;a;b Qmn;b þ Z~mn:a:b
2 sqr
Imn;b ð3:27Þ

For the second term of (3.19), an analogue expression is obtained as shown in

(3.27).
 
Vmsqr0 ;a  Vnsqr ~ ~ ~2
0 ;a ¼ 2 Rmn;a;c Pmn;c þ Xmn;a;c Qmn;c þ Zmn:a:c Imn;c
sqr
ð3:28Þ

Finally, for the last term of (3.19), the voltage drop is given by:

    Pmn;a  jQmn;a
~Vn0 ;a  ~
Vn;a ¼ Rmn;a þ jXmn;a ð3:29Þ
~ 
Vn;a

~ mn;a þ jX
For this segment it is considered that Rmn;a þ jXmn;a ¼ R ~mn;a , reaching
the expression presented in (3.30).
76 C. F. Sabillón et al.

 
Vnsqr ~ ~ ~ 2 sqr
0 ;a  Vn;a ¼ 2 Rmn;a Pmn;a þ Xmn;a Qmn;a þ Zmn:a Imn;a
sqr
ð3:30Þ

Therefore, the whole voltage drop in circuit mn is defined by (3.31), which

considers coupling effects between the phases.
Xn   o
sqr
Vm;f sqr
 Vn;f ¼ 2 R~ mn;fh Pmn;f þ X
~mn;fh Qmn;f þ Z~mn;fh
2 sqr
Imn;f 8mn 2 L; f 2 F
h2F
ð3:31Þ

In addition, the active and reactive power balances for each node are represented
by (3.32) and (3.33), respectively. The calculation of the circuit current is given by
(3.34).
P P  
Pkm;f  Pmn;f þ PLmn;f þ PG
m;f ¼ Pm;f
D
8m 2 N; f 2 F ð3:32Þ
km2L mn2L

P P  
Qkm;f  Qmn;f þ QLmn;f þ QG
m;f ¼ Qm;f
D
8m 2 N; f 2 F ð3:33Þ
km2L mn2L

sqr sqr
Vn;f Imn;f ¼ P2mn;f þ Q2mn;f 8mn 2 L; f 2 F ð3:34Þ

To determine the power loss ðPL and QL Þ, the complex power loss is initially
expressed as
! !
X Pmn;h þ jQmn;h Pmn;f þ jQmn;f
SLmn;f ¼ Zmn;f ;h ð3:35Þ
~
Vm;h ~
Vm;f
h2Xf

Equation (3.35) can be also written as

   
X Pmn;h þ jQmn;h Pmn;f þ jQmn;f
SLmn;f ¼ Zmn;f ;h   ð3:36Þ
~
Vm;f ~
Vm;h \ hm;f  hm;h
h2Xf

Furthermore, replacing (3.22) in (3.36), it is obtained:

   
X Pmn;h þ jQmn;h Pmn;f þ jQmn;h
SLmn;f  Z~mn;f ;h ð3:37Þ
~
Vn;f ~
Vn;h
h2Xf

Later, (3.37) and (3.38) are reached by separating the real and imaginary parts of
(3.37). These constraints represent the power loss in an unbalanced network.
3 Mathematical Optimization of Unbalanced Networks … 77

P
PLmn;f ¼ ~ mn;f ;h ðPmn;f Pmn;h þ Qmn;f Qmn;h Þ
R Vn;f Vn;h
h2Xf 8mn 2 L; f 2 F ð3:38Þ
~mn;f ;h ðQmn;f Pmn;h þ Pmn;f Qmn;h Þ
þX Vn;f Vn;h

P
QLmn;f ¼ ~mn;f ;h ðPmn;f Pmn;h þ Qmn;f Qmn;h Þ
X Vn;f Vn;h
h2Xf 8mn 2 L; f 2 F ð3:39Þ
~ mn;f ;h ðQmn;f Pmn;h Pmn;f Qmn;h Þ
þR Vn;f Vn;h

Finally, (3.40) represents the complete NLP model for the steady-state operation
of an unbalanced network.

min a
ð3:40Þ
subject to ð3:31Þ  ð3:34Þ; ð3:38Þ; and ð3:39Þ

The NLP model (3.40) contain nonlinearities [in constraints (3.34), (3.38), and
(3.39)] that must be
 addressed to reach
 a LP model. Equation (3.34) has a product
sqr sqr
of two variables Vn;f and Imn;f in the left-hand side and the square of two
 
variables Pmn;f and Qmn;f in the right-hand side. The left-hand side of (3.34) is
sqr
linearized replacing the variable Vn;f ~n;f
by an estimated value V 2
. On the other hand,
the right-hand side is approximated using a piecewise linearization technique (see
Appendix 1). Hence, (3.41) shows the complete linear expression used to
approximate (3.34).
   
~n;f
V 2 sqr
Imn;f ¼ f Pmn;f ; Pmn;f ; K þ f Qmn;f ; Qmn;f ; K 8mn 2 L; f 2 F ð3:41Þ

Furthermore, as represented by (3.38) and (3.39), power loss can be approxi-

mated by any of the following options:
Option–A: Use actual, historic, or estimated values for the voltages and the power
flows, in order to reach approximated values for PL and QL in (3.32) and (3.33),
disregarding (3.38) and (3.39), as shown in (3.42) and (3.43).

P P  
Pkm;f  ~ Lmn;f þ PG
Pmn;f þ P m;f ¼ Pm;f
D
8m 2 N; f 2 F ð3:42Þ
km2L mn2L

P P  
Qkm;f  ~L
Qmn;f þ Q mn;f þ Qm;f ¼ Qm;f 8m 2 N; f 2 F
G D
ð3:43Þ
km2L mn2L
78 C. F. Sabillón et al.

In not fully observable distribution networks, a two-stage approach is recom-

mended to estimate the power loss P ~ Lkm;f and Q ~ L . In the first stage, the LP is
km;f
solved disregarding the power loss, i.e., P~ km;f and Q
L ~ L are equal to zero. Later, the
km;f
solution of stage one is used to initialize stage two and the LP model is once again
solved.
If option A is chosen, the complete LP model will have the following form:

min a
ð3:44Þ
subject to ð3:31Þ; and ð3:41Þð3:43Þ

Option–B: Use Taylor’s approximation

 around an estimated
 point for the
   
power flows and voltages Pmn;f ; Pmn;h ; Vm;f and Vm;h . Let the functions
gðPmn;f ; Pmn;h ; Qmn;f ; Qmn;h ; Vm;f
 
; Vm;h Þ and hðPmn;f ; Pmn;h ; Qmn;f ; Qmn;h ; Vm;f
 
; Vm;h Þ
be equal to the right part of (3.38) and (3.39), respectively. Equations (3.45) and
(3.46) show the Taylor’s approximation used to determine the power loss.

 
 
PLmn;f ¼ g þ @P@gmn;f  ðPmn;f  Pmn;f Þ þ @P@gmn;h  ðPmn;h  Pmn;h Þ
 
 
þ @Q@gmn;f  ðQmn;f  Qmn;f Þ þ @Q@gmn;h  ðQmn;h  Qmn;h 8mn 2 L; f 2 F
 
 
þ @V@gm;f  ðVm;f  Vm;f

Þ þ @V@gm;h  ðVm;h  Vm;h

Þ
ð3:45Þ
 
 
QLmn;f ¼ h þ @P@hmn;f  ðPmn;f  Pmn;f Þ þ @P@hmn;h  ðPmn;h  Pmn;h Þ
 
 
þ @Q@hmn;f  ðQmn;f  Qmn;f Þ þ @Q@hmn;h  ðQmn;h  Qmn;h Þ 8mn 2 L; f 2 F
 
 
þ @V@gm;f  ðVm;f  Vm;f

Þ þ @V@gm;h  ðVm;h  Vm;h

Þ
ð3:46Þ

Hence, the complete LP model is given by (3.47).

min a
ð3:47Þ
subject to ð3:31Þð3:33Þ; ð3:41Þ; ð3:45Þ; and ð3:46Þ
3 Mathematical Optimization of Unbalanced Networks … 79

3.2.3 Performance and Accuracy

Two LP formulations to determine the steady-state operation point of an unbal-

anced network via mathematical optimization were presented. These representa-
tions are the base of the complete mathematical optimization framework for the
distribution network operation. Different objectives related to technical and/or
economic constraints can be pursued embedding these formulations in multi-period
and multi-scenario optimization (see Appendix 2). Therefore, the quality of studies
developed hereinafter will rely on their level of accuracy.
The performance and accuracy of both formulations is evaluated in the
IEEE123-node test system [11]. All LP models were written in the mathematical
language AMPL [12], and solved using CPLEX [3]. The case study had the fol-
lowing characteristics:
• The whole conventional demand of the distribution network was 1.62 MVA
(40.61%), 1.05 MVA (26.36%), and 1.32 MVA (33.03%), connected to phases
A, B, and C, respectively.
• All loads were connected in wye configuration.
• All demands were considered as constant power loads.
The LP formulations presented (3.17) and (3.44) are expressed as shown in
(3.48) and (3.49), aiming to minimize the total active power generation; considering
nominal voltage at the substation node. Moreover, as the accuracy of both unbal-
anced formulations depends on the precision of the assumed operation point (i.e.,
 
V re ; V im for the current-based formulation, and PL and QL for the power-based
formulation), a two-stage approach was used for both formulations to obtain a better
approximation for the operation point (see Appendix 3).
P  re Gre 
min VS;f IS;f þ VS;f
im Gim
IS;f
f 2F ð3:48Þ
subject to : ð3:6Þð3:9Þ; ð3:15Þ; andð3:16Þ
P
min PG
S;f
f 2F ð3:49Þ
subject to ð3:31Þ and ð3:41Þð3:43Þ

A comparison of the two methods (Current-based and power-based power flow)

was made analyzing the voltage magnitude profile obtained from each when
compared to the one obtained from the solution of a conventional power flow. To
solve this conventional power flow, the specialized software OpenDSS [13] was
selected. For both formulations, Table 3.1 shows the maximum error percentage in
the voltage magnitude, for each phase, compared against the OpenDSS; as well as
the minimum voltage magnitude in the system.
Although both formulations show high accuracy when compared with OpenDSS
results, it can be seen from Table 3.1, that the current-based LP formulation out-
performs the power-based representation. This is an important fact to be taken
 80

Table 3.1 Comparison of the LP formulations against OpenDSS

Phase A B C
LP formulation Current-based Power-based Current-based Power-based Current-based Power-based
Error (%) 0.01 0.02 0.01 0.05 0.11 0.30
Vmin (pu) 0.8996 0.8997 0.9599 0.9595 0.9241 0.9223
C. F. Sabillón et al.
3 Mathematical Optimization of Unbalanced Networks … 81

under consideration by the distribution network optimizer when choosing one LP

formulation for an electric distribution network optimization algorithm.
Nevertheless, the complexity due to troublesome constraints associated with
operational limits and the inclusion of different technologies and devices should be
thoroughly analyzed.

3.3 Operational Constraints

The LP formulations presented in (3.17) and (3.44) are expected to serve as the
central engine of optimization analysis, mathematically formalizing decision–
making processes regarding several objectives. Hereby, an essential component of
any optimization strategy or algorithm related to electricity distribution is the ser-
vice quality. Although several electrical quantities can be measured and limited to
guarantee good quality in the service, voltage magnitude limits and thermal limits in
conductors and transformers are the most commonly used in steady-state studies to
ensure the proper operation of the electric distribution network. Therefore, the
mathematical representation of these limits and their inclusion in the LP formula-
tions are presented below.

3.3.1 Voltage Magnitude

Ensuring a good quality service in the electric distribution network, the voltage
magnitude is limited within a range established by regulatory policies. This range is
mathematically expressed as shown in (3.50), in terms of the minimum and max-
imum values of the voltage magnitude ðV and V,  respectively). Therefore, the
inclusion of (3.50) in the LP formulations is presented below.
 
V  ~
V  V
 ð3:50Þ

Current-Based Representation—Voltage Magnitude Limits: The current-based

representation is written in terms of the currents and voltages real and imaginary
parts. Hence, (3.50) is rewritten in (3.51), limiting the square power of the voltage
magnitude. Nevertheless, (3.51) presents nonlinearities which have to be dealt with
in order to include this limit in the LP model.

V 2  Vn;f
re2
þ Vn;f
im2 2
V 8n 2 N; f 2 F ð3:51Þ
82 C. F. Sabillón et al.

Fig. 3.4 Constraints for θb*

voltage limits in the
current-based formulation V
V
ϕ

ϕ
θa*

θc*

To reach a linear approximation for (3.51), a set of linear constraints limiting the
feasible region for the voltage real and imaginary components is implemented.
Figure 3.4 depicts the sets of lines used to approximate the upper and lower voltage
limits (red lines and blue lines, respectively). In each phase, the upper limit is
replaced by a set of 2  ‘ line segments; where ‘ lines are built clockwise starting
from an estimated operation angle h , and ‘ lines are built counterclockwise. For the
lower limit, a single line is used for each phase also built around h . Equation (3.52)
and (3.53) are the mathematical expressions for the linear approximations of the
upper and lower voltage limits, respectively.
þ
Þsinðu
h i
sinðun;f n;f ;i Þ  cosðu  
im
Vn;f  þ
cosðun;f
;i
Þcosðu Þ
re
Vn;f V n;f ;i Þ þ V sinðun;f ;i Þ 8i 2 ‘. . .‘; n 2 N; f 2 F
;i n;f ;i

ð3:52Þ

sinðuon;fþ Þsinðuo Þ
h i
im
Vn;f  n;f
cosðuon;fþ Þcosðuo Þ
re
Vn;f  V cosðuon;fþ Þ þ V sinðuon;fþ Þ 8n 2 N; f 2 F
n;f

ð3:53Þ
þ    
where un;f ;i ¼ hn;f þ ‘i /; un;f ;i ¼ hn;f þ ð‘i  1Þ/; hn;f is the operation angle for bus
n in phase f; ‘i is the i-element from the set of line segments; and / is the angle of
the arc corresponding to each line segment. Finally, uon;fþ;i ¼ hn;f þ /; and

uo
n;f ;i ¼ hn;f  /.
3 Mathematical Optimization of Unbalanced Networks … 83

Power-Based Representation—Voltage Magnitude Limits: On the other hand, to

include the voltage magnitude limits in the power-based representation, (3.50) is
rewritten in order to limit the square power of the voltage magnitude, as shown in
(3.54).

sqr
V 2  Vn;f 2
V 8n 2 N; f 2 F ð3:54Þ

3.3.2 Circuit Currents

Thermal limits are often the main constraint in electric distribution networks.
Hence, a proper operation of the electric distribution network must maintain the
current in all circuits within the conductor thermal limitations; i.e., the magnitude of
the current in all circuits must be held under their ampacities. Constraint (3.55)
shows the general mathematical expression for the current limit, which is added to
each LP formulation as

~
I   I ð3:55Þ

Current-Based Representation—Circuit Current Limits: Similar to the voltage

magnitude limit, (3.55) is squared and expressed in terms of the real and imaginary
parts of the circuit currents, as shown in (3.56). Later, the nonlinearities are avoided
through a piecewise linearization technique, reaching a linear expression for the
circuit current limit in terms of the real and imaginary parts of the current (3.57).

re2
Imn;f þ Imn;f
im2
 Imn
2
8mn 2 L; f 2 F ð3:56Þ
   
re
f Imn;f ; I; K þ f Imn;f
im 
; I; K  Imn
2
8mn 2 L; f 2 F ð3:57Þ

Power-Based Representation—Circuit Current Limits: Analogous to (3.50),

(3.58) must be added to the power-based LP formulation to limit the current in each
circuit.

sqr
Imn;f  Imn
2
8mn 2 L; f 2 F ð3:58Þ
84 C. F. Sabillón et al.

3.3.3 Transformer Capacity

In steady-state studies of the electric distribution network where different voltage

levels are taken into account, thermal limits regarding the distribution and sub-
station transformers are a key aspect. Therefore, the inclusion of the transformer
capacity in the mathematical optimization framework for the distribution network
operation is presented below. Hence, the mathematical expression (3.59) limiting
the transformer apparent power must be fulfilled for each transformer.
 
~
S  S ð3:59Þ

Hereby, let TRN be the set of nodes where a transformer is installed; while PTR
}
and QTR
} the active and reactive power, respectively, for transformer }. Thus, the
general expression for each transformer capacity is presented in (3.60), and lin-
earized in (3.61) through piecewise linearization technique.
2
P}TR 2 þ QTR
}  S}
2 TR
8} 2 TR ð3:60Þ

2
f ðPTR
} ; S} ; KÞ þ f ðQ} ; S} ; KÞ  S}
TR TR TR TR 8} 2 TR ð3:61Þ

Current-Based Representation—Transformer Capacity: Besides (3.61), for the

current-based formulation, PTR TR
} and Q} have to be calculated. Equations (3.62) and
(3.63) calculate, for transformer }, the active and reactive
 powers
 in terms an
re=im
operation point for the real and imaginary voltages V};f and the currents
 
re=im
I};f at the secondary windings.

P  
} ¼
PTR I};f þ V};f 8} 2 TR
re re im im
V};f I};f ð3:62Þ
f 2F

P  
} ¼
QTR V};f I};f þ V};f 8} 2 TR
re im im re
I};f ð3:63Þ
f 2F

Power-Based Representation—Transformer Capacity: Likewise, for the

power-based formulation, (3.64) and (3.65) show the mathematical expression for
} and Q} in terms of the active an reactive power flowing through each phase of
PTR TR

the transformer P};f and Q};f , respectively).
3 Mathematical Optimization of Unbalanced Networks … 85

P
} ¼
PTR P};f 8} 2 TR ð3:64Þ
f 2F

P
} ¼
QTR Q};f 8} 2 TR ð3:65Þ
f 2F

Finally, the complete LP formulations, considering operational limits are pre-

sented in (3.66) and (3.67), for the current-based and power-based representations,
respectively.

min a
subject to : ð3:6Þ  ð3:9Þ; ð3:15Þ; ð3:16Þ; ð3:52Þ; ð3:53Þ; ð3:57Þ; andð3:61Þ  ð3:63Þ
ð3:66Þ

min a
subject to : ð3:31Þ; ð3:41Þ  ð3:43Þ; ð3:54Þ; ð3:58Þ; ð3:61Þ; ð3:64Þ; andð3:65Þ
ð3:67Þ

3.4 Load Representation

Nowadays, electric distribution networks must deal with social, technical and
environmental challenges in order to successfully satisfy present-day consumers.
The world’s electrical energy consumption is expected to have an annual growth
rate of about 2.2% until 2040 [14], which will substantially impact the operation of
future networks. Hence, network operators are continuously challenged as they are
in charge of meeting customer demands and optimize energy sources, while
guaranteeing a reliable service. As a result, improvements in the network load
modelling are continuously demanded within the grid operator efforts for predicting
system behavior.
Besides the rapid growth of the conventional demand, electric distribution net-
works face issues related to the progressive integration of new technologies. The
rise of new loads (e.g., electric vehicles), which cannot be pigeonholed in tradi-
tional classifications, require a special attention. Thus, this section presents the load
modelling and the inclusion of special loads within the mathematical optimization
framework.

3.4.1 Type of Loads: Voltage Dependent Load Models

In electric distribution networks, loads are traditionally classified as residential,

industrial or commercial. This rough classification was elaborated in order to group
86 C. F. Sabillón et al.

loads that share features related to usage patterns. These well-studied patterns are
commonly affected by factors such as location, weather, cultural habits, and type of
human works. However, since large concentrated loads called for detailed classi-
fications and special representation in order to determine operational characteristics,
studies have been performed aiming to achieve a more precise categorization for
different types of loads [15]. Hereby, a new classification following the voltage
dependency of the actual loads becomes more important in the representation as the
network gets closer to individual loads. The demand of a distribution network is
classified into loads that can be represented as
• constant power loads,
• constant impedance loads,
• constant current loads, or
• a combination of those.
Although load modelling in electric distribution networks is a well-studied topic,
approached by several researches related to voltage and angular system stability,
this issue has to be also taken into account on decision-making algorithms related to
the steady-state operation of the grid. As discussed in [16–18], the effectiveness of
several mathematical models in electric distribution networks is highly dependent
on the accuracy of the load representation. Dependence on the voltage magnitude
and frequency is considered in the load models; mathematically, this dependence
can be represented by static and dynamic load models described by the traditional
ZIP model. In this regards, two static models are commonly studied for the rep-
resentation of the active and reactive demanded powers ðPD and QD , respectively):
the polynomial load model and the exponential load model shown in (3.68) and
(3.69), respectively [19].
 2

Zo V Io V
P ¼P P
D o
þP o þP Po
V o2 V
 2
 ð3:68Þ
Zo V Io V
Q ¼Q Q
D o
þQ o þQ Po
V o2 V
 a
V 1 þ Kpf ðfr  fro Þ
P ¼P
D o
Vo fro
 b ð3:69Þ
V 1 þ Kqf ðfr  fro Þ
Q ¼Q
D o
Vo fro

where Po ; Qo , and V o , are the nominal active and reactive power and bus voltage,
respectively. PZo ; PIo , and PPo , are the active power percentage of the total load
classified as constant impedance, constant current, and constant power, respec-
tively. Likewise, QZo ; QIo , and QPo , are the reactive power percentage for constant
impedance, constant current, and constant power, respectively. In the polynomial
load model (3.68), the loads are treated as a combination of constant impedance
3 Mathematical Optimization of Unbalanced Networks … 87

constant current and/or constant power; hence the sum of these coefficients will
represent the total load, as shown in (3.70). On the other hand, in the exponential
load model (3.69) the load voltage dependency is generalized, and the demanded
active and reactive powers vary according to the voltage exponents a and b. These
voltage exponents depend on the type and composition of the load.

PZo þ PIo þ PPo ¼ 1

ð3:70Þ
QZo þ QIo þ QPo ¼ 1

Moreover, fr represents the frequency of the bus voltage and fro represents the
nominal frequency. The coefficients Kpf and Kqf are the frequency sensitivities for
the active and reactive power loads, respectively. Nevertheless, for the exponential
model the effects associated with frequency may be disregarded, as shown in (3.71).
Hence, with an appropriate adjustment of the constants a and b the model can be
restricted to the steady-state analysis case (i.e., dependence directly on the voltage
magnitude). Appropriate values for these constants may be found in previous
works, such as [20].
 
V a
PD ¼ Po
Vo
 b ð3:71Þ
V
QD ¼ Qo o
V

In some decision-making processes for electric distribution networks, the load

voltage dependency is a key aspect of the suitable representation of the network
operation, e.g., volt-var control. Hence, they must be included in the LP formula-
tions presented in (3.17) and (3.44), where the demanded active and reactive
powers were considered as constant values.
Current-Based Representation—Polynomial load model: In order to include the
load voltage dependency in the LP problem presented in (3.17), the values of the
demanded powers have to be replaced by (3.68) in (3.13) and (3.14). Therefore, the
expressions representing the functions g and h are presented in (3.72) and (3.73).

rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi !
Pon;f Vn;f
re
PZo   PIo  
n;f n;f
g¼ re2
Vn;f þ Vn;f
im2
þ o re2
Vn;f þ Vn;f
im2 þ PPo
re2
Vn;f þ Vn;f
im2 Vno2 Vn n;f

rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi !
Qon;f Vn;f
im
QZo   QIo   ð3:72Þ
n;f n;f
þ re2
Vn;f þ Vn;f
im2
þ o Vn;f þ Vn;f þ Qn;f
re2 im2 Po
re2
Vn;f þ Vn;f
im2
Vno2 Vn
8n 2 N; f 2 F
88 C. F. Sabillón et al.

rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi !
Pon;f Vn;f
im
PZo   PIo  
n;f n;f
h¼ re2
Vn;f þ Vn;f
im2
þ o re2
Vn;f þ Vn;f
im2 þ PPo
re2
Vn;f þ Vn;f
im2
Vno2 Vn n;f

rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi !
Qon;f Vn;f
re
QZo   QIo   ð3:73Þ
n;f n;f
 re2
Vn;f þ Vn;f
im2
þ o re2
Vn;f þ Vn;f
im2 þ QPo
re2
Vn;f þ Vn;f
im2 Vno2 Vn n;f

8n 2 N; f 2 F

Current-Based Representation—Exponential load model: Likewise, to include

the exponential load model in (3.17), (3.13) and (3.14) have to be rewritten, as
shown in (3.74) and (3.75).

re
Vn;f
 2 2
an;f2 1 im
Vn;f
 2 2
bn;f2 1
g ¼ Pon;f a
re
Vn;f þ Vn;f
im
þ Qon;f ;t b
re
Vn;f þ Vn;f
im
8n 2 N; f 2 F
Vo n;f Vo n;f
ð3:74Þ

im
Vn;f
 an;f2 1 re
Vn;f
 2 bn;f 1
2 2
im2 2
h ¼ Pon;f a
re
Vn;f þ Vn;f
im
Qon;f ;t b
re
Vn;f þ Vn;f 8n 2 N; f 2 F
Vo n;f Vo n;f
ð3:75Þ

Power-Based Representation—Polynomial load model: For the power-based

linear representation shown in (3.44), the expressions for the demanded power
(3.68) are directly added to the formulation; notwithstanding, the nonlinearities of
these equations must be dealt with, as shown in (3.76) and (3.77).

 sqr sqr 
Zo Vn;f Io Vn;t ð3:76Þ
n;f ¼ Pn;f Pn V o2 þ Pn V  V o þ Pn
PD 8n 2 N; f 2 F
o Po
n n;t n

 sqr sqr 
Zo Vn;f Io Vn;t ð3:77Þ
n;f ¼ Qn;f Qn V o2 þ Qn V  V o þ Qn
QD 8n 2 N; f 2 F
o Po
n n;t n

Hence, the complete LP formulation, considering the polynomial model for the
load voltage dependency, is given by (3.78).

min a
ð3:78Þ
subject to : ð3:31Þð3:33Þ; ð3:41Þ; ð3:45Þ; ð3:46Þ; ð3:74Þ; andð3:75Þ
3 Mathematical Optimization of Unbalanced Networks … 89

Power-Based Representation—Exponential load model: Finally, to include the

exponential load model in the power-based LP formulation, expressions (3.71) are
added. To avoid the nonlinearities associated to these expressions, (3.71) is
rewritten as shown in (3.79) and (3.80).

V sqr an;f2
PD ¼ Pon;f n;f
8n 2 N; f 2 F ð3:79Þ
n;f Vno2

V sqr bn;f2
QD ¼ Qon;f n;f
8n 2 N; f 2 F ð3:80Þ
n;f Vno2

Later, Taylor’s approximation is applied to linearize (3.79) and (3.80).

V sqr an;f2 sqrð

an;f  
an;f Pon;f 2 1 Þ sqr sqr
PD
n;f ¼ Pon;f n;f
Vno2
þ 2 an;f Vn;f Vn;f  Vn;f 8n 2 N; f 2 F
ð Þ
Vno 2

ð3:81Þ

V sqr bn;f2 bn;f  

bn;f Qon;f sqr 1 sqr sqr
QD
n;f ¼ Qon;f n;f
Vno2
þ 2 bn;f Vn;f 2
Vn;f  Vn;f 8n 2 N; f 2 F
ð Þ
Vno 2

ð3:82Þ

Therefore, the complete LP formulation, considering the exponential load

voltage dependency model, is given by (3.83).

min a
ð3:83Þ
subject to : ð3:31Þð3:33Þ; ð3:41Þ; ð3:45Þ; ð3:46Þ; ð3:81Þ; and ð3:82Þ

3.4.2 Special Loads: Plug-In Electric Vehicles

A large number of EVs is expected to be integrated to the transport sector in the

upcoming years, as an to environmental concerns related to the reduction of
greenhouse gas emissions [21]. From the point of view of the customers, EVs
represent an economical option in response to high fuel costs. On the other hand,
for the electric distribution network, EVs represent an additional load which need to
be attended, increasing the conventional demand in several ways, depending on the
charging place [22]. Hence, EVs are new loads in networks which have to be taken
into account in optimization studies for the grid.
EVs recharge their batteries from the distribution network, and an uncontrolled
charging of large fleets can cause overloads, voltage limit violations, and excessive
90 C. F. Sabillón et al.

energy losses [23]. Hence, the EV charging coordination (EVCC) problem have to
be tackled as part of the distribution network operation, and has received much
attention in recent years [24, 25]. Furthermore, the ability of EVs to inject power
into the grid (also known as vehicle-to-grid (V2G) technology), providing ancillary
services to the network, also represent a highly studied subject [25, 26].
In order to include the EVCC within the electric distribution network opti-
mization framework, let EV be the set of EVs plugged into the grid. PEV e is the
power injected/drawn by EV e, and it is equal to the sum of the maximum charging
and discharging powers P  EV
e
þ  EV
and P e

, respectively) multiplied by the binary
variables ye and ze , which represent the charging or discharging state, as shown in
(3.84). Moreover, (3.85) ensures only one action for the EV (e.g., charging, dis-
charging, or idle). Due to the binary nature of these variables, a MILP model is
obtained as the result of their inclusion.

PEV  EV þ ye  P
e ¼ Pe
 EV
e ze 8e 2 EV ð3:84Þ

ye þ ze  1 8e 2 EV ð3:85Þ

On the other hand, EVs storage

 EV  capacity also needs to be taken into account, i.e.,
maximum energy limit E e and, for V2G applications, the maximum depth of
discharge ðDoDÞ must be always fulfilled. Hence, if an EV is charged/discharged
e during a time interval Dt, (3.86) ensures that the state of charge
constantly at PEV
(SOC) is always maintained between the pre-established limits.
 
 eEV DoD  EeEVi þ DtðP
min EeEVi ; E  EV
e
þ
ye gEV
e
þ  EV
P e ze gEV
e
  eEV
ÞE 8e 2 EV
ð3:86Þ
þ 
where EeEVi is the initial SOC for EV e; while, gEV
e and gEV
e are the charging and
discharging efficiencies, respectively.
The interaction of the EVs with the grid is integrated in the steady-state oper-
ation as follows:
Current-Based Representation—EV: Eqs. (3.87) and (3.88) represent
  the EV
active and reactive powers in terms of the voltage operation point Ve where the
 
EV is plugged, and the EV current IeEVre . Considering that EVs will only
exchange active power. Moreover, (3.89) and (3.90) are the extensions of (3.8) and
(3.9) taking into account the EV current injection in each node.

e ¼ V e Ie
PEV re EVre
þ Veim IeEVim 8e 2 EV ð3:87Þ

0 ¼ Vere IeEVim þ Veim IeEVre 8e 2 EV ð3:88Þ

3 Mathematical Optimization of Unbalanced Networks … 91

 
P P P P im
Vm;f
Gre
Im;f þ re
Ikm;f  re
Imn  Bkm;f þ Bmn;f 2
km2L P mn2L km2L mn2L 8m 2 N; f 2 F
¼ Im;f
Dre
þ IeEVre ce;m;f
e2EV
ð3:89Þ
 
P P im P P re
Vm;f
Gim
Im;f þ im
Ikm;f  Imn  Bkm;f þ Bmn;f 2
km2L P mn2L km2L mn2L 8m 2 N; f 2 F
¼ Im;f
Dim
þ IeEVim ce;m;f
e2EV
ð3:90Þ

where cx;m;f is a binary parameter that takes a value of 1 if the device x is connected
at node m and phase f.
Power-Based Representation—EV: For this formulation, (3.91) represents the
active power balance in each node, taking into account the EV active power
injection/consumption.

P P   P EV
Pkm;f  Pmn;f þ PLmn;f þ PG
m;f ¼ Pm;f þ
D
Pe ce;m;f 8m 2 N; f 2 F
km2L mn2L e2EV
ð3:91Þ

Therefore, the complete MILP formulations for an unbalanced network, con-

sidering EV operation is presented in (3.92) and (3.93) for the current-based and
power-based formulations, respectively.

min a
ð3:92Þ
subject to : ð3:6Þð3:9Þ; ð3:15Þ; ð3:16Þ; andð3:84Þð3:90Þ

min a
ð3:93Þ
subject to : ð3:31Þ; ð3:41Þð3:43Þ; ð3:84Þð3:86Þ; andð3:91Þ

3.5 Distributed Generation

Since the decade of 2000s, distributed generation has continuously grown among
electric distribution networks, motivated by economic, environmental, technical,
and market related features [27, 28]. Due to the flexibility of DG as a power source,
distribution networks have been transformed from a passive network to an active
network. Nowadays, DG plays an important role in the operation, structure and
92 C. F. Sabillón et al.

design of networks; therefore, several researches have been developed to model the
integration of DG units in the network operation [29–31].
DG units are integrated into the electric distribution network in places that were
not originally adapted to connect them can create several problems for distribution
networks in terms of stability and power quality; particularly, when large amounts
of DG units are connected to high impedance networks. In addition, integrating
renewable sources of DG, such as wind or solar power, can mean new challenges to
the network operation. Furthermore, according to the capacity of the DG units, the
network can become an active one, attending loads without the need of the energy
purchased from the main grid. Therefore, the inclusion of the DG in the study of the
network operation is imperative [2].
On the other hand, DG can also offer several advantages to the electric distri-
bution network, i.e., improving system reliability, reducing energy losses, reducing
transmission and distribution line costs, and alleviating congestion in the grid.
Moreover, the installation of small-scale DG units, close to loads, may delay or
avoid investments in additional transmission or distribution infrastructure. In
addition, certain types of DGs also have the ability to offer ancillary services, such
as reactive power support, voltage control, and frequency control.
Typically, in mathematical representations for DG units, the models of syn-
chronous generators (SiGs), induction generators (IGs), and doubly-fed induction
generators (DFIGs) are disregarded. DG units are commonly modelled by a simple
representation and coupling elements are not detailed. Hence, a simple mathe-
matical representation is presented in (3.94)–(3.97) for the generation limits of DG
units.

ðPDG 2 DG 2 DG 2 ð3:94Þ

n Þ þ ðQn Þ  ðSn Þ 8n 2 DG

n  Pn tanðarccosðpfn ÞÞ
QDG DG DG
8n 2 DG ð3:95Þ

n  Qn  Qn
QDG 8n 2 DG ð3:96Þ
DG DG

PDG
n 0 8n 2 DG ð3:97Þ

where DGN represents the set of nodes in which a DG unit is connected. PDG n and
QDG
n are the active and reactive powers of DG unit n; while, pf DG
n is the minimum
power factor, QDG  DG are the minimum and maximum reactive power limits,
and Q
n n
and, SDG
n is the maximum apparent power. Therefore, (3.94) shows the nonlinear
representation for the apparent power limit, and it is approximated in (3.98) via a
piecewise linearization technique. Constraints (3.95) and (3.96) limit the reactive
power in terms of the power factor and the maximum and minimum reactive power
limits, respectively. Finally, (3.97) ensures non-negativity for the active power of
the DG unit.
3 Mathematical Optimization of Unbalanced Networks … 93

(a) P Stator current limit

(Q2 DG , P2 DG)
(Q3 DG , P3 DG)
(Q1 DG , P1 DG)

Under- (Q4 DG , P4 DG)

excitation limit
Rotor
current
limit

Q
Q DG Q DG

(b) P
(Q2 DG , P2 DG) (Q3 DG , P3 DG)
Rotor
Stator current
current limit limit

(Q1 DG , P1 DG) (Q4 DG , P4 DG)

Q
Q DG Q DG

Fig. 3.5 Capability curves: a DFIG and b SiG [29]

   
; DG ; K þ f QDG ; QDG ; K  ð 2
f PDG
n Pn n n n Þ
SDG 8n 2 DG ð3:98Þ

On the other hand, an improved and realistic model for DG units, considering the
capability curves (see Fig. 3.5) for SiGs and DFIGs is presented in [29]. These
types of generators are widely used in DG applications, e.g., wind turbines,
biomass-based CHP generation systems, and small hydroelectric plants.
Figure 3.5 defines the points ðQDG 1;n ; P1;n Þ; ðQ2;n ; P2;n Þ; ðQ3;n ; P3;n Þ; and
DG DG DG DG DG

ðQ4;n ; P4;n Þ, which are used to obtain linear expressions for the DG operation
DG DG

constraints, as

PDG
n  ðQDG
n  Qn Þ 8n 2 DG ð3:99Þ
DG
PDG 1;n
Q1;n QDG
DG
n

2;n P1;n
PDG DG

n 
PDG ðQDG
n  Q2;n Þ þ P2;n 8n 2 DG ð3:100Þ
DG DG
QDG
2;n
QDG
1;n
94 C. F. Sabillón et al.

Table 3.2 Linearization points for the linearization of distributed generators capability curves
[29]
SiG DFIG
1;n ; P1;n Þ
ðQDG DG the intersection between the half of the arc of the armature current
under-excitation and armature current limit between points ðQDG n
; 0Þ and
limits ðQDG ; PDG
Þ
2;n 2;n

2;n ; P2;n Þ
ðQDG DG the intersection between the armature the intersection between the
current limit and the P axis armature current and field current
limits,
3;n ; P3;n Þ
ðQDG DG half of the arc of the armature current the intersection between the field
limit between points ðQDG2;n ; P2;n Þ and
DG current limit and the P axis
ðQ4;n ; P4;n Þ
DG DG

4;n ; P4;n Þ
ðQDG DG the intersection between the armature the half of the arc of the field current
current and field current limits limit between points ðQDG 3;n ; P3;n Þ and
DG

ðQ ; 0Þ
DG
n

3;n P2;n
PDG DG

n 
PDG ðQDG
n  Q3;n Þ þ P3;n 8n 2 DG ð3:101Þ
DG DG
QDG
3;n
QDG
2;n

PDG
 DG
n 
PDG ðQDG
n  Qn Þ 8n 2 DG ð3:102Þ
4;n
DG  DG
Q4;n Qn

Table 3.2 shows how these points were obtained for each type of generator.
Moreover, the power injection of the DG units connected in the system is
integrated in the electric distribution network steady-state formulations as follows:
Current-Based Representation—DG Units: Assuming three-phase DG units,
(3.103) and (3.104) represent the DG unit active
 and reactive powers in terms of the
DG
voltage operation point and their currents In;f . Finally, the current injection due
to DG units is included in the current balance for each node, as:

8n 2 DG;
n =3 ¼ Vn;f In;f
PDG þ Vn;f ð3:103Þ
re DGre im DGim
In;f
f 2F

8n 2 DG;
n =3 ¼ Vn;f In;f
QDG þ Vn;f ð3:104Þ
re DGim im DGre
In;f
f 2F
3 Mathematical Optimization of Unbalanced Networks … 95

X X
Gre
Im;f þ Im;f
DGre
þ re
Ikm;f  re
Imn
km2L mn2L
! ð3:105Þ
X X im
Vm;f 8m 2 N;
 Bkm;f þ Bmn;f ¼ Im;f
Dre

km2L mn2L
2 f 2F
X X
Gim
Im;f þ Im;f
DGim
þ im
Ikm;f  im
Imn
km2L mn2L
! ð3:106Þ
X X re
Vm;f 8m 2 N;
 Bkm;f þ Bmn;f ¼ Im;f
Dim

km2L mn2L
2 f 2F

Power-Based Representation—DG Units: For this formulation, active and

reactive powers are included in the power balance as:
P P  
Pkm;f  Pmn;f þ PLmn;f þ PG
m;f þ Pm =3 ¼ Pm;f
DG D
8m 2 N; f 2 F
km2L mn2L
ð3:107Þ
P P  
Qkm;f  Qmn;f þ QLmn;f þ QG
m;f þ Qm =3 ¼ Qm;f
DG D
8m 2 N; f 2 F
km2L mn2L
ð3:108Þ

Therefore, the complete LP formulations for an unbalanced network, considering

DG units is presented by:

min a
subject to : ð3:6Þð3:9Þ; ð3:15Þ; ð3:16Þ; ð3:95Þð3:98Þ; and ð3:103Þð3:106Þ
ð3:109Þ

min a
subject to : ð3:31Þ; ð3:41Þ  ð3:43Þ; ð3:95Þ  ð3:98Þ; ð3:107Þ; and ð3:108Þ
ð3:110Þ

for the current-based and power-based formulations, respectively.

3.5.1 Renewable DG

Different from the dispatchable DG, in which the electric distribution network
operator controls the active and reactive powers injection for each DG unit,
renewable generation depends on availability of renewable resources (e.g., wind
speed and solar irradiance). Nowadays, renewable DG has taken an important role
96 C. F. Sabillón et al.

in the decentralization of energy production [32]. Due to difficulties related to DG

forecasting when it operates from renewable energy sources, additional consider-
ations must be taken into account to properly include this type of DG into a
distribution network optimization framework. Moreover, with the increase in
penetration of these technologies, issues related to voltage profiles, energy losses,
restoration actions, and network reinforcements have to be addressed.
In this regard, (3.111) represents an additional constraint for renewable DG
units, and models an active power curtailment. This technique is used in order to
avoid undesired levels of power injection from renewable sources which can lead to
voltage rises and high energy losses [33, 34].

^ DG
P ~ DG
n ¼ Pn þ Pn
DG
8n 2 DG ð3:111Þ

where P^ DG ~ DG
n and Pn are the maximum available power and the power curtailment
^ DG
for DG unit n. Under this optimization scheme, P n will depend on the availability
related to the renewable energy source (e.g., wind speed and solar irradiance);
hence, multi-scenario approaches are mainly used to tackle this problems.

3.6 Energy Storage Devices

Energy storage systems have been foregrounded as an answer to conciliate

time-difference between excessive generation and peak demand. In recent years,
energy storage devices prices have declined, which in turn, raised the usage of these
technologies in the electric distribution networks. For electric distribution,
battery-based energy storage systems (BESS) are the most common type of storage.
This is because other storage technologies such as super capacitors and flywheels
are characterized by their high energy cost and are primarily applied on high power,
short duration applications. Hence, due to the growth in the utilization of BESS and
their constant interaction with renewable DG and EVs, their inclusion in the dis-
tribution network steady-state operation must be addressed [18, 35, 36].

3.6.1 BESS Operation

The BESS power drawn or injected from/to the grid must be taken into account in
the steady-state operation of the system. Thus, let SD be the set of BESSs plugged
into the grid. PSD
u is the power injected/drawn by BESS u, and it is equal to the sum
of two non-negative variables that represent the ESS charging and discharging
þ
powers PSD u and PSD
u , respectively), as shown in (3.112). Moreover, (3.113) and
þ
(3.114) limit the variables PSD u and PSD , in terms of the BESS maximum
 u
charging/discharging power Pu  SD
and the binary variables wu and xu .
3 Mathematical Optimization of Unbalanced Networks … 97

Equation (3.115) ensures only one action for the BESS (e.g., charging, discharging,
or idle). Due to the integer nature of these variables, once they are included in the
distribution network optimization framework, one obtains a MILP problem.
SD þ ð3:112Þ
u ¼ Pu
PSD  PSD
u 8u 2 SD

0  PSD
u
þ  SD
P u wu 8u 2 SD ð3:113Þ

0  PSD
u
 SD
P u xu 8u 2 SD ð3:114Þ

wu þ xu  1 8u 2 SD ð3:115Þ

On the other hand, BESS have physical  SDconstraints

 regarding their storage
capacity, i.e., the maximum energy limit E  u and the maximum DoD must be
u is maintained during a time interval Dt, (3.116)
always fulfilled. Hence, if the PSD
keeps the BESS energy level between the pre-established limits.

 uSD DoD  EuSDi þ DtðPSD

E u
þ þ
gu  PSD
u g  SD
u Þ  Eu 8u 2 SD ð3:116Þ

where EuSDi is the initial SOC for BESS u; while, guþ and g
u are the charging and
discharging efficiencies, respectively.
The interaction of the BESSs with the grid is integrated in the steady-state
operation as follows:
Current-Based Representation—BESS: Initially, in (3.117) and (3.118) the
active and reactive powers injected or
 consumed
 by the BESS are expressed in
terms of the voltage operation point Vu where the BESS is connected and the
 
BESS current IuSDre . Considering that BESSs will only inject/drawn active power.
Moreover, (3.119) and (3.120) are the extensions of (3.8) and (3.9) taking into
account the BESS current injection in each node.

u ¼ Vu Iu
PSD re SDre
þ Vuim IuSDim 8u 2 SD ð3:117Þ

0 ¼ Vure IuSDim þ Vuim IuSDre 8u 2 SD ð3:118Þ

!
X X X X im
Vm;f
Gre
Im;f þ re
Ikm;f  re
Imn  Bkm;f þ Bmn;f
2
km2L
X
mn2L km2L mn2L 8m 2 N; f 2 F
¼ Dre
Im;f þ IuSDre cu;m;f
u2SD
ð3:119Þ
98 C. F. Sabillón et al.

!
X X X X re
Vm;f
Gim
Im;f þ im
Ikm;f  im
Imn  Bkm;f þ Bmn;f
2
km2L
X
mn2L km2L mn2L 8m 2 N; f 2 F
¼ Dim
Im;f þ IuSDim cu;m;f
u2SD
ð3:120Þ

where cx;m;f is a binary parameter that takes a value of 1 if the device x is connected
at node m and phase f.
Power-Based Representation—BESS: For this formulation, (3.121) represents
the active power balance in each node, taking into account the BESS active power
injection/consumption.

P P   P SD
Pkm;f  Pmn;f þ PLmn;f þ PG
m;f ¼ Pm;f þ
D
Pu cu;m;f 8m 2 N; f 2 F
km2L mn2L u2SD
ð3:121Þ

Therefore, the complete MILP formulations for an unbalanced network, con-

sidering BESS is presented in (3.122) and (3.123) for the current-based and
power-based formulations, respectively.

min a
ð3:122Þ
subject to : ð3:6Þð3:9Þ; ð3:15Þ; ð3:16Þ; and ð3:112Þð3:121Þ

min a
ð3:123Þ
subject to : ð3:31Þ; ð3:41Þð3:43Þ; ð3:112Þð3:116Þ; and ð3:121Þ

3.7 Voltage and Reactive Power Control Devices

Voltage optimization and reactive power control have been widely used in power
systems as tools to improve energy efficiency and quality [16, 17]. In electric
distribution networks, the management of voltage magnitudes variations together
with the reactive power flows is known as volt-var control (VVC). The main
objective of the VVC is to determine control actions for the devices related to
voltage management and reactive power flow management. The classical devices
controlled within a VVC scheme are on-load tap changers (OLTCs), voltage reg-
ulators (VRs), and switched capacitor banks (SCBs). Hence, the optimization of the
VVC will pursue a proper distribution network operation, while maximizing or
3 Mathematical Optimization of Unbalanced Networks … 99

minimizing an objective imposed by the distribution network operator, e.g., power

loss reduction, minimization of voltage deviation, or maximization of energy
efficiency.
In this framework the mathematical modeling for the optimization of the VVC
problem is presented. A solution for the VVC problem will provide the number of
enabled/disabled modules in every SCB, and the tap position for the OLTCs and
VRs. Hence, the mathematical representation for the switchable SCBs, OLTCs, and
VRs, are shown for both LP formulations. Due to the integer nature of the variables
that model the operation of the VVC devices, the obtain formulations corresponds
to a MILP problem.

3.7.1 Capacitor Banks

The inclusion of switchable SCBs in the electric distribution network operation will
represent an injection of reactive power that will depend on the number of SCB
modules enabled. Thus, let CBN be the set of nodes where a three-phase SCB is
installed. Bn is an integer variable that represents the number of modules enabled
from the SCB connected at node n; B  n is the maximum number of SCB modules;
Qcb
n is the reactive power delivered; and Qesp
n is the reactive power capacity of each
module. Equation (3.124) represents the reactive power injected by the modules of
the switchable SCBs, while the maximum number of operating modules for each
BC is modeled by (3.125).

n ¼ Bn Qn
Qcb esp
8n 2 CB ð3:124Þ

n
0  Bn  B 8n 2 CB ð3:125Þ

Furthermore, in a multi-period optimization where the SCB operations permitted

along the entire time period must be limited, (3.126) must be taken into account.
P  
Bn;t  Bn;t1   Dcb 8n 2 CB ð3:126Þ
t2T

where T is the set of time intervals; Bn;t is the number of modules enabled from the
SCB connected at node n in time t; and Dcb is the maximum number of operations
allowable over the time period.
The reactive power injection Qcb n is included in the steady-state operation of the
distribution network as follows:
Current-Based Representation—Capacitor Banks: The active and reactive power
of the switchable SCBs are represented by (3.127) and (3.128), considering that the
value for the active power injection of every SCB will always be equal to zero.
100 C. F. Sabillón et al.

Moreover, the current balances in each node are updated to include the injection
due to the SCB reactive power, as shown in (3.129) and (3.130).

0 ¼ Vn;f In;f þ Vn;f

re cbre im cbim
In;f 8n 2 CB; f 2 F ð3:127Þ

Qcb
n
3 ¼ Vn;f In;f þ Vn;f
re cbim im cbre
In;f 8n 2 CB; f 2 F ð3:128Þ
!
X X X X im
Vm;f
Gre
Im;f þ re
Ikm;f  re
Imn  Bkm;f þ Bmn;f ¼ Im;f
Dre
 In;f
cbre

km2L mn2L km2L mn2L

2
8n 2 N; f 2 F
ð3:129Þ
!
X X X X re
Vm;f
Gim
Im;f þ im
Ikm;f  im
Imn  Bkm;f þ Bmn;f ¼ Im;f
Dim
 In;f
cbim

km2L mn2L km2L mn2L

2
8n 2 N; f 2 F
ð3:130Þ

cbre cbre
where In;f and In;f are the real and imaginary parts of the current injected in phase
f, by the SCB connected and node n.
Power-Based Representation—Capacitor Banks: For this formulation, the Qcb is
included in the reactive power balance as shown in (3.131).

P P  
Qkm;f  Qmn;f þ QLmn;f þ QG
m;f ¼ Qm;f  Qm
D cb
8m 2 N; f 2 F
km2L mn2L
ð3:131Þ

3.7.2 On-Load Tap Changers and Voltage Regulators

OLTCs and VRs are the devices in charge of controlling the voltage magnitudes in
the electric distribution network, within a VVC environment. These devices adjust
their input voltage through tap changing, and their operation can be represented
under the same mathematical formulation. Thus, let RTL be the set of circuits
where a VR is installed. tpmn;f is the integer variable that defines the tap position for
the VR installed in circuit mn, in phase f; while, Tpmn;f is the maximum number of
taps; and %Rmn is the regulation percentage.
Independent of the steady-state formulation adopted, (3.132) represents the
minimum and maximum limits of the tap position. Analogue to CB, the number of
3 Mathematical Optimization of Unbalanced Networks … 101

tap changes permitted along the entire time period must be limited in a multi-period
optimization; hence, under such scenario (3.133) must be taken into account.

Tpmn  tpmn;f  Tpmn 8mn 2 RT; f 2 F ð3:132Þ

P  
tpmn;f ;t  tpmn;f ;t1   Dvr 8mn 2 RT; f 2 F ð3:133Þ
t2T

Current-Based Representation—OLTCs and VRs: In this formulation, (3.134)

and (3.135) represent the real and imaginary regulated voltage; while, (3.136) and
(3.137) represent the real and imaginary regulated current on each VR.

re
Vn;f ¼ ð1 þ %Rmn tpmn;f =Tpmn ÞVm;f
re
8mn 2 RT; f 2 F ð3:134Þ
im
Vn;f ¼ ð1 þ %Rmn tpmn;f =Tpmn ÞVm;f
im
8mn 2 RT; f 2 F ð3:135Þ
re
Ikm;f ¼ ð1 þ %Rmn tpmn;f =Tpmn ÞImn;f
re
8mn 2 RT; f 2 F ð3:136Þ
im
Ikm;f ¼ ð1 þ %Rmn tpmn;f =Tpmn ÞImn;f
im
8mn 2 RT; f 2 F ð3:137Þ

Equations (3.134)–(3.137) represent the operation of VRs and OLTCs in terms

of the real and imaginary voltages and currents of the electric distribution network.
Nevertheless, the nonlinearities presented have to be addressed, i.e., the product of
the decision variables tpmn;f and, Vm;f or Imn;f on the real and imaginary compo-
nents. In this regard, the integer number of steps is represented as a set of binary
variables btmn;f and the products tpmn;f Vm;f ; tpmn;f Imn;f are substituted by auxiliary
c c
variables Vmn;f ;k and Imn;f ;k , respectively.
A linear extension for (3.134)–(3.137) is presented in (3.138)–(3.151), where
(3.138) and (3.139) represent the calculation of the regulated voltage, and, (3.140)
and (3.141) the calculation of the regulated current. Constraint (3.142) associates
the set of binary variables with the tap integer variable. Equations (3.143) and
c c
(3.144), and (3.145) and (3.146), define the auxiliary variables Vmn;f ;k and Imn;f ;k
respectively while (3.147) and (3.148), and (3.149) and (3.150) describe their
limits. Finally, (3.151) represents the sequencing of the binary variable btmn;f .

Pmn
2Tp
%Rmn cðreÞ
re
Vn;f ¼ ð1  %Rmn ÞVm;f
re
þ Tpmn Vmn;f ;k 8mn 2 RT; f 2 F ð3:138Þ
k¼1

Pmn
2Tp
%Rmn cðimÞ
im
Vn;f ¼ ð1  %Rmn ÞVm;f
im
þ Tpmn Vmn;f ;k 8mn 2 RT; f 2 F ð3:139Þ
k¼1
102 C. F. Sabillón et al.

Pmn
2Tp
%Rmn cðreÞ
re
Ikm;f ¼ ð1  %Rmn ÞImn;f
re
þ Tpmn Imn;f ;k 8mn 2 RT; f 2 F ð3:140Þ
k¼1

Pmn
2Tp
%Rmn cðimÞ
im
Ikm;f ¼ ð1  %Rmn ÞImn;f
im
þ Tpmn Imn;f ;k 8mn 2 RT; f 2 F ð3:141Þ
k¼1

Pmn
2Tp
btmn;f ;k  Tpmn ¼ tpmn;f 8mn 2 RT; f 2 F ð3:142Þ
k¼1
 
 re cðreÞ    btmn;f ;k Þ
Vm;f  Vmn;f ;k   Vð1 8mn 2 RT; f 2 F; k ¼ 1. . .2Tpmn ð3:143Þ
 
 im cðimÞ    btmn;f ;k Þ
Vm;f  Vmn;f ;k   Vð1 8mn 2 RT; f 2 F; k ¼ 1. . .2Tpmn ð3:144Þ
 
 cðreÞ  
Vmn;f ;k   Vbtmn;f ;k 8mn 2 RT; f 2 F; k ¼ 1. . .2Tpmn ð3:145Þ
 
 cðimÞ  
Vmn;f ;k   Vbtmn;f ;k 8mn 2 RT; f 2 F; k ¼ 1. . .2Tpmn ð3:146Þ
 
 re cðreÞ 
Imn;f  Imn;f ;k   Imn ð1  btmn;f ;k Þ 8mn 2 RT; f 2 F; k ¼ 1. . .2Tpmn ð3:147Þ
 
 im cðimÞ 
Imn;f  Imn;f ;k   Imn ð1  btmn;f ;k Þ 8mn 2 RT; f 2 F; k ¼ 1. . .2Tpmn ð3:148Þ
 
 cðreÞ  
Imn;f ;k   Imn btmn;f ;k 8mn 2 RT; f 2 F; k ¼ 1. . .2Tpmn ð3:149Þ
 
 cðimÞ  
Imn;f ;k   Imn btmn;f ;k 8mn 2 RT; f 2 F; k ¼ 1. . .2Tpmn ð3:150Þ

btmn;f ;k  btmn;f ;k1 8mn 2 RT; f 2 F; k ¼ 1. . .2Tpmn ð3:151Þ

Power-Based Representation—OLTCs and VRs: For the power-based formu-

sqr
lation, Vm;f is altered by the square of the regulation ratio, which is expressed in
terms of the regulation percentage, the tap integer value, and the maximum tap, as
shown in (3.152).

 2
sqr
Vn;f
tp
¼ 1 þ %Rmn Tpmn;f sqr
Vm;f 8mn 2 RT; f 2 F ð3:152Þ
mn
3 Mathematical Optimization of Unbalanced Networks … 103

In order to cope with the nonlinearities observed in (3.152), tp2mn;f is represented

sqr
as a set of binary variables btmn;f , and the product tp2n;f Vm;f is represented using the
c
auxiliary variables Vmn;f , as shown in set (3.153)–(3.157).

Pmn
2Tp h   i
sqr %Rmn ð2k1Þ%Rmn
Vn;f ¼ þ 2ð1  %Rmn Þ Vmn;f
c
;k
k¼1
Tpmn Tpmn
8mn 2 RT; f 2 F
sqr 2
þ Vm;f ð1  %Rmn Þ
ð3:153Þ
sqr
V 2 ð1  btmn;f ;k Þ  Vm;f  Vmn;f
c
;k 8mn 2 RT; f 2 F; k ¼ 1. . .2Tpmn ð3:154Þ
sqr
Vm;f  Vmn;f
c 2
;k  V ð1  btmn;f ;k Þ 8mn 2 RT; f 2 F; k ¼ 1. . .2Tpmn ð3:155Þ

V 2 btmn;f ;k  Vmn;f
c 2
;k  V btmn;f ;k 8mn 2 RT; f 2 F; k ¼ 1. . .2Tpmn ð3:156Þ

btmn;f ;k  btmn;f ;k1 8mn 2 RT; f 2 F; k ¼ 2. . .2Tpmn ð3:157Þ

Therefore, the complete MILP formulations for VVC optimization considering

operational limits, are presented in (3.158) and (3.159), for the current-based and
power-based representations, respectively.

min a
subject to : ð3:6Þ  ð3:9Þ; ð3:15Þ; ð3:16Þ; ð3:52Þ; ð3:53Þ; ð3:57Þ; ð3:61Þ  ð3:63Þ; ð3:124Þ; ð3:125Þ;
ð3:127Þ  ð3:130Þ; ð3:132Þ; and ð3:138Þ  ð3:151Þ
ð3:158Þ

min a
subject to : ð3:31Þ; ð3:41Þ  ð3:43Þ; ð3:54Þ; ð3:58Þ; ð3:61Þ; ð3:64Þ; ð3:65Þ; ð3:124Þ; ð3:125Þ;
ð3:131Þ; ð3:132Þ; and ð3:153Þ  ð3:157Þ
ð3:159Þ

3.8 Mathematical Framework Application in Control

Approaches

Two control applications are presented in this section to assess the presented
mathematical optimization framework. Initially, the EVCC problem is tackled using
the current-based formulation, as shown in [25]. Later, a voltage control using the
power-based LP formulation is shown, solving a VVC scheme to reduce voltage
deviation and guarantee proper operation of the electric distribution network.
104 C. F. Sabillón et al.

3.8.1 Electric Vehicle Charging Coordination Problem

The EV charging coordination problem consists in determining the optimal

schedule for charging the EV batteries aiming an economical operation of the
electric distribution network, while maintaining a suitable and efficient system
operation. Hereby, a multi-period MILP formulation was proposed in [25] to solve
the optimal charging coordination of EVs in unbalanced distribution networks
considering V2G technology and DG. The steady-state operation of the grid is
represented using the current-based formulation. The MILP formulation is
embedded in a step-by-step control method that considers randomness in EV
arrival, departure, and initial SOC.
The multi-period approach studies a specific time period which is divided into
several time intervals. The control method finds an optimal schedule for the energy
exchange between EV batteries and the grid. This method solves the proposed
MILP model at the beginning of each time interval, constructing a step-by-step
solution over the entire time period. The solution presents a charging schedule for
each EV, which is generated between arrival and departure, ideally dispatching a
fully charged battery.
The objective function of the EVCC problem, presented in (3.160), seeks to
minimize the cost of the energy provided by the substation and the DG units as well
as to reduce energy curtailment in EVs (if an EV cannot be completely charged, the
unserved energy is considered as an energy curtailment).
XX   XX X
min aG
S;t Dt V re Gre
I
S;f ;t S;f ;t þ V im Gim
I
S;f ;t S;f ;t þ a DG
n;t Dt PDG
n;t þ bEeSH
f 2F t2T n2N t2T e2EV

ð3:160Þ

where aGS;t and an;t are the energy costs at the substation and for each DG unit in
DG

time interval t, respectively. EeSH is the energy curtailment for EV e, while b is the
EV curtailment cost (typically a high value to avoid curtailment).
Furthermore, the steady-state operation of the distribution network was modeled
using (3.6), (3.7), (3.15), (3.16), (3.161), and (3.162). Constraints (3.52), (3.53),
and (3.57), were used represent the operational limits. The DG units were modeled
using (3.95), (3.96), (3.97), (3.103), and (3.104); an additional limit for the active
power was also employed. Finally, the operation of the EVs was represented by
(3.84)-(3.88), and (3.163).
 
P re P re P P im
Vm;f
Gre
Im;f þ Im;f
DGre
þ Ikm;f  Imn  Bkm;f þ Bmn;f 2
P km2L mn2L km2L mn2L
ð3:161Þ
¼ Im;f
Dre
þ IeEVre ce;m;f
e2EV
8m 2 N; f 2 F
3 Mathematical Optimization of Unbalanced Networks … 105

 
P P P P re
Vm;f
Gim
Im;f þ Im;f
DGim
þ  im
Ikm;f im
Imn  Bkm;f þ Bmn;f 2
P km2L mn2L km2L mn2L
ð3:162Þ
¼ Im;f
Dim
þ IeEVim ce;m;f
e2EV
8m 2 N; f 2 F
P
 eEV ¼ EeEVi þ
E  EV
DtðP þ
ye;t gEV þ  EV
P ze;t gEV Þ þ EeSH 8e 2 EV
e e e e ð3:163Þ
t2T

The proposed model was tested in the IEEE 123-node test system [11], and the
following considerations were taken into account:
• Phases A, B, and C of the electric distribution network were charged with 1.42
MVA (40.7%), 0.915 MVA (26.2%), and 1.155 MVA (33.1%), respectively.
• The time period was set from 18:00 to 08:00 h, divided into half-hour time
intervals.
• Two types of EV batteries were considered: 50 kWh Tesla EVs and 20 kWh
Nissan Leafs. The charging maximum power was 10 and 4 kW, and for
EV-V2Gs the discharging maximum power was 5 and 2 kW, respectively.
• Hourly energy cost and load variation were considered.
• The arrival and departure time intervals were generated based on the two
chi-squared probability functions with 8 and 4 degrees of freedom.
• The initial SOC of the EVs was generated using the normal-based probability
function with mean value and the standard deviation of 15 and 10, respectively.
• The minimum voltage limit was set at 0.90 pu.
• The maximum current was 500 A for all feeders.
• 400 EVs were plugged into the grid.
• 40% of EVs were considered to have V2G technology.
The model was implemented in the mathematical programming language AMPL
[12] and solved with the commercial solver CPLEX [3]. Initially, the dumb charge
case is presented. Here, the EV recharge was done without any charging coordi-
nation, i.e., the EV batteries started an uninterrupted charging process as soon as
they were plugged into the electric distribution network. Later, several control
scenarios were analyzed. Figure 3.6 shows the energy exchange between the EVs
EV Power Consumption (kW)

400
EV Power Consumption (kW)

400 4000
Number of EVs Connected

4000
Number of EVs Connected

3500 350
3500 350
3000 300
3000 300
2500 250
2500 250
2000 200
2000 200 150
1500
1500 150 1000 100
1000 100 500 50
500 50 0 0
0 0 -500 100
18 19 20 21 22 23 24 01 02 03 04 05 06 07 08 18 19 20 21 22 23 24 01 02 03 04 05 06 07 08
Hour Hour
Dumb Charge Coordinated Charging

Fig. 3.6 EV active power exchange [25]

106 C. F. Sabillón et al.

1.00 700 Maximum Current Limit

Voltage Magnitude (p.u.)
0.98

Current Magnitude (A)

Dumb Charge
0.96 600 Coordinated Charging
0.94 500
0.92 400
0.90
0.88 300
0.86 Dumb Charge 200
0.84 Coordinated Charging
Minimum Voltage Limit 100
0.82
0.80 0
18 19 20 21 22 23 00 01 02 03 04 05 06 07 08 18 19 20 21 22 23 00 01 02 03 04 05 06 07 08

Hour Hour

Lower Voltage Limits Current Limit

Fig. 3.7 Operational limits [25]

and the grid for the dumb charge and a coordinated charge scenario, in which all
EVs were considered ‘Tesla EV’. For the coordinated charge case, the objective
function was reduced by 25% when compared to the dumb charge case, and no
curtailment was presented.
The power related to the charging and discharging of EVs is shown in red and
blue, respectively. Following convention, the EV charging power and the EV-V2G
discharging power are given in positive and negative values, respectively. It can be
seen that without any coordination (Dumb Charge), the EV batteries were contin-
uously charged upon arrival; hence, the peak load for this case was between 21:00
and 22:00. In the coordinated charge case, this peak was shifted to the low-cost time
intervals. This represented a reduction of almost 1 MW in the total active power
demand.
Figure 3.7 presents the voltage and thermal limits for these cases. For the dumb
charge, voltage and current limit violations were presented. These breaches were
avoided when the EV charging control was enabled. Hence, it is stated that the
EVCC in the distribution network is beneficial not only for peak load reduction, but
also for maintaining the proper operation of the grid.
Therefore, the step-by-step methodology based on the current-based MILP
formulation was proved efficient to find an optimal charging schedule for EVs in
unbalanced network considering V2G technology.

3.8.2 Voltage Control Problem

High voltage drops along a radial distribution feeder lead to elevated energy losses.
Hence, control methods for voltage optimization are crucial in daily distribution
network operation. The application of the power-based formulation in the voltage
control method is evaluated using the IEEE 13-node test system [11].
Figure 3.8 shows the IEEE 13-node test system, which has nominal conven-
tional demand of 1.31 MVA (34.2%), 1.16 MVA (30.3%), and 1.36 MVA (35.5%),
connected to phases A, B, and C, respectively. For this test, all loads are considered
to be connected in wye-configuration and classified as constant power. Moreover,
3 Mathematical Optimization of Unbalanced Networks … 107

Fig. 3.8 IEEE 13-node test 01

system
05 04 02 03

09 08 06 01

10 11

Table 3.3 Voltage profile of Bus\Phase A B C

the IEEE 13-node test system,
under nominal demand (pu) 1 1.0000 1.0000 1.0000
2 0.9555 0.9695 0.9347
3 0.9523 0.9675 0.9319
4 – 0.9591 0.9378
5 – 0.9557 0.9389
6 0.9221 0.9723 0.8585
7 0.9138 0.9734 0.8545
8 0.9200 – 0.8546
9 – – 0.8508
10 0.9138 – –
11 0.9221 0.9723 0.8585

the nominal voltage is 4.16 kV, and the voltage magnitude at the substation is fixed
at 1.0 pu. Working under nominal demand, the test system shown in Fig. 3.8,
presents the voltage profile shown in Table 3.3.
In order to improve the voltage profile of the electric distribution network, the
minimum and maximum voltage limits are set at 0.93 and 1.05 pu, respectively.
An OLTC and two switchable SCB are added to the grid in order to fulfill voltage
limit requirements. It is desired to use these devices to keep the voltage between the
established limits, while minimizing the voltage deviation at each bus for 4 different
load levels, 100, 70, 50, and 30% of the nominal demand (i.e., it is expected to
maintain the voltage magnitude of all buses as close as possible to the nominal
value).
Hence, the control actions for each device should be determined in order to fulfill
the operational constraints related to the voltage limit, while minimizing voltage
deviation. In this matter, consider:
• The OLTC is installed at the substation and controls the voltage magnitude at
node 1.
• The OLTC can vary the input voltage magnitude in a 5% regulation ratio,
distributed in 8 tap positions (±4).
108 C. F. Sabillón et al.

• The switchable SCBs are installed at node 7 and 9.

• Each switchable SCB has 6 500 kvar, modules.
To determine the control actions for the OLTC and the BCs, the power-based LP
formulation is chosen. Initially the voltage deviation ðwÞ is expressed
 as the
2
absolute value of the difference between nominal voltage Vnom and the square
 
sqr
value of the bus voltage Vn;f , as shown in (3.164). Thus, (3.164) is linearized in
(3.165) and (3.166) taking advantage that the deviation is minimized in the
objective function.
 
 2 sqr 
Vnom  Vn;f  ¼ wn;f 8n 2 N; f 2 F ð3:164Þ

sqr
2
Vnom  Vn;f  wn;f 8n 2 N; f 2 F ð3:165Þ
 
sqr
 Vnom
2
 Vn;f  wn;f 8n 2 N; f 2 F ð3:166Þ

The LP formulation presented in Sect. 3.7 is used, and the complete MILP is
shown as
PP
min wn;f
n2N f 2F
subject to : ð3:31Þ; ð3:41Þ  ð3:43Þ; ð3:54Þ; ð3:58Þ; ð3:61Þ; ð3:64Þ; ð3:65Þ; ð3:124Þ; ð3:125Þ;
ð3:131Þ; ð3:132Þ; ð3:153Þ  ð3:157Þ; ð3:165Þ; and ð3:166Þ
ð3:167Þ

The mathematical formulation was written in the mathematical language AMPL

[12], and solved using CPLEX [3]. The model was solved for the four load levels,
finding the best configuration for each case. Table 3.4 presents for each case the tap
position for the OLTC, the number of enabled modules for the SCBs, and the total
deviation. Besides, it can be seen that the voltage limits were fulfilled for each case
in which the devices were taken into account.

Table 3.4 Summary of the results for the Voltage Control Problem
Case Loading OLTC-Tap SCB-7 SCB-9 Voltage Voltage
(%) modules modules deviation limits
active/ active/ (pu)
total total
w/o 100 – – – 4.0177 Breached
devices
I 100 +3 6/6 6/6 1.2943 Fulfilled
II 70 +2 6/6 6/6 0.7122 Fulfilled
III 50 +1 6/6 5/6 0.4937 Fulfilled
IV 30 +1 1/6 3/6 0.2783 Fulfilled
3 Mathematical Optimization of Unbalanced Networks … 109

Table 3.5 Voltage profile of the IEEE 13-node test system, after voltage control implementation
(pu)
Phase A B C
Bus\Case w/o devices I w/o devices I w/o devices I
1 1.0000 1.0375 1.0000 1.0375 1.0000 1.0375
2 0.9555 0.9980 0.9695 1.0067 0.9347 1.0017
3 0.9523 0.9949 0.9675 1.0048 0.9319 0.9991
4 – – 0.9591 0.9967 0.9378 1.0046
5 – – 0.9557 0.9934 0.9389 1.0056
6 0.9221 0.9693 0.9723 1.0076 0.8585 0.9594
7 0.9138 0.9622 0.9734 1.0092 0.8545 0.9566
8 0.9200 0.9672 – – 0.8546 0.9601
9 – – – – 0.8508 0.9608
10 0.9138 0.9613 – – – –
11 0.9221 0.9693 0.9723 1.0076 0.8585 0.9594

To better illustrate the solution found through mathematical optimization, the

voltage profile for the solution determined by the MILP formulation under nominal
demand, is shown in Table 3.5. It can be seen that the MILP formulation found a
solution which guarantee the compliance of the voltage limits.
Voltage limit violations were highlighted in red. The efficiency of the
power-based MILP formulation was proven as the solutions for all cases improved
the voltage profile, keeping all voltage magnitudes in between the limits, while
minimizing the voltage deviation.

3.9 Comparative Overview and Discussion

The formulations presented, efficiently model the steady-state operation of unbal-

anced networks; constituting a mathematical framework that can be used by
planners and operators as a tool inside optimization methods and algorithms,
aiming to optimize specific goals. Although both formulations target the same
objective, the planner/operator can choose the one that better accommodates and fits
the problem that he is aiming to tackle. In order to make this decision, the following
considerations must be addressed and well-thought:
1. Although both formulations show high accuracy determining the steady-state
operation point, the current-based model slightly outperforms the power-based,
as shown in Sect. 3.2.3.
110 C. F. Sabillón et al.

2. Due to its representation in terms of the real and imaginary parts of voltages and
currents, modeling operational limits within the current-based formulation
requires a high number of constraints. This fact conveys to higher computational
burden, which can slow down the solution process especially in large-scale test
systems.
3. In optimization methods or algorithms applied over non-stressed test systems,
where operational limits are not a concern to the optimizer (e.g., demand
response or market-based optimizations), the current-based formulation may
highlight as a better option.
4. As mentioned, both formulations can handle optimizations algorithms taking
into account smart grid devices plugged into the network. Nevertheless, the
mathematical representation of the considered smart grid devices can sway the
formulation choice; e.g., in a volt-var approach, the power-based formulation is
recommended, as the volt-var devices influence directly over the voltage
magnitude and reactive power flows.
5. Finally, the load behavior is also an important feature to take into account as it
impacts directly in the operation point estimation. Both LP formulations rely on
the accuracy of the estimated operation point. In this regard, in little observable
distribution networks, estimating voltages is an easier task to the planner/
operator than estimating power flows along the grid. Hence, the current-based
formulation will suit better to this application.
It is important to remark that every optimization problem targeted in electric
distribution network will bring specific considerations that have to be analyzed to
make the best choice.

Appendix 1: Piecewise Linearization Technique

The piecewise linearization is a technique in which a nonlinear function is

approximated using a set of piecewise linear functions [37]. Widely used in engi-
neering, this technique is often employed to cope with quadratic nonlinearities,
helping to reach LP models. Typically, a function f is defined in order to calculate
the square value of a variable r, represented as r þ þ r and limited by the interval
 . This type of function has a general structure, as
½0; r

X
K
; KÞ ¼
f ðr; r /r;k Dr;k ð3:168Þ
k¼1

r ¼ r þ  r ð3:169Þ
3 Mathematical Optimization of Unbalanced Networks … 111

X
K
r þ þ r ¼ Dr;k ð3:170Þ
k¼1

=K
0  Dr;k  r 8k 2 K ð3:171Þ

/r;k ¼ ð2k  1Þ

r=K 8k 2 K ð3:172Þ

The parameter /r;k is calculated to compute the contribution of Dr;k in each step
of the discretization. The parameter r represents the maximum value of r, while K
is the number of discretizations used in the linearization.
It is important to remark that this approach is limited to maximizing strictly
concave functions or minimizing convex functions. If the application of this
technique under different conditions is desired, the inclusion of binary variables and
additional constraints is mandatory.

Appendix 2: Multi-period and Multi-scenario Extension

Typically, optimization analyses in electric distribution network operation are done

along a time window in which several control actions have be defined and they may
be dependent among them; this is known as multi-period optimization. For
example, the day-ahead operation planning is typically divided in one-hour time
windows, and the decisions from one hour may or may not affect the decisions
regarding the next time intervals. Thus, mathematical formulations for the distri-
bution network operation should be able to handle multi-period optimization
analyses. In this regard, the LP formulations presented can be easily adapted to
handle several time intervals. Hence, a new index associated to the time interval is
added to the variables that represent the distribution network operation (e.g.,
voltages, currents, and power flows).
Furthermore, adaptability to multi-scenario optimizations is also required in an
optimization framework for electric distribution network to model the uncertainty in
the grid. The multi-scenario optimization is a method usually employed to solve
stochastic programming problems in which some of the variables or parameters are
of uncertain nature (e.g., EV behavior, renewable DG availability, and demand
variations). The uncertainties are represented through a set of scenarios and each
one with an associated probability, i.e., a multi-scenario model will provide an
optimal solution on average, considering all the scenarios simultaneously.
Analogue to the multi-period case, a new index is added to the uncertain variables
associated to each scenario. Thereby, the objective function of the problem is
calculated as the expected value due to the inclusion of the probabilities related to
each scenario.
112 C. F. Sabillón et al.

Appendix 3: Estimated Steady-State Operation Point

As mentioned in Sect. 3.2.3, the accuracy of the presented three-phase formulations

relies on the precision of the estimated operation point. High quality estimations
will minimize the error corresponding to some approximations in voltage magni-
tudes and some linearization techniques (e.g., Taylor’s linearization). In order to
obtain a suitable estimated operation point, the followings techniques might be
employed:
1. A two-stage approach, in which a first stage solves the LP model using a flat
start (e.g., assuming nominal voltages and disregarding power). Later, the
solution of the first stage is used to initialize the second stage in which the LP
model is once again solved from the already calculated operating point.
2. Using historical data, where historical data is used in order to determine the
estimated values. Typically, the operator’s knowledge and experience are cru-
cial to select previous operating points which have occurred under similar
loading and generation scenarios.
3. Using the previous time interval operating point is another technique for the
estimation of the operating point. This approach is commonly used on small
time interval optimization approaches in which abrupt changes in the demand
are not expected (e.g., EVCC problems).
It is important to remark that the estimation of the operation point is an important
issue to be taken into account when applying the presented formulations.
Furthermore, the technique chosen to determine the estimated operation point will
depend on the information available and the characteristics of the problem that is
being tackled.

References

1. J.A. Momoh., Electric Power System Applications of Optimization, 2nd edn. (CRC Press,
Boca Raton, 2009)
2. L. Aleixo et al., A general framework for active distribution network planning, in CIGRE
Symposium 2013, no. April 2013 (2013), pp. 1–8
3. “IBM ILOG CPLEX V12.1 User’s Manual for CPLEX.” CPLEX Division, Incline Village,
NV (2009)
4. MOSEK ApS, The MOSEK optimization tools manual—version 6.0. Copenhagen, 2009
5. G. O. Inc., Gurobi Optimizer reference manual, www.Gurobi.com, vol. 6 (2014), p. 572
6. R.H. Byrd, J. Nocedal, R.A. Waltz, Knitro: an integrated package for nonlinear optimization,
in Large–Scale Nonlinear Optimization, ed. by G. Di Pillo, M. Roma (Springer US, Boston,
MA, 2006), pp. 35–59
7. P. Bonami, J. Lee, BONMIN Users’ Manual, (2007)
8. D. Shirmohammadi, H.W. Hong, A. Semlyen, G.X. Luo, A compensation-based power flow
method for weakly meshed distribution and transmission networks. IEEE Trans. Power Syst.
3(2), 753–762 (1988)
3 Mathematical Optimization of Unbalanced Networks … 113

9. R.G. Cespedes, New method for the analysis of distribution networks. IEEE Trans. Power
Deliver. 5(1), 391–396 (1990)
10. J.F. Franco, M.J. Rider, R. Romero, A mixed-integer linear programming model for the
electric vehicle charging coordination problem in unbalanced electrical distribution systems.
IEEE Trans. Smart Grid 6(5), 2200–2210 (2015)
11. IEEE/PES, Distribution Test Feeders, 123-bus Feeder.
12. R. Fourer, D.M. Gay, B.W. Kernighan, AMPL: a modeling language for mathematical
programming, 2nd edn. (Brooks/Cole-Thomson Learning, Pacific Grove, CA, 2003)
13. EPRI, Open Distribution System Simulator, (2012)
14. International energy outlook 2013. (2013)
15. P.L. Dandeno et al., System load dynamics-simulation effects and determination of load
constants. IEEE Trans. Power Appar. Syst. PAS-92 (1973)
16. A. Padilha-Feltrin, D. Quijano, J.R. Mantovani, Volt-VAR multiobjective optimization to
peak-load relief and energy efficiency in distribution networks. IEEE Trans. Power Deliver.
30(2), 618–626 (2015)
17. H. Ahmadi, J.R. Martí, H.W. Dommel, A framework for volt-VAR optimization in
distribution systems. IEEE Trans. Smart Grid 6(3), 1473–1483 (2015)
18. C. Sabillon, O. Melgar Dominguez, J. Franco, M. Lavorato, M.J. Rider, Volt-VAR control
and energy storage device operation to improve the electric vehicle charging coordination in
unbalanced distribution networks. IEEE Trans. Sustain. Energy (2017)
19. L.M. Korunovic, S. Sterpu, S. Djokic, K. Yamashita, S. M. Villanueva, J.V. Milanovic,
Processing of load parameters based on Existing Load Models, in 2012 3rd IEEE PES
Innovative Smart Grid Technologies Europe (ISGT Europe) (2012), pp. 1–6
20. Task force on load representation for dynamic performance IEEE, Bibliography on load
models for power flow and dynamic performance simulation. IEEE Trans. Power Syst. 10 (1),
523–538 (1995)
21. N. Anglani, F. Fattori, G. Muliere, Electric vehicles penetration and grid impact for local
energy models, in 2012 IEEE International Energy Conference and Exhibition
(ENERGYCON) (2012), pp. 1009–1014
22. K. Clement-Nyns, E. Haesen, J. Driesen, The impact of charging plug-in hybrid electric
vehicles on a residential distribution grid. IEEE Trans. Power Syst. 25(1), 371–380 (2010)
23. D. Wu, D.C. Aliprantis, K. Gkritza, Electric energy and power consumption by light-duty
plug-in electric vehicles. IEEE Trans. Power Syst. 26(2), 738–746 (2011)
24. A. O’Connell, A. Keane, D. Flynn, Rolling multi-period optimization to control electric
vehicle charging in distribution networks, in 2014 IEEE PES General Meeting|Conference
Exposition (2014), p. 1
25. C. Sabillon Antunez, J.F. Franco, M.J. Rider, R. Romero, A new methodology for the optimal
charging coordination of electric vehicles considering vehicle-to-grid technology. IEEE
Trans. Sustain. Energy 7(2), 596–607 (2016)
26. A.T. Al-Awami, E. Sortomme, Coordinating vehicle-to-grid services with energy trading.
IEEE Trans. Smart Grid 3(1), 453–462 (2012)
27. T. Ackermann, G. Andersson, L. Söder, Distributed generation: a definition. Electr. Power
Syst. Res. 57(3), 195–204 (2001)
28. K. Qian, C. Zhou, Y. Yuan, X. Shi, M. Allan, Analysis of the environmental benefits of
Distributed Generation, in IEEE Power and Energy Society 2008 General Meeting:
Conversion and Delivery of Electrical Energy in the 21st Century, PES (2008)
29. A.C. Rueda-Medina, J.F. Franco, M.J. Rider, A. Padilha-Feltrin, R. Romero, A mixed-integer
linear programming approach for optimal type, size and allocation of distributed generation in
radial distribution systems. Electr. Power Syst. Res. 97, 133–143 (2013)
30. J.A.P. Lopes, N. Hatziargyriou, J. Mutale, P. Djapic, N. Jenkins, Integrating distributed
generation into electric power systems: a review of drivers, challenges and opportunities.
Electr. Power Syst. Res. 77(9), 1189–1203 (2007)
31. M. Bollen, F. Hassan, Integration of Distributed Generation in the Power System (2011)
114 C. F. Sabillón et al.

32. Y.M. Atwa, E.F. El-Saadany, MMa. Salama, R. Seethapathy, Optimal renewable resources
mix for distribution system energy loss minimization. IEEE Trans. Power Syst. 25(1), 360–
370 (2010)
33. S. Weckx, C. Gonzalez, J. Driesen, Combined central and local active and reactive power
control of PV inverters. IEEE Trans. Sustain. Energy 5(3), 776–784 (2014)
34. R. Tonkoski, L.A.C. Lopes, T.H.M. El-Fouly, Coordinated active power curtailment of grid
connected PV inverters for overvoltage prevention. IEEE Trans. Sustain. Energy 2(2), 139–
147 (2011)
35. M. Ross, R. Hidalgo, C. Abbey, G. Joos, Analysis of energy storage sizing and technologies,
in EPEC 2010—IEEE Electrical Power and Energy Conference: “Sustainable Energy for an
Intelligent Grid” (2010)
36. M.S. ElNozahy, T.K. Abdel-Galil, M.M.A. Salama, Probabilistic ESS sizing and scheduling
for improved integration of PHEVs and PV systems in residential distribution systems. Electr.
Power Syst. Res. 125, 55–66 (2015)
37. J.F. Franco, M.J. Rider, M. Lavorato, R. Romero, A mixed-integer LP model for the
reconfiguration of radial electric distribution systems considering distributed generation.
Electr. Power Syst. Res. 97, 51–60 (2013)
Chapter 4
Multi-stage Primary-Secondary
Planning Considering Wholesale-Retail
Markets

Mehrdad Setayesh Nazar, Alireza Heidari

and Mahmood Reza Haghifam

Abstract This chapter presents an approach for Integrated Distributed Generation

and primary-secondary network Expansion Planning (IDGNEP) in the presence of
wholesale and retail markets. The presented method uses a unified model to explore
the impacts of retail market participants on the IDGNEP procedure. While the
theory and practice of IDGNEP have advanced over the years, the Non-Utility
Retail Market Participants (NURMPs) and Customers’ Active MicroGrids
(CAMGs) introduce some other resources which can also be included in distribu-
tion network planning exercises. An electric distribution network may interchange
energy with wholesale/retail market participants and downward CAMGs. When the
volume of the energy interchanged between the network and NURMPs/CAMGs is
comparable with the volume of electricity delivered to the end users, the IDGNEP
results may considerably be different from the condition that no energy is inter-
changed. The presented model of IDGNEP is a Mixed Integer Non Linear
Programming (MINLP) problem and the introduced algorithm decomposes the
IDGNEP problem into multi sub-problems to achieve an optimal expansion plan-
ning of a network, in which the investment and operational costs are minimized,
while the reliability of the network is maximized. Demand Side Management
(DSM) programs, Distribution Automation (DA) investment alternatives and
NURMP and CAMGs contribution scenarios which may significantly change the
network’s resources are considered in IDGNEP formulation. The algorithm was
successfully tested for an urban distribution network.

M. S. Nazar (&)
Faculty of Electrical Engineering, Shahid Beheshti University, Tehran, Iran
e-mail: [email protected]
A. Heidari
School of Electrical Engineering and Telecommunication,
University of New South Wales, Sydney, Australia
e-mail: [email protected]
M. R. Haghifam
School of Electrical and Computer Engineering, Tarbiat Modares University,
Tehran, Iran
e-mail: [email protected]

© Springer Nature Singapore Pte Ltd. 2018 115

F. Shahnia et al. (eds.), Electric Distribution Network Planning,
Power Systems, https://doi.org/10.1007/978-981-10-7056-3_4
116 M. S. Nazar et al.

Keywords Distribution network expansion planning
Genetic algorithm (GA)

Optimization Active microgrids

4.1 Introduction

Distribution network electric loads are usually supplied through a common electric
distribution network and each end user is coupled with the main utility grid (de-
noted as ‘main grid’) through the point of common coupling as shown in Fig. 4.1
[1]. However, many of the end users may have Distributed Generation
(DG) facilities that supply energy to their energy-intensive industries and they may
behave as dispatchable loads by reducing their electricity withdrawal from the
distribution network and increasing the utilization of their electricity generation
systems. In addition, the main grid may transact electricity with upward wholesale
electricity market and its downward CAMGs [1].
Based on the end users’ electrical load group characteristics, land ownership and
operational constraints, the main grid can be segmented into different operational
zones. In addition, for an open access electric distribution network, CAMGs and
end users of different zones can transact energy with each other and they may form
various power exchange patterns. However, any retail market electricity transac-
tions between end users is analysed and approved by the Distribution System
Operator (DSO) in advance and then the transactions can be performed. Energy
interchanging between a network and wholesale and retail markets may change the

Fig. 4.1 Schematic diagram of the large primary/secondary electric distribution network
4 Multi-stage Primary-Secondary Planning Considering … 117

network’s resources, costs and reliability. Thus, the optimal resource planning and
operation of an energy interchanging network may be different from the ordinary
ones [2–5].
The DG and primary-secondary Network Expansion Planning (DGNEP) prob-
lem consists of determining of the location, the capacity, and the time of installation
of generation devices and network devices, depend on load growth conditions,
reliability criteria, DSM programs, DA investment alternatives and NURMPs/
CAMGs contribution scenarios.
The DGNEP is logical in light of demands and electric distribution network
optimal operation. However, the main operation decisions are critical due to the
two-way DSO and NURMPs/CAMGs interactions what will happen now to what
will happen later based on the networks constraints.
This book chapter is on the Integrated DGNEP (IDGNEP) algorithm that con-
siders the wholesale market price uncertainties, and NURMPs/CAMGs bid/offer
price scenarios. Further, it uses the optimal reconfiguration procedure to investigate
the adequacy of network resources under normal and contingent conditions.

4.2 Problem Modelling and Formulation

The wholesale market is assumed as a mandatory power pool. The DSO submits
purchase bids to the wholesale market as a wholesale market participant. In this
two-sided auction model, the independent system operator runs an auction market,
settles the market and returns the settlements to the wholesale market participants.
The retail market is assumed as a voluntary power market, in which the
NURMPs can contract bilaterally with the each other. The residual load is supplied
by the DSO. The DSO may use the UDGs or purchase energy from the NURMPs/
CAMGs and/or wholesale market to supply the residual load [4, 6, 7].
A NURMP can be classified as Non-Utility Distributed Generation (NUDG) and
Dispatchable/Non-Dispatchable Load (DL/NDL) as shown in Fig. 4.2.
The NUDG and DL/NDL can be categorized as [4]:
(1) A NUDG that might be dispatchable by paying an appropriate capacity and
energy fee is known as Dispatchable NUDG (DNUDG). The NUDG can
participate in the retail market and the generation cost of this type of NURMP
can be formulated as (4.1):
!
X
Np
CDNUDG ¼ a:CapDNUDG þ bi :PDNUDG
gi :si :r ð4:1Þ
i¼1

(2) A NUDG that might be non-dispatchable technically or economically is known

as Non-Dispatchable NUDG (NDNUDG). The NDNUDG can participate in the
retail market and the power generation cost of this type of DG can be formu-
lated as (4.2):
118 M. S. Nazar et al.

Fig. 4.2 Schematic diagram of the considered wholesale/retail market and active microgrid

!
X
Np
CNDNUDG ¼ gi :PNDNUDG
gi :si :r ð4:2Þ
i¼1

(3) A NURMP that reduces its electric demand can be represented as DL or NDL
based on the technical and economic parameters. The DL/NDL can participate
in the retail market and the costs of load reduction in DL and NDL are for-
mulated as (4.3) and (4.4), respectively:
!
X
Np
CDL ¼ c:CapDL þ vi :PDL
d i :si :r ð4:3Þ
i¼1

!
X
Np
CNDL ¼ 1i :PNDL
d i :si :r ð4:4Þ
i¼1

A customer’s active microgrid can be classified as Customer’s Dispatchable

MicroGrid (CDMG) and Customer’s Non-Dispatchable MicroGrid (CNDMG).
The CDMG and CNDMG can be categorized as
4 Multi-stage Primary-Secondary Planning Considering … 119

(1) A CAMG that might be dispatchable is known as CDMG. The CDMG sells its
power to the DSO at its point of common coupling and it can’t participate in the
retail market. The generation/load reduction cost of this type of CAMGs can be
formulated as (4.5):
!
X
Np
CCDMG ¼ -i :PCDMG
gi :si :r ð4:5Þ
i¼1

(2) A CAMG that might be non-dispatchable is known as CNDMG. The CNDMG

may have intermittent power generation units and it can’t sell its power to the
DSO or participate in the retail market. This type of CAMG is modelled as an
embedded generation of a load point.
The DSO makes optimal decisions throughout IDGNEP horizon with incom-
plete information and it determines the optimal values of problem decision variables
that consist of the location, the capacity, and the time of installation the UDGs and
network devices. The DSO uses an estimated data of NURMPs/CDMGs location,
type and capacity to determine optimal generation schedules of its generation units,
electricity transactions with wholesale market and NURMPs/CDMGs, and
contingency-based load shedding alternatives.
The IDGNEP problem is subject to the three sources of uncertainty: wholesale
electricity market prices, NURMPs/CDMGs contribution scenarios, and network
contingencies.
Thus, the uncertainty can be modelled as a multi-stage decision making problem
based on a scenario driven algorithm. An iterative four stage optimization proce-
dure is presented that minimizes the total network costs based on the wholesale
market price scenarios at the first stage. At the second stage, the algorithm deter-
mines the optimal device allocations and capacity selections. At the third stage, the
algorithm finds the optimal NURMPs/CDMGs contribution scenarios to maintain
network reliability at the third stage. Finally, at the fourth stage, the optimal
restoration problem investigates the adequacy of network resources under contin-
gent conditions.

4.2.1 First Stage Problem Formulation

At the first stage, the DSO determines the number of wholesale market scenarios for
each bi-annual periods and it estimates the electric loads, upward wholesale market
prices, and power exchanges with upward network and NURMPs/CDMGs con-
tribution scenarios. Based on the wholesale market price scenarios, the first stage
minimizes the cost allocation problem that consists of whole investment and
operational costs and customer interruption cost for the bi-annually periods of
planning years. The objective function of the first stage problem can be written as
(4.6):
120 M. S. Nazar et al.

Nyear X
X Nzone X
Min C1 ¼ ½ CijX :uijX
i¼1 j¼1 X
XSc
WM
þ WWM Purchased energy  uPurchase
WM ijk
Scenario
:MCPijk :Eijk
WM

k¼1
X Sc
NURMP
þ WNURMP Purchased energy  uPurchase
NURMPijk
Scenario
:priceNURMP
ijk :Eijk
NURMP

k¼1
X Sc
CDMG
þ WCDMG Purchased energyk  uPurchase
CDMGijk
Scenario
:priceCDMG
ijk :Eijk
CDMG

k¼1
X Outage
N Critical
þ WCIC  CICijk 
k¼1
X ¼ fSub; Feed; RPS; UDG; DSM; DAg
ð4:6Þ

The first stage objective function can be decomposed into three groups:
(1) substation, feeder, RPS, UDG, DSM and DA investment plus aggregated
operation costs, (2) the costs of purchased energy from upward network and
NURMP and CDMG as the second group that is calculated in the third stage
problem, (3) customer interruption cost as the third group. The first term (invest-
ment term) of (4.6) is described in the second stage problem and the customer
interruption cost term is determined in the fourth stage problem. The Decision
Variable Set (DVS) of the first stage problem can be written as (4.7):

DVS1 ¼ ½uX ; uPurchase Scenario

; uPurchase Scenario
; uPurchase Scenario

WM NURMP CDMG
ð4:7Þ
X ¼ fSub; Feed; RPS; UDG; DSM; DAg

The technical constraints are categorized into: device loading constraints, the
entire load centres to be served constraints and dc load flow constraints.
The dc load flow constraints for network can be represented as (4.8):

f1 ðx; u; zÞ ¼ 0 ð4:8Þ

where x, u, z are problem variables, controls and network topology, respectively.

Technical constraints can be compactly represented as (4.9):

g1 ðx; u; zÞ  0 ð4:9Þ

The first stage results can be used in the second stage to determine where and
when an investment is needed. For every first stage scenario, the second stage
problem optimizes the characteristics of network devices and their technical
parameters for each period of the planning years.
4 Multi-stage Primary-Secondary Planning Considering … 121

4.2.2 Second Stage Problem Formulation

At the second stage, the DSO estimates the network’s electric loads, upward
wholesale market prices, and power exchanges with upward network and
NURMPs/CDMGs for quarter of year periods. At this stage, the DSO optimizes the
network device selection and allocation parameters. The second stage objective
function can be presented as (4.10):

Nyear X
X Nzone X X
Min C2a ¼ ½ CijC :wijC þ CijD :wijD 
i¼1 j¼1 C D

C ¼ fNsub; N Fr; N UDG; N DSM; N DA; N RPSg ð4:10Þ

D ¼ fN NURMP; N CDMGg
a 2 First stage problem state space

The second stage objective function can be decomposed into two groups:
(1) substation, feeder, UDG, DSM, DA and RPS allocation and capacity selection,
(2) the NURMP and CDMG contribution scenarios as the second group that is
calculated in the third stage problem.
The DVS of the second stage problem can be presented as (4.11):

DVS2 ¼ ½wC ; wD 
C ¼ fNsub; N Fr; N UDG; N DSM; N DA; N RPSg ð4:11Þ
D ¼ fN NURMP; N CDMGg

The constraints can be categorized into: voltage constraints, device loading

constraints, power balance at the network nodes under normal and contingent
conditions, the entire load centres to be served constraints, uniqueness parameter
selection constraints, operation of radial network under normal and contingent
conditions and reliability constraints.
Technical and uniqueness parameter selection constraints can be presented as

g2 ðx; u; zÞ  0 ð4:12Þ

Load flow constraints are:

f2 ðx; u; zÞ ¼ 0 ð4:13Þ
122 M. S. Nazar et al.

4.2.3 Third Stage Problem Formulation

Based on the second stage results, the contribution scenarios of NURMP/CDMGs is

investigated for cost minimization. The third stage problem explores the monthly
NURMP bid/offer state space and finds the best NURMPs/CDMGs contribution
scenarios [2]. The DSO supplies the residual load of the retail market, thus, the third
stage objective function can be stated as (4.14):

Nyear XX
X Nzone Np X
Min C3b ¼ ½ CijkK :/ijkK
i¼1 j¼1 k¼1 K

þ WPurchased energyk  ðMCPijk þ Trans: service priceijk Þ:Eijk

WM

K ¼ fN UDG; N NDNUDG; N DNUDG; N DL; N NDL; N CDMGg
b 2 Second stage problem state space
ð4:14Þ

The third stage objective function consists of UDG, NURMP and CDMG
contribution scenarios. The DSO uses the scenario driven information to describe
NDNUDG, DNUDG, DL, NDL and CDMG contribution scenarios. By selection of
the best NURMP/CDMG contribution scenarios, the DSO optimizes the decision
variables of (4.14).
The DVS of the third stage problem can be presented as (4.15):

DVS3 ¼ ½wK 
K ¼ fN UDG; N NDNUDG; N DNUDG; N DL; N NDL; N CDMGg
ð4:15Þ

The third stage technical selection constraints can be presented as

g3 ðx; u; zÞ  0 ð4:16Þ

Load flow constraints are:

f3 ðx; u; zÞ ¼ 0 ð4:17Þ

If the NURMPs/CDMGs contribution scenarios are fixed, the feasibility of

network restoration can be investigated. This problem is a slave problem of the third
stage problem that optimizes the network resource coordination in contingent
conditions.
4 Multi-stage Primary-Secondary Planning Considering … 123

4.2.4 Fourth Stage Problem Formulation

The fourth stage problem tries to find the monthly optimal resource coordination in
contingent conditions. The control variables of the network can be categorized as
[2, 8]:
1. Discrete control variables of the network resources such as switching of tie
switches and capacitors, and
2. Continuous control variables of the network resources such as UDG, DNUDG,
CDMG and DL.
The fourth stage objective function is introduced as

Nyear X Noutage
X Nzone X X
Min Cc4 ¼ ½CICijk þ DCOPðNÞ ijk 
i¼1 j¼1 k¼1 N
ð4:18Þ
N ¼ fUDG; DNUDG; CDMG; DLg
c 2 Third stage problem state space

The fourth stage procedure investigates the adequacy of network resources for
restoration of the most important loads. It tries to switch the tie switches and
capacitors and optimize the network resource coordination. For new topology, the
optimal coordination of resource coordination problem can be solved by custom
optimal power flow. The optimization constraints are voltage drop, line loading and
load flow constraints. The DVS of the fifth stage can be written as (4.19):

DVS4 ¼ ½Cb1 ; TS1 ; . . .; CbN RPS

; TSN TS
 ð4:19Þ

Cbi ¼ 1 if the i th element shunt is used; else it equals 0: ð4:20Þ

TSi ¼ 1 if the i th tie switch is used; else it equals 0: ð4:21Þ

Technical and radial operation constraints can be compactly represented as

y0n
5 ðx; u; zÞ  0 8n 2 f0; 1; . . .; Noutageg ð4:22Þ

Load flow constraints are formulated as

f50n ðx; u; zÞ ¼ 0 8n 2 f0; 1; . . .; Noutageg ð4:23Þ

124 M. S. Nazar et al.

4.3 Solution Algorithm

The mentioned model of IDGNEP is a scenario driven MINLP problem.

The IDGNEP sub-problems are nonlinear and non-convex, if the first stage problem
parameters are fixed. The fourth stage problem uses discrete control variables such
as capacitor and line switching. It also generates new state spaces for solving the
optimal restoration. Every effective switching will generate a new state space based
on the load variations. The fourth-stage problem has a great state space and its
solution algorithm must be able to effectively search this space. For optimization
procedure, a Genetic Algorithm (GA) with variable fitness functions is used. The
rates of the operators are adapted in a deterministic, reinforcement-based manner.
The behaviour of each operator (that is, the specific way it operates) is modified by
changing its parameter values. Figure 4.3 depicts the flowchart of the introduced
multi-stage optimization algorithm. At first, the first stage problem is optimized for
each bi-annual period of the planning years for wholesale market price scenarios.
Then, the second stage problem optimality is investigated. At the third stage, the
NURMP/CDMG contribution scenarios are optimized. Finally, at the fourth stage a
feasible and optimal restoration problem solution is investigated.
In order to map the possible solutions of the problem, a binary basis codification
is employed. The first, second, third and fourth stage GA Population (GAP) of
decision variables can be presented as

GAPi ¼ DVSi 8i ¼ f1; 2; 3; 4g ð4:24Þ

To improve the performance and speed of the specified GA, a list of suitable
candidates is selected for the first generation of the chromosomes. This population
could be generated using engineering experience rules. In order to map the possible
solutions of problem, a binary basis codification is employed.
Two operators, namely crossover and mutation, are applied to the first genera-
tion and as a result new chromosomes are generated. Figure 4.4 depicts the
crossover process for the first stage problem. The crossover operator is employed
for each of the decision variable set. Figure 4.5 depicts the mutation process for the
first stage problem.
For implementation of operational constraints in the optimization process, a
penalty factor representation is used. The final optimization fitness function of the
multi-stage problem can be written as

Max Z ¼ M  C  W:gðu; x; zÞ  W 0 :f ðu; x; zÞ ð4:25Þ

where Z and M are objective function and high number vectors, respectively.
W and W′ are weight factor vectors that can be increased linearly through iterations
from zero to a very high number.
The Weighted Reliability Index (WRI) is used for stopping criteria, defined as
4 Multi-stage Primary-Secondary Planning Considering … 125

Start

Reading Existing and Future

Distribution System Data Retail Market Simulation and Determination of
NURMP Contribution Scenarios

Determine the Distribution System Resources for Each Bi -annual

Determination of Third Stage Problem Initial
Period of Planning
Population

Determination of Load Centres Parameters

Third Stage Problem Objective
Function Optimization

Determination of Distribution System Energy and Peak Load

Simulation of Wholesale Market Yes

Any
Penalty Function
Constraints
Calculation
Violation?
Determination of Wholesale Market Prices for Each Bi-annual Period

No
Estimation of Electric Loads, Upward Wholesale Market Prices, and Power Exchanges with Upward
Network and NURMP/CDMG Contribution s Optimal Solution of Third Stage
Problem

Determination of First Stage Problem Initial Population

Determination of Fourth Stage Problem Initial
Population
First Stage Problem Objective
Function Optimization

Coding of Topology Chromosomes

Yes

Any Fourth Stage Problem Objective

Penalty Function
Constraints Function Optimization
Calculation
Violation?

Decoding of Topology Chromosomes for

No Tie switch and Capacitor Switching

Optimal Solution of First Stage Problem

Yes
Any
Penalty Function
Constraints
Calculation
Violation?
Determination of Second Stage Problem Initial
Population No

Yes
End of Topology Change ?

Estimation of Electric Loads, Upward Wholesale Market Prices, and Power Exchanges with Upward
No
Network and NURMPs And CDMGs for quarter of year periods

Optimal Solution of Fourth Stage Problem

Determination of Second Stage Problem Initial Population

Second Stage Problem Objective

Function Optimization No

Satisfaction of Stopping Criteria of Fourth

Stage Problem

Yes
Any
Penalty Function
Constraints
Calculation
Violation? Yes
No

No
Optimal Solution of Second Stage Problem End of Wholesale Market
Price Scenarios?

Estimation of Monthly E lectric Loads, Upward Wholesale Market Prices, and Power Exchanges with
Upward Network and NURMPs And CDMGs
Yes

Save The Optimal

Generation and Primary-
Secondary Distribution
Stop
Expansion Planning Data
for Specified Wholesale
Market Price

Fig. 4.3 Flowchart of introduced algorithm for solving of IDGNEP

126 M. S. Nazar et al.

Fig. 4.4 Crossover process of the first stage problem chromosomes

Fig. 4.5 Mutation process of the first stage problem chromosomes

0 0
WRI ¼ wf1  SAIDI þ wf2  SAIFI ð4:26Þ

where

SAIFI ¼ Total number of customer interruptions=total number of customers served:

ð4:27Þ

SAIDI ¼ Sum of customers' interruption duration=total number of customers:

ð4:28Þ
4 Multi-stage Primary-Secondary Planning Considering … 127

4.4 Numerical Results

The discussed algorithm was applied to an urban electric distribution network. This
network was a part of a city network with 45,000 customers at the horizon year. The
selected region will have about 7000 residential and commercial customers at the
horizon year. Its primary network voltage is 20 kV and its secondary network
voltage is 400 V. The time horizon was chosen 7 years into the future. The average
repair time of the permanent faults is assumed as 2 h and the weight factors are 1.0.
The stopping criterion was selected as WRI < 8 for all the buses based on the
utility reliability standard with wf1′ = wf2′ = 1/2 or the number of
iterations > 2000.
By wholesale/retail market simulation, the most important NURMP contribution
scenarios and the residual load of the network were determined for different periods
and the NURMPs and CAMGs groups were recognized. The DSO investigates the
optimum NURMP/CAMGs contribution scenarios for supplying of residual load at
the normal and contingent conditions. By wholesale market simulation, different
wholesale market price categories were calculated, which their values are shown in
Monetary Units (MUs) in Table 4.1. Table 4.2 shows the total costs of NURMPs/
CAMGs contribution. Table 4.3 shows the final load forecasting results considering
NURMPs/CAMGs contribution.
The second stage problem determined the optimum allocation of network sub-
stations and feeder routing. Table 4.4 shows the final transformer capacity selection
results that were determined in the second stage. This problem consists of using of
the existing devices, which their life cycle is not expired, but the devices with
greater capacity are needed. For example, a pole mounted transformer, which
supplies a load, but the capacity of the existing transformer is not enough for
supplying the load, must be changed by another one with greater capacity. The
older transformer can be used in other substations to supply an appropriate load.
Table 4.5 depicts the final optimized IDGNEP costs. The transformer and feeder
investment costs takes on a value 1921 billion MUs, which is decomposed in 159.1,
138.0, 129.2, 135.6, 107.8, 571.6 and 679.4 billion MUs for the first, second, third,
fourth, fifth, sixth and seventh year of planning, respectively. Further, the DA and
DSM and RPS investment costs takes on a value 4.792 billion MUs, which is
decomposed in 1.0859, 1.7294, 0.5185, 0.5594, 0.2425, 0.2983 and 0.3585 billion
MUs for the first, second, third, fourth, fifth, sixth and seventh year of planning,
respectively. Finally, the NURMP and CAMG contribution costs takes on a value
8798 billion MUs, which is decomposed in 775.267, 899.48, 1063, 1261.2, 1504.6,
1576.8 and 1718 billion MUs for the first, second, third, fourth, fifth, sixth and
seventh year of planning, respectively. The UDGs contribution factor is about 0.25,
0.28, 0.32, 0.35, 0.38, 0.41 and 0.43 for the first, second, third, fourth, fifth, sixth
and seventh year of planning, respectively.
The fourth stage problem investigates the adequacy of network resources for
restoration of the most important loads. The algorithm tries to switch the tie
switches and capacitors and find a new set of network resources. The final and
 128

Table 4.1 Estimated upward wholesale market price and network energy consumption
Year First year Second Third year Fourth year Fifth year Sixth year Seventh
year year
Estimated first scenario wholesale market 7.50E + 01 7.80E + 01 8.25E + 01 8.53E + 01 8.81E + 01 9.24E + 01 9.63E + 01
price (MU/MWh)
Estimated first scenario energy 7.83E + 07 8.22E + 07 8.63E + 07 9.07E + 07 9.52E + 07 1.00E + 08 1.05E + 08
consumption (kWh)
Estimated second scenario wholesale market 9.23E + 01 9.59E + 01 1.01E + 02 1.05E + 02 1.08E + 02 1.14E + 02 1.18E + 02
price (MU/MWh)
Estimated second scenario energy 7.46E + 07 7.91E + 07 8.31E + 07 8.72E + 07 9.16E + 07 9.62E + 07 1.01E + 08
consumption (kWh)
M. S. Nazar et al.
Table 4.2 Total costs of NURMPs/CAMGs contribution
Type of First year Second year Third year Fourth year Fifth year Sixth year Seventh year
NURMP cost (Million cost (Million cost (Million cost (Million cost (Million cost (Million cost (Million
MUs) MUs) MUs) MUs) MUs) MUs) MUs)
DNUDG 225,520.84 261,668.25 310,848.66 384,383.45 476,444.07 502,622.19 547,858.19
NDNUDGG 31,083.62 36,070.26 42,561.34 50,540.28 62,851.62 65,716.39 69,093.95
DL 204,208.7472 238,021.6392 284,517.7848 327,717.3744 378,447.804 393,343.5096 429,672.7188
NDL 54,581.72 62,166.38 70,874.21 80,873.68 92,346.59 97,118.3 105,858.94
CDMG 226,408.1 263,589.4 311,137.7 368,826.7 438,987.6 460,783.1 503,111.6
4 Multi-stage Primary-Secondary Planning Considering …

CNDMG 33,464.28 37,967.61 43,095.52 48,942.75 55,592.55 57,299.47 62,456.42

129
Table 4.3 The final load forecasting results considering NURMPs/CAMGs contribution
130

Load Load in first Load in second Load in third Load in fourth Load in fifth Load in sixth Load in seventh
centre year (kVA) year (kVA) year (kVA) year (kVA) year (kVA) year (kVA) year (kVA)
1 130.8 126.9 145.1 148.3 162.0 167.6 177.7
2 305.2 326.5 345.2 375.4 394.7 410.9 437.1
3 135.6 134.7 141.7 155.6 157.5 176.4 171.5
4 253.2 266.3 284.8 306.6 321.2 332.0 360.1
5 190.1 215.0 218.0 225.8 240.4 259.8 273.7
6 317.7 325.6 356.2 367.1 387.3 421.5 438.0
7 294.7 319.0 343.4 354.1 370.6 392.4 419.3
8 53.6 57.0 60.2 75.7 66.9 72.1 83.7
9 221.3 237.6 241.8 263.9 275.7 295.9 310.5
10 82.6 86.0 99.1 100.4 100.2 105.6 116.3
11 278.1 297.2 312.4 329.4 356.1 367.3 384.4
12 217.3 239.1 252.0 251.6 276.6 289.5 307.4
13 312.6 330.2 358.9 375.1 399.5 410.1 434.8
14 158.2 162.1 180.8 188.5 199.0 212.0 226.4
15 343.2 357.2 387.6 399.4 429.9 455.6 473.9
16 784.5 835.1 884.3 935.0 997.2 1046.3 1109.5
17 569.7 606.0 646.2 675.6 721.2 762.8 797.8
18 618.9 695.8 748.4 786.8 825.7 879.6 918.0
19 61.4 74.1 71.9 72.6 85.9 78.5 92.6
20 288.4 303.7 324.9 342.8 348.5 376.6 388.7
21 40.8 54.1 52.8 44.5 51.8 51.2 62.7
22 366.7 383.4 406.5 434.6 461.8 479.9 509.6
23 40.3 45.9 43.2 44.9 43.0 52.3 48.4
(continued)
M. S. Nazar et al.
Table 4.3 (continued)
Load Load in first Load in second Load in third Load in fourth Load in fifth Load in sixth Load in seventh
centre year (kVA) year (kVA) year (kVA) year (kVA) year (kVA) year (kVA) year (kVA)
24 271.0 274.1 303.8 315.1 329.8 357.4 368.3
25 38.2 36.5 31.8 40.3 34.1 41.6 50.9
26 212.4 232.7 239.6 258.0 275.5 280.5 297.8
27 73.4 76.3 92.2 83.0 88.4 101.1 103.0
28 167.1 182.9 184.9 202.3 205.2 225.4 225.6
29 493.4 529.4 558.5 580.1 628.0 656.5 690.0
30 318.0 336.3 356.6 367.0 391.7 420.4 433.0
31 825.6 870.9 928.5 968.4 1032.3 1083.3 1154.4
32 886.1 942.0 995.1 1055.6 1112.3 1166.9 1234.2
33 371.4 394.5 414.3 437.3 472.8 492.6 520.1
34 213.3 218.5 228.5 249.8 255.7 275.6 290.9
35 131.9 140.8 153.5 162.1 169.5 176.9 190.8
36 145.5 155.8 164.2 174.4 174.7 188.8 196.5
37 165.0 181.4 193.4 202.1 212.8 222.0 236.5
4 Multi-stage Primary-Secondary Planning Considering …

38 496.6 522.2 559.9 580.8 619.5 650.1 686.3

39 320.4 329.8 353.0 375.0 396.9 411.7 435.3
40 489.1 519.5 544.3 575.1 604.5 638.9 670.3
41 616.0 646.5 678.9 724.2 764.8 803.0 854.5
42 162.7 183.8 193.4 192.2 214.5 213.0 233.9
43 496.5 520.9 556.7 588.2 615.1 657.8 688.3
44 306.1 336.3 343.4 373.2 384.9 409.9 438.6
45 214.5 225.4 238.5 255.5 255.6 271.9 293.1
46 168.3 172.5 186.6 195.7 210.6 211.4 236.2
(continued)
131
Table 4.3 (continued)
132

Load Load in first Load in second Load in third Load in fourth Load in fifth Load in sixth Load in seventh
centre year (kVA) year (kVA) year (kVA) year (kVA) year (kVA) year (kVA) year (kVA)
47 81.7 80.0 78.0 82.8 95.5 95.5 104.8
48 83.7 78.9 90.7 82.4 88.1 99.6 110.1
49 152.7 156.0 160.6 174.1 186.4 197.3 200.7
50 494.7 530.0 552.0 592.0 623.8 649.4 690.5
51 877.6 931.6 986.9 1052.1 1104.5 1179.0 1237.9
52 64.7 72.4 68.1 72.3 72.1 77.0 83.1
53 162.3 181.5 182.2 203.4 202.3 224.1 237.4
54 141.9 152.3 165.5 172.4 186.3 196.2 201.4
55 133.3 142.6 145.1 166.9 162.8 184.4 183.9
56 489.9 526.5 552.7 581.0 618.0 650.8 691.6
57 214.3 225.8 232.2 244.1 259.0 277.2 295.4
58 170.5 180.0 192.1 200.8 201.8 217.6 229.4
59 64.7 71.4 70.2 81.0 84.1 77.6 82.0
60 133.5 142.0 155.6 164.3 172.0 179.2 193.0
61 81.3 79.3 86.3 86.0 97.5 100.4 105.2
62 496.8 532.4 560.4 582.2 624.3 652.0 690.8
63 217.5 223.9 229.6 253.9 263.1 277.5 288.0
64 152.1 159.8 170.5 176.5 179.2 199.1 197.7
65 42.7 43.7 54.7 48.3 56.9 56.9 60.2
66 167.5 178.2 179.5 192.4 202.0 222.0 229.8
67 882.9 938.9 992.2 1046.4 1105.7 1171.5 1241.8
68 137.8 150.7 149.1 165.0 161.7 171.0 181.1
69 80.8 74.1 86.7 94.1 101.1 95.7 109.2
(continued)
M. S. Nazar et al.
Table 4.3 (continued)
Load Load in first Load in second Load in third Load in fourth Load in fifth Load in sixth Load in seventh
centre year (kVA) year (kVA) year (kVA) year (kVA) year (kVA) year (kVA) year (kVA)
70 170.5 176.3 182.2 197.6 215.0 223.6 233.6
71 39.1 47.4 50.7 52.9 48.7 54.7 54.3
72 151.1 148.3 164.8 168.9 181.4 185.3 207.4
73 128.7 149.8 153.3 163.0 171.4 178.2 183.0
74 494.7 531.3 555.3 592.0 618.8 654.3 694.3
75 214.1 221.8 228.4 247.7 266.8 280.8 292.9
76 889.5 941.9 993.0 1048.3 1111.9 1173.2 1232.9
77 494.0 520.1 554.7 587.0 620.9 661.0 685.2
78 149.9 156.3 157.2 168.8 178.3 196.5 206.1
79 82.3 76.8 84.1 91.8 97.9 94.1 108.0
80 162.8 175.5 185.6 202.7 213.9 213.3 229.6
4 Multi-stage Primary-Secondary Planning Considering …
133
Table 4.4 The final transformer capacity selection results
134

Load First year Second year Third year Fourth year Fifth year Sixth year Seventh year
centre capacity (kVA) capacity (kVA) capacity (kVA) capacity (kVA) capacity (kVA) capacity (kVA) capacity (kVA)
1 200 200 200 200 200 200 250
2 400 400 500 500 500 500 630
3 200 200 200 200 200 250 250
4 315 315 400 400 400 400 500
5 250 315 315 315 315 315 400
6 400 400 500 500 500 500 630
7 400 400 500 500 500 500 500
8 100 100 100 100 100 100 100
9 315 315 315 315 400 400 400
10 100 200 200 200 200 200 200
11 400 400 400 400 500 500 500
12 315 315 315 315 400 400 400
13 400 400 500 500 500 500 630
14 200 200 250 250 250 250 315
15 500 500 500 500 630 630 630
16 1000 1000 1250 1250 1250 1250 1600
17 800 800 800 800 1000 1000 1000
18 800 1000 1000 1000 1000 1250 1250
19 100 100 100 100 200 100 200
20 400 400 400 500 500 500 500
21 50 100 100 100 100 100 100
22 500 500 500 630 630 630 630
23 50 100 100 100 100 100 100
(continued)
M. S. Nazar et al.
Table 4.4 (continued)
Load First year Second year Third year Fourth year Fifth year Sixth year Seventh year
centre capacity (kVA) capacity (kVA) capacity (kVA) capacity (kVA) capacity (kVA) capacity (kVA) capacity (kVA)
24 400 400 400 400 400 500 500
25 50 50 50 50 50 50 100
26 250 315 315 315 400 400 400
27 100 100 200 100 200 200 200
28 200 250 250 250 250 315 315
29 630 630 800 800 800 800 1000
30 400 400 500 500 500 500 630
31 1000 1250 1250 1250 1250 1600 1600
32 1250 1250 1250 1250 1600 1600 1600
33 500 500 500 630 630 630 630
34 315 315 315 315 315 400 400
35 200 200 200 200 200 250 250
36 200 200 200 250 250 250 250
37 200 250 250 250 315 315 315
4 Multi-stage Primary-Secondary Planning Considering …

38 630 630 800 800 800 800 1000

39 400 400 500 500 500 500 630
40 630 630 800 800 800 800 800
41 800 800 800 1000 1000 1000 1250
42 200 250 250 250 315 315 315
43 630 630 800 800 800 800 1000
44 400 400 500 500 500 500 630
45 315 315 315 315 315 400 400
46 200 250 250 250 250 250 315
(continued)
135
Table 4.4 (continued)
136

Load First year Second year Third year Fourth year Fifth year Sixth year Seventh year
centre capacity (kVA) capacity (kVA) capacity (kVA) capacity (kVA) capacity (kVA) capacity (kVA) capacity (kVA)
47 100 100 100 100 200 200 200
48 100 100 200 100 200 200 200
49 200 200 200 250 250 250 250
50 630 630 800 800 800 800 1000
51 1250 1250 1250 1250 1600 1600 1600
52 100 100 100 100 100 100 100
53 200 250 250 250 250 315 315
54 200 200 200 250 250 250 250
55 200 200 200 200 200 250 250
56 630 630 800 800 800 800 1000
57 315 315 315 315 315 400 400
58 250 250 250 250 250 315 315
59 100 100 100 100 100 100 100
60 200 200 200 200 250 250 250
61 100 100 200 200 200 200 200
62 630 630 800 800 800 800 1000
63 315 315 315 315 315 400 400
64 200 200 250 250 250 250 250
65 100 100 100 100 100 100 100
66 200 250 250 250 250 315 315
67 1250 1250 1250 1250 1600 1600 1600
68 200 200 200 200 200 250 250
69 100 100 200 200 200 200 200
(continued)
M. S. Nazar et al.
Table 4.4 (continued)
Load First year Second year Third year Fourth year Fifth year Sixth year Seventh year
centre capacity (kVA) capacity (kVA) capacity (kVA) capacity (kVA) capacity (kVA) capacity (kVA) capacity (kVA)
70 250 250 250 250 315 315 315
71 50 100 100 100 100 100 100
72 200 200 200 200 250 250 250
73 200 200 200 200 250 250 250
74 630 630 800 800 800 800 1000
75 315 315 315 315 315 400 400
76 1250 1250 1250 1250 1600 1600 1600
77 630 630 800 800 800 800 1000
78 200 200 200 200 250 250 250
79 100 100 100 200 200 200 200
80 200 250 250 250 315 315 315
4 Multi-stage Primary-Secondary Planning Considering …
137
Table 4.5 Final IDGNEP results
138

Costs First year Second Third year Fourth year Fifth year Sixth year Seventh
cost year cost cost cost cost cost year cost
Total transformer and feeder installation 1.60E + 08 1.40E + 08 1.30E + 08 1.40E + 08 1.10E + 08 5.70E + 08 6.80E + 08
Costs (1000 MUs)
Total DA and DSM and RPS installation 1.1E + 06 1.7E + 06 5.2E + 05 5.6E + 05 2.4E + 05 3.0E + 05 3.6E + 05
costs (1000 MUs)
Total DNUDGs and NDNUDGs 2.57E + 08 2.98E + 08 3.53E + 08 4.35E + 08 5.39E + 08 5.68E + 08 6.17E + 08
contribution costs (1000 MUs)
Total DLs and NDLs contribution 2.59E + 08 3.00E + 08 3.56E + 08 4.09E + 08 4.71E + 08 4.90E + 08 5.36E + 08
costs (1000 MUs)
Total CAMGs contribution 2.61E + 08 3.03E + 08 3.54E + 08 4.18E + 08 4.95E + 08 5.18E + 08 5.66E + 08
costs (1000 MUs)
Total UDG energy generation (kWh) 4.31E + 07 5.08E + 07 6.39E + 07 7.21E + 07 8.52E + 07 9.58E + 07 1.10E + 08
Total energy purchased from wholesale 1.03E + 08 9.99E + 07 9.78E + 07 9.07E + 07 8.74E + 07 8.41E + 07 8.71E + 07
market (kWh)
M. S. Nazar et al.
4 Multi-stage Primary-Secondary Planning Considering … 139

optimum topology of the network has 2463 independent single short and long time
failures. Based on the fourth stage procedure, the stopping criterion was selected as
WRI. The optimized network topology is shown in Fig. 4.6 for the seventh year of
planning years. The optimal tie switch allocations are also illustrated. Under normal
conditions, the sectionalizers are normally opened and can be used for optimal
restoration. The corresponding WRI index is shown in Fig. 4.7 for the seventh year
of planning years. As Fig. 4.7 shows, the maximum WRI index takes on a value
8.72 at bus 5 for the seventh year of planning years.

Appendix

The notations used throughout this chapter are listed below:

E WM Total energy purchased from the wholesale market to supply
residual load
E NURMP Total energy purchased from the NURMPs to supply
residual load
E CAMG Total energy purchased from the CAMGs to supply residual
load
CapDNUDG Generation capacity prepared by DNUDG
CapDL Load reduction capacity prepared by DL
CX Present worth of cost allocation of X set
CC Present worth of device installation cost of C set
CD Present worth of NURMP/CDMG contribution cost of D set
CK Present worth of contribution cost of K set
COPðNÞ Present worth of operation costs of N set
N Fr Number of primary/secondary feeder type candidates for
installation
MCP Marginal clearing price of the wholesale market
Nyear Number of planning years
N Critical Outage Number of network critical outages
Np Number of load curve periods
Nzone Number of distribution network zones
N RPS Number of RPS installation candidates
N DSM Number of DSM installation candidates
N DA Number of DA installation candidates
N_UDG Number of UDG installation alternatives
N_NURMP Number of NURMP installation scenario at the second stage
problem
N CDMG Number of CDMG contribution alternatives
N_NDNUDG Number of NDNUDG contribution alternatives
N_DNUDG Number of DNUDG contribution alternatives
N_DL Number of DL contribution alternatives
140 M. S. Nazar et al.

Fig. 4.6 The optimized topology for the seventh year of planning years
4 Multi-stage Primary-Secondary Planning Considering … 141

Fig. 4.7 The WRI at the seventh year of planning years

N_NDL Number of NDL contribution alternatives

Nsub Number of primary/secondary substations candidates
priceNURMP Price of energy purchased from the NURMPs to supply
residual load
priceCDMG Price of energy purchased from the CDMGs to supply
residual load
Trans_Service_price Transmission service price of upward transmission network
which delivers energy to the DSO
W Weighting Factor
WM_Sc Number of wholesale market price scenarios
NURMP_Sc Number of NURMP contribution scenarios
CDMG_Sc Number of CDMG contribution scenarios
a Generation capacity fee of DNUDG
b Energy generation fee of DNUDG
v Load reduction energy fee of DL
uPurchase
WM
Scenario Decision variable of purchasing energy from the wholesale
market
uPurchase
NURMP
Scenario Decision variable of purchasing energy from NURMP
uCDMG
Purchase Scenario Decision variable of purchasing energy from CMDG
c Load reduction option fee of DL
g Load reduction energy fee of NDL
r Present worth factor
1 Load reduction fee of DL
s Time duration of NURMP contribution
- CDMG power generation contribution
142 M. S. Nazar et al.

w Decision variable for device installation or UDG/DNUUDG/

CDMG/DL contribution
/ Decision variable for UDG/DNUUDG/CDMG/DL contri-
bution coordination problem

References

1. S. Chowdhury, S.P. Chowdhury, P. Crossley, Microgrids and active distribution networks, in

IET Renewable Energy Series (2009)
2. M.S. Nazar, M.R. Haghifam, M. Nazar, A scenario driven multiobjective Primary-Secondary
Distribution System Expansion Planning algorithm in the presence of wholesale–retail market.
Int. J. Electr. Power Energy Syst. 40, 29–45 (2012)
3. C.L.T. Borges, V.F. Martins, Multistage expansion planning for active distribution networks
under demand and distributed generation uncertainties. Int. J. Electr. Power Energy Syst. 36,
107–116 (2012)
4. M.S Nazar, M.R. Haghifam, Multiobjective electric distribution system expansion planning
using hybrid energy hub concept. Electr. Power Syst. Res. 79, 899–911, (2009)
5. P.S. Georgilakis, N.D. Hatziargyriou, A review of power distribution planning in the modern
power systems era: Models, methods and future research. Electr. Power Syst. Res. 121, 89–100
(2015)
6. A. Satchwell, R. Hledik, Analytical frameworks to incorporate demand response in long-term
resource planning. Utilities Policy 28, 73–81 (2014)
7. S. Sekizaki, I. Nishizaki, T. Hayashida, Electricity retail market model with flexible price
settings and elastic price-based demand responses by consumers in distribution network. Int.
J. Electr. Power Energy Syst. 81, 371–386 (2016)
8. M.M. Aman, G.B. Jasmon, A.H.A. Bakar, H. Mokhlis, M. Karimi, Optimum shunt capacitor
placement in distribution system: a review and comparative study. Renew. Sustain. Energy
Rev. 30, 429–439 (2014)
Chapter 5
Multi-agent Based Planning Considering
the Behavior of Individual End-Users

Jan Kays

Abstract The volatile feed-in of distributed generation based on renewable energy

sources as well as new and intelligent loads and storages require an appropriate
consideration in the distribution grid planning process. With the conventional
planning method being dependent on extreme scenarios, the consideration is very
limited. Therefore, a new planning tool based on the concept of a multi-agent
system is presented. In this system, every network user is represented by an agent,
allowing not only the consideration of the volatile feed-in characteristics of
renewable energy sources but also of the dependencies between the network users
and their environment. Every network user is modeled as an agent of its own,
guaranteeing the preservation of its individual character. Within this chapter, a
system overview is given and the agent design process demonstrated on the
example of the household load agent and the storage agent, including negotiations.
This multi-agent system generates time series for all relevant system variables,
defining detailed input parameters in the distribution grid planning process. The
probabilities of occurrence of loading situations can be derived from the time series.
For the first time, this allows for a detailed determination of the conditions in the up
to now rarely measured medium and low voltage grids. As a consequence, new
assumptions for the planning process are derivable, permitting a demand- and
future-oriented grid planning and avoiding over-dimensioning of the grids.

Keywords Distribution grid planning Multi agent system
Time series


Storage systems Distributed energy resources

J. Kays (&)
Amprion GmbH, Dortmund, Germany
e-mail: [email protected]
J. Kays
TU Dortmund University, Dortmund, Germany

© Springer Nature Singapore Pte Ltd. 2018 143

F. Shahnia et al. (eds.), Electric Distribution Network Planning,
Power Systems, https://doi.org/10.1007/978-981-10-7056-3_5
144 J. Kays

5.1 Introduction

The requirements concerning the distribution grid planning have become more
challenging in the last years. In contrast to transmission systems that have been
planned precisely and computer-aided for over four decades already, the electric
distribution network planning was simpler, due to several reasons. With unidirec-
tional power flows from the higher voltage layers to the customers in the past, the
needed grid structure has been rather simple. The distribution system operators
(DSO) are faced to a large number of network areas, with many customers and
assets. Neither the monitoring of every device, nor the installation of complicated
protection systems had been justifiable from an economic point of view.
Additionally, the collection of a vast amount of measurements was just impossible,
when electronic data processing was not as powerful as today. For this reason, most
of the low-voltage grids and with them most customers’ power consumption are not
measured [1].
Besides these reasons, the low deployment of distributed generation (DG) units
and the inflexible demand have led to a deterministic planning practice, which is
known as fit and forget [2]. Based on deliberated planning guidelines and the
planners’ experiences, the grid performance is evaluated on the basis of extreme
scenarios for the combination of maximum load/minimum feed-in and maximum
feed-in/minimum load. Without integrating the probabilities of occurrence for
certain loading situations, this conventional planning [3] has its difficulties in
finding an efficient solution to face the present challenges in the distribution grids.
With the evolution from passive to active distribution grids, new methods of
planning and optimizing the distribution grids are mandatory [4]. Many different
approaches have been developed in the past to facilitate the distribution grid
planning process. An overview to different development trends is given in [3] as
well as in [5–7], showing that optimization approaches often deal with an optimized
placement and sizing of DG units. Considering the planning of the grid itself,
approaches use inter alia Monte Carlo Simulations as in [8] or Particle Swarm
Optimization (PSO) methods, like in [9], to determine necessary grid reinforcement
measures. Others like [2, 10, 11] use probabilistic approaches to face uncertainties
of the loading situations. The uncertainty of occurrence of future supply tasks is
considered in [12], enabling the DSO to assort the priority of grid measures.
Although the utilization of time series supports the evolution to active distribution
networks [4] as well as the integration [13] and financial analysis of Smart Grid
applications into the planning, most of the optimization approaches are based on
input data that are derived from the conventional extreme scenarios.
Nevertheless, some approaches, as [14, 15], utilize at least measurements of
installed smart meters to derive loading situations. Others like [16] analyze the
positioning of storage systems on the basis of profiles. A battery management
system for demand response purposes has been developed in [17]. Intelligent loads
and their resulting impact on the system are analyzed in [18] or [19], as well as the
optimized loading of plug-in electric vehicles is outlined in [20]. In [21],
5 Multi-agent Based Planning Considering the Behavior of … 145

an agent-based model is combined with a PSO algorithm to determine necessary

grid measures on the basis of time series. An agent-based system for the analysis of
Smart Grid applications in a test feeder is outlined in [22]. A test bed for simula-
tions of energy markets with high penetration of photovoltaics and loads with price
sensitivity is published in [23]. The interaction of electric vehicles with DSOs is
outlined in [24], implementing a multi agent system. An agent-based system for
smart grid and market for the analysis of negotiations and synergies on a daily basis
has been presented in [25]. Another agent-based system is presented in [26],
combing electric vehicles, photovoltaic units and domestic loads to generate load
profiles, as well as socio-demographic information. A simulation framework for
market-based control in emergency situations in the distribution grids focuses on
the operational aspects in [27]. The reviewed approaches either use extreme sce-
narios for optimization problems or time series that usually do not include func-
tional dependencies between the network users that are existent in Smart Grid
applications. Therefore, the impact of Smart Grid and Smart Market applications is
difficult to port into the mid- and long term planning of the grids.
As a consequence, a new simulation environment on the basis of a multi-agent
system has been developed and is presented in this chapter as key element of a new
planning process, where probabilities of occurrence become more important.
Within this simulation environment, the functional and chronological dependencies
of the network users can be combined, resulting in detailed time series. These time
series, containing implicitly the consequences of the dependencies during the
simulation, constitute new input data for the following planning process. Especially
the probability of occurrence and duration of loading situations are derivable and
support the decision process for reinforcement measures. Including all relevant
information, the generated time series can replace the conventional extreme sce-
narios to allow for adapted distribution grid planning.
The implementation and application of the developed simulation environment is
demonstrated in a test grid with a variation of scenarios of a given supply task in
this chapter. To start with, the applicability of multi-agent systems in the distri-
bution grid planning is outlined. Then, the developed multi-agent simulation
environment with a brief description of the most important aspects is introduced.
An exemplary agent design for household load as well as storage systems is per-
formed afterwards. The simulation environment is applied in a test grid.
A conclusion and outlook closes the chapter.

5.2 The Application of Multi-agent Systems

in the Distribution Grid Planning

When simulating electrical networks, the challenge occurs in the individual

behavior of the network users, influencing each other. In the grids, many similar
network users, like loads or DG units and storages, are existent, having complex
146 J. Kays

interdependencies. With network users reacting to different input parameters, like

market prices or the behaviors of others, the system gets emergent characteristics.
Agent-based systems offer a method to meet these challenges by disassembling the
complex problem to small parts that describe elements of the system.
A multi-agent system (MAS) is defined as a system that is built up of two or
more (intelligent) agents, without having a superior system target [28]. These agents
are either small entities of hardware or software in a simulation environment. In
contrast to particle swarm optimization (PSO), MAS use various agents with
individual characteristics and egoistic objectives. Therefore, MAS allow for a very
realistic modeling of complex situations, whose overall structure is based on
interaction of individuals. The internal behavior of agents is for the others in the
multi-agent system a black box. They are only dependent on some input data x and
an internal parameter set u to produce an output vector y. Consequently, the
functional description of the agent behavior is defined as in (5.1) [29].

y ¼ f ðx; uÞ ð5:1Þ

Power systems present a good area of application of MAS, because they usually
extend over large areas and many diverse and individually behaving participants
and elements take part in them. Especially the ownership structure of distributed
generation units with different operation philosophies and decisions implies an
individual modeling of each unit. The concept of local data access and management
in agents avoids the processing of large data. The usefulness of application in
electric distribution networks depends on the challenges that need to be faced. With
focus on these networks, the implementation of MAS makes sense, if [28]:
• interactions between entities are essential, like in control systems or power
plants;
• a large group of entities interacts and the modeling of the complete explicit
system behavior is impossible;
• sufficient local data for analysis and decision is available;
• new functionality needs to be integrated into an existing system;
• new functionality is to be integrated over time.
Therefore, MAS are already applied in electric distribution networks to simulate
or coordinate different tasks. In [21], a smart grid test feeder is coordinated with an
agent system. Additionally, MAS are used to control Smart Grids and µ-grids in
[30, 31] as well as the voltage within a distribution grid in [32]. The integration of
storage systems for in-house applications is realized with a MAS in [33]. The
analysis of price effects on the grid is performed in [34] and time series, though
neglecting interdependencies, are generated in [22]. Additional MAS applications
in Smart Grids are outlined in [35].
Detailed input parameters are often a weak point within approaches that improve
the distribution grid planning. Furthermore, the consideration of interdependencies
between network users and external parameters is usually not integrated due to the
complexity in large systems. Therefore, the concept of agent-based systems is an
5 Multi-agent Based Planning Considering the Behavior of … 147

adequate solution to solve the aforementioned shortcomings in planning simulation

systems. The decomposition of the complex electric distribution networks using a
multi-agent system modeling allows for an individual modeling of all network users
within a complex environment considering interdependencies between all network
users.
Here, agents represent single network users, modeling their own objectives and
realistic behaviors. Based on available environmental parameters and internal
desires, every agent derives the most probable behavior of the represented network
user. Existing interdependencies amongst the network users and dependencies on
environmental parameters can be considered while designing the agents’ behaviors.
Besides the exchange of relevant output parameters of the agents, the more complex
interdependencies include the necessity for collaboration as well as negotiation
between the agents. As a consequence, the system is able to simulate the manifold
interplay between the network users, the grid and the electricity market.
Additionally, the modular design concept of multi-agent systems facilitates the
flexible analysis of different future supply tasks by easily adding or removing
network users in the simulation environment.

5.3 Simulation Environment

The fact that there are many different types of distribution network users compli-
cates the detailed and realistic modeling of their behavior for the analysis of
occurring loading situations in the grid. The probabilistic network planning allows
for taking into account the probability of loading situations. There, the application
and appraisal of measured time series enable the derivation of frequency distribu-
tion functions. But especially in the low and medium voltage level, measurements
are often unavailable, and thus the necessary input for the probabilistic modeling is
not satisfying. Additionally, neither the input data for the distribution grid planning
process nor the available planning methods are able to consider interdependencies
between the network users. With the conventional distribution grids developing to
smart grids with intelligent interacting users, this aspect becomes more important in
the future but cannot be covered by existing methods.
The decomposition of complex problems is one of the main benefits of
agent-based systems that are described previously. Therefore, the setup of an
agent-based simulation environment is a feasible solution for facing the complex
modeling problem of distribution grids, including the interdependencies between
the network users. The modular characteristics of agent-based systems enables the
individual and close to reality modeling of all existing network users, taking into
account their real behavior, desires and objectives. The simulation of the agents’
behaviors results in time series for all relevant information in the distribution grid.
These time series can be analyzed subsequently, and recommendations for grid
extensions are derivable.
148 J. Kays

5.3.1 Structures of the Simulation Environment

Aiming for getting realistic time series of asset loadings and nodal voltages in the
distribution grid, the simulation environment needs an adequate representation of
the network users within the analyzed grid. Therefore, every participant is modeled
with an individual agent, which preserves characteristic properties of the repre-
sented user. These agents are modeled straightforwardly, fragmenting the complex
modeling problem into small entities.
An overview of the agents with their tasks, their dependencies as well as their
provided parameters to others is given in Table 5.1. Focusing on the simulation

Table 5.1 Overview of the modeled agents

Agent Task Dependent on Provides
type
Grid Performs power flow Nodal power value Nodal voltage UðtÞ,
agent (1) calculation, determines asset vectors Pn;load ðtÞ and asset loading (lines,
loading, initiation of voltage Pn;gen ðtÞ from (2) transformers),
control concepts
Node Collection of nodal data Power consumption Nodal power balance,
agent (2) (power consumption/feed-in), Pload ðtÞ aggregated consumption
supports nodal negotiations Power feed-in Pgen ðtÞ Pn;load ðtÞ and feed-in
of connected users Pn;gen ðtÞ
(6/7/8/9)
Time Keep simulation synchronous Acknowledgement of Simulation time step t
agent (3) grid agent (1)
Weather Provides weather data Time (3) Weather data
agent (4) (temperature, solar
radiation, wind speed,
wind direction)
Market Represents external market Time (3) Market price pðtÞ
agent (5) behavior, provides the market
price pðtÞ
Load Representation of residential Time (3), weather (4) Power consumption
agent (6) loads, either with load profiles Pload ðtÞ
or probabilistic modeling
Electric representation of electric time (3), nodal power power consumption
vehicle vehicles, including driving balance Pn;load ðtÞ and Pload;EV ðtÞ
agent (7) behavior and different Pn;gen ðtÞ (2)
charging objective functions
DG unit Representation of the DG Time (3), weather (4) Power feed-in Pgen ðtÞ
agent (8) units, in the low voltage grid
mainly PV units;
Storage Represents small battery Nodal power balance Power consumption or
agent (9) storage systems with different Pn;load ðtÞ and Pn;gen ðtÞ feed-in Pstorage ðtÞ
objective functions (2), market price pðtÞ
(5)
5 Multi-agent Based Planning Considering the Behavior of … 149

3 4 5

MV

1 6
LV

7 2

Fig. 5.1 Low voltage network user and influencing external variables

within low voltage (LV) networks, the relevant network users and external data
sources for the simulation are also depicted in Fig. 5.1. However, the systematic
representation of the elements is too extensive to be outlined completely in the
following. For details of the elements, the consideration of [36] is recommended,
the load agent, the DG agent as well as the storage agent are briefly outlined in the
following subsection.
The resulting general system concept is depicted in Fig. 5.2 and explained in the
following. Dependent on global information about the time, weather and market
price, the elements, which are connected to a node, determine their behavior. If
there are requests for negotiations on the nodal level, the agents initiate a negoti-
ation and determine a solution. Negotiations are necessary, if conflicting situations
occur while the agents pursue their objectives. In the following, the complex power
flow calculation in the grid agent determines the current loading situation in the
grid. If DG units or storage systems have implemented voltage control algorithm or
power flow control mechanisms, their agents react on the grid agent’s calculation
results. So, negotiations and control algorithms are also feasible on a grid level.
Principally possible, the resulting power balance in the analyzed grid can influence
the market price for electricity in the market agent. This influence causes a new
determination of the nodal elements’ behavior, etc. Therefore, depending on the
implementation, the feedback to the market agent can lead to a negotiation of the
market price between the market agent on a global level and the consumers and
producers on a nodal level.
The chosen agent-based design of the distribution system allows for a maximum
of flexibility in the implementation of negotiations, control algorithms and depen-
dencies of the network users within and between the different identified levels. The
utilization of the system in the planning process enables the determination of rel-
evant loading situations in dependency of the assumed interactions between the
150 J. Kays

Initialisation phase

Allocation of global services,

e.g. weather, market price

Grid level
Global level

Nodal level
Nodal elements determine their
behaviour

Request for nodal Yes

negotiations?
No

Grid control Yes

algorithms?
No

Influence on Yes
global parameters?

No

No
End of simulation?

Yes
Post processing

Fig. 5.2 General system concept

network users. Additionally, the simulation results can be aggregated and applied in
the analysis of higher voltage levels. As a consequence of the modular construction
of the system and its elements, any level of detail can be analyzed in the post
processing of the simulation. The significance of the results is only influenced by
the modeling approach of the elements. Consequently, for example, if the relevant
data is available, the residential loads can be further decomposed to single domestic
loads like washing machines, which can still negotiate the price for electric energy
with the market agent on the global level.

5.4 Network Users as Agents

This subsection describes the exemplary modeling of relevant agents in the simu-
lation environment. The way of representing the network users allows for keeping
their individual characteristics. At first, the necessary information for the agent
is analyzed, followed by the description of the defined objectives and its
5 Multi-agent Based Planning Considering the Behavior of … 151

implementation in a logical behavior. A complete and detailed description of all

agents in the simulation environment is given in [36] as well as in [37], where an
additional agent model of heat pumps is outlined.

5.4.1 Household Load Agent

5.4.1.1 Defining Relevant Input and Output Parameters

Within the handled data, some parameters will not change during the simulation
and are therefore static, while dynamic parameters will change. The static param-
eters are stored in the setup database of the MAS. An overview of the relevant static
parameters for household agents is given in Table 5.2. The probability density
function (PDF) data is required for the probabilistic calculation of the demanded
power of a household. Following the analysis of given smart meter data in [38],
generalized extreme value (GEV) distributions are used for the calculation. The
necessary parameters to describe GEV distribution functions are the shape n, the
location l and the scale r. For every PDF of every quarter-hour interval of working
days, Saturdays and Sundays they are stored in tables n, l and r in the database.
During the simulation, some dynamic input variables are necessary for the
household agent to adapt its behavior to the current situation. Besides the time
t from the time agent, this is the market price for energy pðtÞ, provided by the
market agent and relevant in case of sensitivity of the household behavior to the
market price.
Based on the static and changing input parameters, the agent determines its
output parameters, which are the active power PLoad;res ðtÞ and the reactive power
QLoad;res ðtÞ. These values may differ to the internal variables PLoad ðtÞ and QLoad ðtÞ,
due to the consideration of objectives that are described in the following.
Afterwards, the agent sends these values to subscribed agents and stores them in a
result database for further analysis purposes. Embedded in the simulation envi-
ronment, the household agent communicates exclusively with the node agent of its
node and the market agent, if it is sensitive to changing market prices.
The calculation of the demanded power can easily be adapted to other distri-
bution functions, load profiles or specific characteristics of individual loads in the
network within the agent. Therefore, the necessary diversity of loads in an area can
be met.

Table 5.2 Static input Parameter Variable Unit

parameters for household
loads Annual energy consumption Ea kWh
Power factor cosðuÞ –
Connected node Node name –
Sensitivity to market price s mp –
PDF data tables n, l, r –
152 J. Kays

5.4.1.2 Objectives and Behavior of the Household Agent

When receiving a new time step message from the node agent, the household agent
determines its power consumption. This is done on a probabilistic basis. Let wðtÞ be
a random number in R þ with the distribution function in (5.2). Then, the resulting
active PLoad ðtÞ and reactive power consumption QLoad ðtÞ is calculated with (5.4)
and (5.5).

ZwðtÞ
F ðwðtÞÞ ¼ f ðxÞdx ð5:2Þ
1

with

f ðxÞ ¼ GðxjnðtÞ; lðtÞ; rðtÞÞ

(    1 )
x  lðtÞ nðtÞ
f ðxÞ ¼ exp  1 þ nðtÞ ð5:3Þ
rðtÞ

Ea
PLoad ðtÞ ¼ w
ð5:4Þ
1000 kWh
QLoad ðtÞ ¼ PLoad ðtÞ
tanðuÞ ð5:5Þ

Due to the limited power consumption of households, PLoad ðtÞ is limited to a

maximum of 20 kW, if the random generated value is higher. Otherwise, unrealistic
maximum loadings in the grid could occur and adulterate the simulation results due
to the characteristic of the used PDF.
The household agent represents the domestic load behavior. Therefore it
objectives for satisfying at least the probabilistically calculated power demand
PLoad ðtÞ in any case. The market price sensitivity smp , which is the indication for the
participation in demand side management, can be either true or false. It is set in the
database for every load agent either manually or randomly. If smp is true, the agent
will try to reproduce the probable household customer behavior in minimizing the
household’s expenditures for electrical energy in every time step. Then, the
resulting PLoad;res ðtÞ becomes a function of the calculated power demand PLoad ðtÞ
and the market price pðtÞ. Considering these objectives, the household agent
implements a resulting behavior, whose overview is depicted in Fig. 5.3.
5 Multi-agent Based Planning Considering the Behavior of … 153

Setup / Initialisation phase

Receive new time step t

Determination of power
consumption PLoad(t) and QLoad(t)

Yes Sensitive to market price No

smp= true

Consideration of market price p(t)

Adaption of power consumption

Forwarding PLoad,res(t) to the node agent,

store variables in result database

Fig. 5.3 Behavior of the household agent

5.4.2 Storage Agent

The designed storage agent is a representation of storages from a system operator’s

point of view. Therefore, the implemented battery model is rather simple, without
the consideration of a specific technology (as the DSO does not know either),
lifetime preservation algorithms or costs for the storage system operator. However,
more detailed models can be also integrated within the simulation system.

5.4.2.1 Defining Relevant Input and Output Parameters

Some parameters, which the storage agent needs to know, will never change during
the simulation. As they are internal and static, they are stored in a database that is
read-out during the initialization phase. These parameters are listed in Table 5.3.
Additionally, the relevant internal variables changing in every simulation time step
are given in Table 5.4. The required external parameters are, depending on the
pursued charging strategy, the nodal power consumption PLoad ðtÞ and power
154 J. Kays

Table 5.3 Static parameters Variable Unit

for storage agents
Rated peak power Pr;BSS kW
Storage capacity EBSS kWh
Maximum charging power Pcharge;max kW
Maximum discharging power Pdischarge;max kW
Efficiency gBSS –
Maximum depth of discharge DoDmax –
Connected node Node name –
Length of time interval s h
Power factor cosðuÞ

feed-in PPV ðtÞ as well as the market price pðtÞ (see also Table 5.1). The resulting
consumption or feed-in PBSS;res ðtÞ and QBSS;res ðtÞ are the output variables of the
storage agent. The static input parameters, stored in the setup database, are only
retrieved in the initialization phase.

5.4.2.2 Objectives of the Storage Agents

Representing the real situation, the storage system operator might pursue different
operation strategies. Three different main objective functions for the storage system
operation can be identified: the network oriented behavior to reduce the grid
loading, the self-consumption maximization of self-produced renewable energy of
attached DG units and the market driven behavior to maximize the owner’s
revenues.
When applying the network-oriented behavior, the storage system is used to
reduce the peaks of local distributed generation, namely PV units. It implies bal-
ancing load and generation locally. The high PV feed-in in noon hours in combi-
nation with low load consumption stresses the grid and raises the voltage level.
Therefore, the storage system charges, if a given threshold Pth;PV for the PV feed-in
PPV ðtÞ is exceeded:

PPV ðtÞ [ Pth;PV ! PBSS ðtÞ [ 0 ð5:6Þ

When the sum of local loads rises above the feed-in sum, the storage system will
discharge the batteries to supply the loads:

jPLoad ðtÞj [ jPPV ðtÞj ! PBSS ðtÞ\0 ð5:7Þ

The second possible strategy aims for maximizing the self-consumption of a DG

unit operator. Looking at one node, this means charging when generation is
dominant at a period and discharging when the load is higher. If the feed-in of the
5 Multi-agent Based Planning Considering the Behavior of … 155

DG unit is independent from the local load, the storage system can balance these
differences, trying to minimize the purchase of energy from the grid:

PD ðtÞ þ PBSS ðtÞ ¼ 0 ð5:8Þ

with

PD ðtÞ ¼ PG ðtÞ  PLoad ðtÞ ð5:9Þ

The resulting decision on this objective is defined as

8
< 0\PBSS \Pcharge;max ; if PD ðtÞ [ 0
PBSS ðtÞ ¼ Pdischarge;max \PBSS \0; if PD ðtÞ\0 ð5:10Þ
:
0; otherwise

The third objective of the storage system behavior uses the variability of the
price for electrical energy pðtÞ to maximise the revenues. Consequently, the storage
will charge at low price periods and discharge at high price periods, depending on
given price limits pupper limit and plower limit that are assumed on economic
objectives:
8
< Pcharge;max ; ifpðtÞ\plower limit
PBSS ðtÞ ¼ 0; otherwise ð5:11Þ
:
Pdischarge;max ; ifpðtÞ [ pupper limit

5.4.2.3 The Implemented Behavior

Having three different objective functions, the storage agent implements an

advanced behavior structure like a finite state machine. The general structure of the
agent’s performance during the simulation is pictured in Fig. 5.4.
When the agent is set up in the initialization phase, the information about its
objective function, either market price driven, self-consumption maximization or
network oriented is taken from the database. After having received the new time
step t from the node agent, this is taken into account. If the storage agent behaves

Table 5.4 Dynamic Variable Unit

parameters of storage agents
(Dis-)charging power PBSS ðtÞ kW
State of charge (SoC) SoCðtÞ %
Charging level (SoC absolute) ESoC ðtÞ kWh
Available capacity Ecap;av ðtÞ kWh
156 J. Kays

Setup / Initialisation phase

Receive new time step t

Market oriented

Receive market price p(t) Receive node balance

PΔ (t)=Pgen(t)-Pload(t) from
node agent
p(t) > pupper limit
Yes No Other storage No
agents present?
p(t) < plower limit No
Yes
Yes Identify desired activity

Negotiation process

No Allowed to
operate?
Yes
Determine resulting activity

PΔ (t)< 0 Yes
Self-con-
sumption
Network
No PΔ (t)> 0 oriented
Yes
No PPV(t)> Pth,PV
Yes
Discharge Charge Discharge
Do nothing Charge storage
storage storage storage

Update internal data, write to database, send power values to node agent

Fig. 5.4 Internal behavior structure of the storage agent

5 Multi-agent Based Planning Considering the Behavior of … 157

market price driven, it will process the received market price. Depending on the
given price limits pupper limit and plower limit the storage system will charge or
discharge according to (5.11). If the available capacity for (dis-) charging is
insufficient, the power will be reduced (5.12).

Ecap;av ðtÞ
PBSS ðtÞ ¼ ð5:12Þ
s

Acting either network-oriented or aiming for maximization of the

self-consumption, the agent requests the local feed-in PGen ðtÞ and consumption
PLoad ðtÞ to derive the nodal balance PD ðtÞ. Additionally, the presence of further
storage or EV agents is checked. If further storing agents are present, the desired
behavior is identified and a negotiation amongst them is started to coordinate their
behavior before charging. Being the winner of the negotiation process or the only
storage system at the node, the agent determines its charging behavior depending on
the considered objective functions which are described with (5.6)–(5.10). Finally,
after having updated internal data, the time step results are stored into a database
and the node agent is informed about its consumed or injected power.

5.4.2.4 Negotiation Between Two Storage Agents Acting

at the Same Node

If more than one storage agent with network oriented behavior is connected to a
node, a negotiation process between these agents is essential. Otherwise, with all
storage agents behaving uncontrolled in the same way, the grid situation would not
be improved but probably worsened. For example, if all storage systems are
charging to reduce the feed-in peak of a PV unit, the voltage could exceed the lower
voltage interval limit instead of the upper limit in the case of no charging storage
systems. The implemented negotiation process between all storage agents that are
connected at the same node solves this problem. Because of the storage agents
having the same network oriented behavior, they do not react egoistically but
cooperatively during the negotiation and demonstrate implicitly the agent basic
characteristic of social ability. The implemented message exchange is depicted in
Fig. 5.5. Focusing on the interactions between storage agents, the messaging with
the other agents assigned to the node is only indicated in the figure. However, the
amount of agents, including electric vehicles or other network users is not limited.
After informing the storage agents about the new time step and the reception of
request messages, the node agent sends the nodal power balance to the storage
agents (1). Without knowing if any further storage agents are present in the current
time step, they check the receiver list of the sent propose message (a). If the list
contains more than one addressee, a negotiation is started. Otherwise, the pro-
ceeding with only one storage agent occurs. One of the storage agents becomes the
chief negotiator to coordinate the negotiation (storage agent SA 1 in Fig. 5.5). It
queries the other storage agents (here SA 2), whether they want to charge or
158 J. Kays

node
SA 1 SA 2
agent

1 PROPOSE
a a
proceeding

2 QUERY_IF
3 INFORM_IF b
c 4 PROPOSE
5 AGREE
e d
6 INFORM
7 INFORM

Fig. 5.5 Negotiation process with two storage agents

discharge in the current time step (2). Based on the internal parameters and the
current nodal power balance, SA 2 determines its desired behavior (b) and informs
SA 1 about its intention (3). Additionally, SA 2 adds a charging priority to the
message, which is derived in dependence on the current SoCðtÞ. Also having
determined a charging behaviour on its own, SA 1 evaluates all received intended
behaviors with the corresponding charging priorities in a non-discriminating
manner (c). Then, the chief negotiator sends messages to all storage agents with the
proposed behavior (4). Following the priority ranking, the storage agents are
allowed to perform their intended behavior until the common target is reached. On
this basis, the storage agents check, if their desires have been considered (d).
Although able to behave completely selfish, the agents are implemented to be
constructive. This constructive behavior is implicitly assumed in the pursued
objective that aims to support the grid. Additionally, this ensures the convergence
of the negotiation. However, the chief negotiator waits (e) until all others agents
have agreed to the determined solution (5). Then, all storage agents inform the node
agent about their deduced power consumption or injection (6). Finally the node
agent takes all collected power data and forwards the aggregated values to the grid
agent (7).

5.5 Simulation Example

For the demonstration of the multi-agent simulation environment, a low voltage test
grid is utilized. It aims for representing the current situation in a small rural set-
tlement or village in Germany and is based on real grid data from internal grid
5 Multi-agent Based Planning Considering the Behavior of … 159

Fig. 5.6 Illustration of the low voltage test grid for the initial and future scenario

Table 5.5 Line parameters in the test grid

R [X/km] X [X/km] B [µS/km] Imax [A]
Cable 95 mm2 0.32 0.069 185.4 235
Cable 150 mm2 0.207 0.069 216.8 300
Overhead line 70 mm2 0.443 0.072 163.4 195

studies. Historically grown and extended, the grid serves an area with spacious
distributed detached houses. This leads to an inhomogeneous application of grid
equipment with long feeders, consisting of cables with different cross-sections or
overhead lines. The schematic illustration of this grid is given in Fig. 5.6.
Supplied from the 20 kV medium voltage grid via a 400 kVA local transformer,
the grid consists of four feeders with different characteristics (line parameters see
Table 5.5). The feeder F1 has a total length of about 850 m and an additional
branch that has a length of 380 m. The first part is realized as a cable with a
cross-section of 150 mm2, the last part with an older one with a cross-section of
only 95 mm2. Except some longer distances, the average distance between the
connected 38 private customers is around 30 m. The overall length of F2 results in
700 m with 28 residential loads being connected to, having an average distance of
25 m between them. The cable type of F2 has a cross-section of 95 mm2. The 32
households connected to feeder F3, which is designed with a 150 mm2
cross-section cable, have an average distance of 25 m, too. Feeder F4 supplies a
smaller group of 17 customers that are more spatially distributed. Therefore, the
average distance between them is about 60 m and the branch has the longest
160 J. Kays

distance of 100 m. In contrast to the other feeders, F4 is realized as a 70 mm2

overhead line. The installed PV units have a installed rated capacity of either 10 or
30 kW and are operated with cosðuÞ ¼ 1. Their complete installed capacity is
330 kW. This set of input parameters defines the initial scenario.
Originating from the initial scenario, the future scenario assumes a development
of the grid’s supply task. Against the background of a furthermore increasing
installation of PV units, an additional installation of new PV units with 10 and
30 kW and a power factor of cosðuÞ ¼ 0:95 resp. cosðuÞ ¼ 0:9 is assumed for
some households in this area.

5.5.1 Conventional Grid Analysis

In the first instance, the test grid’s performance at both scenarios is checked and
evaluated with the conventional grid planning method. The grid’s performance is
analyzed for two extreme situations in every supply scenario with commercial
network calculation software (DIgSILENT PowerFactory). The maximum load
scenario assumes 100% of the rated load power and 0% of PV feed-in. The feed-in
scenario assumes 10% of the rated load power and 85% of PV power. The resi-
dential loads have a rated consumption of 2 kW at peak load with a power factor of
cosðuÞ ¼ 0:97. The reactive power behaviour of the PV units is set according to the
scenario description above.
The grid is able to serve the supply task of the initial scenario without any
problems. However, the additional PV installation in the future scenario causes
severe problems. Besides an overloading of the local transformer with 115% of its
rated power, over-voltages, up to 1.09 pu, appear in all feeders. Therefore, exten-
sive reinforcement measurements, i.e. transformer exchange and new parallel cables
in the feeders, are necessary (see Table 5.6).

5.5.2 Time Series Based Analysis

Now, the test grid’s performance is analyzed in the scenarios with the help of time
series, which are generated with the developed multi-agent system. While some
information, like the grid topology and PV unit nominal power, are already
available on the basis of the test grid definition, some agents need additional

Table 5.6 Necessary Investment Subtotal (€)

investment in the grid
1 Transformer upgrade 10,000 €/unit 10,000
1450 m of 150 mm2 cable 60,000 €/km 87,000
Total sum 97,000
5 Multi-agent Based Planning Considering the Behavior of … 161

Table 5.7 Assumptions for the required parameters of the PV agent

Parameter Symbol Assumption
Module aE Random value based on Gaussian distribution, expectation 0° S,
azimuth interval [−90°, 90°]
Module cE Random value based on Gaussian distribution, expectation 38.5°,
elevation interval [0°, 60°]
Inverter gINV Random value based on uniform distribution on interval [0.95,
efficiency 0.99]
Albedo A Set to 0.2

120

100 max(future) initial scenario

transformer loading [%]

max(future, 3h) future scenario

80 max(future, 6h)
future scenario, 3h average
future scenario, 6h average
60
future scenario, conventional
max(initial)
40

20

0
0 2.000 4.000 6.000 8.000
time step [h]

Fig. 5.7 Transformer loading duration curves for the scenarios

information to perform the simulation. The household agents require an annual

energy consumption, which is, in this case, homogeneously assumed to be Ea ¼
4000 kWh and the basis for probabilistic hourly power consumption. Similarly, all
household agents are not sensitive to the spot market price. Additional parameters
for the PV agents are listed in Table 5.7. Because of them being inaccessible to the
DSO, they have to be estimated for the simulation.
The simulated time period covers the year 2011. A processing of the resulting
time series to duration curves facilitates the derivation of the maximum and min-
imum loading as well as the analysis of the frequency of occurrence of loading
situations. The resulting duration curves for both scenarios are illustrated in the
diagram of Fig. 5.7.
In contrast to the depicted loading situation in the conventional analysis with
extreme scenarios, the maximum loading of the transformer is reduced significantly
in the test grid simulation with the multi-agent system. With the transformer’s
maximum loading being below 95% of the rated power in the future scenario,
162 J. Kays

0,4 1

cumulative distribution function

initial scenario
probability of occurrence future scenario
upper voltage
after reinforcement 0,8
0,3 boundary

0,6
0,2
lower voltage 0,4
boundary
0,1
0,2

0 0
0,9 0,92 0,94 0,96 0,98 1 1,02 1,04 1,06 1,08 1,1
nodal voltage [pu]

Fig. 5.8 Histogram and distribution function of the nodal voltage at last node of feeder F4

a transformer exchange to a higher rated type is not necessary. If the 6 h average is

considered, the maximum loading is even reduced to 81%.
Besides the transformer overloading, the nodal voltages at the feeders’ ends
exceed the permitted voltage interval in the feed-in situation of the future scenario
in the conventional grid analysis. The time series analysis reveals that only feeder
F4 has voltage problems. The resulting histogram and distribution function of the
appearing nodal voltage at the last node in F4 is depicted in Fig. 5.8.
The analysis of the test grid with the multi-agent simulation identifies an overall
grid extension necessity of 360 m of new 150 mm2 cable. Assuming investment
costs of 60,000 €/km, the induced investment in the test grid amounts to 21,600 €.
That is 75,400 € or 77% less than the resulting investment that can be identified via
the conventional extreme scenario based method.

5.5.3 Analysis of the Influence of New Network Users

on the Grid

For the demonstration of the objective functions’ effects of storage systems, every
PV unit in the test grid is equipped with a storage system. A 4 kW battery is
installed at each 10 kW PV units and a 12 kW battery at the 30 kW PV units. The
analyzed objective functions try either to maximize the self-consumption, to min-
imize the impact on the grid or to maximize the revenues due to fluctuating market
prices. The resulting voltage histogram is depicted in Fig. 5.9, demonstrating the
impact of the analyzed objective functions on the grid. Exemplarily it reveals that
the market-driven behavior is only a problem in a few hours during a year, where
countermeasures can be defined.
5 Multi-agent Based Planning Considering the Behavior of … 163

1
future scenario
cumulative distribution function market oriented

future scenario/self consumption

0,8 network oriented

network oriented reaches 1

market oriented reaches 1

self-consumption
0,6

reaches 1
0,4

0,2

0
0,9 0,92 0,94 0,96 0,98 1 1,02 1,04 1,06 1,08 1,1
nodal voltage [pu]

Fig. 5.9 Cumulative distribution function of the voltage at node for different storage objective
functions

This analysis can be performed for all innovative network users with different
interdependencies and objective functions. Therefore, it allows for the first time a
holistic analysis of the impact of innovative and functional dependent grid elements
and users on the distribution grid.

References

1. C. Baudot, G. Roupioz, A. Billet, Modernizing distribution network management with linky

smart meters—lessons learned in greenlys project, in Proceedings Of CIRED 2015 (Lyon,
France, 15–18 June, 2015)
2. G. Celli et al., A Comparison of distribution network planning solutions: traditional
reinforcement versus integration of distributed energy storage, in IEEE PowerTech 2013
(Grenoble, France, 16–20 June 2013)
3. S. You et al., An overview of trends in distribution network planning: A movement towards
smart planning, in T&D Conference and Exposition, 2014 IEEE PES (Chicago, USA, 14–17
April, 2014)
4. Cigré Task Force C6.19, Planning and Optimization Methods for Active Distribution Systems
(2014)
5. A. Keane et al., State-of-the-art techniques and challenges ahead for distributed generation
planning and optimization. IEEE Trans. Power Syst. 28(2), 1493–1502 (2013)
6. P.S. Georgilakis, N.D. Hatziargyriou, Optimal distributed generation placement in power
distribution networks: models, methods, and future research. IEEE Trans. Power Syst. 28(3),
3420–3428 (2013)
7. S. Prabhakar Karthikeyan et al., A review on soft computing techniques for location and
sizing of distributed generation systems, in 2012 International Conference on, Computing,
Electronics and Electrical Technologies (ICCEET) (21–22 March, 2012)
8. P. Wiest et al., New Hybrid planning approach for distributions grids with a high penetration
of RES, in Proceedings of CIRED 2015 (Lyon, France, 15–18 June, 2015
9. I. Ziari et al., Optimal distribution network reinforcement considering load growth, line loss,
and reliability. IEEE Tran. Power Syst. 28(2), 587–597 (2013)
164 J. Kays

10. V. Klonari etal., Probabilistic analysis tool of the voltage profile in low voltage grids, in
Proceedings of CIRED 2015 (Lyon, France, 15–18 June, 2015)
11. D.F. Frame, G.W. Ault, S. Huang, The uncertainties of probabilistic LV network analysis, in
IEEE Power and Energy Society General Meeting 2012 (San Diego, USA, 22–26 July, 2012)
12. C. Engels, L. Jendernalik, M. Osthues, H. Spitzer, ‘Smart planning’—an integrated approach
for distribution system planning to cope with its future requirements, in Proceedinds of
CIRED (Stockholm, Sweden, 10–13 June, 2013)
13. R.F. Arritt, R.C. Dugan, Value of sequential-time simulations in distribution planning. IEEE
Trans. Ind. Appl. 50(6), 4216–4220 (2014)
14. E.Tønne, J.A. Foosnæs, T.Pynten, Power system planning in distribution networks today and
in the future with Smart Grids, in Proceedings of CIRED (Stockholm, Sweden, 10–13 June,
2013)
15. G. Roupioz, X. Robe, F. Gorgette, First use of smart grid data in distribution network
planning, in Proceedings of CIRED (Stockholm, Sweden, 10–13 June, 2013)
16. M. Nick et al., On the optimal placement of distributed storage systems for voltage control in
active distribution networks, in IEEE PES Innovative Smart Grid Technologies (ISGT
Europe) (Berlin, Germany, 14–17 Oct, 2012)
17. N. Siebert et al., Scheduling demand response and Smart Battery flexibility in a market
environment: results from the Reflexe demonstrator project, in IEEE PowerTech 2015
(Eindhoven, Netherlands, June 29–July 2, 2015)
18. R. Roche et al., A multi-agent model and strategy for residential demand response
coordination, in IEEE PowerTech 2015 (Eindhoven, Netherlands, June 29–July 2, 2015)
19. N.G. Paterakis et al., Distribution system operation enhancement through household
consumption coordination in a dynamic pricing environment, in IEEE PowerTech 2015
(Eindhoven, Netherlands, June 29–July 2, 2015)
20. S. Bashash, H.K. Fathy, Cost-optimal charging of plug-in hybrid electric vehicles under
time-varying electricity price signals. IEEE Trans. Intell. Transp. Syst. 15(5), 1958–1968
(2014)
21. F. Boulaire et al., A hybrid simulation framework to assess the impact of renewable
generators on a distribution network, in Proceedings of the 2012 Winter Simulation
Conference (WSC) (Berlin, Germany, 9–12 Dec, 2012)
22. P. Jahangiri et al., Development of an agent-based distribution test feeder with smart-grid
functionality, in IEEE Power and Energy Society General Meeting 2012 (San Diego, USA,
22–26 July, 2012)
23. C. Chengrui et al., Agent-based simulation of distribution systems with high penetration of
photovoltaic generation, in Power and Energy Society General Meeting, 2011 IEEE (24–29
July, 2011)
24. J. Hu et al., Multi-agent based modeling for electric vehicle integration in a distribution
network operation. Electr. Power Syst. Res. 136, 341–351 (2016)
25. T. Pinto et al., Smart grid and electricity market joint simulation using complementary
multi-agent platforms, in IEEE PowerTech 2015 (Eindhoven, Netherlands, June 29–July 2,
2015)
26. L. Hattam, D.V. Greetham, Green neighbourhoods in low voltage networks: measuring
impact of electric vehicles and photovoltaics on load profiles. J. Mod. Power Syst. Clean
Energy 5(1), 105–116 (2017)
27. E.F. Bompard, B. Han, Market-based control in emerging distribution system operation. IEEE
Trans. Power Delivery 28(4), 2373–2382 (2013)
28. S.D.J McArthur et al., Multi-agent systems for power engineering applications—part I:
concepts, approaches, and technical challenges. IEEE Trans. Power Syst. 22(4), 1743–1752
(2007)
29. C. Rehtanz, Autonomous Systems and Intelligent Agents in Power System Control and
Operation (Springer, Berlin, New York, 2003). ISBN 3540402020
5 Multi-agent Based Planning Considering the Behavior of … 165

30. A. Prostejovsky et al., Demonstration of a multi-agent-based control system for active electric
power distribution grids, in IEEE International Workshop on Intelligent Energy Systems
(IWIES) 2013 (Vienna, Austria, 14 Nov, 2013)
31. F.I. Hernandez et al., Active power management in multiple microgrids using a multi-agent
system with JADE, in International Conference on Industry Applications (INDUSCON), 2014
11th IEEE/IAS (Juiz de Fora, Brazil, 7–10 Dec, 2014)
32. E. Polymeneas, M. Benosman, Multi-agent coordination of DG inverters for improving the
voltage profile of the distribution grid, in IEEE PES General Meeting 2014 (National Harbor,
USA, 27–31 July, 2014)
33. E.A.M. Klaassen et al., Integration of in-home electricity storage systems in a multi-agent
active distribution network, in IEEE PES General Meeting 2014 (National Harbor, USA, 27–
31 July, 2014)
34. I. Pisica et al., A multi-agent model for assessing electricity tariffs, in Innovative Smart Grid
Technologies Conference Europe (ISGT-Europe), 2014 IEEE PES (Istanbul, Turkey, 12–15
Oct, 2014)
35. G.H. Merabet et al., Applications of multi-agent systems in smart grids: a survey, in
International Conference on Multimedia Computing and Systems (ICMCS) 2014 (Marrakech,
Marocco, 14–16 April, 2014)
36. J. Kays, Agent-based Simulation Environment for Improving the Planning of Distribution
Grids, Ph.D. thesis, TU Dortmund University, Dortmund, Germany, 2014, ISBN
9783868446623
37. A. Seack, Time-series Based Distribution Grid Planning Considering Interaction of Network
Participants with A Multi-agent System, Ph.D. thesis, TU Dortmund University, Dortmund,
Germany, 2016
38. J.Kays, A. Seack, C. Rehtanz, Consideration of smart-meter measurements in a multi-agent
simulation environment for improving distribution grid planning, in Innovative Smart Grid
Technologies Conference (ISGT), 2016 IEEE PES (Minneapolis, USA, 6–9 Sept, 2016)
Chapter 6
Optimal Siting and Sizing of Distributed
Generations

Karar Mahmoud and Yorino Naoto

Abstract Recently, the penetration of distributed generations (DG) has been

obviously increased in electric distribution networks throughout the world. DGs are
small scale generators connected near load centers in networks, thereby avoiding
losses in transmission systems and releasing system capacity. At present, there are
many types of DG, such as wind power, solar power, fuel cell, biomass,
micro-turbines, and diesel engines. DG can play an important role in improving the
performance of the networks; therefore, allocating DG optimally is one of the most
crucial subjects in DG planning. In this chapter, the DG allocation problem is
studied, and an efficient method is presented for accurately solving this optimization
problem. The proposed method combines between analytical expressions and an
optimal power flow (OPF) algorithm to determine the optimal locations, sizes and
the best mix of various DG types for minimizing the total real power loss in electric
distribution networks. The proposed analytical expressions are general for directly
calculating the optimal sizes of any combination of multi-type DG technologies.
The optimal power factors of the various units can be analytically computed,
thereby contributing positively to loss reduction. The 69-bus test system is used to
test the proposed method. The effectiveness of the proposed method is demon-
strated for determining the optimal mix of various combinations of different DG
types.

Keywords Electric distribution networks
DG location
DG size

Power loss Optimal allocation

K. Mahmoud (&)
Electrical Engineering Department, Aswan University, Aswan, Egypt
e-mail: [email protected]
Y. Naoto
Department of Electrical and Computer Engineering, Hiroshima University,
Hiroshima, Japan
e-mail: [email protected]

© Springer Nature Singapore Pte Ltd. 2018 167

F. Shahnia et al. (eds.), Electric Distribution Network Planning,
Power Systems, https://doi.org/10.1007/978-981-10-7056-3_6
168 K. Mahmoud and Y. Naoto

6.1 Introduction

The needs for reliable and efficient electric distribution networks have motivated the
research on renewable energy sources (RES). RESs, such as wind turbines, pho-
tovoltaic (PV), and biomass systems, are clean sources with low investment costs
[1]. In electric distribution networks, the penetration of distributed generations
(DGs) has steadily increased because of their benefits. DG units are allocated near
load centers, reducing the stress on the transmission systems and saving costs [2–5].
The introduction of DG units into electric distribution networks has great
impacts on the operation, stability, and protection of the systems. These impacts
vary depending on the selected locations, sizes and types of the DG units [6–8]. As
most of the losses in the entire power systems are normally dissipated in electric
distribution networks, considering DG impacts on losses is important when allo-
cating the DG units. The total active losses are significantly affected by DG, where
they can be decreased/increased according to the DG allocation [9, 10]. Voltage rise
and reverse power flow are common technical problems associated with the inte-
gration of DG in electric distribution networks. These technical problems con-
straints the allowable penetration of DG. An efficient method is required to allocate
DG in an optimal way with considering overall system and DG constraints.
The allocation of DG units in MV electric distribution networks is considered an
important issue for system planners. The optimal DG allocation aims at determining
the best locations and sizes of DGs to optimize the network operation. Recently,
many methods have been presented in the literature for solving the optimal DG
allocation problem in networks. These methods are categorized as follows:
(1) numerical-based methods, (2) heuristic-based methods, and (3) analytical-based
methods. Examples of numerical-based methods are gradient search [11], linear
programming [12], optimal power flow [13], and exhaustive search [14, 15]. These
methods can determine the optimal DG sizes at candidate locations. To determine
the optimal locations of DGs, these numerical-based are applied to solve the
optimal DG sizes at all possible combinations of DG locations. The heuristic-based
methods utilize artificial intelligence algorithms, e.g., genetic algorithms [16, 17],
particle swarm optimization [18], harmony search [19], and tabu search [20]. These
heuristic-based methods have the ability to deliver near-optimal solutions of DG
sizes and locations but require exhaustive computational efforts. Regarding
analytical-based methods, they are simple, easy to be applied, and computationally
fast. The analytical-based methods simplify the DG allocation problem by con-
sidering only uniformly distributed load types or single DG placements [21, 22]. In
[23], an analytical method is presented to deliver the optimal locations of DG units,
and their sizes are optimally calculated using the Kalman filter method. In [24, 25],
a method based on the concept of load centroid to optimally allocate multiple DGs.
An analytical method for allocating a single DG unit is proposed in [26], and then it
is improved in [27, 28] by considering the reactive power capability of multiple DG
units.
6 Optimal Siting and Sizing of Distributed Generations 169

This book chapter is on the optimal siting and sizing of multi-type DG units in
electric distribution networks. In this chapter, two efficient methods for the optimal
allocation of multi-type DG for loss minimization are presented. The first method is
based on analytical expressions that directly can calculate the optimal sizes of
multi-type DG units and evaluate the corresponding DG benefits. A second hybrid
method is presented which combines the first analytical method and an optimal
power flow (OPF) algorithm for solving the DG allocation problem. The presented
methods are accurate, general for multi-type DGs, and valid for radial and meshed
systems. The performance of the proposed methods is tested and validated using the
69-bus test system.

6.2 DG Models

According to the output scheme of DG units, they can be classified to three models:
(1) DG Model A, (2) DG Model B, and (3) DG Model C. First, for DG Model A, its
active power is not specified and needed to be optimally computed. Second, unlike
DG Model A, the variable in the DG Model B is reactive power, not the active
power generation. Third, the DG Model C model has two variables includes both
active and reactive power generation, which means that this model is more complex
than the other two models to be optimally solved. The mathematical representations
of these units are described in Fig. 6.1. As seen in the figure, the different RES
technologies have their interfaced devices to the main grid. For each DG type, once
the state variable (active and/or reactive power generation), the interfaced device
and the configuration of DG technology can be properly designed [29, 30]. It is
important to mention that considering the reactive power capability of DG is
important to simulate the real situation of employing the DG reactive power

DG Models

DG Type:
DG Type 1: Unspecified P DG Type 2: Unspecified Q DG Type 3: Unspecified PQ
Min Max
DG Active Power Characteristic: Min
PDGi Max
≤ PDGi ≤ PDGi
Spec
PDGi = PDGi PDGi ≤ PDGi ≤ PDGi
QDGi = QDGiSpec
QDGi
Min Max
≤ Q DGi ≤ Q DGi QDGi
Min Max
≤ Q DGi ≤ Q DGi
DG Reactive Power Characteristic:

Static Power Converter

Interface Devices to Utility: Squirrel Cage Induction
Permanent Magnet Doubly Fed Induction
Machine Synchronous Machine
Synchronous Machine Machine

DG Technology: Internal Combustion

Micro Turbine PV Arrays Fuel Cell Wind Turbine Gas Turbine
Engine

Fig. 6.1 Steady state models of different DG technologies

170 K. Mahmoud and Y. Naoto

Fig. 6.2 Generation Reactive Power

capability curve of PV
inverters

PDG

PF DG Active power

Q DG

injection/rejection for voltage regulations. For example, Fig. 6.2 shows the gen-
eration capability curve of PV inverters, where the red circle indicates the rated
power of the interfaced inverter.

6.3 DG Impacts on Electric Distribution Networks

DGs have enormous impacts on electric distribution networks according to their

locations, sizes, and types. For instance, consider a DG unit is added to a network
supplied from the distribution substation (Fig. 6.3). Figure 6.4 shows the variation
of power loss with active and reactive powers of the DG unit. At each DG power
factor, as the active power of DG is increased, the active losses are reduced to a
minimum value and increased again after exceeding a specific DG penetration level
(optimal DG penetration). Therefore, to minimize the total systems losses, the
optimal power factors of the DG units are required to be accurately computed.
Regarding voltage variation with DG, Fig. 6.5. describes the impact of DG
penetration on the voltage profile. The penetration level of DG can be defined as the
ratio of the total size of DGs to the total load demand in the system. Normally,
the voltage of distribution feeders drops with increasing the distance from the
distribution substation. However, if a DG unit with high penetration is added,
6 Optimal Siting and Sizing of Distributed Generations 171

k m

Substation Distribution System DG

PS P DG

Fig. 6.3 Electric distribution network with DG

211

Active Losses (Kw)

206
QDG (Mvar)

201
0.75

0
0.2
0.4
0.6
0.8
1
1.2
1.4
1.6

0
1.88
2
2.2
2.4
26
2.6

1
2.8

PD
DG (MW)

Fig. 6.4 The variation of the active losses with active and reactive powers of DG

Voltage

Voltage with excessive DG penetration

max
V

rated
V

min Voltage with optimal DG peneteration

V
Voltage without DG

Distance from Substation

Fig. 6.5 The variation in voltage profile with the penetration level of DG

voltage profile will increase with distance from the substation. For example, the
penetration level of the PV units is changed by increasing the number of arrays of
the PV units. An optimal penetration of DG is required to ensure that the voltage
profiles along the distribution feeder do not exceed lower/upper limits.
172 K. Mahmoud and Y. Naoto

Besides losses and voltage profile, overall constraints of the electric distribution
network must be considered when allocating DG.

6.4 Description of the DG Allocation Problem

The optimal placement of DG is a complex optimization as the number of

alternative solutions (i.e., possible locations and sizes of units) is huge and the
electric distribution network is nonlinear. Figure 6.6 shows an example of a
network where various components are needed to be placed at some of their
recommended sites. These recommended sites for each unit type can be listed
according to many factors including fuel distribution, investor strategies, weather
conditions (for renewable DG) and etc. Figure 6.7 shows a PV system where its
optimal size can be specified by the number of PV models. Therefore, the target
of optimal placement is to determine the best set of locations for the various units,
according to their recommendation locations. For instance, consider installing
NDG DG units in a network with NB nodes that are eligible for the installation.

19 20 21 22 26 27 28 29 30 31 32 33

1 6 11
2 3 4 5 7 8 9 10 12 13 14 15 16 17 18

23 24 25

Fig. 6.6 Example of an electric distribution network

6 Optimal Siting and Sizing of Distributed Generations 173

Fig. 6.7 PV system model PV array

PCU Grid

The DG units of type i, whose number is NDGi, can be installed only in their
corresponding nodes NBi. Thus

X
N DGT X
N DGT

NDG ¼ NDGi ; NB ¼ NBi ð6:1Þ

i¼1 i¼1

where NDGT is the number of DG types to be installed. The number of possible

combination of DG sites in this case can be computed from
!
NY
DGT

NCom ¼ CNNDGi
Bi
ðNDG !Þ ð6:2Þ
i¼1

To determine the optimal combination, it is required to evaluate all of these

possible combinations of the DG sites. Note that the number of the possible
combinations is high, especially when placing different DG types in large-scale
distribution networks. This excessive number of combinations will not only
increase the complexity of the optimization problem but also degrade the compu-
tational performance. A fast method is required to determine the optimal combi-
nation among all of these combinations. The main requirements of the DG
allocation method can be listed as
• Accurate (proper DG locations with optimal capacities).
• Generic formulations for multi-type DG allocation.
• High computational speed (especially when allocating multiple DGs in
large-scale systems).
174 K. Mahmoud and Y. Naoto

6.5 Combined Analytical-OPF Method

6.5.1 Losses with DG

The basic formula for calculating the total active power loss Ploss is expressed as
X  
Ploss ¼ uj P2j þ Q2j ð6:3Þ
j2/

in which
Rj
uj ¼
Vj2

where Pj and Qj the active and reactive power flows, respectively, through the
distribution line j. / is a set of system lines, Vj is voltage magnitude of the receiving
bus of the line, and Rj is line resistance. Consider adding a DG or a capacitor, which
injects Pg and/or Qg , at a certain bus in a network, the variation in the reactive
power loss can be linearly estimated, whereas it can be computed by
X   X  2  2 
Ploss;DG ¼ uj P2j þ Q2j þ uj Pj  Pg þ Qj  Qg ; a[b ¼ /
j2a j2b

ð6:4Þ

where a and b represent two different sets of lines whose power flows are not
affected and whose power flows are affected by adding the DG, respectively. The
above equation can be modified to be in general form for expressing the effect of
integrating multiple DG units and capacitors at a set of locations w on reactive
power loss as follows
0 !2 !2 1
X   X X X
Ploss;DG ¼ uj P2j þ Q2j þ u j @ Pj  Xij Pgi þ Qj  Xij Qgi A
j2a j2b i2w i2w

ð6:5Þ

The X matrix can be built based on the radial structure of networks. The binary
matrix X for a small-scale system shown in Fig. 6.8. when adding PV and wind
units, respectively, at buses 11 and 7 can be expressed as follows

System Buses
1 2 3 4 5 6 7 8 9 10 11 12
" # ð6:6Þ
1 1 1 0 1 1 1 0 0 0 0 0 7 Wind Bus
X¼
1 1 1 0 0 0 0 0 1 1 1 0 11 PV Bus
6 Optimal Siting and Sizing of Distributed Generations 175

Wind
turbines
Path for wind power

5 6 7 8
1 2 3 4
Slack
Node

9 10 11 12

Path for PV power

PV

Fig. 6.8 Power flows after adding PV and wind units to an electric distribution network

Equation (6.5) could be expressed in different ways with active or reactive

power injection of DG with respect to the power factor PFg as in (6.7) and (6.8).

Qgi ¼ Mgi Pgi ð6:7Þ

where
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
1  PFgi2
Mgi ¼ ð6:8Þ
PFgi

6.5.2 Analytical Expressions for Optimal DG Sizing

The objective of the placement of DG units in electric distribution networks is to

minimize the total power loss by selecting best locations and optimal sizes of these
units. Since the losses can be represented by (6.5), the objective function can be
expressed as minimization of Ploss;DG . The variable for this optimization problem is
the active Pg and reactive Qg powers of the units. As the variations of Ploss;DG with
Pg and Qg are equal to zero at the optimal point,

@Ploss;DG
¼ 0; 8m 2 W ð6:9Þ
@Pgm
176 K. Mahmoud and Y. Naoto

Fig. 6.9 Optimal DG sizing process

@Ploss;DG @Ploss;DG
¼ ; 8m 2 W ð6:10Þ
@Qgm @Pgm

The above two equations can be written for each unit; therefore, their number is
equal to double the number of the units to be placed. For the set of (6.9) and (6.10),
they can be arranged in matrix form to be as given in (6.11) and (6.12), respectively.
With employing the new (6.11) and (6.12), a direct optimal solution of Pg and Qg for
all units (i.e., optimal power factors) can be delivered. Once the combination of the
DG and capacitor locations is defined, the optimal sizing for all units can be com-
puted, as shown in Fig. 6.9. The parameters of XYWU matrices in (6.11) and (6.12)
can be completely computed directly from the power flow result for the base case
without requiring iterative processes. The proposed formulae have been established
based on the radial structure of power distribution systems. In order to apply the
proposed method to radial systems, it is required to break all system loops and then
the DG allocation problem is solved for the resulted radial system [31, 32].
2 3 2 31 2 3
PgW1 XW1 ;W1 XW1 ;W2


XW1 ;W3 YW1
6 PgW2 7 6 XW2 ;W1 XW2 ;W2


XW2 ;WN 7 6 YW2 7
6 7 6 7 6 7
6 .. 7 ¼ 6 .. .. .. .. 7 6 .. 7 ð6:11Þ
4. 5 4 . . . . 5 4. 5
PgWN XWN ;W1 XWN ;W2


XWN ;WN YWN
2 3 2 3 2 31 2 3
QgW1 PgW1 UW1 ;W1 UW1 ;W2


UW1 ;W3 WW 1
6 QgW2 7 6 PgW2 7 6 UW2 ;W1 UW2 ;W2


UW2 ;WN 7 6 WW 2 7
6 7 6 7 6 7 6 7
6 .. 7 ¼ 6 .. 7  6 .. .. .. .. 7 6 .. 7 ð6:12Þ
4. 5 4. 5 4 . . . . 5 4. 5
QgWN PgWN UWN ;W1 UWN ;W2


UWN ;WN WW N
6 Optimal Siting and Sizing of Distributed Generations 177

where
X X  
Xn;m ¼ Xnj uj Xmj ð1 þ MDGm MDGn Þ; Ym ¼ Xmj uj Pj þ MDGm Qj
j2b j2b
X X  
Un;m ¼ Xnj uj Xmj ; Wm ¼ Xmj uj Pj  Qj
j2b j2b

6.6 Solution Process

The proposed method for determining the optimal mix of DG involves a combi-
nation of the proposed analytical expressions and OPF [31–34]. The objective
function of OPF is set to be the minimization of the losses with considering equality
and inequality constraints (6.13)–(6.17). Since the analytical expressions are gen-
eral for optimally solving any combination of sites where various units are placed,
they can be employed to evaluate all possible combinations of the sites. This
evaluation process is essential to select the optimal combination of unit sites (i.e.,
the optimal mix). The computation burden of the evaluation process is greatly
improved using the proposed analytical expressions, as the optimal solution can be
directly computed using (6.11) and (6.12). The benefits of employing the OPF
formulation are to apply system constraints for the optimal combination obtained by
the analytical expressions and slightly correct the unit sizes to the exact optimal
solution. The flowchart which illustrates the solution process of the proposed
method is given in Fig. 6.10. The backward/forward sweep power flow method
presented in [35] is employed as a power flow solver. As clear in the figure, the
proposed analytical expressions are needed to calculate the optimal mix, while OPF
is employed once for considering system constraints. This combination between
these two formulations is efficient; since the proper optimal combination can be
obtained with the analytical expressions, and the optimal solution can be accurately
computed with including various constraints via OPF.
Minimize:

X
NLine
F¼ PiL ð6:13Þ
j¼1

Subject To: (1) Equality constraints

X
N    
PS  PD  Vj jVi j Gij cos hij þ Bij sin hij ¼ 0 ð6:14Þ
i¼1
178 K. Mahmoud and Y. Naoto

Start

Read Data

Data Structure

Power Flow Solver

Proposed Method Solver

Optimal DG Sizing

Optimal DG Siting
Analytical
OPF Solver -OPF Method?
Analytical
Minimize: PLoss,DG (x,u )
Subject To H(x,u) = 0
Power Flow Solver
G(x,u) ≤ 0

Print Results

End

Fig. 6.10 Flowchart of the proposed method for DG allocation

X
N    
QS  QD  Vj jVi j Gij sin hij  Bij cos hij ¼ 0 ð6:15Þ
i¼1

(2) Inequality constraints

DGi  PDGi  PDGi

Pmin for i ¼ 1; 2; . . .NDG ð6:16Þ
max

Vimin  Vi  Vimax for i ¼ 1; 2; . . .N ð6:17Þ

6.7 Results and Discussions

6.7.1 Assumptions

• A single DG unit can be placed on each bus in the test system;

• The maximum allowed number of DG to be connected is three units;
6 Optimal Siting and Sizing of Distributed Generations 179

• The specified generation values of DG Types A and DG type B are considered

to be zero.
• For DG Type C, its power factor is equal to 0.90 lagging.
• The maximum penetration of DG is 100%.
• The minimum and maximum limits of voltages are 0.9 and 1.05, respectively.

6.7.2 Test System

The 69-bus test system is used to test out the proposed method for DG placement.
Figure 6.11 shows the system, which is a preferable benchmark test system for
several allocation approaches, where the data are given in [36]. This system consists
of 68 load buses and a slack bus, and the active and reactive power loss in the base
case are 225 kW and 102 kvar, respectively. The proposed method has been
implemented by C++ programming. This analysis aims at demonstrating the
effectiveness of the proposed method for solving the allocation problem of different
DG types. Assume that an area is recommended for each type of DG. To do so, the
69-bus distribution test system is divided into four different areas as
• Area-A: This area contains candidate locations of DG type A.
• Area-B: This area contains candidate locations of DG type B.
• Area-C: This area contains candidate locations of DG type C.
• Area-D: No DG is allowed to be installed in this area.
Table 6.1 shows different studied cases with different combinations of DG types.
The first case (Case 0) is the base case without DG while the other seven cases
(Case 1–Case 7) involve installing three DG units of different types, as illustrated in
the table.

Table 6.1 Number of DG Cases DG Type A DG Type B DG Type C

for different cases
Case 0 – – –
Case 1 2 1 –
Case 2 2 – 1
Case 3 1 2 –
Case 4 – 2 1
Case 5 – 1 2
Case 6 1 – 2
Case 7 1 1 1
180 K. Mahmoud and Y. Naoto

36 37 38 39 40 41 42 43 44 45 46

Area C
47 48 49 50 51 52 66 67 68 69

Area D Area A
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

53 54 55 56 57 58 59 60 61 62 63 64 65

28 29 30 31 32 33 34 35

Area B

Fig. 6.11 The 69-bus test system

6.7.3 Analyses

In this subsection, the benefits of allocating DGs to the test system are discussed.
Table 6.2 shows the computed DG sizes and their corresponding buses with using
the proposed analytical-OPF method. Note that the calculated DG locations and
sizes are almost different for all the cases at which the DG numbers are similar
(three units) for the cases. The only difference between the cases is the combina-
tions of the different DG types. These results imply that the type of DG has a
significant influence on the DG allocation problem. The major differences between
the eight cases can be listed as
• Total Losses: Figure 6.12 shows the losses in kW and kVA for the cases after
allocating the DG units into the 69-bus test system. It is obvious that there are
significant reductions of the total losses for all the cases of DG allocation with
respect to the base case, but these reductions are different for the cases. Note that
DG type C has a higher effect on loss reduction compared with the other DG
types. For example, Case 5 has a high value of loss reduction due to its two DG
units of type C. This feature is reasonable since this DG type can inject active
and reactive power, thereby contributing heavily in reducing losses. In addition,
6 Optimal Siting and Sizing of Distributed Generations 181

Table 6.2 Results for the 69-bus test system

Cases DG Type A DG Type B DG Type C Total DG
Bus Size (kVA) Bus Size (kVA) Bus Size (kVA) size (kVA)
Case 0 – – – – – – –
Case 1 9 2388.12 61 1314.6 – – 4154.0
18 451.373
Case 2 12 370.6 – – 51 2535.9 3219.2
21 312.8
Case 3 9 2839.5 53 652.6 – – 4685.6
61 1193.5
Case 4 – – 61 938.9 51 2640.3 3785.7
64 206.5
Case 5 – – 61 1138.2 51 2007.7 3850.3
68 704.3
Case 6 21 302.1 – – 51 2129 3225.2
66 793.8
Case 7 9 2249.6 61 1270.1 68 655.4 4175.2

Active Losses (kW) Losses (KVA)

300

250

200
Losses

150

100

50

0
Base Case Case 2 Case 6 Case 4 Case 1 Case 3 Case 5 Case 7
Cases

Fig. 6.12 The losses for the different cases

it is noted that Case 7 which involves different three types of DG yields the
highest loss reduction.
• The Total Size of DGs: The total size of DGs is important in the DG allocation
as it can be employed to estimate the installation cost of DG. The higher DG
size, the higher cost of DG. The total size of DG units for the case are shown in
Table 6.2. It is clear from the table that Case 1 and Case 3 at which DG type C is
not included, have the highest capacity. This trend means that the installation
costs for these two cases are relatively high. However, this trend can be an
advantageous feature if the penetration DG is required to be maximized.
• Voltage Profile: Figure 6.13 shows the voltage profile for the different cases.
Table 6.3 compares the minimum voltage, maximum voltage, and the value of
182 K. Mahmoud and Y. Naoto

1.02

1
Base case
0.98 Case 1
Voltage (pu)

Case 2
0.96
Case 3

0.94 Case 4
Case 5
0.92 Case 6
Case 7
0.9
2
36
47
31
39
5
41
43
46
6
51
53
11
12
55
14
17
20
23
26
58
61
64
Bus

Fig. 6.13 Voltage profile for the different cases

Table 6.3 Voltage for the different cases

Item Base case Case 1 Case 2 Case 3 Case 4 Case 5 Case 6 Case 7
Maximum 0.9092 0.9481 0.9307 0.9488 0.9461 0.9461 0.9313 0.9484
voltage
Maximum 1.0000 1.0000 1.0000 1.0002 1.0005 1.0006 1.0001 1.0005
voltage
Voltage 0.0993 0.0193 0.0345 0.0236 0.0278 0.0224 0.0336 0.0207
deviation

voltage deviation for the different cases. It is clear that the voltage profile is
significantly improved for the cases of DG installation compared with Case 0.
It is important to mention that the proposed method solves the DG allocation
problem optimally, and the total losses are reduced for all cases. The proposed
method is general, and so it can be applied for solving other cases and electric
distribution networks. The proposed method is a helpful tool for optimizing the
networks with DG and selecting the optimal mix of the available DG technologies
to maximize benefits. Note that the proposed method is very effective for solving
the allocation problem of multi-type DG units compared with existing analytical
methods in the literature. This superiority is accomplished as the proposed method
has high accuracy rates with fast computational speed, and it can directly compute
the optimal power factors of different DG types [31, 32].
6 Optimal Siting and Sizing of Distributed Generations 183

References

1. L. Willis, W.G. Scott, Distributed Power Generation: Planning and Evaluation, vol.
10 (Marcel Dekker, New York, 2000)
2. O.M. Toledo, D.O. Filho, A.S.A.C. Diniz, J.H. Martins, M.H.M. Vale, Methodology for
evaluation of grid-tie connection of distributed energy resources—case study with
photovoltaic and energy storage. IEEE Trans. Power Syst. 28(2), 1132–1139 (2013)
3. A.G. Exposito, A.J. Conejo, C.A. Canizares, Electric Energy Systems Analysis and Operation
(CRC Press, 2009)
4. X. Zhang, G.G. Karady, S.T. Ariaratnam, Optimal allocation of CHP-based distributed
generation on urban energy distribution networks. IEEE Trans. Sustain. Energy 5(1) (2014)
5. X. Zhang, R. Sharma, Y. He, Optimal energy management of a rural microgrid system using
multi-objective optimization, in Proceedings of IEEE PES Innovative Smart Grid Technology
Conference (Washington DC, USA, 2012), pp. 1–8
6. R.A. Walling, R. Saint, R.C. Dugan, J. Burke, L.A. Kojovic, Summary of distributed
resources impact on power delivery systems. IEEE Trans. Power Del. 23, 1636–1644 (2008)
7. T. Ackermann, V. Knyazkin, Interaction between distributed generation and the distribution
network: operation aspects, in Proceedings on 2002 IEEE T&D Conference (2002),
pp. 1357–1362
8. D. Singh, K.S. Verma, Multiobjective optimization for DG planning with load models. IEEE
Trans. Power Syst. 24(1), 427–436 (2009)
9. Y.M. Atwa, E.F. El-Saadany, M.M.A. Salama, R. Seethapathy, Optimal renewable resources
mix for distribution system energy loss minimization. IEEE Trans. Power Syst. 25(1), 360–
370 (2010)
10. P.S. Georgilakis, N.D. Hatziargyriou, Optimal distributed generation placement in power
distribution networks: models, methods, and future research. IEEE Trans. Power Systems 28
(3), 3420–3428 (2013)
11. N.S. Rau, Y.-H. Wan, Optimum location of resources in distributed planning. IEEE Trans.
Power Syst. 9(4), 2014–2020 (1994)
12. A. Keane, M. O’Malley, Optimal allocation of embedded generation on distribution networks.
IEEE Trans. Power Syst. 20(3), 1640–1646 (2005)
13. G.P. Harrison, A.R. Wallace, Optimal power flow evaluation of distribution network capacity
for the connection of distributed generation. IEE Proc. Gener. Transm. Distrib. 152(1), 115–
122 (2005)
14. D. Zhu, R.P. Broadwater, K.S. Tam, R. Seguin, H. Asgeirsson, Impact of DG placement on
reliability and efficiency with time-varying loads. IEEE Trans. Power Syst. 21(1), 419–427
(2006)
15. D. Singh, R.K. Mirsa, D. Singh, Effect of load models in distributed generation planning.
IEEE Trans. Power Syst. 22(4), 2204–2212 (2007)
16. K. Vinothkumar, M.P. Selvan, Fuzzy embedded genetic algorithm method for distributed
generation planning. Electr. Power Compon. Syst. 39(4), 346–366 (2011)
17. K.H. Kim, Y.J. Lee, S.B. Rhee, S.K. Lee, S.K. You, Dispersed generator placement using
fuzzy-GA in distribution systems, in Proceedings of IEEE Power Engineering Society
Summer Meeting (July 2002), pp. 1148–1153
18. W. Prommee, W. Ongsakul, Optimal multiple distributed generation placement in microgrid
system by improved reinitialized social structures particle swarm optimization. Eur. Trans.
Electr. Power 21(1), 489–504 (2011)
19. M.F. Shaaban, Y.M. Atwa, E.F. El-Saadany, DG allocation for benefit maximization in
distribution networks. IEEE Trans. Power Syst. 28(2), 639–649 (2013)
20. M.E.H. Golshan, S.A. Arefifar, Optimal allocation of distributed generation and reactive
sources considering tap positions of voltage regulators as control variables. Eur. Trans. Electr.
Power 17(3), 219–239 (2007)
184 K. Mahmoud and Y. Naoto

21. H.L. Willis, Analytical methods and rules of thumb for modeling DG-distribution interaction,
in Proceedings of IEEE Power Engineering Society Summer Meeting (July 2000), pp. 1643–
1644
22. C. Wang, M.H. Nehrir, Analytical approaches for optimal placement of distributed generation
sources in power systems. IEEE Trans. Power Syst. 19(4), 2068–2076 (2004)
23. S.-H. Lee, J.-W. Park, Selection of optimal location and size of multiple distributed
generations by using Kalman filter algorithm. IEEE Trans. Power Syst. 24(3), 1393–1400
(2009)
24. T. Xiao-bo, W. Xue-hong, A new method of distributed generation optimal placement based
on load centroid, in 2011 Proceedings of IEEE Power and Energy Engineering Conference
(APPEEC), pp. 1–5
25. A. Elmitwally, A new algorithm for allocating multiple distributed generation units based on
load centroid concept. Alexandria Eng. J. 52(4), 655–663 (2013)
26. N. Acharya, P. Mahat, N. Mithulananthan, An analytical approach for DG allocation in
primary distribution network. Int. J. Elect. Power Energy Syst. 28(10), 669–678 (2006)
27. D.Q. Hung, N. Mithulananthan, R.C. Bansal, Analytical expressions for DG allocation in
primary distribution networks. IEEE Trans. Energy Convers. 25(3), 814–820 (2010)
28. D.Q. Hung, N. Mithulananthan, Multiple distributed generators placement in primary
distribution networks for loss reduction. IEEE Trans. Ind. Electron. 60(4), 1700–1708 (2013)
29. M.J.E. Alam, K.M. Muttaqi, D. Sutanto, A three-phase power flow approach for integrated
3-wire MV and 4-wire multigrounded LV networks with rooftop solar PV. IEEE Trans.
Power Syst. 28(2), 1728–1737 (2013)
30. K. Mahmoud, M. Abdel-Akher, Analysis of hybrid photovoltaic and wind energies connected
to unbalanced distribution systems, in Proceedings of 2010 IEEE International Conference on
Power and Energy (PEcon), (Kuala Lumpur, 2010), pp. 79–84
31. K. Mahmoud, N. Yorino, A. Ahmed, Optimal distributed generation allocation in distribution
systems for loss minimization. IEEE Trans. Power Syst. 31(2), 960–969 (2016)
32. K. Mahmoud, N. Yorino, A. Ahmed, Power loss minimization in distribution systems using
multiple distributed generations. IEEJ Trans. Elec. Electron. Eng. 10(5), 521–526 (2015)
33. K. Mahmoud, Y. Naoto, Optimal combination of DG technologies in distribution systems, in
Power and Energy Engineering Conference (APPEEC), 2015 IEEE PES Asia-Pacific (2015)
34. K. Mahmoud, Optimal integration of DG and capacitors in distribution systems, in Power
Systems Conference (MEPCON), 2016 Eighteenth International Middle Eastern IEEE (2016),
pp. 651–655
35. K. Mahmoud, Y. Naoto, Robust quadratic-based BFS power flow method for multi-phase
distribution systems. IET Gener. Transm. Distrib. 10(9), 2240–2250 (2016)
36. M.E. Baran, F.F. Wu, Optimum sizing of capacitor placed on radial distribution systems.
IEEE Trans. Power Del. 4(1), 735–743 (1989)
Chapter 7
Battery Energy Storage Planning

Mahdi Sedghi, Ali Ahmadian, Ali Elkamel, Masoud Aliakbar Golkar

and Michael Fowler

Abstract Rechargeable grid-scale batteries are suitable and mature technology for
energy storage in active distribution networks. Battery energy storage (BES) units
have many advantages and are used for several purposes in electric systems and
distribution grids. They are used not only for peak shaving and voltage regulation,
but also for reliability enhancement and dispatching the renewable-based dis-
tributed generation (DG) sources. However, BES technologies are still expensive
and need to be employed optimally to prevent excess investment cost. Optimal
planning of BES is a complex approach that determines the type, location, capacity
and power rating of energy storage units. The optimization should handle the
uncertain conditions and it requires to develop the appropriate models and methods.
There are many effective components that should be addressed. These components
influence the results of the optimal planning and make it more complicated. In this
chapter the optimal BES planning methodologies are presented. Firstly the opti-
mization problem is formulated considering different economic perspectives. Then
the approaches and strategies for solving the combinatorial problem are described.
In this way, both the probabilistic and possibilistic methods and models are
displayed. In addition, the most important components and factors that affect the

M. Sedghi (&)
M. A. Golkar

Faculty of Electrical Engineering, K. N. Toosi University
of Technology, Tehran, Iran
e-mail: [email protected]
M. A. Golkar
e-mail: [email protected]
A. Ahmadian
Department of Electrical Engineering, University of Bonab, Bonab, Iran
e-mail: [email protected]
A. Elkamel
M. Fowler
Department of Chemical Engineering, University of Waterloo,
Waterloo, ON, Canada
e-mail: [email protected]
M. Fowler
e-mail: [email protected]

© Springer Nature Singapore Pte Ltd. 2018 185

F. Shahnia et al. (eds.), Electric Distribution Network Planning,
Power Systems, https://doi.org/10.1007/978-981-10-7056-3_7
186 M. Sedghi et al.

optimal planning are characterized and analyzed, including conventional DGs,

renewable-based DGs, capacitor banks, plug-in electric vehicles, etc.

Keywords Battery energy storage
Optimal planning
Active distribution

network

7.1 Introduction

The energy storage systems (ESSs) are widely used in active distribution networks
for several purposes. The ESSs absorb energy and store it for a period of time and
then inject it to supply energy in scheduled times. They can be installed throughout
the electric grid and employed in heating and cooling networks, stand-alone and
grid-connected electric systems, uninterruptible power supply (UPS) systems, etc.
Hence, they provide a valuable opportunity for grid operators to manage the net-
works optimally. The ESSs can store energy in off-peak hours (when the electricity
price is low) and inject it in peak-load times (when the electricity price is high) [1].
So, they are used for the power profile peak shaving and valley filling in efficient
manner and such they provide an ‘energy arbitrage’ service. This service can
address the generation, transmission and distribution expansion concerns, and also
it defers the reinforcement necessity of the infrastructure. The stored energy
enhances the reliability, especially in radial distribution networks and stand-alone
systems, as it can restore the loads in outage times [2]. Furthermore, they can
provide voltage regulation service and also a valuable tool for grid operators to
accommodate the supply and/or demand-side variability [3]. Several technologies
have been introduced for energy storage and they can be implemented on large and
small scales in distributed and centralized manners within electric systems [4].
While some technologies are mature or in near maturity, some of them are still in
the early development stages and will require additional attention before their
potential is fully realized [5]. Research and development works are currently
underway with the primary goals of realizing technology cost reductions and
improving the performance of existing, new and emerging storage technologies. In
addition, the non-technical barriers should be addressed for deployment of ESSs by
governments and industry stakeholders.
Generally, the ESSs are categorized in thermal and electricity (heat or cold)
groups based on their outputs. Both groups can work as generators and consumers,
making them useful for energy management. Based on input and output energy
types of ESSs, they can provide several services including electricity to electricity,
heat to heat, electricity to heat, and heat to electricity applications. Using these
services, the ESSs provides a valuable opportunity for energy management
improvement both in energy supply and demand sides [6].
Due to their characteristics, each type of the ESSs can be utilized for individual
application. For example, the high power ESSs are needed for frequency regulation
service to maintain a constant main frequency in utility application, while a more
7 Battery Energy Storage Planning 187

sustained power ESSs are needed for energy management service in which the
demand is met and the supply is not wasted. There are six parameters for energy
storage systems which make distinguished of different technologies. Specific
energy, energy density, specific power, power density, operating cost, efficiency,
and cycle life are the main parameters which play important role in technology type
choosing for different applications [5]. Specific energy is the energy storage
capacity per unit mass while the energy density denotes energy storage capacity per
unit volume. The operating cost is the cost of one cycle of charge/discharge in rated
power, and it consists of maintenance, heating, labor, etc. Specific power charac-
terizes the capability of the discharge power per unit mass. The efficiency is defined
as the ratio of total energy injected to the load over total energy absorbed by the
ESS in a single charge/discharge cycle. The cycle life determines the number of
charge/discharge cycles which could be fulfilled until the useful capacity of the ESS
brings down to 80% of its initial capacity. The importance of every parameter
depends on the application.
Among the energy storage technologies, the batteries have been widely con-
sidered to be utilized in electric distribution networks. Different battery types
consist of lead-acid, NaS, Ni-Cd, lithium-ion, Zn-Br, vanadium redox flow and
ZEBRA, some of which are commercialized for utility applications [7]. Each of
these battery types have advantages and disadvantages in comparison with each
other, making it suitable for individual application.
Although the ESSs have several advantages in distribution networks, however
the optimal planning and scheduling of them are the main prerequisites for optimal
exploitation of them. Generally, the battery energy storage (BES) can be imple-
mented in the most buses of the distribution networks as the batteries have less
environmental and non-technical constraints. However, the electrical considerations
such as power follow, power loss, voltage regulation and etc. affect on optimal
location of batteries [8]. Similarly, the optimal scheduling of batteries depends on
electrical and non-electrical considerations of networks such as power price, power
loss, power follow, renewable energies availability, flexible load and generation
availability, etc. As a result, the batteries should be optimally scheduled and
planned in distribution networks to prevent excess operation and investment costs
[9]. For these purposes, the effective components in generation and consumption
sides as well as electric grid should be modeled properly. In generation sides the
components such as conventional distributed generations (DGs), wind and solar
energy based DGs, etc., and in the consumption side the flexible and non-flexible
load demands should be modeled appropriately.
The BES planning approach should determine the optimal type, location, size
and power rating of the batteries in electric systems and grids. A lot of research has
focused on the optimal sizing of BES in electric systems including renewable-based
resources [10–14]. The optimization of the BES location, capacity and power rating
in distribution network is shown in [15–17] while the optimal typing has been
investigated in [7, 9]. With respect to the literature, there are several methods and
models for optimal BES planning that is displayed here.
188 M. Sedghi et al.

In this chapter, firstly the BES planning problem is defined. It is formulated in

two ways, considering the owner of storage that invests in BES. In order to solve
this problem, the various methods and techniques employed for optimal BES
planning are presented in continue. These methods and techniques include
• Multi-layer methods (including master/slave based strategy and optimal power
flow based strategy)
• Probabilistic optimal power flow
• Possibilistic optimal power flow
• Monte Carlo simulation
• Point estimate method
• Fuzzy load flow
After introducing the problem and solution techniques, the results of optimal
BES planning presented in literature are discussed and analyzed. The discussion
represents how the results of case studies are influenced by the most effective
components in active distribution networks. These components which affect the
optimal BES planning can be categorized in three main groups of
• The other facilities of distribution network
• Storage capabilities
• Techno-economic factors
The components which belong to these groups are shown in Fig. 7.1. The impact
of effective components on BES are comprehensively compared at the end.

The other Storage capabilities

facilities
Reactive
Islanding
Conventional power
mode of
DGs (injection/
operation
absorption)
Renewable-
based DGs
Optimal
BES
Plug-in
electric Planning
vehicles Techno-economic factors

Capacitor Installation
banks Durability
cost

Electricity
Tap changer Efficiency
price
equipped
transformers
Environmental impacts

Fig. 7.1 The most effective components which influence the optimal BES planning
7 Battery Energy Storage Planning 189

7.2 Optimal BES Planning

The long-term BES planning determines the optimal type, location, capacity and
power rating of grid-scale batteries. In other words, they are the decision variables
in optimal BES planning. The optimal solution should minimize the cost function
(or maximize the benefit) considering the technical constraints. The costs usually
include the investment cost, operation cost and reliability cost. The optimal solution
should satisfy the voltage magnitude constraint, maximum allowed capacity con-
straint and power balance constraint. It is possible to model the technical constraints
as a penalty factor in the objective function. So the unpractical solutions are
penalized in this way.
With respect to the economic perspectives, there are two kinds of planning as
follow:
• scenario-I: The distribution utility invests in BES; and,
• scenario-II: An independent organization rather than distribution company
(DisCo) is the owner of storage.
In the first case, a cost objective function is defined to be minimized subject to
the technical constraints. While in the second state, one is faced to two objectives
that should maximize the benefits of two different companies simultaneously.
Therefore two objective functions are defined in the latter case. Both definitions are
presented and formulated below.

7.2.1 Definitions

7.2.1.1 Objective Function(s)

In the first scenario a total cost function is determined to be minimized subject to

the technical constraints. Generally the objective function consists of investment
cost, operation cost, reliability cost and maybe a penalty factor, as

Min: f ¼ IC þ OC þ RC þ M  PF ð7:1Þ

where M is a large fixed value for penalization of the infeasible solutions.

The investment cost contains three terms: installation cost related to the battery
capacity, installation cost related to the battery power rating, and replacement cost
of batteries, as
nN h
X i
IC ¼ xb  CCap
INS
þ xb  CPow
INS
þ xb  CBatt
REP
ð7:2Þ
b¼1
190 M. Sedghi et al.

The operation cost includes the cost of purchased power by DisCo and the
operation and maintenance (O&M) cost of batteries, over the project period, as

X
nN
 
OC ¼ CEAR ðTÞ þ xb  CBatt
OM
ðTÞ ð7:3Þ
b¼1

It should be noted that CEAR ðTÞ contains the cost of power loss as well.
The reliability cost is the cost of outage which arises from all the probable failure
events in distribution network. RC is a function of energy not supplied (ENS) as
well as the type of interrupted loads. It is noticeable that RC denotes the cost of
ENS in (7.1). So it is possible to combine the costs with the same unit in a single
objective function. If the other reliability indices with different units are employed,
different weights should be used in the objective function to consider the impor-
tance of the different reliability indices.
The penalty factor is corresponding with the number of violated constraints.
In the second scenario, two profit functions i.e. fDisCo and fStOwn are defined.
fDisCo and fStOwn are the profit functions of DisCo and storage owner, respectively.
The profit of the storage owner is equal to the difference between the income and
cost, as

Max: fStOwn ¼ CIN

ST
 CCOS
ST
ð7:4Þ

The storage owner purchases the electrical energy from DisCo in light-load
periods, and then, it sells the electricity to the DisCo in peak-load times. So, the
income of storage owner can be formulated as
ST
CIN ¼ CSel
ST
 CPur
ST
ð7:5Þ

The cost of storage owner is obtained from

nN h
X i XnN
 
ST
CCOS ¼ xb  CCap
INS
þ xb  CPow
INS
þ xb  CBatt
REP
þ xb  CBatt
OM
ðTÞ ð7:6Þ
b¼1 b¼1

The profit of DisCo is corresponding with the reduction of the cost due to the
installation of storage. Moreover, the DisCo is responsible for satisfying the tech-
nical constraints. Hence the penalty factor should be taken into account in DisCo’s
profit, as shown below.
DC
CwithoutST  CwithST
DC
Max: fDisCo ¼ ð7:7Þ
1 þ PF
7 Battery Energy Storage Planning 191

These cost functions are presented in Eqs. (7.8) and (7.9).

DC
CwithoutST ¼ CE;withoutST
AR
ðTÞ þ RCwithoutST ð7:8Þ

DC
CwithST ¼ CE;ST
DC
ðTÞ þ CE;withST
AR
ðTÞ þ RCwithST ð7:9Þ

In fact the DisCo purchases the electricity from the storage owner with a specific
tariff.
If all the technical constraints are satisfied, the penalty factor is equal to zero, and
hence, PF = 0 in (7.7). Otherwise PF > 0, so that the profit of the DisCo is reduced
considerably.
In scenario-II the Pareto set is used to assign the global best solution similar to
the work shown in [18].
It is notable that (7.1) represents a static BES planning. Generally the optimal
BES planning may be defined as a multistage expansion planning where the
planning horizon can be divided into several stages, so that the elements to be
installed in each stage should be determined. In other words, the installation date is
a decision variable as well. In this situation, the objective function is defined as

X
N
F¼ fs ð7:10Þ
s¼1
where

fs ¼ ICðsÞ þ OCðsÞ þ RCðsÞ þ PF ð7:11Þ

In this definition, the investment and operation cost formulations are modified as
follow:
nN h
X i
IC ¼ xs;b  CCap
INS
ðTs Þ þ xs;b  CPow
INS
ðTs Þ þ xs;b  CBatt
REP
ðTs Þ ð7:12Þ
b¼1

X
nN
 
OCðsÞ ¼ CEAR ðTs Þ þ xs;b  CBatt
OM
ðTs Þ ð7:13Þ
b¼1

Similar to the investment and operation costs, the reliability cost is a function of
the number and duration of stages as well. In all the situations, the technical
constraints should be satisfied at all the operation times, as formulated below.

7.2.1.2 Technical Constraints

There are several technical constraints that should be considered in optimal storage
planning. In an acceptable solution, the voltage level of all the nodes must be
limited to the allowed boundary; the constraint of maximum allowed capacity
192 M. Sedghi et al.

should not be violated for all the equipment; and the active/reactive power balance
should be satisfied. These technical constraints are formulated below.

Vmin  Vn;t  Vmax ; 8n 2 AN ; 8t 2 AT ð7:14Þ

Sumin  Sut  Sumax ; 8u 2 AEQ ; 8t 2 AT ð7:15Þ

X X X X
s;t þ
SSS d;t ¼
SDG k;t þ
SST l;t þ St
SLD ; 8t 2 AT ð7:16Þ
LOSS

s2ASS d2ADG k2AST l2ALD

In (7.16), the real part of SST

k;t is a negative value during discharge period.
In addition, the state of charge (SOC) of the batteries are updated as

SOCk;t þ PST
k;t =gdis ; k;t  0
if PST
SOCk;t þ 1 ¼ ð7:17Þ
SOCk;t þ gch  PSTk;t ; otherwise

7.2.1.3 Penalization

The optimal solution should satisfy the technical constraints. In this way the can-
didate solutions in which the constraints are violated, should be penalized. The
penalty factor is proportional to the total number of violations in every solution.
However, the penalization should be corresponding with the probability of the
violation, as well. For example, the violations with 1 and 100% probability should
be considered in different manner. Hence, the penalty factor is defined as

X
nVIO
PF ¼ pðvÞ ð7:18Þ
v¼1

p(v) is obtained from the probabilistic/possibilistic load flow (PLF) analysis that is
described in the next subsections.

7.2.2 Optimal Power Flow (OPF)

The type, location, capacity and power rating of energy storage units are the main
decision variables in optimal battery planning. However, the long-term optimization
should be accomplished considering the optimal charge/discharge cycles. In real
conditions an optimal scheduling i.e. OPF is required to be taken into account.
Unlike the common OPF which is performed for a snapshot, here the OPF is an
uncertain probabilistic/possibilistic approach. So the intermittent renewable power
and the stochastic load profiles are taken into consideration in probabilistic/
possibilistic optimal power flow (POPF).
7 Battery Energy Storage Planning 193

Optimal battery energy storage planning (Main layer)

Decision variables: Type, location, capacity and power rating

POPF

Optimal power flow (OPF)

Decision variables: active/reactive power and SOC of the batteries

Probabilistic/possibilistic load flow (PLF)

Using MCS or PEM or fuzzy analysis

Fig. 7.2 Multi-layer approach for optimal BES planning (master/slave strategy)

In order to employ the POPF in optimal battery planning, there are two strategies
as follow: the Master/slave strategy, and, the OPF-based strategy. The master/slave
strategy is based on a multi-layer method. The optimal planning is as the main
optimization layer that considers the results of POPF as a sub-layer. The main layer
determines the optimal type, location, capacity and power rating as decision vari-
ables, while the active/reactive power and SOC are the decision variables of the
sub-layer. For every given type, location, capacity and power rating, the POPF is
executed completely to specify the optimal active/reactive power and SOC.
On the other hand, the implementation of POPF requires executing PLF analysis.
Therefore the PLF appears as a sub-layer in POPF as shown in Fig. 7.2.
Similar to the master/slave strategy, the OPF-based strategy contains the PLF
analysis as a sub-layer, however, the location, capacity and power rating are not
predetermined in a master layer. In OPF-based strategy the POPF is executed
considering stochastic active/reactive power and SOC. Then the candidate capacity,
power rating and the locations are determined according to Eqs. (7.19)–(7.21).

SCap ðnÞ ¼ max½SOCðt; nÞ; 8n 2 AN ð7:19Þ

t2AT

qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
PRat ðnÞ ¼ max ½Pðt; nÞ2 þ ½Qðt; nÞ2 ; 8n 2 AN ð7:20Þ
t2AT

 
AST ¼ n 2 AN jSCap ðnÞ [ 0 ð7:21Þ

So the candidate location, capacity and power rating are functions of the POPF
program. In both strategies it is required to execute the probabilistic or possibilistic
load flow analyses that are described in the next subsections.
194 M. Sedghi et al.

7.2.2.1 Probabilistic Load Flow

The probabilistic load flow (PLF) is based on the basic load flow analysis whose
main equations are

X
n
Pp ¼ Vp Vq ðGpq cosðdpq Þ þ Bpq sinðdpq ÞÞ; p ¼ 1; 2; . . .; n ð7:22Þ
q¼1

X
n
Qp ¼ Vp Vq ðGpq sinðdpq Þ  Bpq cosðdpq ÞÞ; p ¼ 1; 2; . . .; n ð7:23Þ
q¼1

There are several load flow methods to solve the above equations in active
distribution networks as shown in [19]. However, in PLF the active and reactive
power as well as the voltage and current magnitudes and angles are stochastic
variables [20].
The methodologies for PLF can be categorized to three types of analytical
methods, numerical methods and approximate methods. The analytical methods are
very complex, while the numerical ones are much simpler but time consuming. The
approximate methods are not very complex and does not take a long time, however,
the results are less precise compared with the other methods. Despite of this fact,
the accuracy of the approximate methods is usually acceptable in optimal storage
planning. Here, two well-known methods i.e. Monte Carlo simulation (MCS) and
point estimate method (PEM) are briefly presented.

7.2.2.2 Monte Carlo Simulation (MCS)

MCS, as a numerical method, is the simplest approach for PLF analysis. It is based
on deterministic load flow and random sampling. Random samples are selected
considering the probability distribution function (PDF) of input stochastic variables.
The load flow analysis is frequently executed considering the selected samples and
the results are saved. Finally the PDF of output stochastic variables are estimated.
Figure 7.3 represents the flowchart of the MCS implementation. Although the MCS
is very simple and accurate, but it needs a large computation effort. So the
approximate methods are more suitable in real conditions.

7.2.2.3 Point Estimate Method (PEM)

The PEM has been used widely for PLF analysis in electric system studies [21]. It is
based on Taylor series expansion. Suppose that the load flow nonlinear functions
are presented as Y = h(X). The PDF of input and output variables are shown with
fX(x) and fY(y), respectively. fX(x) and h are known functions, while the goal is to
determine the PDF fY(y).
7 Battery Energy Storage Planning 195

Start

Inputs: Network properties and PDF

of load demands and power generation
of DGs

Generate samples using distribution function

parameters

=1

Choose the k-th random sample for each generation and load power at
network buses

Run deterministic load flow using k-th sample inputs

Store magnitude and angle of voltages and lines current (or other necessary
outputs) in k-th row of the output matrix B

No Is convergence
= +1
criterion satisfied?

Yes

Assign or fit each column of matrix B to a distribution function

Outputs: PDFs of magnitude and angle of

buses voltage and lines current (or other
necessary outputs)

End

Fig. 7.3 The flowchart of the MCS implementation for PLF analysis
196 M. Sedghi et al.

The PDF fX(x) is constricted in n points i.e. xi (i = 1, 2, …, n), so that it can be

written as

X
n
fX ðxÞ ¼ Pi dðx  xi Þ ð7:24Þ
i¼1

In n-PEM, n points are considered for contraction. The more the number of
points, the more accurate the results would be.
As an example, in 2-PEM the PDF fX(x) is approximated according to

fX ðxÞ ffi P1 dðx  x1 Þ þ P2 dðx  x2 Þ ð7:25Þ

This approach is illustrated in Fig. 7.4.

In order to establish the approximation, the first three moments of the actual
function and the approximated function should be equal with each other [22].
The jth central moment of fX(x) is defined as

Zþ 1
Mj ðXÞ ¼ ðx  lX Þ j fX ðxÞdx; j ¼ 1; 2; . . . ð7:26Þ
1

To normalize the approximated function, variable ni is defined as

x i  lX
ni ¼ ; i ¼ 1; 2 ð7:27Þ
rX

The moments of fX(x) are then

M0 ¼ P1 þ P2 ¼ 1 ð7:28Þ

M1 ¼ n1 P1  n2 P2 ¼ 0 ð7:29Þ

Fig. 7.4 The 2-PEM

approach for PLF analysis
7 Battery Energy Storage Planning 197

M2 ¼ r2X ðn21 P1 þ n22 P2 Þ ¼ r2X ð7:30Þ

M3 ¼ r3X ðn31 P1  n32 P2 Þ ¼ kX r3X ð7:31Þ

Using the Taylor series expansion of h(X) about µX, and considering the
approximation shown in (7.25), one gets [22]

P1 þ P2 ¼ M0 ðXÞ ¼ 1 ð7:32Þ

M1 ðXÞ
P1 n1 þ P2 n2 ¼ ¼ kX;1 ð7:33Þ
rX

M2 ðXÞ
P1 n21 þ P2 n22 ¼ ¼ kX;2 ð7:34Þ
r2X

M3 ðXÞ
P1 n31 þ P2 n32 ¼ ¼ kX;3 ð7:35Þ
r3X

Solving the system of four equations with four unknowns results in

rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
kX;3 kX;3 2
n1 ¼ þ 1þð Þ ð7:36Þ
2 2
rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
kX;3 kX;3 2
n2 ¼  1þð Þ ð7:37Þ
2 2
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
kX;3 k 2
2  1 þ ð X;3 2 Þ
P1 ¼  qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ð7:38Þ
k 2
2 1 þ ð X;3 2 Þ

qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
kX;3 k 2
2 þ 1 þ ð X;3 2 Þ
P2 ¼ qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ð7:39Þ
k 2
2 1 þ ð X;3 2 Þ

The locations of xi (i = 1,2) are calculated using (7.27). Finally the kth moment
of Y is determined as

E(Y k Þ ffi P1 ½hðx1 Þk þ P2 ½hðx2 Þk ð7:40Þ

The mean value and variance of Y are obtained from

E(YÞ ffi P1 hðx1 Þ þ P2 hðx2 Þ ð7:41Þ

198 M. Sedghi et al.

r2Y ¼ EðY 2 Þ  ½EðYÞ2 ð7:42Þ

Similarly, in n-PEM (n > 2), the locations xi (i > 2) and the corresponding
probabilities Pi (i > 2) are estimable. Then the kth moment of Y is calculated as
n n
X o
EðY k Þ ffi Pi ½hðxi Þk ð7:43Þ
i¼1

The PDF of the output stochastic variables are extracted using the moments
of Y. The maximum entropy and gram-charlier are the most well-known methods
employed for this estimation [23]. The flowchart of 2-PEM application is repre-
sented in Fig. 7.5.

7.2.2.4 Possibilistic OPF

Another POPF i.e. possibilistic OPF is based on fuzzy load flow (FLF) analysis. It
is useful when there is no exact model or precise information about the uncertain
parameters. Implementation of the FLF is simpler than PLF analysis. Similar to the
probabilistic OPF, the possibilistic OPF can be handled using master/slave or
OPF-based strategies, however, the FLF layer is used instead of the PLF layer.
The FLF analysis is based on the conventional load flow methods e.g. backward/
forward sweep load flow, but the input and output variables are fuzzy values. Hence
it is required to use fuzzy operators in computations [8]. The fuzzy values can be
defined using several types of membership functions. The triangular and trapezoidal
membership functions are the most common functions which has been widely used
in FLF analyses. These membership functions are shown in Fig. 7.6.
A triangular fuzzy value is introduced using three parameters i.e. f1, f2 and f3.
e ¼ ða1 ; a2 ; a3 Þ and B
If A e ¼ ðb1 ; b2 ; b3 Þ are two fuzzy numbers, then the main
fuzzy operators are defined as follow [8]:

eB
A e ¼ ða1  b1 ; a2  b2 ; a3  b3 Þ ð7:44Þ

e:B
A e ¼ ða1 b1 ; a2 b2 ; a3 b3 Þ ð7:45Þ

e B
A= e ¼ ða1 =b3 ; a2 =b2 ; a3 =b1 Þ ð7:46Þ

The trapezoidal fuzzy values are shown using four parameters i.e. l1, l2, l3 and l4.
Suppose that C e ¼ ½c1 ; c2 ; c3 ; c4  and D
e ¼ ½d1 ; d2 ; d3 ; d4  are two trapezoidal
fuzzy numbers, then the main fuzzy operators are defined as follow [24]:
7 Battery Energy Storage Planning 199

Start

Inputs: Network properties and PDF

of load demands and power generations
of DGs

Calculate mean and variance of random variables of load demands and DGs

Calculate moments of random variables of load demands and DGs

Calculate ,3 using Eq. (1.35)

Calculate 1 and 2 using Eqs. (1.23), (1.36) and (1.37)

Calculate P 1 and P 2 using Eqs. (1.38) and (1.39)

Calculate ℎ( ) that are nodes voltage and lines current using deterministic
load flow (for generation/consumption power in 1 and 2 operation points)

Calculate moments of nodes voltage and lines current using Eq. (1.40)

Calculate mean, variance and skewness of voltages and currents

Estimate PDFs of voltages and currents using maximum Entropy method or

Gram-Charlier (if necessary)

Outputs: Properties of PDFs of magnitude

and angel of voltages and currents (or other
necessary outputs)

End

Fig. 7.5 The flowchart of 2-PEM implementation for PLF analysis

200 M. Sedghi et al.

Fig. 7.6 a The triangular and b the trapezoidal membership functions used in FLF analysis

Fig. 7.7 Multi-linearization

scheme of membership
function for FLF analysis

eD
C e ¼ ½ c 1  d1 ; c 2  d2 ; c 3  d3 ; c 4  d4  ð7:47Þ

e D
C: e ¼ ½minðc1 d1 ; c1 d4 ; c4 d1 ; c4 d4 Þ; minðc2 d2 ; c2 d3 ; c3 d2 ; c3 d3 Þ;
ð7:48Þ
maxðc2 d2 ; c2 d3 ; c3 d2 ; c3 d3 Þ; maxðc1 d1 ; c1 d4 ; c4 d1 ; c4 d4 Þ

e D
C= e ¼ ½minðc1 =d1 ; c1 =d4 ; c4 =d1 ; c4 =d4 Þ; minðc2 =d2 ; c2 =d3 ; c3 =d2 ; c3 =d3 Þ;
maxðc2 =d2 ; c2 =d3 ; c3 =d2 ; c3 =d3 Þ; maxðc1 =d1 ; c1 =d4 ; c4 =d1 ; c4 =d4 Þ
ð7:49Þ

When the membership functions are nonlinear and complex, the computation
effort increases. Therefore the membership functions are simplified using lin-
earization techniques. It is usually preferred to linearize the membership functions
about several points, but not about a single point [25]. In this way the simplification
becomes more accurate. Figure 7.7 represents a multi-linearization scheme for a
fuzzy membership function.
7 Battery Energy Storage Planning 201

7.3 Effective Components

As shown in Fig. 7.1, there are several components that affect the optimal BES
scheduling and planning. These components provide opportunities and threats for
battery storage and they influence the penetration of battery units in active distri-
bution networks. They are described and analyzed below.

7.3.1 Conventional DGs

Conventional dispatchable DGs such as micro-turbines and diesel generators can be

used for several purposes such as peak shaving, reliability enhancement and voltage
regulation. In fact, they are rivals for storage units in grid-connected as well as
stand-alone electric systems. However, they are basically different technologies
with various costs and benefits. In active distribution networks, the storage units
usually absorb active power in light-load times and they inject the power to the grid
in peak-load periods. While the conventional DGs are used only to inject the power
in normal-load and/or peak-load times [26–28]. So storage units are more flexible to
smooth the load profile. Moreover, in presence of renewable-based DGs, the excess
energy can be stored in storage units, while without storage it should be injected to
the upstream network. In such times, the electrical energy price may be low and it is
not optimal to sell the energy to the grid.
In stand-alone systems, diesel generators are widely used to meet load and
generation especially in peak-load periods. If the renewable-based DGs are
employed in such systems, the excess energy should be curtailed in off-peak times
when the load is less than the generated power. So in this case, the batteries are
much more beneficial [29].
Using the BES usually increases the power loss in distribution network due to
the energy storage efficiency which is less than 100% in actual conditions [8].
However, the cost of power loss is reduced thanks to the cheap electrical energy in
light-load periods [17].
In the competition between the BES and conventional DGs, the most important
factors are the investment cost, fuel price and the electrical energy price in
light-load times. The energy storage is an expensive technology with higher cost for
installation/replacement. However, it is more economical than DGs if the fuel price
increases or the electrical energy price reduces in light-load times.
Finally, from the environmental point of view, employing BES instead of
conventional DGs reduces the pollution. The reason is that it reduces the required
capacity of the fossil fuel power plants and provides the techno-economic condi-
tions for increasing the penetration of renewable energy.
202 M. Sedghi et al.

7.3.2 Renewable-Based DGs

Renewable energies such as wind and solar are intermittent resources which cannot be
dispatched without energy storage units. The combination of renewable-based DG
and BES is very interesting for stand-alone electric systems in remote areas.
Additionally it can be beneficial in grid-connected systems for peak shaving, relia-
bility enhancement, voltage regulation and dispatching the intermittent power. The
required capacity of storage increases as the penetration of renewable energy increases
in active distribution networks [9, 17]. Traditionally using the BES in grid-connected
systems is taken into attention if the penetration of renewable-based DGs is high
[15, 30–32]. The reason is that the energy storage technology is very expensive. In
distribution networks without intermittent energy resources, it is not economical to
install BES units, unless they are used for several objectives simultaneously [27].

7.3.3 Plug-in Electric Vehicles (PEVs)

The PEVs can be connected to distribution networks in grid-to-vehicle (G2V) or

vehicle-to-grid (V2G) modes of operation. In uncoordinated G2V manner, they are
appeared as new uncertain loads, so that more storage units are required to support
the grid. However, less capacity of storage is needed, if the PEVs are coordinated in
G2V mode of operation. The reason is that the coordinated PEVs are usually
charged in light-load periods when the storage units should be charged as well. So
the grid capacity should be optimally apportioned between the PEVs and storage. In
the case of smart V2G, the PEVs can play the role of storage in distribution network
and the storage application is challenged critically.
In near future, both G2V and V2G modes of operation are possible, as a part of
vehicles are plugged into the grid in uncoordinated manner [28]. Hence, in order to
reinforce the network, the BES units should be employed beside the DG resources,
as a global optimal solution [8]. In these situations, the share of BES and DG is a
function of PEVs penetration as shown typically in Fig. 7.8.

Fig. 7.8 The optimal portion of BES and DG as a function of PEVs penetration [33]
7 Battery Energy Storage Planning 203

As can be seen, the optimal capacity of BES and DG is a nonlinear function of

the PEVs penetration. With increasing the penetration of PEVs up to 25%, the
capacity of storage should be increased, and after that, the storage capacity is
reduced. The growing of BES and DG capacity are in reverse with each other. The
reason is that the V2G (with a high penetration) can treat as stationary storage unit,
and hence, more DG capacity (instead of storage) is required to support the grid
against those PEVs that are operated in uncoordinated manner.
In addition, the uncertainty behaviour of PEVs is another effective factor which
influences the optimal capacity of storage units. Let’s consider the following two
scenarios:
• scenario-A: Optimal planning of BES considering the deterministic model of
PEVs; and,
• scenario-B: Optimal planning of BES considering the uncertain model of PEVs.
The optimal capacity of storage in scenario-B is more than that in scenario-A
[8, 33]. So the uncertain manner of PEVs results in more required BES capacity in
active distribution network.

7.3.4 Tap-Changer Equipped Transformers

The BES units can be used to inject the power to the upstream grid and sell the
electrical energy if it is beneficial considering the electricity price profile. However,
reverse power to high voltage (HV) network results in raising the voltage level in
medium voltage (MV) or low voltage (LV) feeders. Therefore the number or
capacity of BES units is limited due to the violation of technical constraints [34]. In
this state, absorbing more reactive power in BES units increases the power rating
cost (but not necessarily the capacity cost) and it is not efficient enough in reverse
power flow conditions. Reducing the tap position of HV/MV transformers, which
are equipped to on-load tap-changer, is more effective to decrease the voltage level
of the grid. As a result, the penetration of BES units is allowed to be increased
thanks to regulating the tap position. The on-load tap changer and the BES units
embedded in distribution grid should be operated in coordinated manner.
It is notable that using tap changer for voltage regulation reduces the lifetime of
the HV/MV transformer, and hence, the reliability of the system decreases [35]. The
failure event, due to the unsuccessful change of tap position, causes a wide outage
in distribution network.
204 M. Sedghi et al.

7.3.5 Capacitor Banks

Capacitor banks have been used widely in passive distribution networks to increase
the voltage level of long MV feeders. However, in modern distribution networks
containing DG sources and flexible active/reactive power flow capability of storage
units, it may not be needed to install capacitor banks seriously. Without capacitor
banks, more storage capacity is required and the cost function increases [35].
Consequently the combination of storage and capacitor is more beneficial. In this
case, the optimal siting and sizing of both technologies i.e. the BES units and
capacitors, should be performed simultaneously.

7.3.6 Battery Capabilities

BES units have two important capabilities which are very helpful for distribution
grids. These capabilities influence not only the conditions of the network, but also
the optimal operation, siting and sizing of storage units. These effects are presented
below.

7.3.6.1 Reactive Power

The grid-scale BES units can inject/absorb not only the active power, but also the
reactive power. So they can be operated in four zones of active-reactive power
surface [36]. However, the electrical current of the battery and the converter which
connect the dc source to the ac grid, is limited. The apparent current has two
components on the real and imaginary axes, as shown in Fig. 7.9.
The apparent current can be stated as

Ist2 ¼ Ist;re
2
þ Ist;im
2
ð7:50Þ

For a given voltage, the allowed active-reactive power zone where the storage
can be operated, is inside a circle with the radius of apparent power, as shown in
Fig. 7.10.

Fig. 7.9 The vector of

storage current (Ist) and its Im.
components on real (Re.) and
Ist
imaginary (Im.) axes
Ist,im

Ist,re
Re.
7 Battery Energy Storage Planning 205

Fig. 7.10 The active/reactive

power limit in a BES unit

According to this figure, the maximum allowed value of QðtÞ is a function of

PðtÞ.
If the BES is not allowed to inject/absorb reactive power, it is forced to regulate
the voltage only using the active power injection/absorption. As a result, more
number or capacity of BES is necessary [17]. So utilizing the reactive power
capability reduces the required capacity of storage, and hence, the investment cost
decreases.
The BES units that are allowed to exchange reactive power with the network, are
preferred to be located near the weak nodes from voltage level point of view. In
such nodes, the variance of voltage is usually high, and the flexible active/reactive
capability of BES improves the voltage regulation.

7.3.6.2 Islanding Mode Operation

The distributed energy storage units are capable to reduce the ENS within a dis-
tribution grid. When a device e.g. a feeder section or transformer is failed, it is
disconnected from the grid using normally closed switches (NCSs). In this case, the
BES unit can be employed to restore a part of the load in islanding mode of
operation, as shown in Fig. 7.11.

Failure BES DG
HV/MV event

Grid
NCS NCS NCS

Vital load

Fig. 7.11 Islanding mode operation of storage in order to reduce ENS cost
206 M. Sedghi et al.

Start of islanding
Charge battery
using DG power.
Inputs: failure rates, data
of storage, DG and load No
Is storage Yes Curtail the
fully charged? excess power.
t=1
Yes
Is storage available No Is storage No
considering SOC existing?
constraint?
Yes

, + , ≥
,> , + , No , + , No
AND are constraints
AND are constraints
t=t+1 satisfied?
satisfied?

Yes
Yes
Curtail the Yes Is storage
excess power. fully charged?
No
No
Is islanding Store excess energy
finished? in the battery.

Yes
Update SOC of the
Calculate outage cost battery.
considering ENS and
outage duration.

Update ENS and

Output: the outage cost outage duration.

Update ENS and

End of islanding
outage duration.

Fig. 7.12 The flowchart of BES operation algorithm in islanding mode with renewable-based DG
(in jth node of the grid)

The islanding mode operation of BES may take place for several hours during
which the islanding mode algorithm is used for operation. The goal of algorithm is
to minimize the outage cost. If the storage SOC is not enough to restore the vital
load, DG sources inject the power to the islanded area. Due to their intermittency,
the renewable-based DGs cannot restore the loads alone. Therefore they should be
coordinated with the BES units if possible. The excess energy of the DG sources
can be used to charge the BES, so that the battery is sustained to feed the loads in
the next hours of islanding operation. If the battery is fully charged, the excess
energy of DG should be curtailed. All the technical constraints as well as the SOC
limitations must be satisfied within the islanding mode operation, otherwise the load
cannot be restored and the ENS is not reduced. The algorithm which is used for
islanding mode of operation is shown in Fig. 7.12. In this algorithm the solution
feasibility in contingent cases and adequacy of installed devices in system’s con-
tingencies are evaluated to consider the cost of ENS properly in the planning
algorithm.
7 Battery Energy Storage Planning 207

The capability of storage to reduce the ENS, influences both the optimal siting
and sizing of battery. The ENS cost is a function of failure rate and load type. The
storage is preferred to be placed at the end of long feeders where the outage is more
probable compared with the other locations. On the other hand, the outage cost in
commercial and industrial nodes is more than that in residential regions. Hence the
storage is suggested to be installed near the vital loads. Consequently the final
decision should be taken considering all the effective factors.
When the islanding mode operation is taken into account in addition to the other
objectives e.g. peak shaving and voltage regulation, the optimal capacity of battery
increases and the cost function is reduced [17]. As a result, the application of BES is
justified and it becomes more beneficial. It should be noted that using BES in
grid-connected systems is not appropriate from economic perspectives, if it is
employed only for reliability enhancement in islanding mode of operation [27].
When the reliability enhancement is considered as an objective beside the other
objectives e.g. peak shaving, the optimal power flow of the battery is changed as
well. Consider the following scenarios:
• scenario-A: Charge/discharge battery optimally only for peak shaving; and,
• scenario-B: Charge/discharge battery optimally not only for peak shaving, but
also for reliability improvement.
In scenario-A the battery is charged/discharged to minimize the cost of pur-
chased energy and power loss. In scenario-B, it is tried not only to minimize the
cost of purchased energy and power loss, but also to increase the average SOC as
much as possible. In the latter strategy, more back-up energy is available to support
the grid against the probable outages. For example, Fig. 7.13 represents the average
SOC of a typical battery in two scenarios.

Fig. 7.13 A typical charge/discharge power and SOC of a battery in two scenarios
208 M. Sedghi et al.

As can be seen, the battery is charged faster in scenario-B, and hence, the surface
below the SOC profile in scenario-B is more than that in scenario-A. Therefore
more back-up energy is available in scenario-B.

7.3.7 Techno-economic Factors

There are some important factors that influence the optimal planning of BES units
in grid-connected systems. These factors are summarized as follow:
• Installation cost;
• Durability;
• Efficiency;
• Electricity price; and,
• Environmental impact.
The installation cost is the most important factor that determines the usefulness
of the grid-scale batteries. It is an uncertain parameter in real conditions and also in
near future. Unlike the uncertain conditions of the other effective components e.g.
the power of renewable-based DGs, the uncertainty of the investment cost reduces
the optimal capacity of batteries [9]. In other words, the uncertain economic con-
ditions challenge the justification of BES. The durability of battery is usually
corresponded to the investment cost, as it affects the replacement cost. Development
of battery technology results in more durability of the energy storage. So the
penetration of batteries increases in the grid-connected systems as well as the
stand-alone electric systems.
The storage efficiency is the overall efficiency of the battery and converter. It is
an important parameter which determines the power loss of energy storage.
However, the impact of storage efficiency on optimal planning is less significant
than the installation cost and the durability effects [9].
The electricity price is a very important parameter that makes the BES applicable
or inapplicable from economic point of view. In order to characterize the impact of
electricity price, two parameters are introduced as follow:
• Price factor (PRF); and,
• Average price P .
The price factor and the average price are defined in Eqs. (7.51) and (7.52):
PL
Ce
PRF ¼ LL ð7:51Þ
Ce
PT
Ce ðtÞ
P¼ t¼1
ð7:52Þ
T1
7 Battery Energy Storage Planning 209

If the battery is used only for peak shaving, PRF and P should be large enough
to justify the storage application in distribution grid. The optimal capacity of battery
rises as the price factor and average price increase.
The environmental impact of batteries is the other factor that influences the
penetration of BES in distribution networks. The grid-scale batteries become more
popular if they address the environmental concerns.
The impact of effective components on BES are summarized in Table 7.1.

Table 7.1 The impact of effective components on BES

Effective component/factor Influence on BES Description
Increasing conventional DGs Reducing the Dispatchable DGs are competitors for
optimal BES
penetration of
BES
Increasing renewable-based Increasing the BES makes the intermittent power of
DGs optimal DGs dispatchable to support the grid
penetration of
BES
Increasing Uncoordinated Increasing the BES is required seriously
PEVs G2V optimal
penetration penetration of
BES
Coordinated Reducing the Coordinated G2V reduces the light-load
G2V optimal times
penetration of
BES
Smart V2G Reducing the Smart V2G is a serious competitors for
optimal BES
penetration of
BES
Increasing Increasing the More capacity of BES is required in
uncertainty of optimal uncertain technical conditions
PEVs behavior penetration of
BES
On-load tap changer Increasing the Being more important in reverse power
availability optimal conditions
penetration of
BES
Using capacitor banks Reducing the Preventing excess investment cost
optimal
penetration of
BES
Reactive power capability Reducing the Preventing excess investment cost
optimal capacity
of BES
Islanding mode capability Also effective on the optimal location of
BES considerably
(continued)
210 M. Sedghi et al.

Table 7.1 (continued)

Effective component/factor Influence on BES Description
Increasing the
optimal capacity
of BES
Investment Installation cost Reducing the Being much more important than BES
cost increment optimal efficiency
penetration of
BES
Increasing Reducing the Indicating the uncertain economic
uncertainty optimal conditions
penetration of
BES
Durability increment Increasing the Reducing the replacement cost
optimal
penetration of
BES
Efficiency increment Reducing the Not very effective on BES penetration
power loss of compared with the other factors e.g.
BES installation cost and durability
Electricity Average price Increasing the The BES penetration increases if both of
price increment optimal them increase
Price factor penetration of
increment BES
Environmental concerns Increasing the Depending on the environmental impacts
increment optimal of BES and renewable energy
penetration of penetration growth
BES

Appendix

The notations used throughout this chapter are listed below:

f objective function
IC investment cost
OC operation cost
RC reliability cost
PF penalty factor
M a large fixed value
xb binary decision variable associated to installation of the bth battery
unit
nN number of all the nodes in distribution network
INS the installation cost related to the battery capacity
CCap
INS the installation cost related to the battery power rating
CPow
REP the replacement cost of batteries
CBatt
7 Battery Energy Storage Planning 211

CEAR the cost of purchased active/reactive power at the high voltage/

medium voltage (HV/MV) substations
T period of project
OM the O&M cost of batteries
CBatt
ST the income of the storage owner
CIN
ST the cost of the storage owner
CCOS
ST the cost of power which the storage injects to the grid
CSel
ST the cost of power that the storage absorbs from the grid
CPur
DC the cost function of the DisCo when no storage unit is installed in the
CwithoutST
network
DC the cost function of DisCo after installing the storage units
CwithST
AR the cost of purchased active/reactive power at the HV/MV substations
CE;WithoutST
without any embedded storage unit
RCwithoutST the reliability cost without storage
AR the cost of purchased active/reactive power at the HV/MV substations
CE;WithST
with embedded storage units
RCwithST the reliability cost with storage
DC the cost of purchased power from storage units
CE;ST
F the objective function of the multistage expansion planning
fs the objective function of the sth stage
N number of all the stages
xs;b binary decision variable associated to installation of the bth battery
unit at sth stage
Ts duration of the sth stage
AR the cost of purchased active/reactive power in HV/MV substations at
CE;s
sth stage
Vn;t the voltage magnitude in nth node of the network at tth hour
Vmin the allowed minimum voltage magnitude
Vmax the allowed maximum voltage magnitude
Sut the power of equipment u at tth hour
Sumin the allowed minimum power of equipment u
Sumax the allowed maximum power of equipment u
SSS
s;t
the power of the sth HV/MV substation at tth hour
SDG
d;t the generated power of dth DG unit at tth hour
SST
k;t
the charge power of the kth storage unit at tth hour
SLD
l;t the load power of the lth node at tth hour
SLOSS
t
the total power loss in distribution network at tth hour
AEQ the set of all the equipment
AN the set of all the nodes
AT the set of all the hours
AST the set of all the installed storage units
ALD the set of all the load nodes
SOCk;t the SOC of the kth storage unit at time t
212 M. Sedghi et al.

PST
k;t
the active power of the kth storage unit at time t
gch the charge efficiency of the storage units
gdis the discharge efficiency of the storage units
pðvÞ the probability of the vth violation
nVIO the number of all probable violations in a candidate solution
SCap ðnÞ the battery capacity in nth node
PRat ðnÞ the power rating of storage in nth node
SOCðt; nÞ the SOC of battery in nth node at time t
Pðt; nÞ the active power of the battery in nth node at time t
Qðt; nÞ the reactive power of the battery in nth node at time t
Pp the active power of pth node
Qp the reactive power of pth node
Vp the voltage magnitude in pth node
dpq the angle between the voltages of the nodes p and q
Gpq the real elements of the grid’s admittance matrix
Bpq the imaginary elements of the grid’s admittance matrix
Pi the corresponding probability of xi
dð:Þ Dirac’s delta function
lX the mean value of X
rX the standard deviation of X
kX the skewness of X
Ist;re the real component of Ist
Ist;im the imaginary component of Ist
SR the apparent power of storage
PðtÞ the active power of storage at time t
QðtÞ reactive power of storage at time t
PL the average electricity price in peak-load period
Ce
LL the average electricity price in light-load times
Ce
Ce ðtÞ the electricity price at tth hour
T1 the number of hours over one year

References

1. P. Poonpun, W.T. Jewell, Analysis of the cost per kilowatt hour to store electricity. IEEE
Trans. Energy Conv. 23(2), 529–534 (2008)
2. E. Naderi, I. Kiaei, M.R. Haghifam, NaS technology allocation for improving reliability of
DG-enhanced distribution networks, in Proceedings of IEEE International Conference on
Probabilistic Methods Applied to Power Systems, Singapore (2010), pp. 148–153
3. M.N. Kabir, Y. Mishra, G. Ledwich, Z. Xu, R.C. Bansal, Improving voltage profile of
residential distribution systems using rooftop PVs and battery energy storage systems. Appl.
Energy 134, 290–300 (2014)
4. B. Zakeri, S. Sanna, Electrical energy storage systems: a comparative life cycle cost analysis.
Renew. Sustain. Energy Rev. 42, 569–596 (2015)
7 Battery Energy Storage Planning 213

5. X. Luo, J. Wang, M. Dooner, J. Clarke, Overview of current development in electrical energy

storage technologies and the application potential in power system operation. Appl. Energy
137, 511–536 (2015)
6. Technology Roadmap Energy storage, [Online] Available: https://www.iea.org/publications
7. M. Daghi, M. Sedghi, M. Aliakbar-Golkar, Optimal battery planning in grid connected
distributed generation systems considering different technologies, in Proceedings of 20th
Iranian Electrical Power Distribution Conference, Zahedan, Iran (2015), pp. 138–142
8. A. Ahmadian, M. Sedghi, M. Aliakbar-Golkar, Fuzzy load modeling of plug-in electric
vehicles for optimal storage and DG planning in active distribution network. IEEE Trans.
Veh. Technol. 66(5), 3622–3631 (2017)
9. M. Daghi, M. Sedghi, A. Ahmadian, M. Aliakbar-Golkar, Factor analysis based optimal
storage planning in active distribution network considering different battery technologies.
Appl. Energy 183, 456–469 (2016)
10. R. Anindita, S.B. Kedare, S. Bandyopadhyay, Optimum sizing of wind-battery systems
incorporating resource uncertainty. Appl. Energy 87, 2712–2727 (2010)
11. O. Ekren, B.Y. Ekren, Size optimization of a PV/wind hybrid energy conversion system with
battery storage using simulated annealing. Appl. Energy 87, 92–98 (2010)
12. J.M. Lujano-Rojas, R. Dufo-Lopez, J.L. Bernal-Agustin, Optimal sizing of small wind/battery
systems considering the DC bus voltage stability effect on energy capture, wind speed
variability, and load uncertainty. Appl. Energy 93, 404–412 (2012)
13. V. Carpentiero, R. Langella, A. Testa, Hybrid wind-diesel stand-alone system sizing
accounting for component expected life and fuel price uncertainty. Electr. Power Syst. Res.
88, 69–77 (2012)
14. J. Wang, F. Yang, Optimal capacity allocation of standalone wind/solar/battery hybrid power
system based on improved particle swarm optimization algorithm. IET Renew. Power Gener.
7(5), 443–448 (2013)
15. Y.M. Atwa, E.F. El-Saadany, Optimal allocation of ESS in distribution systems with a high
penetration of wind energy. IEEE Trans. Power Syst. 25(4), 1815–1822 (2010)
16. J. Tant, F. Geth, D. Six, P. Tant, J. Driesen, Multiobjective battery storage to improve PV
integration in residential distribution grids. IEEE Trans. Sustain. Energy 4(1), 182–191 (2013)
17. M. Sedghi, A. Ahmadian, M. Aliakbar-Golkar, Optimal storage planning in active distribution
network considering uncertainty of wind power distributed generation. IEEE Trans. Power
Syst. 31(1), 304–316 (2016)
18. M.A. Abido, Multiobjective particle swarm optimization for environmental/economic
dispatch problem. Elect. Power Syst. Res. 79(7), 1105–1113 (2009)
19. M. Sedghi, M. Aliakbar-Golkar, Analysis and comparison of load flow methods for
distribution networks considering distributed generation. Int. J. Smart Elect. Eng. 1(1), 27–32
(2012)
20. P. Chen, Z. Chen, B. Bak-Jensen, Probabilistic load flow: a review, in Proceedings of 3rd
International Electric Utility Deregulation and Restructuring and Power technology
conference, Nanjing, China (2008), pp. 1586–1591
21. S. Chun-Lien, Probabilistic load-flow computation using point estimate method. IEEE Trans.
Power Syst. 20(4), 1843–1851 (2005)
22. G. Verbic, C.A. Canizares, Probabilistic optimal power flow in electricity markets based on a
two-point estimate method. IEEE Trans. Power Syst. 21(4), 1883–1893 (2006)
23. T. Williams, C. Crawford, Probabilistic load flow modeling comparing maximum entropy and
Gram-Charlier probability density function reconstructions. IEEE Trans. Power Syst. 28(1),
272–280 (2013)
24. Y. Deng, X. Ren, Fuzzy modeling of capacitor switching for radial distribution systems, in
Proceedings of IEEE Power Engineering Society Winter Meeting, Columbus, OH, USA
(2001), pp. 830–834
25. J. Hao, L. Shi, G. Xu, Y. Xie, Study on the fuzzy AC power flow model, in Proceedings of
5th World Congress Intelligent Control and Automation, Hangzhou, China (2004), pp. 5092–
5096
214 M. Sedghi et al.

26. W. Ouyang, H. Cheng, X. Zhang, L. Yao, Distribution network planning method considering
distributed generation for peak cutting. Energy Convers. Manage. 51(12), 2394–2401 (2010)
27. M. Sedghi, M. Aliakbar-Golkar, M.R. Haghifam, Distribution network expansion considering
distributed generation and storage units using modified PSO algorithm. Elect. Power Energy
Syst. 52, 221–230 (2013)
28. M. Sedghi, M. Aliakbar-Golkar, Optimal storage scheduling in distribution network
considering fuzzy model of PEVs, in Proceedings of 18th Conference on Electric Power
Distribution 30 Apr–1 May 2013, pp. 1–6
29. D. Suchitra, R. Jegatheesan, M. Umamaheswara Reddy, T.J. Deepika, Optimal sizing for
stand-alone hybrid PV-wind power supply system using PSO, in Proceedings of International
Conference on Swarm, Evolutionary and Memetic Computing (2013), pp. 617–629
30. M. Ghofrani, A. Arabali, M. Etezadi-Amoli, M.S. Fadali, A framework for optimal placement
of energy storage units within a power system with high wind penetration. IEEE Trans.
Sustain. Energy 4(2), 434–442 (2013)
31. M. Ghofrani, A. Arabali, M. Etezadi-Amoli, M.S. Fadali, Energy storage application for
performance enhancement of wind integration. IEEE Trans. Power Syst. 28(4), 4803–4811
(2013)
32. H. Kihara, A. Yokoyama, K.M. Liyanage, H. Sakuma, Optimal placement and control of
BESS for a distribution system integrated with PV systems. Int. Council Elect. Eng. 1(3),
298–303 (2011)
33. M. Sedghi, Optimal Battery Planning in Active Distribution Networks Considering Plug-in
Electric Vehicles Uncertainty, Ph.D. thesis, K. N. Toosi University of Technonlgy, Tehran,
Iran, 2015
34. A. Ahmadian, M. Sedghi, M. Aliakbar-Golkar, A. Elkamel, M. Fowler, Optimal probabilistic
based storage planning in tap-changer equipped distribution network including PEVs,
capacitor banks and WDGs: a case study for Iran. Energy 112, 984–997 (2016)
35. O. Anuta, N. Wade, J. McWilliams, Coordinated operation of energy storage and on-load tap
changer on a UK 11 kV distribution network, in Proceedings of 22nd International
Conference on Electricity Distribution (CIRED), Stockholm, Sweden, June 2013, pp. 1–4
36. A. Gabash, P. Li, Flexible optimal operation of battery storage systems for energy supply
networks. IEEE Trans. Power Syst. 28(3), 2788–2797 (2013)
Chapter 8
Optimal Distributed Generation
Placement Problem for Power
and Energy Loss Minimization

Aggelos S. Bouhouras, Paschalis A. Gkaidatzis

and Dimitris P. Labridis

Abstract This chapter introduces the Optimal Distributed Generation Placement

problem towards power and energy loss minimization. Several solving methods are
applied in order for the most suitable to emerge. Apart from technical and DG
constraints, recent raised issues due to high Distributed Generation penetration like
the reverse power flow effect is considered as well. The load and generation
variability and their impact in integrating Renewable Energy Sources are examined,
aided by the use of Capacity Factors implementation. In addition, the impact of
Optimal Distributed Generation Placement problem in conjunction with Network
Reconfiguration and Optimal Energy Storage Systems Placement is introduced
aiming to examine how joined management schemes could be efficiently combined
in order to maximize the potential loss and energy reduction.

Keywords ODGP Loss minimization Distributed generation




Optimization Heuristics Reverse power flow Capacity factors



Load/generation variability ESS Network reconfiguration

A. S. Bouhouras (&)
P. A. Gkaidatzis
D. P. Labridis

School of Electrical and Computer Engineering,
Aristotle University of Thessaloniki, AUTH, Thessaloniki, Greece
e-mail: [email protected]; [email protected]
P. A. Gkaidatzis
e-mail: [email protected]
D. P. Labridis
e-mail: [email protected]
A. S. Bouhouras
Department of Electrical Engineering, University of Applied Sciences,
TEIWM, Kozani, Greece

© Springer Nature Singapore Pte Ltd. 2018 215

F. Shahnia et al. (eds.), Electric Distribution Network Planning,
Power Systems, https://doi.org/10.1007/978-981-10-7056-3_8
216 A. S. Bouhouras et al.

8.1 Introduction

The penetration of Distributed Generation (DG) units in Electric Distribution

Networks has been considered as an efficient way to exploit the benefits of sus-
tainable energy, promoted by distributed energy resources. In most cases, appro-
priate consideration of DG installation can highly benefit the electric distribution
networks in terms of loss reduction, voltage-profile, and reliability improvement [1,
2]. However, high penetration of DG units could potentially cause problems to
several operational characteristics, especially due to reverse power flow, leading to
excessive losses and feeder overloading [3, 4]. Regarding DG placement, the final
decision lies on the owners and investors, depending on site and fuel availability or
climatic conditions. Notwithstanding the merits of installing DG and exploiting it in
order to solve networks problems, the fact remains that, in most cases, the
Distribution Network Operator (DNO) has neither significant control, nor influence
over the DG location and size. Still, DG placement affects critically the operation of
a network, below a certain limit. Thus, optimization tools which provide both
optimal locations and capacity of DG units to be installed should be highly
appreciated by DNOs. The Optimal Distributed Generation Placement (ODGP)
problem generally deals with the determination of the location and appropriate
sizing of DG units to be installed into existing electric distribution networks, subject
to networks’ and DG operational, as well as investment constraints.
In this chapter, a comparative analysis of several promising methods is initially
presented, such as analytical or heuristic ones and their merits and drawbacks are
pointed out, when contemplating power loss reduction via ODGP approach.
The DG units are considered capable of both active and reactive power production.
Secondly, using the most suitable of them, the ODGP towards power loss reduction
is solved by taking into consideration a possible reverse power flow effect [5, 6].
Thirdly, as a first step for the integration of an optimal combination of Renewable
Energy Resources (RESs) into an electric distribution network, a method is
demonstrated, considering concurrently the geographical characteristics of the area,
where the examined network is placed, the different weather conditions and the
availability of RESs, by the introduction of Capacity Factors (CFs), while trying to
keep problem complexity at a minimum.
Furthermore, the ODGP towards energy loss reduction is coped, initially taking
into account the impact of load composition variation while considering the DG
units having constant power output, and then with variable power output resem-
bling the function of several RESs such as Wind Turbines, or Photovoltaics.
Finally, the cooperation of ODGP with Network Reconfiguration (NR) towards
power loss reduction is presented and an initial effort regarding the cooperation of
ODGP with the Optimal Energy Storage System Placement (OESSP) problem.
8 Optimal Distributed Generation Placement Problem … 217

8.2 ODGP Towards Power Loss Minimization—Problem

Formulation

8.2.1 Objective Function—Constraints

The ODGP problem is a mixed-integer-non-linear-constrained (MINLC) opti-

mization problem; mixed integer because both the power of the DGs installed
(sizing) and their position (siting) are requisites; non-linear, due to the power flow
equations needed to solve the problem. As an optimization problem, various
objectives can be found in literature, such as cost minimization, benefit maxi-
mization, greenhouse gas emission reduction, either solved individually, or as a
multi-objective approach [7–9]. In this section, power loss minimization is to be
contemplated, formulated as

X
nl h  i
Floss ¼ min gi;j Vi2 þ Vj2  2Vi Vj cos hi  hj ð8:1Þ
k¼1

where:
gi;j is the conductance between buses i and j, respectively,
nl is the total number of branches of the network,
Vi , Vj are the voltage magnitudes of buses i and j, respectively, and
hi , hj are the voltage angles of buses i and j, respectively.
The constraints of the problem can be separated to obligatory and occasional. As
obligatory constraints, the power flow Eqs. (8.2a), (8.2b) and the technical con-
straints of the electric distribution network (8.3), (8.4) are considered, as they must
always be met. They are expressed as:
Power Flow Constraints:

X
nb     
PG;i  PD;i  jVi jVj Yi;j  cos ui;j  hi þ hj ¼ 0 ð8:2aÞ
j¼1

X
nb     
QG;i  QD;i þ jVi jVj Yi;j  sin ui;j  hi þ hj ¼ 0 ð8:2bÞ
j¼1

DN Contraints:

Vimin  Vi  Vimax ð8:3Þ

Sk  Smax
k ð8:4Þ
218 A. S. Bouhouras et al.

where:
PG;i , QG;i is the active and reactive power generation at bus i, respectively,
PD;i , QD;i is the active and reactive power demand at bus i, respectively,
nb is the total number of buses of the network,
Yi;j is the magnitude of bus admittance element i,j,
ui;j is the angle of bus admittance element i,j
Vimin , Vimax are the voltage lower and upper limits of bus i, respectively, and
Smax
k is the thermal limit of line k, by terms of apparent power.
As occasional constraints, technical constraints regarding DG units and/or their
penetration level are considered. They are classified as occasional because they can
be present on occasion, defined by the aspect of the problem examined and not
necessarily mandatory, as the previous ones. They can be expressed as:
DG constraints:

min  Sm  Smax
SDG ð8:5Þ
DG DG

DG
pfmin  pfmDG  pfmax
DG
ð8:6Þ

Penetration Constraints:

X
nDG

m  g
STotal
SDG ð8:7Þ
Load

m¼1

where:
SDG
m is the power of a DG unit,
pfmDG is the power factor of a DG unit,
SDG DG
min , Smax are the limits of power for a DG unit, respectively,
DG DG are the limits of power factor of a DG unit, respectively,
pfmin , pfmax
nDG is the total number of DG units to be installed,
g is a percentage indicating the desired DG penetration level, and
SLoad
Total
is the total load installed in the DN.

8.2.2 Penalty Function—Terms

In general, constrained problems can be solved using deterministic, or stochastic

algorithms. However, deterministic approaches such as feasible direction and
generalized gradient descent, require strong mathematical properties of the objec-
tive function, such as continuity and differentiability. Moreover, solving the ODGP
problem by analytical methods could prove to be complex and time-consuming
[10], or be restrained to solutions including only one DG unit. In cases, where these
8 Optimal Distributed Generation Placement Problem … 219

properties are absent, evolutionary computation offers reliable alternative methods.

Since most evolutionary approaches were primarily designed to address uncon-
strained problems, constrained handling techniques are usually required to detect
only feasible solutions. The most common of those techniques is the use of a
penalty function. In spite of its drawbacks, it performs rather efficiently, provided a
proper calibration of the penalty parameters is undertaken [11, 12]. According to
this approach, the constraints expressed via penalty terms are incorporated into the
objective function in order to formulate the penalty function that penalizes any
infeasible solutions as:

Pð xÞ ¼ f ð xÞ þ Xð xÞ ð8:8Þ
n o
Xð xÞ ¼ q g2 ð xÞ þ ½maxð0; hð xÞÞ2 ð8:9Þ

where:
Pð xÞ is the Penalty function,
f ð xÞ is the objective function, in this case the Floss , as expressed in (8.1),
X ð xÞ is the penalty term,
q is the penalty factor,
gð x Þ refers to the equality constraints, in this case as defined in (8.2a), (8.2b), and
hð x Þ refers to the inequality constraints, in this case as defined in (8.3)–(8.7).
Thus, in case of the ODGP problem and using, for the sake of argument, only the
obligatory constraints, the updated Penalty Function could be expressed as:

Pð xÞ ¼ minðFloss þ XP þ XQ þ XV þ XL Þ ð8:10Þ

where XP and XQ refer to the equality constraints of

( )
X
nb X
nb     
XP ¼ qP PG;i  PD;i  jVi jVj Yi;j  cos ui;j  hi þ hj ð8:11aÞ
i¼1 j¼1

( )
X
nb X
nb     
XQ ¼ qQ QG;i  QD;i þ    
jVi j Vj Yi;j sin ui;j  hi þ hj ð8:11bÞ
i¼1 j¼1

and XV and XL to inequality constraints of

X
nb
  2 X
nb
  2
XV ¼ qV max 0; Vimin  Vi þ qV max 0; Vi  Vimax ð8:12Þ
i¼1 i¼1

X
nl   2
XL ¼ qL max 0; Sk  Smax
k ð8:13Þ
k¼1
220 A. S. Bouhouras et al.

As can be easily deduced, any other constraints such as (8.5), (8.6) or (8.7) can
be incorporated in (8.10) via the same process.

8.3 ODGP Towards Power Loss Minimization—Solving

Methods

According to a literature survey on the subject, a great amount of scientific research

has been undertaken with the aim of solving the ODGP problem [13]. Several
promising methods have emerged such as analytical ones [14–17], heuristics [5, 6,
18–33], or combination of the above, solving siting and sizing individually, but in a
sequential order [34–36].
However, as stated earlier, ODGP is a MINLC optimization problem. The
conventional approaches utilizing analytical methods could be intricate and
time-consuming in this case, or restricted to solving for just one DG unit being
placed. Therefore, over the last few decades, Heuristics such as Particle Swarm
Optimization (PSO) [5, 6, 18], Genetic Algorithm (GA) [9, 19, 20], Artificial Bee
Colony (ABC) [21–23], Cuckoo Search (CS) [24–27], and Harmony Search
(HS) [28–30] have been implemented. Moreover, they have proved quite promising
and still evolving in this field. Some additional mentions could be, for example,
Bacterial Foraging Optimization Algorithm (BFOA) [31], Ant-Lion Optimization
(ALO) [32], Grey Wolf Optimization (GWO) [33], and many more [13], noting
more advancement in solving the ODGP problem. In this section, a small com-
parative analysis will take place in order to determine the most suitable method to
solve the ODGP problem. Three versions of PSO, namely, the Local, Global and
Unified PSO, GA, ABC, CS and HS methods is compared and evaluated. As
analytical methods, the ones presented in [15], namely, Improved Analytical
(IA) method, Loss Sensitivity Factor (LSF) method and Exhaustive Load Flow
(ELF) method is also demonstrated.

8.3.1 Analytical Methods

For calculating the losses for those methods, instead of (8.1) the exact loss formula
is utilized, expressed as:
nb X
X nb
    
Floss ¼ aij Pi Pj þ Qi Qj þ bij Qi Pj  Pi Qj ð8:14Þ
i¼1 j¼1
8 Optimal Distributed Generation Placement Problem … 221

where:
rij   rij  
aij ¼ cos hi  hj ; bij ¼ sin hi  hj ð8:15a; bÞ
Vi Vj Vi Vj

where:
rij þ jxij ¼ Zij is the ijth element of the impedance matrix,
Pi and Pj are the active power injections at ith and jth buses, respectively, and
Qi and Qj are the reactive power injections at ith and jth buses, respectively.

8.3.1.1 IA Method

In IA different formulas are formed according to the DG type to be used, i.e.

injecting only active, and/or reactive, or both. The advantage of the method is that
load flow is required only twice: once at the initial state of the electric distribution
network and once the DG is in place. The drawback is that only a single DG unit is
placed at a time.

8.3.1.2 LSF Method

LSF is based on the linearization of the power flow Eqs. (8.2a), (8.2b). It is most
appropriate for locating the most suitable buses to host DG units by ranking them
according to their LSF values. Then, a DG unit is placed at the bus with the highest
priority and its size is calculated by increasing it in small steps and running load
flow. The merits of the method are its simplicity and directness. However, as in IA
method, only a single DG unit is placed at a time and naturally after the first time,
the solution is biased since some DG units have been already installed.

8.3.1.3 ELF Method

ELF method, also known as repeated load flow solution, requires excessive com-
putational time since all buses are considered in calculation; however, it can lead to
a completely optimal solution. Also, as the number of DG units to be installed
increases to more than one, so does the computational load and indeed does so in an
exponential rate.
222 A. S. Bouhouras et al.

8.3.2 Heuristic Methods

8.3.2.1 GPSO, LPSO, UPSO Methods

PSO was introduced by Eberhart and Kennedy [37]. It was inspired by the social
behavior of bird flocking. A swarm of particles is assigned to explore the solution
space in order to retrieve the optimal solution. Their movement in the solution space
is defined by three key elements:
1. their personal knowledge of the solution space, represented by the Personal Best
parameter,
2. the social knowledge gained by exchanging information among a group of
particles, represented by the Social Best parameter, and finally,
3. their current movement on the solution space, represented by the previous
gained velocity.
Regarding the Social Best, when the information exchange takes place amongst
all particles within the swarm, it is called Global Best, and the respective algorithm
Global PSO (GPSO), whereas if it takes place among smaller formations, called
neighborhoods, it is called Local Best and the respective algorithm Local PSO
(LPSO).
Concerning GPSO, because the particles are instantly aware of the swarm’s best
position, rapid convergence is achieved, therefore better exploitation of the
knowledge gathered regarding the solution space. However, this happens at the
expense of exploration of the solution space, thus resulting in probable local
minima entrapment and therefore not achieving a near-optimal solution.
In contrast, in LPSO the formation of overlapping particle neighborhoods and
the information exchange within them enables for better exploration of the solution
space [38]. However, this happens at the expense of exploitation, thus longer
convergence, since the information exchange is distilled among the various
neighborhoods, instead of the whole swarm.
Therefore, under GPSO, or LPSO, the algorithm is biased towards exploitation,
or exploration, respectively. UPSO, introduced by Parsopoulos and Vrahatis [39],
has been developed as an attempt to harness their merits and, at the same time,
aiming to neutralize their flaws. In this chapter, the UPSO’s Swarm Partitioning
scheme is applied for merging the two versions of PSO, as the most promising [40].
A generic flowchart for PSO is presented in Fig. 8.1.

8.3.2.2 GA Method

GA was introduced by Holland [41]. A simulation of the three fundamental genetic

processes comprises the technique, namely selection, crossover and mutation.
A group of chromosomes is designated in the solution space, here considered a
genetic pool. The most fitted are selected as parents to form the next generation.
8 Optimal Distributed Generation Placement Problem … 223

Fig. 8.1 PSO flowchart Initial

Population

Evaluate
Personal Best

Evaluate
Social Best
Calculate
new Velocity

Calculate
new position

Termination
Criteria No

Yes

Solution Set

Fig. 8.2 GA flowchart Initial

Population

Evaluate

Selection

Crossover

Mutation

Termination
Criteria No

Yes

Solution Set

In this chapter, the roulette-wheel selection scheme is applied. The parents are
stochastically combined to breed offspring that bear combinations of their chro-
mosomes. In addition, a mutation process takes place, where stochastically several
parts of the offspring’s chromosomes are altered. Finally, the best among both
parents and offspring are chosen to constitute the next generation of chromosomes,
as presented in Fig. 8.2.
224 A. S. Bouhouras et al.

8.3.2.3 ABC Method

ABC method was proposed by Karaboga [42]. It was inspired by the intelligent way
bee swarms locate and harness their food. The candidate solution space in that case
is represented by places of potential food sources. The bee colony, divided into
employed, onlooker and scout bees, spreads across it. Employed bees target and
exploit potential food position and inform the onlookers for more potential food
sites. The employed bees then are trying to determine the food potential of those
positions. If an employed bee’s position does not represent a good solution, then the
bee turns into a scout and starts exploring the solution space. The number of the
employed bees is equal to the number of food sources, each of which also represents
a site, being exploited at the moment or to the number of solutions in the popu-
lation, as presented in Fig. 8.3.

8.3.2.4 CS Method

CS method was first introduced by Yang and Deb [43]. It was inspired by the way
some cuckoo species lay their eggs in the nests of other host birds (of other species),
to be nurtured. Each egg in the nest represents a solution, and cuckoo eggs represent
new solutions. The aim is to use the new and potentially better solutions (cuckoos)
to replace the least suitable solutions in the nests. In each iteration, one cuckoo egg
is laid randomly in a selected nest; The nests with high quality eggs will carry over
to the nest generation; Then, in the remaining least suitable nests, a discovery

Fig. 8.3 ABC flowchart Initial

Population

Evaluate

Employed Bee
Phase
Onlooker Bee
Phase

Scout Bee
Phase

Termination
Criteria No

Yes

Solution Set
8 Optimal Distributed Generation Placement Problem … 225

Fig. 8.4 CS flowchart Initial

Population

Evaluate

Cuckoo
Selection

Host Nest
Selection

Worst/Best
Nest Selection

Termination
Criteria No

Yes

Solution Set

operation takes place by the host birds, stochastically retrieving cuckoo laid eggs
and discarding them, therefore ignoring them from further calculations. A generic
flowchart is presented in Fig. 8.4.

8.3.2.5 HS

HS method, introduced in [44], is inspired by the improvisation process of jazz

musicians. Improvisation is a process of searching for the most appropriate har-
mony by trying various combinations of rhythms, under the following three rules:
1. playing any existing rhythm from the memory;
2. playing an altered rhythm from the memory;
3. playing a random rhythm from the possible range.
HS simulates this procedure as:
1. choosing any value from the HS memory;
2. choosing an altered value from the HS memory;
3. choosing a random value from the possible value range.
A generic flowchart is presented in Fig. 8.5.
226 A. S. Bouhouras et al.

Fig. 8.5 HS flowchart Initial

Memory

Evaluate

Memory
Consideration
Local Pitch
Adjustment

Random
Selection

Termination
Criteria No

Yes

Solution Set

8.3.3 Heuristic Methods Evaluation

For the evaluation of all aforementioned solution techniques, the typical 33-bus
system [45] has been employed, as depicted in Fig. 8.6. It is a radial electric
distribution network and has a total load of 3.72 MW and 2.38 MVAr, presenting
initial power loss of 211 kW. Due to their stochastic nature, the methods have been

Fig. 8.6 The 33-bus system

8 Optimal Distributed Generation Placement Problem … 227

applied 1000 times each, and within an ample time of 1000 iterations. Also, they
have been let unrestrained in terms of number of DG units, so as to deduce how
close the optimal solution they could arrive; with the actual optimal solution being
the one with DG units installed in all nodes with nominal capacity equal to the
nodes’ respective load. The installed DG units are considered capable of injecting
active power and injecting/consuming reactive power. Similar results have been
extracted from implementation on other networks, such as the typical 16, 30 and
69-bus systems.
In Table 8.1 solution-related properties for all examined heuristic techniques are
presented: the minimum loss achieved by each technique, the loss reduction per-
centage, the number of DG units installed along with the total DG installation size
in MVA, as provided by the best solution among the 1000 trials that each technique
has reached. Due to the ample time given, every technique has achieved a signif-
icant loss reduction in both systems and the differences are virtually slim, though,
GA seems to be slightly in a bit of a disadvantage, as also confirmed by Fig. 8.7,
where the mean Bus Voltage profile for all the examined heuristic techniques is
shown.
In Table 8.2 convergence related properties are presented, i.e. the average
execution time of one trial, the iteration number required for each technique to

Table 8.1 Heuristics’ solution performance comparison

Method Minimum power Power loss Total DG no. Total size of installed
loss (kW) reduction (%) DG units (MVA) P + jQ
GPSO 0.34 99.84 20 3.64 + j2.32
LPSO 0.22 99.89 20 3.55 + j2.21
UPSO 0.13 99.94 22 3.66 + j3.26
GA 10.77 94.89 21 3.00 + j0.71
ABC 0.52 99.75 17 3.73 + j2.28
CS 0.48 99.77 20 3.82 + j2.32
HS 2.67 98.74 19 3.30 + j2.08

Fig. 8.7 Average voltage

228 A. S. Bouhouras et al.

Table 8.2 Heuristics’ convergence performance comparison

Method 10% 1% 0.1% 93.22% loss Average
tolerance tolerance tolerance reduction execution time
iteration iteration iteration iteration (min)
GPSO 832 983 999 6 6.7
LPSO 845 983 999 7 6.9
UPSO 339 839 900 5 6.9
GA 747 975 999 900 7.3
ABC 846 987 999 19 13.7
CS 909 992 999 43 12.5
HS 684 945 996 9 3.8

Fig. 8.8 Average

convergence

reach in average within 10, 1 and 0.1% tolerance of its final optimal solution,
respectively, e.g. for UPSO given that its optimal solution is 0.133 kW, the 10%
tolerance is 0.146 kW power loss.
In terms of execution time, evidently HS proves to be the fastest with less than
four minutes execution time. Moreover, the iteration number needed for each
technique to reach a certain amount of loss reduction is also presented in Table 8.2.
It is set to 93.22%, regarding the average loss reduction achieved by the least
performing technique, namely being GA. Although all the techniques perform
rather well, it seems, the PSO versions, and especially UPSO, performs better than
the rest, regarding convergence and iteration steps, reaching their final solution in
the least amount of iterations. Therefore, although UPSO is not as efficient as HS in
terms of execution time, it can be argued that it can be applied for less iterations,
thus overcoming this drawback.
This is illustrated in Fig. 8.8, where each technique’s average convergence of the
1000 trials is presented. This is also confirmed by Fig. 8.9, where again each
technique’s average convergence of the 1000 trials is presented, but in a margin of
8 Optimal Distributed Generation Placement Problem … 229

Fig. 8.9 Average

convergence zoom-in

Fig. 8.10 Convergence’s

deviation

less than 1000 iterations, and specifically, within the 10% iteration tolerance of the
best performing technique, being UPSO.
In addition, as shown in Fig. 8.10, the PSO versions, and especially UPSO, have
the lowest convergence of standard deviation along 1000 iterations, meaning that
their 1000 trials do not deviate far from each other, ensuring the robustness of their
solution process and even that less trials are possible.

8.3.4 Heuristic Versus Analytical Methods Evaluation

For a more direct evaluation comparison of the most prominent Heuristic technique,
i.e. UPSO, with the analytical methods presented in this section (IA, LSF, ELF)
again the typical 33-bus system is employed. Three DG units are considered for
installation and capable of injecting only active power. In Table 8.3 the solutions
reached by the four methods are presented.
Based on the results of the previous section, UPSO has been applied 50 times
and with 400 iterations. As can be deduced, UPSO performs rather better than the
analytical ones, in terms of optimal solution, but rather poorly in terms of execution
time. However, as evidenced in the precious section, a Heuristic method is able to
230 A. S. Bouhouras et al.

Table 8.3 Heuristics versus analytical methods solution comparison

Method Minimum power Power loss DG DG size Total DG Time
loss (kW) reduction (%) position (kW) installed (MW) (s)
UPSO 77.9 65.50 13 802 2.95 70
24 1092
30 1054
IA 81.05 61.62 6 900 2.52 0.4
12 900
31 720
LSF 85.07 59.72 18 720 2.43 0.23
25 900
33 810
ELF 74.27 64.83 13 900 2.7 3.06
24 900
30 900

perform with the same efficiency, regardless of the considered number of DGs for
installation, whereas the analytical ones would be either restricted to a small
number of DGs, as in this case, or be biased since installing one DG unit at a time
alters the electric distribution network every time. Moreover, since, the ODGP
problem addresses primarily network planning and operational issues, it can be
argued that time is not considered as important as finding the optimal solution,
resulting in giving priority to the latter.
In conclusion, it is indicated that when contemplating ODGP towards power loss
with a small amount of DG units to be installed, an analytical method might prove a
better option than a heuristic one, in terms of time with only a minor setback in
terms of optimal solution. When, an optimal or near optimal solution is required and
more DG units should be considered for installation, a heuristic method would
prove more suitable.

8.4 ODGP Towards Power Loss Minimization—Reverse

Power Flow

Integration of DG in existing electric distribution networks has been discussed and

studied thoroughly during the last years as a measure of reducing grid’s power loss.
However, the possible impacts of Reverse Power Flow (RPF), caused by extended
DG penetration, on solving the ODGP have not been fully considered.
While reaching optimal solutions for the ODGP problem, recent and forth-
coming massive DG integration brings to light RPF considerations, i.e. power flow
pushed upstream of the network and on neighbouring networks. So far, literature
solves the ODGP problem, towards different optimization functions, either with-
out considering possible RPF to adjacent grids, or by simply not allowing it.
8 Optimal Distributed Generation Placement Problem … 231

However, this solution approach could be proved inadequate; on the one hand, if
RPF is ignored, unfair power displacement to neighbouring grids may occur, or not
recognised; on the other hand, if it is strictly prohibited, it might lead to biased and
sub-optimal solutions, since there are indications that when RPF is included during
the planning process, it could lead to different ODGP solutions that can further
reduce power loss [46–49].
In this section, it is shown that the RPF effect can be integrated into the ODGP
problem as an occasional constraint imposed on the Slack Bus itself. As an alter-
native, an intermediate bus between the Slack Bus and the rest of the electric
distribution network might be inserted, and imposing the constraint on the total
power that flows through it, via its adjacent branches, modifying the network
slightly [50]. The constraint and the corresponding penalty term can be expressed as

Pperm  g%
RPF
Pinit ð8:16Þ
h   i2
XRPF ¼ qRPF max 0; Pperm   g%
RPF
jP init j ð8:17Þ

where:
Pinit is the initial power flowing through the Slack to the network
Pperm is the permitted power flowing through the Slack Bus to/from the network
g%
RPF
is the percentage of the allowed RPF, with respect to the initial Slack Bus
flowing power.
Results from implementation on the typical 30 and 33-bus systems [51], a radial
and a meshed electric distribution network, respectively, are shown in Figs. 8.11,
8.12, 8.13 and 8.14. Power loss reduction is the objective function, while gradually
increasing RPF percentages are considered, and therefore the total permitted DG
capacity to be installed is accomplished. A total number of seven DG units capable
only of active power was considered for both examined networks. Furthermore, it
should be stressed that the 30-bus system has already a 100% DG penetration with
respect to installed load, whereas the 33-bus system has none. Overall, increasing
the RPF results in reduced power loss savings, as can be deduced by Figs. 8.11

Fig. 8.11 RPF impact on

loss reduction (%) in the
33-bus system 0
Loss Reduction (%)

-10
-20
-30
-40
-50
-60
-70
-80
Increasing RPF Levels
232 A. S. Bouhouras et al.

Fig. 8.12 RPF impact on

loss reduction (%) in the
30-bus system
0
-5

Loss Reduction (%)

-10
-15
-20
-25
-30
-35
-40
-45
-50
Increasing RPF Levels

Fig. 8.13 Installed DG 16

capacity in the 33-bus system Installed Capacity of DG 14
12
10
(MW)

8
6
4
2
0

Increasing RPF Levels

Fig. 8.14 Installed DG

Installed Capacity of DG

100
capacity in the 30-bus system 90
80
70
(MW)

60
50
40
30
20
10
0

Increasing RPF Levels

and 8.12. The RPF ranges from 0% (no RPF allowed) up to equal to 250% of the
initial downstream power flow. However, it is also deducted that regardless of the
network’s topology (radial or meshed) and with RPF percentage from 25 to 50% of
the initial downstream power flow, an even better loss reduction is achieved, when
compared to the one with RPF 0%. Moreover, for these RPF percentages the total
DG installed reaches over 100% penetration in both networks, as shown in
8 Optimal Distributed Generation Placement Problem … 233

Figs. 8.13 and 8.14; in the 33-bus system, for the best loss reduction for 25% RPF a
total over 4 MW DG is installed, whereas its total installed load is 3.72 MW and in
the 30-bus system, a nearly additional 40 MW DG is installed, in spite of already
having achieved 100% DG penetration.
In conclusion, RPF existence up to a certain level will not necessarily affect
negatively the ODGP, when considering power loss reduction. Additionally, it
might lead to solutions with greater loss reduction and DG penetration over than
100%. This could benefit other operational aspects of the network, e.g. reliability
improvement and environmental benefits under the installation of Renewable
Energy Sources.

8.5 ODGP Towards Power Loss Minimization—

Renewable Energy Sources

When examining ODGP towards power loss minimization, it is rather difficult to

examine Renewable Energy Sources (RESs) directly and their installation in an
electric distribution network, while keeping problem complexity at minimum, since
their most distinctive feature, stochasticity, is dependent on time. In ODGP towards
power loss minimization only a single state or snapshot of the network is taken into
account. In examining different types of DGs, the most direct distinction from the
network’s point of view are:
• type 1: DG injecting only active power,
• type 2: DG injecting only reactive power,
• type 3: DG injecting active power and injecting/consuming reactive power.
With that in mind, apart from optimal site and size, an aspect of optimal mix of
DG types can be added in the problem formulation. However, the question remains
if it would be possible to examine the integration of RESs in an electric distribution
network even in the current stage, without integrating on time, in other words if an
Optimal Renewable Energy Sources Placement (ORESP) problem can be con-
templated. To that end, several alternatives are offered.
More specifically, one alternative refers to solving the problem separately, i.e., to
find the optimal siting and sizing of DG units in a network, as an ordinary ODGP
problem, and then to determine the RES type, e.g. Photovoltaic or Wind Turbine
[52]. However, no mix of RESs is examined for penetration and an impartial
solution might not be achievable. If an optimal mix is required, an investigation
regarding the different impacts of DGs on power quality and reliability must be
performed; the DG penetration level could be limited by harmonic distortion
because of the nonlinear current injected by inverter-based DG units, as well as by
protection coordination constraints because of the variation in fault current caused
by synchronous-based DG units [53]. Another approach is to implement the ELFs
concept for each bus and each technology of DG [54], though either a possible
234 A. S. Bouhouras et al.

needless computational effort would take place, or the solution could just rely on
approximations. A considerable contribution at this field has been accomplished in
[55], in which the optimal mix of DGs of different technologies has been achieved,
via stochastic models of wind speed and solar irradiance. The candidate nodes for
DG installation are predefined however and thus, only their size is estimated.
Additionally, the different DG technologies could be divided according to their
power output, i.e. whether they can control active/reactive power independently
(PQ mode, or constant power factor mode), or active power and voltage (PV mode,
or variable reactive power mode) [56]. In this latter approach, a simultaneous
solution regarding number, siting and sizing is achieved, under a multi-objective
function that includes active/reactive power loss minimization and voltage profile
improvement. However, the aforementioned distinction between different DG
technologies might not be quite so accurate.
Finally, in this section, the concept of Capacity Factors is implemented [57]. The
basic issue in ORESP is the variations regarding RESs’ power output; it is related to
their technology and the natural resources’ potential and they have to be taken into
account. For example, solar irradiance, wind speed and water availability are
expected to vary among candidate nodes within an electric distribution network,
and these variations could have a significant impact on the optimal siting and sizing
of such RESs, especially when a mix of different RESs is examined. The network’s
nodes are divided into groups, representing different areas with different natural
characteristics and resources, and hence with different CFs. The values of these CFs
express the potential of the respective natural resources available in that node. Thus,
according to their positions, all RESs are assigned their CFs. These CFs are then
included as an additional occasional constraint in the problem formulation, as

X
nres
XCF ¼ qCF CFl ð8:18Þ
l¼1

where nres the number of RESs, and qCF the corresponding penalty factor.
For example, let three RESs technologies to be considered, e.g. Photovoltaic
(PV), Wind Turbine (WT) and Hydro-Plant (HP), for installation in the typical
69-bus system [58]. In order to assign potential for the local natural resources at
each node, the electric distribution network in question has been divided into three
areas, as depicted in Fig. 8.15. At each area, each of the nodes is assigned a value
for the CFs of the three technologies examined, i.e. PV, WT, and HP, respectively.
Assuming that each area is relatively small, it is evident that the nodes within that
area share the same value of the CFs of their respective technologies.
In Table 8.4 a set of typical CFs values for each technology is assigned to each
area, whereas in Tables 8.5 and 8.6 the results of the proposed method are pre-
sented. The results of the ODGP implementation on the same electric distribution
network are also presented for comparison and LPSO was used in both approaches.
Five DG units were considered for installation for the ODGP problem, whereas 5
RESs units for each technology in the ORESP problem. It can be concluded that via
8 Optimal Distributed Generation Placement Problem … 235

Fig. 8.15 The 69-bus system divided into CF areas

Table 8.4 CFs values

Area RES type
PV WT HP
#1 0.10 0.00 0.42
#2 0.10 0.25 0.00
#3 0.15 0.12 0.00

Table 8.5 Overall results

Initial losses (kW) Minimum losses (kW) Loss reduction (%)
ODGP 602.2 148.4 75.357
ORESP 602.2 169.4 71.8698

this method the different geographical characteristics of an area and different

weather conditions leading to availability of RESs, can be taken into account all at
once, by the aid of CFs. Moreover, the corresponding ORESP problem can be
solved while keeping its complexity at minimum and an optimal solution, in terms
not only of siting and sizing, but also of RES type, is achievable.
236 A. S. Bouhouras et al.

Table 8.6 Detail results ODGP ORESP

Bus no. P (kW) Type Bus no. P (kW)
12 503.2 PV 20 420.3
19 376.0 61 23.0
40 718.5 WT 40 723.2
53 1718.8 45 580.8
61 29.48 53 1458.1
56 226.0
59 57.7
HP 12 283.5
Total Total P Total Total P
no. (kW) no. (kW)
5 3346.5 8 3778.2

8.6 ODGP Towards Energy Loss Minimization—

Load/Generation Variation

Although many issues can be examined in ODGP, such as power loss minimization
or reduction, reverse power flow, voltage stability and reliability improvement, the
approach remains incomplete without the time variable. If time is taken into account
though, the problem becomes more complex and time-consuming, than it already is.
Thus, a solving method able to prove the Golden Section between quick conver-
gence time and optimal solution will become more than useful, as that examined in
Sect. 8.3. Thus, the analysis of ODGP towards power loss minimization is useful
and important and also a significant step before examining ODGP towards energy
loss minimization.
Still, an electric distribution network’s load does not remain constant, but varies
over time. Furthermore, the stochasticity of RESs’ generation, and their impact on a
network cannot be examined, when only a single snapshot of the latter is
considered.
For the ODGP towards energy loss, an energy loss minimization objective
function could be
t X
X nl h  i
Feloss ¼ gi;j Vi2 þ Vj2  2Vi Vj cos hi  hj ð8:19Þ
Dt¼1 k¼1

where Dt is the time interval and t is the time period examined.

Regarding the constraints, they remain the same, with the following exception:
the single constraint value retrieved from the single snapshot’s load flow analysis in
power loss minimization approach, now in the energy loss minimization approach,
is replaced by the maximum absolute value retrieved from the time period exam-
ined t, in order to maintain the same order of magnitude in the penalty function.
8 Optimal Distributed Generation Placement Problem … 237

The impact of DG units on energy losses depends on the specific characteristics

of the network, such as demand profile, topology, as well as the relative location of
the generators and whether their output is considered constant or variable.
Incorporating these complexities into an optimization framework for energy loss
minimization is a challenge that has only been partially addressed by a few studies
[59]. In [47] the analysis regarding load and DG power output variations relies on
uniformly distributed loads while these variations refer to a typical daily pattern for
both. Moreover, only the optimal siting of DG units is examined, and one DG unit
is considered for installation. In [60] the case of one wind power unit under both
power output and load demand variations is examined. The analysis yields the
optimal node for the wind power unit installation by considering a sequential
analysis with only one candidate node for DG installation at a time and concludes
that subject to load variations, the optimal location is different when compared to
the operational snapshot. In [55] a probabilistic technique is proposed for optimally
allocating different types of DG technologies. The technique is based on generating
a probabilistic generation-load model. Beta and Rayleigh Probability Density
Functions (PDFs) are used for simulating solar irradiance and wind speed uncer-
tainty, respectively, while IEEE-RTS for the load profile. However, the positions of
the DGs are predetermined, as the number of DGs as well. Other approaches
incorporating load or DG power variations, as in [61], may provide biased solutions
since the installation nodes are predetermined. Furthermore, the analysis in [62]
concludes that the power analysis of one load snapshot is not necessarily adequate
for the overall operation of the electric distribution network in [63] a two-stage
method of optimal siting and sizing of DG units is proposed. Finally, in [64], a
method to address and evaluate the economic benefits of RESs is proposed, when
applied to networks, but the candidate buses are predetermined and the number of
DGs for each type is limited and predefined.

8.6.1 Load Variation

In electric distribution networks, the loads are highly distributed and quite variable.
Thus, detailed modelling is not possible, as yet, and even more difficult due to the
absence of available real data. Thus, mathematical methods are resorted to for-
mulate the load variations. In a first approach, the load of a test network, like the
ones examined so far in this chapter, could be stochastically altered, in order to
create different snapshots of a network, or even more, the load in each node could
be stochastically altered, regarding the current/original load value of the network,
either as its average, or its maximum value.
If the load of an electric distribution network, or even its load composition, is
considered as an average snapshot of the network, then load variations or even load
composition variations could be constructed via a uniform distribution, within a 20
and 50% range of the original snapshot. The loading condition of the IEEE-24 bus
Reliability Test System [65] can be studied as a base case, in order to justify the
238 A. S. Bouhouras et al.

100
Mean value
90 20% variance
50% variance
Load as peak percentage (%) 80

70

60

50

40

30

20

10

0
0 10 20 30 40 50

Fig. 8.16 Load duration curve for the typical IEEE-24 bus reliability test system

load variations modelling for the present analysis. Its hourly, daily, and weekly
peak load factors were used to construct the annual load curve. These peak load
factors have been selected in order to capture the loading conditions that yield the
highest annual energy losses and moreover in order to justify the upper limit for the
load variations (i.e. 50%) adopted in this analysis. This selected variance could
cover a loading composition for the network that refers to the highest load demands
that are expected within a one year time period. Subsequently, the annual load curve
is transformed into a cumulative power curve, as shown by the blue line in
Fig. 8.16 to investigate the loading variability. In the same figure, the mean annual
power, 61.45% of the annual peak power, along with the 20 and 50% variance
limits are also marked in continuous, dotted, and dashed grey lines, respectively. It
is calculated that the 20 and 50% variance cover the 55.08 and 99.40% of the total
annual loading levels, respectively. As proved by Fig. 8.16, the majority of loading
conditions of the network during a one year time period could be captured by load
variations up to 50% of the average load composition of the network.
If the ODGP towards power loss minimization is solved for every snapshot
created, then it would provide an optimal solution for each and every snapshot.
When examining the overall results, it is deduced that some buses appear more
frequently than others, i.e. appear in most of the snapshot’s solutions. This suggests
that some buses emerge as the most critical for siting DG units. Moreover, this
means that the siting stage of the ODGP is insensitive to the load variations, or even
to the load composition variations. Thus, the two stages of ODGP, siting and sizing,
can be examined separately. In Fig. 8.17 the results from the implementation on the
33-bus system are presented, for 2000 snapshots within 20%, and 6000 snapshots
within 50% range of the original load composition solved. The related frequency of
8 Optimal Distributed Generation Placement Problem … 239

Fig. 8.17 Relative frequency of appearance of buses in the 33-bus system

Fig. 8.18 Energy loss

reduction comparison for
1000 snapshots on the 33-bus
system

appearance of each bus for each case (20 and 50%, respectively) is shown. It
appears that buses No. 3, 7, 14, 24, 25 and 30 for both 20 and 50% range variations
emerge as the most prominent for DG installation.
In addition, if their average active and reactive power from the snapshot solu-
tions are to be taken into account for these prominent/critical buses, then they can
present a fixed but adequate solution for the ODGP problem towards energy loss
minimization [66]. In Fig. 8.18, for instance, results from 1000 snapshots within
20% range applied to the 33-bus system are presented. The total energy losses for
the 1000 snapshots without any DG installed is compared to the losses obtained
from the optimal solutions of every snapshot and the fixed solution from the most
prominent buses along with their respective average active and reactive power. It is
demonstrated that the fixed solution’s energy loss reduction is very close to the loss
reduction sum of the optimal solutions of all the snapshots, diverging only slightly
240 A. S. Bouhouras et al.

4
Bus 8
Bus 18
3.5 Bus 30
Slack Bus
3
Active Power (MW)

2.5

1.5

0.5

0
00:00 03:00 06:00 09:00 12:00 15:00 18:00 21:00 00:00
t (h:min)

Fig. 8.19 Daily load profile of various buses of the 33-bus system

by 6% from it. Hence, a very first estimation, if not an adequate solution, regarding
the ODGP towards energy loss is provided.
An alternative approach to load variations would be to consider that the electric
distribution network’s snapshot at hand is the peak load for the time period
examined. Moreover, instead of stochastically reproducing load snapshots to create
a load profile, the network itself can be combined with time-series of standard load
profiles, either real, or synthesized via load forecasting techniques. Thus, in a
straightforward approach, each bus’s load could be multiplied with a normalized
standard load profile and so creating the desired snapshots. However, since the
loads in a network do not necessarily change simultaneously or present the same
pattern, and most importantly the standard load profiles are more or less mea-
surements of the DNO on the substations within its purview and not on the load
buses themselves, a more elaborate scheme can be contemplated. It could be the-
orized that the total network’s load follows the standard load profile’s pattern and
each bus’s load changes in such a way, so that this can be achieved. An example
can be seen in Fig. 8.19, where the load profile of three buses is shown along the
total load profile of the network, as seen from the Slack Bus for a daily time period
of hourly quarter’s intervals, from the implementation of this method on the 33-bus
system. The buses’ load follow their own individual patterns, while the total net-
work load profile is seen from the Slack Bus, and that is observed by the DNO.
Thus, a more realistic approach of the problem has been achieved.
8 Optimal Distributed Generation Placement Problem … 241

Table 8.7 ODGP of different technologies—energy loss reduction results

Technology Energy loss reduction (%)
DG 82.9874
PV 36.5701
WT 50.2236

Table 8.8 ODGP of different DG technologies—detailed solution results

DG PV WT
Bus P (kW) Q (kVar) Bus P (kW) Q (kVar) Bus P (kW) Q (kVar)
no no no
3 248 78.4 3 281.6 81.5 3 395.8 78.6
6 281 94 6 550 204.3 6 358.7 71.9
11 222.6 59.1 9 96.8 22.2 11 312.7 64.1
16 217.6 50.9 11 214.8 46.7 16 387.7 63.2
31 206.1 109.2 16 262.5 52.2 30 388.9 152.3
Total Total P Total Q Total Total P Total Q Total Total P Total Q
no. (kW) (kVar) no. (kW) (kVar) no. (kW) (kVar)
5 1175.5 391.6 5 1405.7 406.9 5 1843.8 430.1

1
initial load
DG
0.8 PV
WT
Active Power (MW)

0.6

0.4

0.2

0
00;00 03;00 06;00 09;00 12;00 15;00 18;00 21;00 00;00
Time (h:min)

Fig. 8.20 Daily load curve without any DG (initial load), and with generic DG, PV and WT

8.6.2 Load/Generation Variation

With respect to RESs’ generation, there are data available both from DG stations
and mathematical tools, such as the Weibull distribution for wind speed, or a
242 A. S. Bouhouras et al.

Beta PDF for solar irradiance modelling. In spite of the more realistic approach of
the problem, it is of interest that still the siting stage of the ODGP is insensitive to
the load variation, load composition variation and perhaps DG technology, indi-
cating that it is more network-topology oriented. Furthermore, as expected, the
energy loss reduction is DG technology dependent. These can be seen in Tables 8.7
and 8.8, where results from an application on the 33-bus system are presented.
A daily period of hourly quarter’s intervals is considered and five DG units capable
of injecting active power with a maximum power factor of 0.95 leading/lagging.
DG units of constant power output, PVs and WTs as renewable technologies have
been applied. In case of PVs real data were used, whereas for WTs synthesized data
were obtained. The corresponding load curves can be seen in Fig. 8.20.
Additionally, with respect to an optimal mix of DG technologies, e.g. PVs and
WTs, it can be argued that the approach proposed in 8.5, is not that far from reality.

Fig. 8.21 The 33-bus system

divided into three CF areas

Area #1 Area #2

Area #3
8 Optimal Distributed Generation Placement Problem … 243

Table 8.9 ORESP towards energy loss minimization using a realistic approach and CF method
ORESP—realistic approach ORESP—CF on peak load
Energy loss 45.1869 Energy loss 28.102
reduction (%) reduction (%)
PV PV
Bus no. P (kW) Q (kVar) Bus no. P (kW) Q (kVar)
3 40.4 33.9
6 440.9 73.1 6 725.7 321.4
11 265.1 66.9 11 424.5 124.3
16 270.7 55.2 16 514.7 82.6
Total no. Total P Total Q Total no. Total P Total Q
(kW) (kVar) (kW) (kVar)
4 1017.1 229.1 3 1664.9 528.3
WT WT
Bus no. P (kW) Q (kVar) Bus no. P (kW) Q (kVar)
2 0 63.8
23 667 41.1 23 193.6 25.2
30 1046.5 149 30 239.7 95.7
Total no. Total P Total Q Total no. Total P Total Q
(kW) (kVar) (kW) (kVar)
3 1713.5 253.9 2 433.3 120.9

More specifically, the 33-bus system is divided in three areas of different weather
and geographical potentials, as depicted in Fig. 8.21, where in area #1 and #2 sun
and wind potential are dominant, respectively, and in area #3 they are competitive.
As earlier, same DG operation regarding active/reactive power and load profile is
assumed. As can be deduced from Table 8.9, if the method developed in Sect. 8.5
is performed for the peak load of the network, the solutions reached are a bit
different, though comparable. It should be stressed, however, that the analysis is
performed in a short time scale, i.e. a daily load curve. However, if the time scale is
extended to a whole year, or years, the solutions might bear more resemblance.

8.7 Combination of ODGP with Other Problems

8.7.1 ODGP and NR

In ODGP the siting and sizing of DG units is the objective whereas in Network
Reconfiguration (NR) an alternative layout is the objective in order to redistribute
the power flow. Both techniques are established as efficient, regarding power loss
reduction.
Despite the significant contribution of each technique towards loss reduction,
when applied individually, it seems that there are quite few studies that try to
244 A. S. Bouhouras et al.

examine the potentials of a combined approach under an efficient application order

for them [67–69]. Both power loss reduction techniques, when applied individually,
affect either the load composition of the electric distribution network (the net power
of the nodes that host DG units is altered in ODGP) or its layout (a reconfigured
topology after the NR application). Thus, when both techniques are applied, the
application order is highly possible to have an impact on the final solution regarding
the overall amount of loss reduction. If it is assumed that the highest possible loss
reduction refers to the ideal 100%, then the contribution of each technique towards
such a solution is affected by the order of their application. For instance, ODGP
could theoretically yield a solution with 100% power loss reduction in the ideal case
with one DG unit with power injection equal to the local load installed at each node.
In this latter case, the further application of NR would be meaningless. On the other
hand, if the ODGP problem refers to the more realistic case of limited available DG
units to be optimally sized and sited, then the application of the NR technique could
yield additional loss reduction and further improve the solution.
If the opposite application order is examined then it is interesting to investigate
how both the siting and sizing of the available DG units to be penetrated in the
electric distribution network would be affected, given that the ODGP problem will
now be applied to an altered network, i.e. with a reconfigured topology while
keeping the same load composition.
Let us consider the following three scenarios of the solving order of ODGP and
NR:
• scenario-1: NR solved first, then ODGP,
• scenario-2: ODGP first, then NR, and,
• scenario-3: both ODGP and NR are concurrently solved.
The results, when implemented in the 69-bus system are presented in
Tables 8.10, 8.11, and 8.12, whereas in Fig. 8.22 the 69-bus system along with its
tie-switches is depicted. Seven DG units are considered for installation and capable

Table 8.10 Scenario-1: NR 1st, ODGP 2nd

NR Initial Sectionalizers Tie switches Loss reduction Final losses
applied losses open closed % (kW)
(kW)
229.8 14,58,62 Tie3-Tie5 54.7 104.1
ODGP Initial Nodes to host Active power Reactive Loss reduction
applied losses DG units of each DG power of each % and final
(kW) unit (kW) DG unit (kVar) losses (kW)
104.1 5 901.7 189.2 93.65%
9 241.6 177.2 6.6
12 427.4 299.6
22 338.3 226.6
40 0 536.4
53 1416.1 938.2
56 318.5 226.7
8 Optimal Distributed Generation Placement Problem … 245

Table 8.11 Scenario-2: ODGP 1st, NR 2nd

ODGP Initial Nodes to host Active power Reactive Loss reduction
applied losses DG units of each DG power of each % and final
(kW) unit (kW) DG unit (kVar) losses (kW)
229.8 2 0 -53.2 97.35%
3 539 340 6.1
9 0 184.9
12 501.2 279.8
19 380.8 251.7
40 717 512
53 1674 1178.8
NR Initial Sectionalizers Tie switches Loss reduction Final losses
applied losses open closed % (kW)
(kW)
6.1 – – 0 6.1

Table 8.12 Scenario-3: ODGP NR concurrently

Candidate Active Reactive Sectionalizers Tie Loss
node to power of power of open switches reduction %
host DG each DG each DG unit closed —final
units unit (kW) (kVar) losses (kW)
57 2021.5 849.8 20,42,46,58,61 Tie1-tie5 68.28%
72.9

of both active and reactive power generation. The UPSO algorithm, as presented in
Sect. 8.3, is utilized. The first scenario seems to be advantageous since the
switching operations rely on the already existent tie-switches and that results in
lower required DG capacity for power loss minimization. In the second scenario, it
is highly possible to be unable to apply the NR technique, especially if the ODGP
technique performs quite well under high power loss reduction by the installation of
the proposed DG units. Finally, in the third scenario since both techniques are
considered concurrently, the problem’s complexity increases exponentially, thus the
algorithm seems unable to provide an adequate solution. It is yet to be investigated,
whether the worth of a better solution in this case is overweighted by the increased
computational burden [70–72].

8.7.2 ODGP and OESSP

ODGP can be targeted towards energy loss reduction, due to the nature of DG units
since they produce electricity even for a certain time period. ESSs, though, present
an entirely different complexion. Moreover, because ESSs’ integration in a more
massive or industrial scale is still in its infancy, cost is still and a more important
246 A. S. Bouhouras et al.

29 30 31 32 33 34 35 36

37 38 39 40 Tie-2

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

60 61
Tie-4

Tie-5
Tie-3

58 59
Tie-1

45 46 47 48 49 50 51 52 53 54 55 56 57

43 44

41 42 62 63 64 65 66 67 68 69 70

Fig. 8.22 The 69-bus system depicted with its tie-switches

Table 8.13 Energy loss reduction from installing DG units along with ESSs
Energy loss (MWh) Energy loss reduction (%)
Initial 2.7647 –
DG 1.5593 43.5997
LS 1.5606 43.5526
EM = 1 MWh 1.5532 43.8203
EM = 2.5 MWh 1.5449 44.1205
EM = 5 MWh 1.5353 44.4678

issue. Thus, when dealing with the OESSP problem, cost or profit objective
functions are considered [73]. Furthermore, it might not be of significant aid
towards energy loss reduction. For example, using the load/generation tools
available from Sect. 8.6.2, two modes of ESSs can be added: load smoothing
(LS) and energy management (EM). The former is used in order to smooth out any
abrupt spikes in load curves and the latter in storing energy during one time period
and providing it at another. PVs and WTs have also been utilised, for a more
thorough approach, provided from the example in Sect. 8.6.2. For LS, the ESSs are
considered to be installed in the buses where the PVs and WTs have been installed,
8 Optimal Distributed Generation Placement Problem … 247

2.5
Initial Load
DG
2 LS
EM = 1MWh
1.5 EM = 2.5 MWh
Active Power (MW)

EM = 5 MWh

0.5

-0.5

-1
00:00 03:00 06:00 09:00 12:00 15:00 18:00 21:00 00:00
Time (h:min)

Fig. 8.23 Load curves for integrating DGs along ESS in LS and EM mode

and for EM near the Slack Bus, since it is theorized that it will be installed by the
DNO. In Table 8.13 the energy losses are presented and in Fig. 8.23 the load curves
for a time period of one day are illustrated. As can be seen, in this configuration the
impact in energy losses of ESS is limited, regardless of mode or size, although great
benefits have been provided for the DNO, from both modes, regarding the load
curves. It should be emphasized though, that OESSP might prove promising in the
field of energy loss minimization over a more extended and elaborate analysis. For
instance, both sizing and siting could be examined concurrently, and ESSs systems
capable of LS and EM operation, or even Frequency Regulation, as well.

References

1. T. Ackermann, V. Knyazkin, Interaction between distributed generation and the distribution

network: operation aspects, in IEEE/PES Transmission and Distribution Conference and
Exhibition (2002), pp. 1357–1362
2. N. Mohandas, R. Balamurugan, L. Lakshminarasimman, Optimal location and sizing of real
power DG units to improve the voltage stability in the distribution system using ABC
algorithm united with chaos. Int. J. Electr. Power Energy Syst. 66, 41–52 (2015)
3. V.H. MendezQuezada, J. RivierAbbad, T. GomezSanRoman, Assessment of energy
distribution losses for increasing penetration of distributed generation. IEEE Trans. Power
Syst. 21(2), 533–540 (2006)
4. K.O. Oureilidis, E.A. Bakirtzis, C.S. Demoulias, Frequency-based control of islanded
microgrid with renewable energy sources and energy storage. J. Mod. Power Syst. Clean
Energy 4(1), 54–62 (2016)
5. P.A. Gkaidatzis, D.I. Doukas, A.S. Bouhouras, K.I. Sgouras, D.P. Labridis, Impact of
penetration schemes to optimal DG placement forlossminimisation. Int. J. Sustain. Energy 36
(5), 473–488 (2017)
248 A. S. Bouhouras et al.

6. A.S. Bouhouras, K.I. Sgouras, P.A. Gkaidatzis, D.P. Labridis, Optimal active and reactive
nodal power requirements towards loss minimization under reverse power flow constraint
defining DG type. Int. J. Electr. Power Energy Syst. 78, 445–454 (2016)
7. M. Esmaili, Placement of minimum distributed generation units observing power losses and
voltage stability with network constraints. IET Gener. Transm. Distrib. 7(8), 813–821 (2013)
8. S. Ge, L. Xu, H. Liu, J. Fang, Low-carbon benefit analysis on DG penetration distribution
system. J. Mod. Power Syst. Clean Energy 3(1), 139–148 (2015)
9. A. Soroudi, M. Ehsan, R. Caire, N. Hadjsaid, Hybrid immune-genetic algorithm method for
benefit maximisation of distribution network operators and distributed generation owners in a
deregulated environment. IET Gener. Transm. Distrib. 5(9), 961 (2011)
10. Y. del Valle, G.K. Venayagamoorthy, S. Mohagheghi, J.-C. Hernandez, R.G. Harley, Particle
swarm optimization: basic concepts, variants and applications in power systems. IEEE Trans.
Evol. Comput. 12(2), 171–195 (2008)
11. K.E. Parsopoulos, M.N. Vrahatis, Particle Swarm Optimization and Intelligence: Advances
and Applications (IGI Global, Hershey, 2010)
12. U. Leeton, D. Uthitsunthorn, U. Kwannetr, N. Sinsuphun, T. Kulworawanichpong, Power
loss minimization using optimal power flow based on particle swarm optimization, in 2010
IEEE International Conference on Electrical Engineering/Electronics Computer
Telecommunications and Information Technology (ECTI-CON) (2010), pp. 440–444
13. P.S. Georgilakis, N.D. Hatziargyriou, A review of power distribution planning in the modern
power systems era: models, methods and future research. Electr. Power Syst. Res. 121, 89–
100 (2015)
14. D.Q. Hung, N. Mithulananthan, R.C. Bansal, Analytical expressions for DG allocation in
primary distribution networks. IEEE Trans. Energy Convers. 25(3), 814–820 (2010)
15. D.Q. Hung, N. Mithulananthan, Multiple distributed generator placement in primary
distribution networks for loss reduction. IEEE Trans. Ind. Electron. 60(4), 1700–1708 (2013)
16. D.Q. Hung, N. Mithulananthan, Loss reduction and loadability enhancement with DG: a
dual-index analytical approach. Appl. Energy 115, 233–241 (2014)
17. P. Prakash, D.K. Khatod, An analytical approach for optimal sizing and placement of
distributed generation in radial distribution systems, in 1st IEEE International Conference on
Power Electronics. Intelligent Control and Energy Systems (ICPEICES-2016) (2016), pp. 1–5
18. T. Kumar, T. Thakur, Comparative analysis of particle swarm optimization variants on
distributed generation allocation for network loss minimization, in 2014 First International
Conference on Networks & Soft Computing (ICNSC2014) (2014), pp. 167–171
19. A.A. Abou El-Ela, S.M. Allam, M.M. Shatla, Maximal optimal benefits of distributed
generation using genetic algorithms. Electr. Power Syst. Res. 80(7), 869–877 (2010)
20. K.-H. Kim, Y.-J. Lee, S.-B. Rhee, S.-K. Lee, S.-K. You, Dispersed generator placement using
fuzzy-GA in distribution systems, in IEEE Power Engineering Society Summer Meeting, vol.
3 (2002), pp. 1148–1153
21. F.S. Abu-Mouti, M.E. El-Hawary, Optimal distributed generation allocation and sizing in
distribution systems via artificial bee colony algorithm. IEEE Trans. Power Deliv. 26(4),
2090–2101 (2011)
22. A.A. Seker, M.H. Hocaoglu, Artificial Bee Colony algorithm for optimal placement and
sizing of distributed generation, in 2013 8th International Conference on Electrical and
Electronics Engineering (ELECO) (2013), pp. 127–131
23. N. Taher, I.T. Seyed, A. Jamshid, T. Sajad, N. Majid, A modified honey bee mating
optimization algorithm for multiobjective placement of renewable energy resources. Appl.
Energy 88(12), 4817–4830 (2011)
24. W.S. Tan, M.Y. Hassan, M.S. Majid, H.A. Rahman, Allocation and sizing of DG using
Cuckoo search algorithm, in 2012 IEEE International Conference on Power and Energy
(PECon) (2012), pp. 133–138
25. M. Zahra, A. Amir, A novel approach based on cuckoo search for {DG} allocation in
distribution network. Int. J. Electr. Power Energy Syst. 44(1), 672–679 (2013)
8 Optimal Distributed Generation Placement Problem … 249

26. W. Buaklee, K. Hongesombut, Optimal DG allocation in a smart distribution grid using

Cuckoo search algorithm, in 2013 10th International Conference on Electrical Engineering/
Electronics, Computer, Telecommunications and Information Technology (ECTI-CON)
(2013), pp. 1–6
27. S. Roy, S. Sultana, P.K. Roy, Oppositional cuckoo optimization algorithm to solve DG
allocation problem of radial distribution system, in 2015 International Conference on Recent
Developments in Control, Automation and Power Engineering (RDCAPE) (2015), pp. 44–49
28. A.Y. Abdelaziz, R.A. Osama, S.M. Elkhodary, Using the harmony search algorithm for
reconfiguration of power distribution networks with distributed generation units. J. Bioinform.
Intell. Control 2(3), 237–242 (2013)
29. S.I. Kumar, N.P. Kumar, A novel approach to identify optimal access point and capacity of
multiple DGs in a small, medium and large scale radial distribution systems. Int. J. Electr.
Power Energy Syst. 45(1), 142–151 (2013)
30. R.S. Rao, K. Ravindra, K. Satish, S.V.L. Narasimham, Power loss minimization in
distribution system using network reconfiguration in the presence of distributed generation.
IEEE Trans. Power Syst. 28(1), 317–325 (2013)
31. A. Mohamed Imran, M. Kowsalya, Optimal size and siting of multiple distributed generators
in distribution system using bacterial foraging optimization. Swarm Evol. Comput. 15, 58–65
(2014)
32. M.J. Hadidian-Moghaddam, S. Arabi-Nowdeh, M. Bigdeli, D. Azizian, A multi-objective
optimal sizing and siting of distributed generation using ant lion optimization technique. Ain
Shams Eng. J. 1–9 (2017)
33. A. Sobieh, M. Mandour, E.M. Saied, M.M. Salama, Optimal number size and location of
distributed generation units in radial distribution systems using Grey Wolf optimizer. Int.
Electr. Eng. J. 7(9), 2367–2376 (2017)
34. M.H. Moradi, M. Abedini, A combination of genetic algorithm and particle swarm
optimization for optimal DG location and sizing in distribution systems. Int. J. Electr. Power
Energy Syst. 34(1), 66–74 (2012)
35. A.J.G. Mena, J.A.M. Garcia, An efficient approach for the siting and sizing problem of
distributed generation. Int. J. Electr. Power Energy Syst. 69, 167–172 (2015)
36. R. Viral, D.K. Khatod, An analytical approach for sizing and siting of DGs in balanced radial
distribution networks for loss minimization. Int. J. Electr. Power Energy Syst. 67, 191–201
(2015)
37. R. Eberhart, J. Kennedy, A new optimizer using particle swarm theory, in Proceedings of the
Sixth International Symposium on Micro Machine and Human Science MHS’95 (1995),
pp. 39–43
38. A.P. Engelbrecht, Computational Intelligence: An Introduction, vol. 115, 2nd edn. (Wiley,
Chichester, 2008), pp. 3–78
39. K.E. Parsopoulos, M.N. Vrahatis, Parameter selection and adaptation in unified particle
swarm optimization. Math. Comput. Model. 46(1–2), 198–213 (2007)
40. P.A. Gkaidatzis, A.S. Bouhouras, D.I. Doukas, K.I. Sgouras, D.P. Labridis, Application and
evaluation of UPSO to ODGP in radial distribution networks, in 2016 13th International
Conference on the European Energy Market (EEM), vol. 2016, July (2016), pp. 1–5
41. J.H. Holland, Genetic algorithms. Sci. Am. 267(1), 66–72 (1992)
42. D. Karaboga, B. Basturk, Artificial bee colony (ABC) optimization algorithm for solving
constrained optimization problems, ed. by P. Melin, O. Castillo, L.T. Aguilar, J. Kacprzyk,
W. Pedrycz, in Proceedings of the Foundations of Fuzzy Logic and Soft Computing: 12th
International Fuzzy Systems Association World Congress (IFSA 2007), Cancun, Mexico, 18–
21 June 2007 (Springer, Berlin, Heidelberg, 2007), pp. 789–798
43. X.S. Yang, S. Deb, Cuckoo search via levy flights, in World Congress on Nature Biologically
Inspired Computing (NaBIC 2009) (2009), pp. 210–214
44. Z.W. Geem, J.H. Kim, G.V. Loganathan, A new heuristic optimization algorithm: harmony
search. Simulation 76(2), 60–68 (2001)
250 A. S. Bouhouras et al.

45. M. Kashem, V. Ganapathy, G. Jasmon, M. Buhari, A novel method for loss minimization in
distribution networks, in International Conference on Electric Utility Deregulation and
Restructuring and Power Technologies (DRPT2000). Proceedings (Cat. No.00EX382), no.
603 (2000), pp. 251–256
46. S. Ghosh, S.P. Ghoshal, S. Ghosh, Optimal sizing and placement of distributed generation in
a network system. Int. J. Electr. Power Energy Syst. 32(8), 849–856 (2010)
47. C. Wang, M.H. Nehrir, Analytical approaches for optimal placement of distributed generation
sources in power systems. IEEE Trans. Power Syst. 19(4), 2068–2076 (2004)
48. M.F. Akorede, H. Hizam, I. Aris, M.Z.A. Ab Kadir, Effective method for optimal allocation
of distributed generation units in meshed electric power systems. IET Gener. Transm. Distrib.
5(2), 276 (2011)
49. R.K. Singh, S.K. Goswami, Optimum siting and sizing of distributed generations in radial and
networked systems. Electr. Power Components Syst. 37(2), 127–145 (2009)
50. D.I. Doukas, P.A. Gkaidatzis, A.S. Bouhouras, K.I. Sgouras, D.P. Labridis, On reverse power
flow modelling in distribution grids, in Mediterranean Conference on Power Generation,
Transmission, Distribution and Energy Conversion (MedPower 2016) (2016), p. 65 (6.)
51. R. Yokoyama, S.H. Bae, T. Morita, H. Sasaki, Multiobjective optimal generation dispatch
based on probability security criteria. IEEE Trans. Power Syst. 3(1), 317–324 (1987)
52. P. Kayal, C.K. Chanda, Placement of wind and solar based DGs in distribution system for
power loss minimization and voltage stability improvement. Int. J. Electr. Power Energy Syst.
53, 795–809 (2013)
53. V.R. Pandi, H.H. Zeineldin, W. Xiao, Determining optimal location and size of distributed
generation resources considering harmonic and protection coordination limits. IEEE Trans.
Power Syst. 28(2), 1245–1254 (2013)
54. A. Keane, M. O’Malley, Optimal distributed generation plant mix with novel loss adjustment
factors, in 2006 IEEE Power Engineering Society General Meeting (2006), 6 pp
55. Y.M. Atwa, E.F. El-Saadany, M.M.A. Salama, R. Seethapathy, Optimal renewable resources
mix for distribution system energy loss minimization. IEEE Trans. Power Syst. 25(1), 360–
370 (2010)
56. C. Yammani, S. Maheswarapu, S. Matam, Optimal placement of multi DGs in distribution
system with considering the DG bus available limits. Energy and Power 2(1), 18–23 (2012)
57. P.A. Gkaidatzis, A.S. Bouhouras, K.I. Sgouras, D.I. Doukas, D.P. Labridis, Optimal
distributed generation placement problem for renewable and DG units: an innovative
approach, in Mediterranean Conference on Power Generation, Transmission, Distribution
and Energy Conversion (MedPower 2016) (2016), p. 66 (7.)
58. S. Soudi, Distribution system planning with distributed generations considering benefits and
costs. Int. J. Mod. Educ. Comput. Sci. 5(October), 45–52 (2013)
59. L.F. Ochoa, G.P. Harrison, Minimizing energy losses: optimal accommodation and smart
operation of renewable distributed generation. IEEE Trans. Power Syst. 26(1), 198–205
(2011)
60. L.F. Ochoa, A. Padilha-Feltrin, G.P. Harrison, Evaluating distributed time-varying generation
through a multiobjective index. IEEE Trans. Power Deliv. 23(2), 1132–1138 (2008)
61. G.N. Koutroumpezis, A.S. Safigianni, Optimum allocation of the maximum possible
distributed generation penetration in a distribution network. Electr. Power Syst. Res. 80(12),
1421–1427 (2010)
62. Y.M. Atwa, E.F. El-Saadany, Probabilistic approach for optimal allocation of wind-based
distributed generation in distribution systems. IET Renew. Power Gener. 5(1), 79 (2011)
63. F. Rotaru, G. Chicco, G. Grigoras, G. Cartina, Two-stage distributed generation optimal
sizing with clustering-based node selection. Int. J. Electr. Power Energy Syst. 40(1), 120–129
(2012)
64. M.F. Shaaban, Y.M. Atwa, E.F. El-Saadany, DG allocation for benefit maximization in
distribution networks. IEEE Trans. Power Syst. 28(2), 939–949 (2013)
65. P. Subcommittee, IEEE reliability test system. IEEE Trans. Power Appar. Syst. PAS-98(6),
2047–2054 (1979)
8 Optimal Distributed Generation Placement Problem … 251

66. A.S. Bouhouras, C. Parisses, P.A. Gkaidatzis, K.I. Sgouras, D.I. Doukas, D.P. Labridis,
Energy loss reduction in distribution networks via ODGP, in International Conference on the
European Energy Market (EEM), vol. 2016–July (2016)
67. B. Pawar, S. Kaur, G.B. Kumbhar, An integrated approach for power loss reduction in
primary distribution system, in 2016 IEEE 6th International Conference on Power Systems
(ICPS) (2016), pp. 1–6
68. W.M. Dahalan, H. Mokhlis, Network reconfiguration for loss reduction with distributed
generations using PSO, in 2012 IEEE International Conference on Power and Energy
(PECon) (2012), pp. 823–828
69. W. Mohd Dahalan, H. Mokhlis, R. Ahmad, A.H. Abu Bakar, I. Musirin, Simultaneous
network reconfiguration and DG using EP method. Int. Trans. Electr. Energy Syst. 25(11),
2577–2594 (2015)
70. A.S. Bouhouras, P.A. Gkaidatzis, D.P. Labridis, Optimal application order of network
reconfiguration and ODGP for loss reduction in distribution networks, in 17 IEEE
International Conference on Environment and Electrical Engineering (EEEIC 2017)
(2017), pp. 1–6
71. A.S. Bouhouras, G.T. Andreou, D.P. Labridis, A.G. Bakirtzis, Selective automation upgrade
in distribution networks towards a smarter grid. IEEE Trans. Smart Grid 1(3), 278–285 (2010)
72. A.S. Bouhouras, D.P. Labridis, Influence of load alterations to optimal network configuration
for loss reduction. Electr. Power Syst. Res. 86, 17–27 (2012)
73. N.D. Hatziargyriou, D. Skrlec, T. Capuder, P.S. Georgilakis, M. Zidar, Review of energy
storage allocation in power distribution networks: applications, methods and future research.
IET Gener. Transm. Distrib. 10(3), 645–652 (2016)
Chapter 9
Optimal Planning of Grid
Reinforcement with Demand
Response Control

Alexandre M. F. Dias and Pedro M. S. Carvalho

Abstract This chapter presents a hybrid methodology based on a local search

algorithm and a genetic algorithm, used to address the multi-objective and multi-
stage optimal distribution expansion planning problem. The methodology is con-
ceived to solve optimal network investment problems under the new possibilities
enabled by the smart grid, namely the new observability and controllability
investments that will be available to enable demand response in the future. The
multi-objective methodology is applied to an existing low-voltage electric distri-
bution network under a congestion scenario to yield a Pareto-optimal set of solu-
tions. The solutions are then projected onto the two investment possibilities
considered: demand control investments and traditional network asset investments.
The projected surface is then analyzed to discuss the merit of demand control with
respect to postponing traditional asset investments.

Keywords Demand response Distribution planning
Information and


communications technology Network optimization

9.1 Introduction

Distributed Generation (DG) and Electric Vehicles (EV) bring new challenges to
the operation of electric distribution networks. New challenges involve dealing not
only with peak load conditions, but also with potential reverse flows due to DG and

A. M. F. Dias (&)
P. M. S. Carvalho

Department of Electrical and Computer Engineering, Instituto Superior Técnico,
University of Lisbon, Lisbon, Portugal
e-mail: [email protected]
P. M. S. Carvalho
e-mail: [email protected]
A. M. F. Dias
P. M. S. Carvalho
Instituto de Engenharia de Sistemas e Computadores - Investigação
e Desenvolvimento (INESC-ID), Lisbon, Portugal

© Springer Nature Singapore Pte Ltd. 2018 253

F. Shahnia et al. (eds.), Electric Distribution Network Planning,
Power Systems, https://doi.org/10.1007/978-981-10-7056-3_9
254 A. M. F. Dias and P. M. S. Carvalho

Fig. 9.1 Graph 4

representation of an electric
distribution network: circles 5
represent network nodes 3 6
5
(graph vertices), while lines
connecting circles represent 7
network branches (graph 6
edges) that connect the 2
network nodes

1 1 2 3 7 8 8

4
10
9
9
10
11

with the new active management possibilities enabled by the Information and
Communications Technologies (ICT) such as demand response (DR) control.
Traditionally, distribution planning does not consider the investment and oper-
ational benefits of control as enabled by ICT. The benefits of additional control-
lability are currently not clearly identified and the impacts of ICT investments are
not well understood.
A multi-objective optimization methodology is used to search for traditional
network investments together with DR control investments, so that a set of
investment projects and their corresponding time periods lead to a minimum overall
value of a set of objective functions. A multi-objective formulation is quite valuable
in this context since it allows projecting the Pareto surface onto the two investment
dimensions of (i) investment on traditional network assets and (ii) investment on
control equipment. Such a projection allows trading-off these two kinds of
investments without speculating on the costs of ICT and control equipment, whose
costs are currently difficult to predict.
Let us represent an electric distribution network by a graph G, where the vertices
represent the network nodes, while the edges represent the existing lines and
transformers (see Fig. 9.1). ICT (vertex) investments can be considered in order to
reduce demand impact through demand and DER control, whereas line/transformer
(edge) reinforcement investments can be considered in order to relieve overloads
and avoid voltage drop or rise beyond predefined acceptable levels. Under such a
solution space, the planning solution is a schedule of projects, both vertex and edge
type projects, that aim to minimize investment and operational costs while
9 Optimal Planning of Grid Reinforcement with Demand … 255

respecting key planning criteria like adequate voltage levels under normal and
contingency situations and loading limits for lines and transformers.
An overview of the existing literature shows that distribution planning used to
rely on a set of methods to decide the location and type of reinforcements needed in
order to cope with the traditional sources of uncertainty, such as expected load
forecast, at minimum cost. Several approaches have been taken in the past to solve
the problem [1–8].
Some models deal with a fixed horizon year and single network solution
topology [9–16] and are thus known as single stage models. Other models deal with
the dynamic nature of demand through time as well as with a sequence of network
solution topologies (one per stage) and are hence known as multistage models [2, 3,
5, 8, 15].
In either single or multistage models, optimization techniques were used to solve
the problem. These involve genetic algorithms [16], Benders’ decomposition [17],
simulated annealing [18], tabu search [19], greedy randomized adaptive search
procedure (GRASP) [20] and game theory [21]. The common output of all these
former approaches is the conception of a plan, i.e., a set of projects where the
system reinforcements and equipment additions are scheduled.
A hybrid optimization strategy is presented in the following sections of this
chapter, combining a local search algorithm with moderate search effort and a
metaheuristic method to broaden the search space and to guide towards a
close-to-optimum solution.
Finally, a realistic case study, where ICT reinforcements are traded off with
conventional grid reinforcements, is analyzed through the application of the
developed strategy.

9.2 Distribution Planning Methodology

9.2.1 Formulation of the Problem

Consider a vector of possible investment projects P ¼ ½p1 ; p2 ; . . .; pN . A decision

schedule can be represented by a vector of timings, t, that index the projects of P,
where t ¼ ½t1 ; t2 ; . . .; tN .
The optimal planning problem can then be formulated as the problem of finding
the optimal timing for each investment project. As a multi-objective problem, the
problem may be formulated as

min f1 ðtÞ; f2 ðtÞ; . . .; fj ðtÞ ð9:1Þ
ðtÞ

s:t: t ¼ ½t1 ; t2 ; . . .; tN 
ti 2 f1; 2; . . .; T þ 1g; i ¼ 1; 2; . . .; N
256 A. M. F. Dias and P. M. S. Carvalho

where fj ðtÞ represents the objective function j to be minimized for the planning
horizon T, and the timing T + 1 is interpreted as the timing to be assigned to the
projects that will not be undertaken within such an horizon.

9.2.2 Solution Approach

As decisions are multi-stage, the decision space is large-scale and the decision
schedules are computationally expensive to evaluate, making the possible effective
solution approaches very confined. Note that the scheduling problem alone is an
NP-Hard problem [22] and therefore the search cannot guarantee the global opti-
mum to be found. The solution approach must therefore be able to provide
close-to-optimal plans involving short-term and longer-term investment decisions
that need to be evaluated thoroughly.
Project schedules can be found by a classical local search algorithm
(Gaussian-like search) with moderate search effort. However, such Gaussian
algorithm is a local optimization approach, and being the stated problem a non-
convex one [23], it does not guarantee close-to-optimum solutions [24]—solutions
typically get trapped in local optima. The Gaussian Search (GS) is sensitive with
respect to the order by which the different investment projects are analyzed [25].
Therefore, to find close-to-optimum schedules a specific Genetic Algorithm (GA) is
presented to learn about the best order for analysis by the GS. The overall solution
approach can then be formulated as a hybrid algorithm.
Thus, the GA is used to find the best order for the GS optimization, while GS is
used to find the best project schedule given an order to analyze the projects. Within
the GS evaluation, objective function values are updated based on the results of a
DR optimized power flow, whose implementation is described in this chapter. The
presented hybrid approach has proven to yield robust solutions in several planning
contexts [25] and has been successfully implemented for a Medium Voltage feeder
[26].
Within the application of this hybrid solution, if a multi-objective function is to
be addressed explicitly, one may define GS and GA selection criteria to address
search as multi-objective and return a set of Pareto non-dominated schedules.
The architecture of the overall algorithm is depicted in Fig. 9.2. The blocks
represent the main methods, while the text and direction of the arrows represent,
respectively, what information and the direction in which such information is
exchanged between those methods. The methods of the hybrid solution approach
are described in the following.
The execution of the algorithm starts with the initialization of a set of possible
orders, O, for project evaluation by the GS. Within the context of this hybrid
approach, a possible order is an order of N projects where each investment project
pi 2 P; i ¼ 1; 2; . . .; N appears exactly once in the order. For example, for N ¼ 6
9 Optimal Planning of Grid Reinforcement with Demand … 257

GA
to learn about
best orders

Local opƟma
{orders} for given
{orders}

GS
to improve
project schedule

project schedule network performance

(investment Ɵmings) (violaƟons, losses)

Grid SimulaƟon
w/ Demand Response
control

Fig. 9.2 Overview of the hybrid optimization methodology (GA and GS combination)

investment projects and an initial set of 41 orders, the initial population of orders
could be the one illustrated in Fig. 9.3.
The population is then subjected to evaluation by the Gaussian Search, with each
order ok of O being evaluated separately. Throughout such evaluation, the optimal
timings (stages) for allocation of the investment projects are continuously and
iteratively updated so that a set of objective functions is minimized. When the set of
optimal timings t of an order remains unchanged from the previous iteration, the
process (GS) stops for that order. A possibility for the initial investment timings for
the projects is T + 1 for all projects, meaning that at the beginning none of the
projects has been allocated to the planning horizon.
More specifically, for a given order ok of O, one iteration of the GS evaluation of
such order corresponds to the update of the optimal investment timings for the
N ¼ 6 projects according to the relative positions of the projects in that order. As a
convention, it is assumed that relative positions are established from left to right
(referring to Fig. 9.3).
In the case of order o1 , this means that the GS starts by analyzing project p3 and
finding the optimal investment timing for such a project (t3 ); then, it proceeds to the
analysis of project p5 and finding its optimal timing t5 . Similarly, the remaining

1
Note that this value is relatively small and was chosen for demonstration purposes. The typical
size of a genetic algorithm population is several tens or even hundreds of individuals.
258 A. M. F. Dias and P. M. S. Carvalho

Fig. 9.3 Example of an initial population of orders. The six projects of each order are highlighted
in different colors for ease of recognition. Each project appears exactly once in each order

projects p2 , p4 , p1 and p6 are analyzed and the end of the order is reached. A certain
set of investment timings t is obtained in this first iteration of the GS. The GS would
then proceed to again analyzing all the projects of order o1 in the aforementioned
sequence, taking into account the optimal timings t of the previous iteration. When
the end of order o1 is reached again, the investment timings set is compared to that
of the previous iteration. The GS procedure is repeated while the sets of investment
timings of the last two iterations compare differently.
The GS determines the optimal investment timing of a project by making use of
an Optimal Power Flow method, which considers, for all the other projects, those
that have been added to the network graph G so far (investment timings between 1
and T) and, for the project being analyzed, all the possibilities of investment timings
(1 to T + 1). Therefore, for the investment timings of all the other projects (un-
changed) and the possibilities for the project under analysis (T + 1), T + 1 sets of
objective function values are obtained.
From these objective function value sets(solutions), the (Pareto) non-dominated
solutions are selected. A solution a ¼ f1 ðta Þ; f2 ðta Þ; . . .; fj ðta Þ is said to be

non-dominated when there is no other solution b ¼ f1 ðtb Þ; f2 ðtb Þ; . . .; fj ðtb Þ that
satisfies b  a, i.e., the value of at least one objective function value in b is less than
the value of the same objective function in a, while the values of the other objective
functions in b are less or equal than the corresponding objective function values in
a. In the latter case, b is said to dominate a and, conversely, a is said to be
dominated by b.
Assume an electric distribution network, with a radial configuration and six
branches (lines or cables) connecting the network nodes. The reinforcement of each
network branch is considered (N ¼ 6), and two objective functions to be minimized
as well: network reinforcement and network losses. For the sake of simplicity, the
cost of reinforcement is 1 unit (year zero), being discounted in the planning horizon
stages with certain inflation and discount rates. Let us also suppose that, during the
GS, some projects had already been analyzed and some of the optimal timings were
between 1 and T (within the planning horizon), meaning that there were investment
projects added to the network graph. When a certain reinforcement project was
analyzed (suppose p4 ), the results of the Optimal Power Flow for the T + 1
investment timing possibilities were the following (T = 3):
9 Optimal Planning of Grid Reinforcement with Demand … 259

t4 f1 ðtÞ f2 ðtÞ
ðt ¼ ½t1 ; t2 ; t3 ; t4 ; t5 ; t6 Þ Present reinforcement cost Network losses
1 3.57 5
2 3.51 5
3 3.46 10
4 2.63 15

The impact of this reinforcement project on losses is the same whether it is

allocated to stages 1 or 2. But, since its present cost decreases with time, the total
reinforcement cost also decreases with time (the investment timings of the other
projects, t1 ; t2 ; t3 ; t5 and t6 remain unchanged). Therefore, the solution for t4 ¼ 2
dominates the solution for t4 ¼ 1. Thus, allocating project p4 to stages 2, 3 or 4
leads to non-dominated solutions, since that postponing project p4 means a decrease
in the reinforcement cost but also a corresponding increase in the network losses.
We are then left with three possibilities for the optimal investment timing t4 . The
criterion or criteria to select a value can vary. For example, one might want to invest
as late as possible or simply choose a non-dominated solution randomly (the latter
is used in the case study of this chapter).
The local optimization by the GS is heavily dependent on the Optimal Power
Flow (OPF) method. The latter gives an indication to the GS of network perfor-
mance (objective function values) considering the investments allocated so far
within the planning horizon, and when the GS needs to decide on the investment
timing of the project being analyzed, it does so based on the results of the OPF
method. How investments affect network performance depends on how they impact
network operation. More specifically, reinforcing a line or cable is different from
ICT infrastructure investment intended to shed consumer load during congestion
periods. The impacts and actions of network investments are simulated by the
selected OPF method.
The OPF method presented in this chapter aims at simulating the actions
facilitated by DR control investment, while still considering the impact of tradi-
tional reinforcement investments. If violations of network line or cable current
ratings or of node voltage levels are detected, it tries to solve them by shifting
(postponing) consumer load demand power. Priority is given to current rating
violations over voltage level violations, since the former are more critical to net-
work operation.
The OPF method ensures that the minimum number of controllable consumers is
affected when trying to solve violations. On the one hand, it tries to solve current
(and later voltage) violations by first looking at problematic branches (nodes) at the
most downstream locations in the network topology and only affecting controllable
consumers downstream such branches (nodes). Only when violations are solved or
when no further reduction in demand power of downstream controllable consumers
is possible it moves upstream the network to solve current (voltage) violations. On
the other hand, in a given time period only the minimum number of consumers is
affected as the demand power of controllable consumers that contribute most to
260 A. M. F. Dias and P. M. S. Carvalho

violations (highest demand power) is reduced first, and only then controllable
consumers with lower demand power may have their demand power reduced in
order to solve violations.
After all the orders of population O are evaluated by the GS, a set of investment
timings t for each order ok is determined, as well as the corresponding objective
function values. Following GS evaluation, individuals (orders) of the population
are selected and then manipulated through genetic operators in order to create a new
population. The process of selection and manipulation is described next.
The orders are selected considering how interior the Pareto fronts to which their
solutions (objective function values) belong is. From a Pareto-optimal point of
view, the more interior the solution front is the better—corresponding to lower
overall values of the objective functions, in the case of a problem formulated as a
minimization. To determine the Pareto fronts to which each solution belongs, the
non-dominated sorting of the NSGA-II [27] is used.
Essentially, the NSGA-II sorting works in the following way: initially, consid-
ering all the solutions obtained, the subset of solutions that are not dominated by
any other solution belongs to the most interior Pareto front of solutions (with rank
equal to one); discarding this first subset of solutions, the remaining solutions that
belong to the next most interior Pareto front (not dominated by any other solution)
are determined, resulting in a second subset of solutions (with rank equal to two).
This process of ranking solutions according to the Pareto front they belong to is
repeated until there are no more solutions to be ranked. There can be as much
Pareto fronts (ranks) as solutions (in the case of one solution per Pareto front).
Following rank determination, the orders are selected for genetic manipulation
using binary tournaments. For each order of the population, an opponent order is
selected. Each of these order-opponent order pairs corresponds to one tournament
round. For each tournament round, it is decided if the existing population order is
kept or is replaced by the opponent order: if the orders solutions have different
ranks, the order with the lowest ranked solution wins; if the solution ranks are
equal, then one order is randomly chosen as winner. In the case the opponent order
wins, it replaces the existing population order. This way of selecting individuals for
genetic manipulation ensures that the global search by the GA is guided towards a
more Pareto-optimal population.
Genetic manipulation follows order selection. The genetic operators to be
applied to the selected orders are order recombination and mutation. The goal of the
recombination process is that information regarding relative project positions in the
orders that lead to more interior Pareto solutions is exchanged between population
orders (note that the selection that precedes manipulation favors lower ranked, more
Pareto-optimal solutions). In turn, the mutation process aims at introducing ran-
domness that might contribute to improve the Pareto-wise quality of the solutions
(i.e., more interior solutions, namely non-dominated solutions).
The recombination works in the following way: the population of orders is
divided into pairs, with each order only being present in one pair; for each pair,
recombination will be undertaken (or not) according to a given probability; if a pair
is to be recombined, a crossover point is chosen; finally, the positions of the projects
9 Optimal Planning of Grid Reinforcement with Demand … 261

(a) (b) (c)

Fig. 9.4 Example of order recombination, starting from the third project position (a): relative
project positions to be respected in each order (b) and offspring orders that result from
recombination (c)

(a) (b)

Fig. 9.5 Example of order mutation: chosen project positions (a) and resulting order (b)

in one order of the pair are altered in order to satisfy the relative project positions
imposed by the other order of the pair, starting from the crossover point of the latter.
As an example, suppose that the orders of Fig. 9.3 have been divided into the
pairs (o1 , o4 ) and (o2 , o3 ). According to the recombination probability, it was
decided that the orders of the pair (o2 , o3 ) were to be recombined. Also, a crossover
point was randomly chosen2 so that recombination takes place starting from the
third project position. The relative project positions in one order of the pair, starting
from the crossover point, impose relative project positions to be respected in the
other order (Figs. 9.4a, b). The orders would have then been changed, with the
resulting project positions being the ones in the offspring orders (indicated in
Fig. 9.4c).
The next step is the mutation of the orders. This process occurs for each order of
the population with a given probability. If it is decided that an order is to be
mutated, then two distinct project positions need to be chosen (e.g. randomly). The
projects in such positions are then swapped. An example of order mutation is shown
in Fig. 9.5.
After genetic manipulation of the orders, a new genetic population is created
(population of the next generation). This population will then be subjected to
evaluation by the GS and to selection and manipulation in order to create another
population. The whole process of population evaluation, selection and manipulation
is repeated until some stop or convergence criteria is met.
The steps of the various methods described (Hybrid Genetic Algorithm,
Gaussian Search, Optimal Power Flow, Binary tournaments, Recombination and
Mutation) are presented below.

2
For N projects, meaningful values of the crossover point are in the range of 2 to N-1.
262 A. M. F. Dias and P. M. S. Carvalho

(M1) Hybrid GA search

Step 1 Initialize a set O of possible orders for project evaluation (population of
project orders).
Step 2 Evaluate the population of orders by running GS (method M2) on each
order k of O (ok).
Step 3 Determine the rank of each order of O according to the non-dominated
sorting described in [27].
Step 4 Select the best orders from the population using binary tournaments
(sub-method S1).
Step 5 Subject the best orders to genetic manipulation: recombination
(sub-method S2) and mutation (sub-method S3).
Step 6 Go back to Step 2 until convergence is achieved.
(M2) Gaussian Search
Step 1 Initialize i (i ¼ 1).
Step 2 Take the ith project of project order ok and choose the optimal timing for
its implementation, ti , according to an OPF method considering DR
control (method M3);
Step 3 If the optimal timing ti is not T + 1, add the project to the network graph
G.
Step 4 Increment i (i i þ 1); go back to Step 2 until the Nth project of ok is
reached.
Step 5 If the end of the project order is reached, go back to Step 1 until the
project timing array t ¼ ½t1 ; t2 ; . . .; tN  remains unchanged from the
previous GS iteration (Steps 1–4).
(M3) DR optimized power flow
Step 1 Run power flow for unconstrained network loading.
Step 2 If the current rating of a network branch (line or cable) is violated, go to
Step 3, otherwise go to Step 7.
Step 3 Select the most downstream branch (or one of the most downstream
branches) with current rating violations.
Step 4 For all consumers downstream the selected branch having a control
enabling device, decrease demand power of the consumers with highest
demand power according to a predefined rate; the shed demand power is
postponed (load shifting); if no (further) decrease of demand power is
possible, go to Step 6.
Step 5 Run power flow. If the current rating of the branch is still violated, go
back to Step 4, otherwise continue.
Step 6 If there is a branch with current rating violations at the same network
topology level of the branch being considered that has yet not been
selected, select that branch; otherwise, if there is at least one branch with
current rating violations at a level upstream the branch that was being
9 Optimal Planning of Grid Reinforcement with Demand … 263

considered, select a branch at the closest level upstream that branch. If a

branch is selected, go back to Step 4.
Step 7 If the voltage level at a node is violated, go to Step 8, otherwise stop.
Step 8 Select the most downstream node (or one of the most downstream
nodes) with voltage level violations.
Step 9 For all consumers downstream the selected node having a control
enabling device, decrease demand power of the consumers with highest
demand power according to a predefined rate; the shed demand power is
postponed (load shifting); if no (further) decrease of demand power is
possible, go to Step 11.
Step 10 Run power flow. If the voltage level of the node is still violated, go back
to Step 9, otherwise continue.
Step 11 If there is a node with voltage violations at the same network topology
level of the node being considered that has yet not been selected, select
that node; otherwise, if there is at least one node with voltage level
violations at a level upstream the node that was being considered, select
a node at the closest level upstream that node. If a node is selected, go
back to Step 9, otherwise stop.
(S1) Binary tournaments
Step 1 For each individual of the population (order ok), randomly select an
opponent order from the same population.
Step 2 For each individual of the population and its opponent order, decide the
winner of the tournament round: if the orders have different ranks, the
order with the lowest rank is the winner, otherwise the winning order is
one of the two orders chosen at random.
Step 3 For each individual of the population (order ok), if the winning order of
the corresponding tournament round is the opponent order, it replaces
the existing order.
(S2) Recombination
Step 1 Randomly divide the population in pairs, with each individual of the
population (order ok) only being present in one single pair.
Step 2 Select the first pair.
Step 3 Decide if the pair is to be recombined with a probability of precomb; if the
pair is to be recombined, go to Step 4, otherwise go to Step 6.
Step 4 Randomly determine the crossover point of the pair.
Step 5 Change the positions of the projects in each order of the pair so that the
exchanged relative positions of projects, beginning from the crossover
point position to the last project position, are respected.
Step 6 Select the next pair and go back to Step 3 until there are no more pairs
left to recombine.
264 A. M. F. Dias and P. M. S. Carvalho

(S3) Mutation
Step 1 Select the first order of the population (order o1).
Step 2 Decide if the order is to be mutated with a probability of pmut; if the
order is to be mutated, go to Step 3, otherwise go to Step 5.
Step 3 Select two distinct project positions at random.
Step 4 Swap the projects in the selected positions.
Step 5 Select the next order of the population and go back to Step 2 until there
are no more orders left to mutate.

9.3 Case Study

A Low-Voltage (LV) electric distribution network is considered as an investment

case study. The network has a nominal voltage of 400 V and is comprised of thirty
six nodes and thirty five cables (underground network with radial topology, see
Figs. 9.6 and 9.7). Several network reinforcement and ICT investments are con-
sidered as possible projects:
15
34
14
35
30 17 16
31
5 13
12 32
4 26 20 33 18
5 13 27
3 6
7 6 12 19
7 28
8 18 23 29 21
19
11 20
24 22
2 21
14 22 25

10 15 27 26
8
1 1 2 3 9 16 29
9 23
32 28 24
30
25
4 31

34 17 35
10
33
11
36

Fig. 9.6 Schematic representation of the network nodes, cables and consumers. Current rating
violations are observed for cables 1 (feeder), 3 and 8, hence their numbers are highlighted in color.
Nodes with connected consumers are colored according to the combined contracted power at the
node: less or equal to 20 kVA (green), greater than 20 kVA and less or equal to 40 kVA (yellow),
greater than 40 kVA (orange)
9 Optimal Planning of Grid Reinforcement with Demand … 265

Fig. 9.7 Geographic view of the LV network of the case study, in which possibilities for
investment in grid reinforcement (R) and DR reinforcements (ICT) are marked. The triangle refers
to the secondary MV/LV substation

• The reinforcement projects are the replacement of three existing weak cables by
185 mm2 Aluminum conductors (reinforcements denoted by R’s in Fig. 9.7).
• The ICT projects are smart meter investments at the premises of every con-
sumer, capable of enabling DR control of such consumers (smart meters denoted
by ICT’s in Fig. 9.7).
ICT investment is considered as a separate and normalized investment. Being
considered as a separate normalized investment (and not minimized together with
traditional investment) allows us to trade-off the two types of investment without
establishing a cost for ICT, which would be very difficult as the use of ICT for DR
control is not yet widespread as a mature technology.
266 A. M. F. Dias and P. M. S. Carvalho

The network has both single- and three-phase consumers (70 single-phase and 30
three-phase consumers) with contracted power values ranging from 1.15 to
20.7 kVA. The most common contracted power values for single-phase consumers
are 3.45 and 6.9 kVA (28% and 22% of all the consumers), while for three-phase
consumers these values are 10.35 and 17.25 kVA (12% and 8% of all the consumers).
Consumer load demand power is calculated for each consumer using historical
smart meter data of a consumer with the same characteristics (identical number of
connection phases and contracted power). Both single- and three-phase consumer
demand power is evenly distributed among the three phases (for balanced load
flow). Unity power factor is assumed. Consumer locations at network nodes are
taken from real network data.
The operational conditions in which the case study is run are
• DR simulation starts at peak load time; whenever there is ICT equipment (smart
meters) installed at the consumers’ premises, the allowed demand power of control
enabled consumers can be decreased in steps of 1.15 kVA, as much as needed in
order to minimize violations of cable current ratings or node voltage levels.
• DR is simulated for a whole day in each stage of the planning horizon. Such
days are assumed to be representative of the corresponding years (stages).
• The MV/LV substation secondary voltage is set to 1.05 pu.
• Monitoring equipment is assumed to be installed at secondary substation level
so that cable currents can be measured and communicated to the ICT control
system. The costs associated with the installation and operation of monitoring
equipment are not considered.
A ten year planning horizon is considered with 11 possible investment stages
(T = 10), being the investment stages evenly distributed in time (one stage per
planning horizon year, plus one stage outside the planning horizon meaning that the
project is not to be undertaken within the horizon). Two objective functions are
optimized (minimized) together:
• Network reinforcement investment
• ICT (smart meter) investment
The inflation and discount rates considered are 3 and 10%, respectively.
The optimization algorithm is run for an initial population of fifty random project
orders and five population generations, and for recombination and mutation prob-
abilities of 80 and 10%, respectively.
From the obtained set of solutions (one optimized project schedule per project
order), the non-dominated solutions were selected, and those with current or voltage
violations were excluded, as they were considered unfeasible. The remaining
non-dominated solutions were projected onto the objectives of ICT investment and
reinforcement investment. Four zones corresponding to four project allocation
modes were identified. The projected non-dominated solutions and the identified
zones are shown in Fig. 9.8.
The solution with highest network reinforcement cost (2400 €) and no ICT
investment corresponds to the reinforcement of all the three cables where current
9 Optimal Planning of Grid Reinforcement with Demand … 267

14
4 3 2 1
12

10
ICT investment (pu)

0
0 500 1000 1500 2000 2500
Network reinforcement (€)

Fig. 9.8 Projection of the obtained non-dominated solutions onto the objectives of ICT
investment and reinforcement investment for the LV electric distribution network

rating violations occur (cables 1, 3 and 8, referring to Fig. 9.6). The reinforcement
of such cables is enough to ensure violation-free operation of the network within the
planning horizon.
As ICT projects are allocated, DR control is possible and the reinforcement
projects are postponed, making the reinforcement cost lower (considering inflation
and discount rates, the present cost is lowered). The transition from zone 1 to 2
occurs when the allocated ICT projects are enough to avoid one reinforcement
project (reinforcement of cable number 8).
Cables number 3 and 8 are almost identical (in length and cross section), and
therefore when the OPF method reduces demand power downstream cable number
8, solving its current rating violations and thus avoiding its reinforcement, it also
removes most of contributions to the violations of the current rating of cable
number 3. As more ICT projects are allocated and more consumer demand power is
shifted by the OPF, the reinforcement of cable number 3 is also avoided and
transition from zone 2 to 3 occurs.
With the reinforcement of cables 3 and 8 avoided, what is left is the rein-
forcement of cable number 1 (feeder cable). Similarly to what happens in zones 1
and 2, allocating more ICT projects and shifting consumer demand allows post-
poning the reinforcement of cable 1. Since there are only ten stages (years) in the
considered planning horizon, the reinforcement of cable 1 can only be postponed up
to the last stage of the planning horizon.
Having enough ICT projects allocated and enough consumer demand power to
be shifted makes the transition from zone 3 to 4 possible—the reinforcement of the
feeder cable (and of the other two cables) is avoided. The optimal solution where all
268 A. M. F. Dias and P. M. S. Carvalho

cable reinforcements are avoided corresponds to about 12 pu of investment in ICT

projects (present cost) and 18 consumers (18% of all consumers) with DR enabling
devices.
In Figs. 9.9 and 9.10, the distribution of the reinforcement and ICT investments
according to their optimal timings for the two extreme solutions is illustrated. In
Fig. 9.9, the allocation of the three reinforcement projects matches the stages when
violations would first occur in each of the cables. On the other hand, it can be
observed from Fig. 9.10 that the allocation of ICT projects is gradual and not
abrupt, varying between 0.9 and 2.0 pu per stage of the planning horizon. This can
also be verified through the cumulative ICT investment cost, whose evolution is
very well fitted to a first-degree polynomial curve.
Figures 9.11, 9.12 and 9.13 correspond to the solutions preceding the transitions
from zone 1 to 2 (Fig. 9.11), from zone 2 to 3 (Fig. 9.12) and from zone 3 to 4
(Fig. 9.13). From such figures, one can observe which reinforcement projects have
been postponed or avoided so far, the limits for postponing reinforcement projects
in the solutions found by the described hybrid approach and, similarly to Fig. 9.10,
that ICT project allocation is gradual.
Present investment cost (€)

3000
Reinf
Cumulative
2000

1000

0
1 2 3 4 5 6 7 8 9 10 11
Investment project number

150

100

50

1 2 3 4 5 6 7 8 9 10 11
Stage

Fig. 9.9 Optimal investment stages and investment costs for reinforcement projects (blue dots and
bars) of the solution where all the necessary reinforcement projects and no ICT projects are allocated
9 Optimal Planning of Grid Reinforcement with Demand … 269

Present investment cost (pu)

15
ICT
Cumulative
10

0
1 2 3 4 5 6 7 8 9 10 11
Investment project number

150

100

50

1 2 3 4 5 6 7 8 9 10 11
Stage

Fig. 9.10 Optimal investment stages and investment costs for ICT projects (green dots and bars)
of the solution where the allocation of enough ICT projects avoids all cable reinforcements

The projection of the solutions in Fig. 9.8 can be looked at from another point of
view: by doing a translation of the horizontal axis, the solutions can be seen from
the perspective of avoided network reinforcement (due to the investment in ICT
equipment) instead of required network reinforcement. The result of this translation
is shown in Fig. 9.14. In this figure, the dashed lines connect the axes origin to the
projection of certain solutions (marked within circles). These are the four solutions
where ICT is most valuable in terms of avoided network reinforcement per installed
ICT unit (€/pu of ICT)—the more horizontal the line connecting the solution
projection to the axes origin, the higher the average ICT unit value.
For the extreme solution where enough ICT investment leads to complete
avoidance of traditional network reinforcement, the current profile of the cables
whose reinforcement is avoided, before and after investment optimization with DR,
is shown in Figs. 9.15 and 9.16. It can be observed from these figures that DR
control affects load demand power (and thus cable current) in two essential periods:
from 14 to 16 h, when the current rating of the feeder cable would be violated, and
from 20 to 24 h, when the ratings of the three cables would also be exceeded (blue
lines). The excess demand power is shifted to later periods, decreasing cable
loading in those two periods, which is offset by a slight increase in cable loading in
the period of 17 to 19 h and a significant increase of such loading in the period of 1
to 5 h.
Changes occur in the load demand profiles of the consumers with DR enabling
devices. Two groups of profile changes are identified: consumers whose profiles are
affected by demand power shed in the period of 20 to 1 h and shifted to the period
270 A. M. F. Dias and P. M. S. Carvalho

2500

Present investment
2000 Reinf
Cumulative

cost (€)
1500

1000

500

0
1 2 3 4 5 6 7 8 9 10 11
4
Present investment

ICT
3 Cumulative
cost (pu)

0
1 2 3 4 5 6 7 8 9 10 11
Investment project

150
number

100

50

1 2 3 4 5 6 7 8 9 10 11
Stage

Fig. 9.11 Optimal investment stages for reinforcement and ICT projects (blue and green dots,
respectively) and investment costs for reinforcement and ICT projects (blue and green bars,
respectively) of the solution preceding the transition from zone 1 to 2 in Fig. 9.8. This solution
corresponds to the limit found for postponing the reinforcement of cable number 8 (stage 10),
while also reinforcing cables number 1 and 3

of 1 to 5 h (named DR pattern number 1) and consumers whose profiles, besides

being affected by the same demand power shifting, are also affected by the post-
poning of demand power in the period of 14 to 17 h to the period of 16 to 19 h
(named DR pattern number 2).
In Table 9.1, for the consumers affected by DR control (eighteen out of the total
of one hundred), their network connection node, contracted power and DR pattern
number are summarized. Figure 9.19 is similar to Fig. 9.6, with the differences
being that the colored network nodes represent the nodes where consumers affected
by DR patterns number 1 (blue) and number 2 (yellow) are located. In Figs. 9.17
and 9.18, two examples of optimized consumer demand profiles due to DR control
(one example per DR pattern) are shown.
By comparing Figs. 9.6 and 9.19, it can be observed that there is a correlation
between nodes with higher combined contracted power and the corresponding
9 Optimal Planning of Grid Reinforcement with Demand … 271

Present investment
2000
Reinf
1500 Cumulative

cost (€)
1000

500

0
1 2 3 4 5 6 7 8 9 10 11

6
Present investment

ICT
Cumulative
cost (pu)

0
1 2 3 4 5 6 7 8 9 10 11
Investment project

150
number

100

50

1 2 3 4 5 6 7 8 9 10 11
Stage

Fig. 9.12 Optimal investment stages for reinforcement and ICT projects (blue and green dots,
respectively) and investment costs for reinforcement and ICT projects (blue and green bars,
respectively) of the solution preceding the transition from zone 2 to 3 in Fig. 9.8. This solution
corresponds to the limit found for postponing the reinforcement of cable number 3 (stage 8), while
reinforcing cable number 1 and avoiding the reinforcement of cable number 8

consumers having DR enabling equipment—most of the nodes concerning con-

sumers with DR equipment (blue and yellow colored nodes in Fig. 9.19) corre-
spond to nodes with a combined contracted power greater than 20 kVA (yellow and
orange colored nodes in Fig. 9.6). This is expected, as more consumers connected
to the same node (higher combined contracted power) means higher aggregate
loading demand power, thus enabling DR control for consumers at such nodes will
decrease upstream cable loading when violations would occur.
It should be noted that these results were obtained for given reference costs of
cable reinforcement and a given load density (respectively, around 40k € per
kilometer of cable replacement and a contracted power density of 1900 kVA per
kilometer of feeder length). For different cable reference costs and contracted power
densities, the valuation of ICT investment would also be different.
If the same load density is considered with varying reference reinforcement
costs, the ICT valuation would be directly proportional to such costs. If not, if the
272 A. M. F. Dias and P. M. S. Carvalho

1500

Present investment
Reinf
Cumulative

cost (€)
1000

500

0
1 2 3 4 5 6 7 8 9 10 11

15
Present investment

ICT
Cumulative
cost (pu)

10

0
1 2 3 4 5 6 7 8 9 10 11
Investment project

150
number

100

50

1 2 3 4 5 6 7 8 9 10 11
Stage

Fig. 9.13 Optimal investment stages for reinforcement and ICT projects (blue and green dots,
respectively) and investment costs for reinforcement and ICT projects (blue and green bars,
respectively) of the solution preceding the transition from zone 3 to 4 in Fig. 9.8. This solution
corresponds to the limit found for postponing the reinforcement of cable number 1 (stage 10),
while avoiding the reinforcement of cables number 3 and 8

network has the same reference costs but different load density, the ICT valuation
would be inversely proportional to the load density. For instance, with a higher load
density (same aggregate contracted power and lower network length) the installa-
tion of ICT infrastructure enables the management of the same load demand power
as in a lower load density situation (higher network length), while the reinforcement
expected to cope with that load demand power would be lower in length (lower in
cost), and in that way the ICT valuation would be lower.
9 Optimal Planning of Grid Reinforcement with Demand … 273

14

12

10
ICT investment (pu)

0
0 500 1000 1500 2000 2500
Avoided network reinforcement (€)

Fig. 9.14 Projection of the obtained non-dominated solutions onto the objectives of ICT
investment and reinforcement investment for the LV electric distribution network

400
Feeder rating
350 Feeder current (unc)
Feeder current (opt)

300
Feeder current (A)

250

200

150

100

50

0
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24
time (h)

Fig. 9.15 Current in the feeder cable (year 10); due to the combination of network monitoring
with DR control as enabled by ICT (peak shifting), feeder cable current does not exceed the cable
current rating (dashed line), allowing its reinforcement to be avoided
274 A. M. F. Dias and P. M. S. Carvalho

300
Cable rating
Cable current (unc)
250 Cable current (opt)

200
Line current (A)

150

100

50

0
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24
time (h)

300
Cable rating
Cable current (unc)
250 Cable current (opt)

200
Line current (A)

150

100

50

0
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24
time (h)

Fig. 9.16 Current in cables number 3 (top) and 8 (bottom), in the last stage of the planning
horizon (year 10); due to the combination of network monitoring with DR control as enabled by
ICT (peak shifting), cable current does not exceed cable current rating (dashed line), allowing the
corresponding reinforcements to be avoided
9 Optimal Planning of Grid Reinforcement with Demand … 275

Table 9.1 Summary of the network connection nodes, contracted power and DR pattern number
of the consumers affected by DR control (eighteen consumers)
Consumer Connection node Contracted power (kVA) DR control pattern
7 6 20.70 2
14 7 13.80 1
19 15 6.90 2
27 16 6.90 2
28 16 6.90 2
33 17 10.35 1
42 18 10.35 1
49 21 17.25 1
51 22 5.75 1
53 22 10.35 1
58 23 17.25 1
65 25 4.60 1
68 25 4.60 1
80 27 6.90 2
90 31 6.90 2
91 31 6.90 2
93 32 6.90 2
100 36 4.60 1

Fig. 9.17 Consumer load demand profile optimization (year-10) as a result of DR as enabled by
ICT (smart meters): consumer 58 as an example of DR pattern number 1
276 A. M. F. Dias and P. M. S. Carvalho

Fig. 9.18 Consumer load demand profile optimization (year-10) as a result of DR as enabled by
ICT (smart meters): consumer 80 as an example of DR pattern number 2

15
34
14
35
30 17 16
31
5 13
12 32
4 26 20 33 18
5 13 27
3 6
7 6 12 19
7 28
8 18 23 29 21
19
11 20
24 22
2
21
14 22 25

10 15 27 26
8
1 1 2 3 9 16 29
9 23
32 28 24
30
25
4 31

34 17 35
10
33
11
36

Fig. 9.19 Schematic representation of the network nodes, cables and consumers affected by DR
control. Current rating violations are observed for cables 1 (feeder), 3 and 8, hence their numbers
are highlighted in color. Nodes with DR enabled consumers are colored according to the DR
control pattern: pattern number 1 (blue) and pattern number 2 (yellow)
9 Optimal Planning of Grid Reinforcement with Demand … 277

Appendix

The notations used throughout this chapter are listed below:

fj Objective function j
P Set of investment projects
O Set of orders for project analysis (population)
pi Project i of P
ok Order k of O (individual of the population)
ti Timing of project pi, ti 2 f1; 2; . . .; T þ 1g
t Decision schedule: indexed array of timings ti for projects pi
N Number of projects
T Number of stages of the planning horizon
G Graph of the electric distribution network

References

1. M.V.F. Pereira, L.M.V.G. Pinto, S.H. Cunha, G.C. Oliveira, A decomposition approach to
automated generation transmission expansion planning. IEEE Trans. Power Syst. PAS-104
(11), 3074–3083 (1985)
2. R. Romero, A. Monticelli, A Hierarchical decomposition approach for transmission network
expansion planning. IEEE Trans. Power Syst. 9(1), 373–380 (1994)
3. R. Romero, A. Monticelli, A zero-one implicit enumeration method for optimizing
investments in transmission expansion planning. IEEE Trans. Power Syst. 9(3), 1385–1391
(1994)
4. G.C. Oliveira, A.P.C. Costa, S. Binato, Large scale transmission network planning using
optimization and heuristic techniques. IEEE Trans. Power Syst. 10(4), 1828–1833 (1995)
5. R. Romero, R.A. Gallego, A. Monticelli, Transmission expansion planning by simulated
annealing. IEEE Trans. Power Syst. 11(1), 364–369 (1996)
6. H. Rudnick, R. Palma, E. Cura, C. Silva, Economically adapted transmission systems in open
access schemes—application of genetic algorithms. IEEE Trans. Power Syst. 11(3),
1427–1440 (1996)
7. R.A. Gallego, A. Monticelli, R. Romero, Comparative studies on non-convex optimization
methods for transmission network expansion planning. IEEE Trans. Power Syst. 13(3),
822–828 (1998)
8. X. Wang, Y. Mao, Improved genetic algorithm for optimal multistage transmission system
planning. IEEE (2001)
9. L.L. Garver, Transmission network estimation using linear programming. IEEE Trans. Power
Syst. PAS-89(1), 1688–1697 (1970)
10. A. Monticelli, A. Santos, M.V.F. Pereira, S.H. Cunha, B.J. Parker, J.C.G. Praça, Interactive
transmission network planning using a least-effort criterion. IEEE Trans. Power App. Syst.
PAS-101(10), 3919–3925 (1982)
11. M.V.F. Pereira, L.M.V.G. Pinto, S.H. Cunha, G.C. Oliveira, A decomposition approach to
automated generation/transmission expansion planning. IEEE Trans. Power Syst. PAS-104
(11), 3074–3083 (1985)
12. R. Romero, A. Monticelli, A hierarchical decomposition approach for transmission network
expansion planning. IEEE Trans. Power Syst. 9(1), 373–380 (1994)
278 A. M. F. Dias and P. M. S. Carvalho

13. R. Romero, A. Monticelli, A zero-one implicit enumeration method for optimizing

investments in transmission expansion planning. IEEE Trans. Power Syst. 9(3), 1385–1391
(1994)
14. G.C. Oliveira, A.P.C. Costa, S. Binato, Large scale transmission network planning using
optimization and heuristic techniques. IEEE Trans. Power Syst. 10(4), 1828–1833 (1995)
15. A. Escobar, R.A. Gallego, R. Romero, Multistage and coordinated planning of the expansion
of transmission systems. IEEE Trans. Power Syst. 19(2), 735–744 (2004)
16. H. Rudnick, R. Palma, E. Cura, C. Silva, Economically adapted transmission systems in open
access schemes—application of genetic algorithms. IEEE Trans. Power Syst. 11(3),
1427–1440 (1996)
17. S. Binato, M.V. Pereira, S. Granville, A new benders decomposition approach to solve power
transmission design problems. IEEE Trans. Power Syst. 16(2), 235–240 (2001)
18. R. Romero, R.A. Gallego, A. Monticelli, Transmission expansion planning by simulated
annealing. IEEE Trans. Power Syst. 11(1), 364–369 (1996)
19. E.L. Silva, J.M.A. Ortiz, G.C. Oliveira, S. Binato, Transmission network expansion planning
under a tabu search approach. IEEE Trans. Power Syst. 16(1), 62–1440 (2001)
20. S. Binato, G.C. Oliveira, J.L. Araújo, A greedy randomized adaptive search procedure for
transmission expansion planning. IEEE Trans. Power Syst. 16(2), 247–253 (2001)
21. J. Contreras, F.F. Wu, A kernel-oriented algorithm for transmission expansion planning. IEEE
Trans. Power Syst. 15(4), 1434–1440 (2000)
22. M. Pinedo, Scheduling—Theory, Algorithms, and Systems (Prentice Hall, 1995). ISBN
0-13-706757-7
23. F.S. Reis, M. Pinto, P.M.S. Carvalho, L.A.F.M. Ferreira, Short-Term Investment Scheduling
in Transmission Power Systems by Evolutionary Computation—DRPT2000 (London, April
2000)
24. F.S. Reis, P.M.S. Carvalho, L.A.F.M. Ferreira, Combining gauss and genetic algorithms for
multi-objective transmission expansion planning. WSEAS Trans. Syst. 3(1), 206–209 (2004)
25. F.S. Reis, P.M.S. Carvalho, L.A.F.M. Ferreira, Reinforcement scheduling convergence in
power systems transmission planning. IEEE Trans. Power Syst. 20(2), 1151–1157 (2005)
26. A. Dias, P.M.S. Carvalho, P. Almeida, S. Rapoport, Multi-objective distribution planning
approach for optimal network investment with EV charging control, in Presented at
PowerTech 2015 (June 2015) [Online], Available: http://ieeexplore.ieee.org/stamp/stamp.jsp?
arnumber=7232674
27. K. Deb, A. Pratap, S. Agarwal, T. Meyarivan, A Fast and elitist multiobjective genetic
algorithm: NSGA-II. IEEE Trans. Evol. Comput. 6(2), 182–197 (2002)
Chapter 10
Simultaneous Network Reconfiguration
and Sizing of Distributed Generation

Wardiah Mohd Dahalan and Hazlie Mokhlis

Abstract This chapter introduces simultaneous optimization concept of Network

Reconfiguration and Distributed Generation sizing. The main objective of the
introduced concept is to reduce the real power loss and improve the overall voltage
profile in the electric distribution network through optimal network reconfiguration
and Distributed Generation sizing, while at the same time satisfy the system
operating constraints. The meta-heuristic methods have been applied in the opti-
mization process due to its excellent capability for searching optimal solution in a
complex problem. The applied meta-heuristics methods are Genetic Algorithm,
Evolutionary Programming, Particle Swarm Optimization, Artificial Bee Colony
and their respective modified types. A detail performance analysis is carried out on
IEEE 33-bus systems to demonstrate the effectiveness of the proposed concept.
Through simultaneous optimization, it was found that power loss reduction is more
as compared to conducting reconfiguration or DG sizing approach alone. The test
result also indicated that Evolutionary Particle Swarm Optimization produced better
result in terms of power loss and voltage profile than other methods.

Keywords Distributed generation Optimization techniques
Power loss



reduction Reconfiguration Meta-Heuristic method

W. M. Dahalan
Department of Electrical Engineering, Universiti Kuala Lumpur,
Kuala Lumpur, Malaysia
e-mail: [email protected]
W. M. Dahalan
Malaysian Institute of Marine Engineering Technology, Perak, Malaysia
H. Mokhlis (&)
Department of Electrical Engineering, University of Malaya,
Kuala Lumpur, Malaysia
e-mail: [email protected]

© Springer Nature Singapore Pte Ltd. 2018 279

F. Shahnia et al. (eds.), Electric Distribution Network Planning,
Power Systems, https://doi.org/10.1007/978-981-10-7056-3_10
280 W. M. Dahalan and H. Mokhlis

10.1 Introduction

The major issues nowadays that most power utilities are trying to achieve in
generation, transmission and distribution systems are to ensure high service quality,
reliability and efficiency of the overall power system. Distribution system is the
final stage in the process of power delivering power from the generation to the
individual customer. It has contributed the greatest amount of power loss in which
resulted to poor voltage magnitude. The performance of electric distribution net-
works becomes inefficient due to the increase in power loss and reduction in voltage
magnitude especially in the heavily loaded network. The studies in [1, 2] report that
70% of the total losses are occurring in the electric distribution networks while
transmission and sub transmission lines account for only 30% of the total losses is
as shown in Fig. 10.1.
One of the well-known techniques to minimize power loss is through network
reconfiguration [3]. This technique can reduce the power loss and improve the
overall voltage profile provided that the most optimum configuration can be
determined. The application of network reconfiguration in the electric distribution
networks can be divided into two categories; planning and operation. In planning,
network reconfiguration is needed to identify the best configuration via changing
the on/off sectionalizing and tie-switches in the network. By doing that, the load
will transfer from heavily loaded feeders to relatively less heavily loaded feeders, so
that the power loss is minimized. Meanwhile, the network reconfiguration during
operation plays an important role in the process of rerouting power supply in the
network due to a fault. In this context, reconfiguration is required to restore power
supply automatically and quickly to un-faulted sections of the system to improve
the system reliability. Instantaneous response to the damage system prevents it from
propagating allowing as many loads as possible to function. Network reconfigu-
ration requires optimization technique to determine the best combination set of
switches to be open. The execution of the process of selection should fulfill the

Fig. 10.1 Illustration of a power system delivering power to customers

10 Simultaneous Network Reconfiguration and Sizing … 281

requirement of optimization (minimize power loss) and satisfying the operating

constraints. The application of reconfiguration network is much simpler and cost
efficient compared to other techniques. In general, reconfiguration have two primary
aim which are to provide the maximum amount of electrical supply to the end
customers and reconfigure the network automatically as soon as there are problems
arises such as fault.
Another technique that able to reduce power loss is by interconnecting local
power supply in electric distribution networks. By having local supply, electrical
power can be delivered to the load in a short distance, which able to reduce power
loss. Local power supply from renewable energy sources such as mini-hydro, wind,
solar and Bio-fuel are nowadays is connected in networks to generate electrical
power. In view of this matter, a new identity appeared in the electric distribution
networks known as “Distributed Generations” (DG). DG is related to the use of
small generating units installed at strategic points on the networks and mainly close
to the load centers. It can be used in an isolated way, supplying the consumer’s
local demand, or in an integrated way, supplying energy to the remaining system
[4]. In general, DG is the generation of electricity by facilities smaller than the
central plants, usually 10 MW or less [5].
From studies, DG penetration is predicted will surpass more than 25% of the
total generation in the foreseeable future [6]. Studies also revealed that the usage of
Renewable Energy DG could reduce 60% of the carbonic pollution from conven-
tional power generation by 2050 [7]. With this regard, changing the environment of
power systems design and operation has caused the need to consider active dis-
tribution network. The integration of distribution system would lead to the
improvement of the voltage profile, load balancing, reliability such as service
restoration and increase energy efficiency. Therefore, it is very crucial to ensure that
the DG size is at the optimal value to maximize its benefits. An inappropriate size of
DG will cause of power loss in the system to be higher than the initial configuration.
Many researchers have employed various methods to overcome the problem of
optimal reconfiguration and DG sizing in the electric distribution networks [8–16].
However, the existing methods have some limitations and drawbacks in their
solutions such as the solution might trap in the locally optimal solution. This is due
to the process of finding the optimum solution used sequential approach (e.g.: find
the optimal DG first and followed by optimal reconfiguration or vice versa).
Moreover, the reconfiguration is a complicated combinatorial and non-differentiable
constrained optimization problem. It involves with many candidate-switching
combinations. These obstacles really put the reconfiguration process in difficulties
to achieve the comprehensive optimal solution and take too long time to reach the
convergence point. Apart from that, the combination switches obtained from their
results sometimes is not in radial which is important characteristic in assist finding
optimal power loss.
Although there are various methods for network reconfiguration, the DG effect
in the network reconfiguration has not been considered widely by researchers.
There are very few researchers who considered network reconfiguration with DG
[17–19]. Most of them have already fixed the size of DG or solve sequentially and
282 W. M. Dahalan and H. Mokhlis

the impact of DG on the distribution system has not been discussed seriously. None
of them tried to solve the reconfiguration and DG problem simultaneously.
The installation of DGs in the distribution system indeed can improve energy
efficiency and voltage profile, and at the same time minimize power interruption
power. However, in order to ensure the effectiveness of DG in the distribution
system, selecting the optimal size of DG plays an important role in giving the
greatest impact on the operations and control of the electric distribution networks.
Thus, the correct size (dispatch value) of DG becomes a vital point for the network
system in order to produce a lower amount of power waste. In literature, the term
DG size is commonly applied to represents the power dispatch value of the DG.
Thus, throughout this chapter, DG size refers to the DG dispatch value.
Considering the existing limitations, this chapter introduces simultaneous opti-
mization concept in finding optimal network reconfiguration and DGs size. The
meta-heuristics methods used in chapter are Genetic Algorithm (GA), Evolutionary
Programming (EP), Particle Swarm Optimization (PSO) and Artificial Bee Colony
(ABC). Meanwhile, the modified versions of these methods which contribute
greatly to this chapter are Modified Genetic Algorithm (MGA), Evolutionary
Particle Swarm Optimization (EPSO), Modified Particle Swarm Optimization
(MPSO) and Simplified Artificial Bee Colony (SABC).

10.2 Optimal Network Reconfiguration and Distributed

Generation Sizing

Electric distribution network reconfiguration can be seen as a combinatorial opti-

mization problem, comprising distribution system planning, loss minimization and
energy restoration. Generally, the network reconfiguration is defined as altering the
topological structure of distribution feeders by changing the opened or closed state
of sectionalization and tie switches (to transfer load from heavily loaded feeders to
relatively less heavily loaded feeders) so that the power loss is minimized and at the
same time constraints are met. These two types of switches are designed for both
protection and configuration management. It is normally being configured radially
for effective coordination of their protective systems. Network is reconfigured to
reduce the system power loss (network reconfiguration for power loss reduction),
improve the voltage profiles and relieve overloads in the network (network
reconfiguration for load balancing) and finally increase energy efficiency of the
system. This operation transfers load from one feeder to another, which will sig-
nificantly improve the operating condition of the overall system. In order to deal
with these problems, several methods such as GA [18], PSO [19], EP [20, 21] and
ABC [14] have been applied in network reconfiguration.
The existing of DG in the system will allow the network to contribute in sup-
plying the most optimum power to the load. However, selecting the optimal size of
DG plays an important role to avoid any drawback to the network. The connection
10 Simultaneous Network Reconfiguration and Sizing … 283

Fig. 10.2 Power loss

dependence on DG size [22]

of high capacity and excess number of DG units to electrical power system will lead
to very high power loss [22]. When DG is accessed to the distribution network, it
can be simplified into 3 different scenarios: in scenario-1, the loads at each bus are
all greater than power generation of each DG. In scenario-2, the total loads are
greater than the total power generation of DG, while in scenario-3, the total loads
are less than the total power generation of DG. For scenario-1, DG’s access can
reduce the power loss of all lines. However, in scenario-2, DGs access may increase
the power loss of some lines, but the total power loss reduces. Meanwhile, in
scenario-3, if the total power generation is less than two times of the total loading,
the influence is the same as the scenario-2, or DGs access will increase the power
loss. However, if the total power generation is in a high proportion of the system, it
will bring down the power quality.
There is a tendency for losses to follow the U-shape trajectory as shown in
Fig. 10.2 [22]. Specifically, losses begin to decrease when connecting small
amounts of DG size until they achieve their minimum level. If the DG increases
then losses begin to rise. Thus, it is worth pointing out that at high DG sizes, losses
can become larger than those without DG connected. In this chapter, the DG size
varies from 0 to 5 MW. According to the U-Shape when the DG size is larger than
the B point value, the power loss in the system has become larger than A, which is
the initial value. This factor makes the optimal size of DG become an important
consideration for the network to have lower power loss value. Thus, it can be seen
that DGs access may reduce or increase the power loss depends on the size of DG
and the network structure. The use of the reconfiguration method in cooperating
with the DG units with appropriate size can help the system to have a much lower
power loss in the distribution system.
284 W. M. Dahalan and H. Mokhlis

10.3 Problem Formulation

Reconfiguration techniques in the distribution network will change the direction of

power flow throughout the network. In this chapter, the main objective for doing the
reconfiguration is to obtain the minimum active power loss in the system based on
active current formulation. Therefore, the objective function is:
( )
X
n
2
Minimise Plosses ¼ I l j kl R l ð10:1Þ
l¼1

where:
I Number of lines in the system.
II Line real active current.
RI Line resistance.
kI is the variable that represents the topology status of the branches (1 = close,
0 = open).
The technical constraints that must be considered for the reconfiguration opti-
mization are:
(a) Distributed Generator operation:

pmin
i  pdg;i  pmax
i ð10:2Þ

where Pmin
i and Pmax
i are the lower and upper bound of DG output and all DG units
shall function within the acceptable limit.
(b) Power injection:

X
k
PDG \ðPLoad þ PLosses Þ; k ¼ no: of DG ð10:3Þ
i¼1

In order to avoid problem in protection setting, extra power from DG units are
not allowed to be injected into the main grid (Substation). At all time the total
power output from DG units should be less than the total load demand in the
electric distribution network. Thus, there will be a power supply from the main grid
to the network at all time.
(c) Power balance:

X
k
PDG þ PSubstation ¼ PLoad þ PLosses ð10:4Þ
i¼1
10 Simultaneous Network Reconfiguration and Sizing … 285

The sum of power generated from DG units and power from substation must be
equal to the summation of power load and power loss. This is to comply with the
principle of equilibrium in power generation and load demand concept.
(d) Voltage bus:

Vmin  Vbus  Vmax ð10:5Þ

The voltage for each bus should operate within the acceptable limit which is in
between 1.05 and 0.95 (±5% of rated value).
(e) Radial configuration:
The radiality of the network should be maintained throughout the reconfigu-
ration process. In order to ensure radial network is maintained, a set of rules has
been adopted for selections of switches [23].
a. All switches that do not belong to any loop are to be closed state.
b. All switches connected to the sources are to be closed state.
c. All switches contributed to a meshed network need to be closed state.
For the implementation of the optimization methods, the variable used for tie
switches represented by S and as for DG size is represented by PDg. The proposed
chromosome or particle can be written as
 
Xim ¼ S1 ; S2 ; . . .; SN ; pDg;1 ; pDg;2 ; . . .; pDg;k ð10:6Þ

where i = 1,2,3…m. The variable m indicates the population size from a set of
random distributions. N = number of tie switches and k = numbers of DG. If the
method only to find the optimum value of DG that can minimize the power loss, the
chromosome or particle can be written as
 
Xim ¼ pDg;1 ; pDg;2 ; . . .; pDg;k ð10:7Þ

10.4 Description of Modified Meta-Heuristic Methods

In this chapter, besides applying conventional GA [18], EP [20], PSO [23] and
ABC [24] methods, its modified version have been applied as well. The modifi-
cation of each method is summarized in Table 10.1. Details description on the
modification can be found in the respective reference.
286 W. M. Dahalan and H. Mokhlis

Table 10.1 Modification of meta-heuristic methods

Method Modification from the conventional method
Modified Basically, the steps involved in MGA are mostly similar to GA steps except a
Genetic slight difference in the mutation process. The chromosomes which consist of
Algorithm tie-switches and DG size are represented in real coded compared to binary
(MGA) [18] coded as represented in conventional method in order to increase the efficiency
and reduce the computational time. Too long string or chromosome will
increase the time consuming in the searching space for the optimum especially
when the system operates in the larger and more complex system. The
advantage of MGA is the acceleration in the searching speed because the
encoding and decoding process is not needed as required in binary-coded.
Furthermore, it is a simple design tool to treat complex constraints because the
method is close to problem spaces
Evolutionary EPSO is developed based on merging two methods PSO and EP. EPSO is
Particle proposed to improve and enhance the convergence speed of conventional PSO.
Swarm The proposed EPSO undergoes the similar steps as the traditional PSO accept
Optimization selection process part where EP employs a selection through the tournament
(EPSO) [19] scheme to choose the survivals for the next generation. Three steps involved
are as
– Combination old and new position
– sort the population based on fitness value
– select the best element from the survival particle (lower value)
With these the particles can move quickly to the optimal point compared to the
conventional PSO
Modified Generally, the steps involved in MPSO are almost similar to conventional PSO.
Particle However, the quality and efficiency of the current PSO has been slightly
Swarm modified. A new parameter (bold) is inserted into the original PSO Equation as
Optimization shown below;    
(MPSO) [23] Vjk þ 1 ¼ x  Vjk þ C1  rand1  Pkbestj  Xjk þ C2  rand2  Gkbest  Xjk
 
þ C3 rand3  Bkbest  X kj
The purpose of the additional new parameter is to avoid the fitness value being
trapped in local optima and increasing the exploration capability of particles in
the search space. Therefore, the exploration and exploitation capability of
MPSO is improved and provide the best solution quality and consistent results
near to the global optimum
Simplified The operation of the SABC is nearly similar to the original ABC. A slight
Artificial Bee modification in term of the searching for new food sources procedure has been
Colony implied on this simplified mode. A new and better concept of changing
(SABC) [24] information between the bees in the population is applied. The following
equations show a new searching area that is used by Employed and Onlooker
Bees in the SABC algorithm
Bi;randðDÞ ¼ SWrandðN Þ;randðDÞ
where Bi is presenting the new searching location by it bees in the current
iteration and SW is the switches. Therefore, the new switches that need to be
opened in next iteration are:
  
SW new ¼ f SW old Bi;rand ðDÞ
The implementation of Bi in the SABC will avoid the unacceptable switch
number appeared during the reconfiguration process compared to original ABC
10 Simultaneous Network Reconfiguration and Sizing … 287

10.5 Implementation of the Proposed Concept

From the base system, five different scenarios are formed to analyze the robustness and
efficiency of the proposed concept based on different meta-heuristic methods (GA, EP,
PSO, ABC, MGA, EPSO, MPSO and SABC). The lists of each scenario are as
• scenario-1: Original network as a base scenario
• scenario-2: Optimal network reconfiguration
• scenario-3: Optimal DG sizing
• scenario-4: Optimal reconfiguration and DG sizing based on Sequential
approach
• scenario-5: Optimal reconfiguration and DG sizing based on Simultaneous
approach.
The parameter used in the simulation of each algorithm depends on the charac-
teristics of the method used. However, the basic data of the network such as bus data
and line data shall be the same for all the methods. The initialization population is
determined by selecting tie switches from the set of the original tie switches as well as
the DG size. Those variables are generated randomly by the program and it is utilized
to compute the power loss in the next step. The number of population is 50 and the
maximum number iteration is set 100 which is applied and used in all the proposed
algorithms. The minimum and maximum voltages are set between 0.95 and 1.05 pu
respectively. The simulation process will stop once the results achieve maximum
number of iteration or convergence level. All the tests and simulations developed in
this work are conducted on a Personal Computer with processor Intel Core Duo CPU
3.07 GHz.
The flow chart of the reconfiguration process is illustrated in the Fig. 10.3. The
range of the DG size varies between 0 and 5 MW which is based on the literature
review [25] where most researchers use similar networks in their work.
The following steps can be used to develop programming codes to apply the
proposed simultaneously optimal network reconfiguration and DG sizing.
Step 1 Randomize N number of switches and DG output. Check either the random
number fulfils all the constraints. If yes then save the opened switches and
DG output. Else delete and re-randomize the new output.
Step 2 Evaluate the fitness functions (f (x)) for successful population:
f(x) = Power loss (Eq. (10.1)).
Step 3 Change the tie switches and sectionalizing switches through simulation
process according to the proposed methods respectively (GA, EP, PSO,
ABC, MGA, EPSO, MPSO and SABC).
Step 4 Evaluate the new fitness function (load flow analysis) and check the
radiality of the output through Graph Theory.
Step 5 Check the stopping criteria, if the iteration number > iter max or all
population give the similar values and then stops. Otherwise go to Step 3.
Step 6 Show the optimal results. End.
288 W. M. Dahalan and H. Mokhlis

Start

Read data of distribuƟon system (bus, load

and branch data), (iter=0)

IdenƟfy iniƟal configuraƟon of Ɵe and secƟonalizing switches, size of DG

(0MW- 5MW)

Evaluate the fitness (load flow analysys)

Change of Ɵe switch and secƟonalizing switch, size of DG using the proposed algorithms

Evaluate the new fitness

Iter= iter + 1

Check the radiality of the results

Stopping criteria is met? No

Iter <=maxiter

Yes

Print output results

End

Fig. 10.3 Flowchart of network reconfiguration process

10.6 Test Results of 33-Bus System

The initial test network of the 33-bus system is shown in Fig. 10.4. The overall
information of 33-bus distribution network is given in Appendix A.1.2. The net-
work consists of 33 buses, 38 lines, 5 tie switches represented by dotted lines and
3 branches (excluding the main branch). The total load of the system is 3715 kW
and 2300 kVar. In scenario-3, scenario-4 and scenario-5, three DGs unit have been
installed and placed at bus number 6, 16 and 25 [26, 27] respectively.
10 Simultaneous Network Reconfiguration and Sizing … 289

Fig. 10.4 Initial configuration of the 33-bus radial distribution system

As mentioned in the previous section, the analysis of reconfiguration involves

the network with and without DG units. In the scenario of a network with DG, the
optimal size of DG units is obtained from the simulation in which both parameters
DG size and the opened switches are adjusted during simulation simultaneously.
The size of each DG is set within the limits of the DG capacity. For example, in this
chapter, the range is set between 0 and 5 MW. The capacity depends on the type of
DG such as medium distribution generation 5 < 50 MW and large distributed
generation 50 < 300 MW [28].
290 W. M. Dahalan and H. Mokhlis

In this chapter, the DG location is assumed to be based on its suitability of

geographical location or any optimal location methods [20, 29]. Tie switch and
sectionalizing switch are considered as the main control variables. Since the process
of randomization at each iteration produces different results, there is a need to do
simulation repeated for several times in order to get the best results. Thus in this
scenario, the simulation is conducted for 30 times. The value which appears the
same for many times shall be chosen and assumed to be the best results. The result
obtained consist of the opened switches, total power loss and optimal DG sizing are
shown as in Eq. (10.6). However, the number of elements in particle or chromo-
some depends on how many tie-switch and DG used in the system. Then, the
network after reconfiguration shows the new opened switches which have produced
the lowest power loss.

10.6.1 Impact of Network Reconfiguration and DG Sizing

on Power Losses

From the analysis conducted on the simulation, the results are discussed in details.
At the initial stage (scenario-1), the network of 33-bus system is run without the
presence of reconfiguration and DG. The network has given the initial total power
loss of 202.3 kW through five initial open switches of 33, 34, 35, 36 and 37 for all
methods used. With regard to Table 10.2 in scenario-2, reconfiguration is employed
in the network of 33-bus system. The impact of reconfiguration of the power loss
reduction can be observed for all methods. The total power loss has been improved
by 34.5% for MGA, MPSO, and EPSO whenever the network reconfiguration is
applied. However, there is a slight difference in SABC method where the power
loss are improved about 32.77%.
Meanwhile, in scenario-3 the network of 33-bus system is operated using DG
which is placed on bus number 6, 16 and 25. The impacts of the DG presence are
then analyzed. The results obtained show greater power loss as compared to
scenario-2. The total power loss has been reduced between 61.9 and 66.5%. In
scenario-4, operation conducted involving both reconfiguration and DG has taken
place. However, reconfiguration process is only done upon obtaining the right size
of DG. In other word, both techniques are run sequentially. The results obtained
reveal even greater power loss reduction as compared to scenario-2 and scenario-3.
Thus, the presence of DG in the reconfiguration process has indeed caused the
reduction of power loss.
The network condition of scenario-5 is almost identical to scenario-4 except that
this time both reconfiguration and DG are being applied simultaneously. In other
words, the switches that will open and the size of DG are determined
10 Simultaneous Network Reconfiguration and Sizing … 291

Table 10.2 The overall performance of simulation results of the 33-bus distribution system
Scenario Method Opened Power loss Loss DG size (MW)
switches (MW) reduction 6 16 25
(%)
scenario-2 GA 6, 10, 14, 136.5 32.53 – – –
17, 28
PSO 7, 10, 28, 136.4 32.58 – – –
14, 32
EP 16, 5, 10, 135.2 33.17 – – –
25, 13
ABC 7, 9, 14, 139.5 31.04 – – –
32, 37
MGA 6, 9, 13, 132.5 34.50 – – –
17, 25
EPSO 7, 10, 13, 130.5 35.49 – – –
16, 25
MPSO 7, 10, 28, 132.43 34.54 – – –
14, 34
SABC 6, 9, 14, 136.0 32.77 – – –
31, 37
scenario-3 GA 33, 34, 110.6 45.33 1.4107 0.902 0.5061
35, 36, 37
PSO 33, 34, 109.6 45.82 1.0038 0.9004 0.5167
35, 36, 37
EP 33, 34, 35, 36, 37 106 47.60 0.7315 0.7224 1.0270
ABC 33, 34, 35, 36, 37 110.5 45.38 0.7540 0.5300 1.5004
MGA 33, 34, 104.0 48.59 1.0190 0.9120 0.5061
35, 36, 37
EPSO 33, 34, 102.5 49.33 0.7310 0.6564 1.1560
35, 36, 37
MPSO 33, 34, 109.2 46.02 1.1488 0.9023 0.5167
35, 36, 37
SABC 33, 34, 109.5 45.87 0.7740 0.5310 1.5004
35, 36, 37
scenario-4 GA 7, 8, 10, 112.0 44.64 1.041 0.905 0.7001
16, 28
PSO 7, 10, 93.5 53.78 1.0439 0.9061 0.7012
14, 28, 32
EP 7, 9, 34, 99.4 50.87 1.0499 0.9098 0.7099
36, 37
ABC 11, 20, 24, 129.7 35.89 1.3004 0.53 0.7054
32, 34
MGA 7, 9, 28, 99.5 50.82 1.048 0.907 0.7001
36, 37
EPSO 7, 9, 33, 94.2 53.44 1.1127 0.918 0.729
36, 37
MPSO 7, 10,14, 97.1 52.00 1.0489 0.9118 0.7312
28, 32
SABC 11, 20, 24, 101.6 49.78 1.2604 0.531 0.774
32, 34
(continued)
292 W. M. Dahalan and H. Mokhlis

Table 10.2 (continued)

Scenario Method Opened Power loss Loss DG size (MW)
switches (MW) reduction 6 16 25
(%)
scenario-5 GA 7, 10, 14, 100.9 50.12 1.1490 0.9427 0.6332
28, 30
PSO 7, 9, 14, 92.3 54.37 1.1523 0.9545 0.6312
28, 32
EP 7, 10, 12, 94.1 53.48 1.1519 0.9378 0.6680
16, 28
ABC 11, 20, 31, 103.9 48.64 1.133 0.9510 0.6220
34, 37
MGA 7, 10, 12, 96.88 52.11 1.1519 0.9335 0.6678
16, 28
EPSO 6, 10, 13, 89.4 55.81 1.1590 0.9747 0.6632
16, 28
MPSO 7, 9, 14, 92.46 54.30 1.1733 0.9651 0.6363
28, 32
SABC 11, 20, 31, 97.5 51.80 1.1019 0.7575 0.7780
34, 37

simultaneously during simulation. The results for the network of 33-bus system
show the greatest improvement on power loss reduction between 31.4 and 51.8%.
The performance of the test of power loss for scenario-5 of the proposed methods is
depicted in Fig. 10.5. From the results obtained in scenario-5, the SABC is
observed to generate highest improvement which is 6.19% as compared to the
original method. This is then followed by EPSO (4.89%), MGA (3.9%) and MPSO
(2.97%). However, in overall results, EPSO still maintain as the method which
produces the lowest power loss due to the changing of switch during the simulation
process and injection of the active power by the DG simultaneously.
After reconfiguration, the optimal sizes of DG of each method have also been
altered as illustrated in Table 10.3. With reference to scenario-5, three DG are
installed at different locations which have been fixed earlier. Once the program is
run, the sizes of DG will vary automatically between the predetermined ranges until
it reaches the optimal values. This can be seen from the Table 10.2 which shows
that the optimum DG size is different of each method. SABC produced the smallest
size of DG which is 2.6374 MW followed by MGA 2.7532 MW. While for MPSO
and EPSO are 2.7747 and 2.7969 MW respectively. If the total size of DG is in a
high proportion of the total load system, it will bring down the power quality. The
analysis indicates that the maximum energy saving is achieved when the DG size is
placed at bus no. 6, 16 and 25 as shown on the diagram. However, the total size of
DG of each method is still within the range.
10 Simultaneous Network Reconfiguration and Sizing … 293

Fig. 10.5 Power losses improvement using the proposed algorithms (scenario-5)

Table 10.3 The performance of the proposed method based on optimal DG sizing
Method DG size (MW)—scenario-5 Total size of DG (MW)
6 16 25
MGA 1.1519 0.9335 0.6678 2.7532
EPSO 1.1590 0.9747 0.6632 2.7969
MPSO 1.1733 0.9651 0.6363 2.7747
SABC 1.1019 0.7575 0.7780 2.6374

Table 10.4 Comparisons of performance of the proposed methods (scenario-5)

30 run times Power loss (kW)
MGA MPSO EPSO SABC
Min solution 96.8 92.4 89.4 97.5
Max solution 98.9 99.8 96.1 99.7
Average 97.4 94.37 92.24 98.3
Standard deviation 0.0009 0.00297 0.00211 0.00107
No. of iteration 35 21 13 30
CPU time (s) 29.4 16.1 12.8 21.5
Original method (s) GA—60 PSO—28.1 EP—16.8 ABC—44.5

Meanwhile, Table 10.4 shows the comparisons of the performance of the pro-
posed method after 30 runs of the 33-bus distribution system. Every repeated
process (run times) was initialized with random new combination switch and DG
size, thus the robustness and efficiency of the algorithm in finding the minimum
power loss can be evaluated.
From the analysis of the results, the minimum solution or the best output
achieved by EPSO is 89.494 kW. However, the maximum solution which indicates
294 W. M. Dahalan and H. Mokhlis

the unfavorable value is produced by MPSO and SABC which are

99.8 and 99.7 kW respectively. MPSO gives the highest standard deviation with
0.00297 among others. Nevertheless, the percentage of differences between the
‘min solution’ and ‘max solution’ of each method is 2.12% until 7.42%, which
gives the smallest standard deviation as compared to the original method.
The analysis of this scenario has proven the simultaneous presence of recon-
figuration and DG together yields a much better rate of power loss because the new
set of switches has been rearranged to create a new configuration system plus the
optimal size of DG. Thus, the new network reconfiguration with DGs of 33-bus
system operated on simultaneously basis is illustrated in Fig. 10.6.
Meanwhile, the maximum number of iterations to reach the optimal value is
35 iterations for MGA, 30 iterations for SABC. Meanwhile MPSO need 21 itera-
tions and only 13 iterations of the EPSO to reach the optimal point. EPSO method
takes only 13 iterations which need 12.8 s to converge while MGA method shows
the longest computing time of 29.4 s compared to other methods. This is due to the
reason that MGA requires more steps before converging. It means the highest
number of iterations, the longer the computing time
The convergence curve summarizes the capability and efficiency of each method
and the speed of the algorithm in reaching the optimal point. Figure 10.7 shows the
convergence characteristics of the proposed method of 33-bus system. With the
updated technique, the value of power loss is improved until the best solution is
reached. From the observation, the EPSO is the fastest (13 iterations) algorithm to
reach the optimal solution followed by MPSO, MGA and SABC.

10.6.2 Impact of Network Reconfiguration and DG Sizing

on Voltage Profile

The impact on the voltage profile for scenario-5 using the proposed method is
depicted in Fig. 10.8. By observing the results, it can be concluded that the voltage
profile has been improved and the average bus voltages have reached 0.950 pu as
compared to the base scenario which is 0.9131 pu In scenario 5, the system which
is operated with reconfiguration and DG simultaneously, the voltage profile of the
system has been improved considerably with a minimum node voltage of
0.977692 pu at point 33 of MPSO method and 0.97698 pu at point 33 of SABC
method. The improvement is about 7.07%. While the minimum node voltage for
MGA and EPSO occurs at the same point which is 0.985983 pu
10 Simultaneous Network Reconfiguration and Sizing … 295

SubstaƟon SubstaƟon
12.66kV
1 12.66kV
1 1
1

2 2

19 3 19 3

4 23 23
20 20 4

5 24 5
21 21 24

6 25 25
22 22 6
DG 25 DG DG 25
33 7 7
DG
33
open open
8 8
26 37
26
9 27 27
9
35 10 35 open
28 10 28
open open open
11 34 29 29
11 34

12 30
12 30
open open
13 31 13 31

Substation 14 32 37
Substation 14 32
Tie line open open
15 Tie line
Bus 33 15
33
Bus
Line 16 Line 16
open DG DG
17 17
18 36 36
18

(a) MGA - 7, 10, 12, 16, 28 (b) MPSO - 7, 9, 14, 28, 32

SubstaƟon SubstaƟon
12.66kV
Substation
12.66kV
Substation
1
1 1
1

2 2

19 19 3
3

23 20 4 23
20 4
open
5 24 21 5 24
21
25 6 25
22 6 22
open DG
DG 25
DG 25
DG 33 7
33 7
37 open
8 8
26 26

9 27 9 27
35 35
10 10 28
28 37
open open
open 29
11 29 11
34 open 34

12 30 12 30

31 13 31
13
open open
14 Substation 14 32
Substation 32
Ɵe line
Ɵe line
15 Bus 15
33 33
Bus
Line 16 line 16
open DG DG
17 17

36 18 36
18

(c) EPSO - 6, 10, 13, 16, 28 (d) SABC - 11, 20, 31, 34, 37

Fig. 10.6 Network Reconfiguration and DG size simultaneously using a MGA, b MPSO, c EPSO
and d SABC (scenario-5)
296 W. M. Dahalan and H. Mokhlis

Fig. 10.7 Convergence characteristics of the SABC, MPSO, MGA and EPSO algorithms

1.005

0.995

0.99
Voltage p.u

0.985

0.98

0.975 SABC MPSO EPSO MGA

0.97

0.965
1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33
Bus Number

Fig. 10.8 Performance of the voltage profile of the proposed methods

Generally, the voltage profile shows slight differences among the methods accept
at point 33. Therefore, the implementation of reconfiguration technique and DG has
given better voltage profile compared to without reconfiguration and DG. The
voltage profile has been improved more effectively whenever the reconfiguration
and DG in the system is operated simultaneously where all bus voltages satisfy the
0.95 pu voltage constraints and near to 1 pu
10 Simultaneous Network Reconfiguration and Sizing … 297

References

1. N. Suresh, Dr. T. Gowri Manohar, Optimal citing of custom power controller in distribution
system for loss reduction, in National Conference on GSSSETW, Oct 2009, Mysore,
Kartanaka
2. K. Prasad, R. Ranjan, N. Sahoo, A. Chaturvedi, Optimal reconfiguration of radial distribution
systems using a fuzzy mutated genetic algorithm. Power Delivery, IEEE Trans. 20(2), 1211–
1213 (2005)
3. N. Gupta, A. Swarnkar, K. Niazi, Reconfiguration of distribution systems for real power loss
minimization using adaptive particle swarm optimization. Electr. Power Compon. Syst. 39(4),
317–330 (2011)
4. C.L. Borges, D.M. Falcão, Impact of distributed generation allocation and sizing on
reliability, losses and voltage profile. Paper presented at the Power Tech Conference
Proceedings, 2003 IEEE Bologna (2003)
5. T.S. Basso, R. DeBlasio, IEEE 1547 series of standards: interconnection issues. Power
Electron IEEE Trans. 19(5), 1159–1162 (2004)
6. O. Javanmardi, M. Nasri, I. Sadeghkhani, Investigation of distributed generation effects on the
voltage profile and power losses in distribution systems. Adv. Electr. Eng. Syst. 1(2), 74–77
(2012)
7. S. Sivanagaraju, Y. Srikanth, E.J. Babu, An efficient genetic algorithm for loss minimum
distribution system reconfiguration. Electr. Power Compon. Syst. 34(3), 249–258 (2006)
8. G. Wang, P. Wang, Y.-H. Song, A. Johns, Co-ordinated system of fuzzy logic and
evolutionary programming based network reconfiguration for loss reduction in distribution
systems. Paper presented at the Fuzzy Systems, 1996. Proceedings of the Fifth IEEE
International Conference on (1996)
9. D. Shirmohammadi, H.W. Hong, Reconfiguration of electric distribution networks for
resistive line losses reduction. Power Deliv. IEEE Trans. 4(2), 1492–1498 (1989)
10. Y. Song, G. Wang, A. Johns, P. Wang, Distribution network reconfiguration for loss
reduction using fuzzy controlled evolutionary programming. Paper presented at the
Generation, Transmission and Distribution, IEE Proceedings (1997)
11. J.Z. Zhu, Optimal reconfiguration of electrical distribution network using the refined genetic
algorithm. Electr. Power Syst. Res. 62(1), 37–42 (2002)
12. L. Ganesan, P. Venkatesh, Distribution system reconfiguration for loss reduction using
genetic algorithm. J. Electr. Syst. 2(4), 198–207 (2006)
13. S. Sivanagaraju, J.V. Rao, P.S. Raju, Discrete particle swarm optimization to network
reconfiguration for loss reduction and load balancing. Electr. Power Compon. Syst. 36(5),
513–524 (2008)
14. C.-T. Su, C.-F. Chang, J.-P. Chiou, Distribution network reconfiguration for loss reduction by
Ant Colony Search algorithm. Electr. Power Syst. Res. 75(2), 190–199 (2005)
15. J. Chakravorty, Network reconfiguration of distribution system using fuzzy controlled
evolutionary programming. Int. J. Eng. Sci. Adv. Technol. 2(2), 176–182 (2012)
16. N. Gupta, A. Swarnkar, K. Niazi, Reconfiguration of distribution systems for real power loss
minimization using adaptive particle swarm optimization. Electr. Power Compon. Syst. 39(4),
317–330 (2011)
17. N. Rugthaicharoencheep, S. Sirisumrannukul, Feeder reconfiguration for loss reduction in
distribution system with distributed generators by Tabu Search. GMSARN Int. J. 3, 47–54
(2009)
18. Y.-K. Wu, C.-Y. Lee, L.-C. Liu, S.-H. Tsai, Study of reconfiguration for the distribution
system with distributed generators. Power Delivery, IEEE Trans. 25(3), 1678–1685 (2010)
19. J. Olamaei, T. Niknam, G. Gharehpetian, Application of particle swarm optimization for
distribution feeder reconfiguration considering distributed generators. Appl. Math. Comput.
201(1), 575–586 (2008)
298 W. M. Dahalan and H. Mokhlis

20. Z. Bingda, Y. Liu, A novel algorithm for distribution network reconfiguration based on
evolutionary programming. Paper presented at the Advanced Power System Automation and
Protection (APAP), 2011 International Conference on (2011)
21. A.C. Nerves, J.C.K. Roncesvalles, Application of evolutionary programming to optimal siting
and sizing and optimal scheduling of distributed generation. Paper presented at the TENCON
2009–2009 IEEE Region 10 Conference (2009)
22. M.P. Lalitha, V.V. Reddy, V. Usha, Optimal DG placement for minimum real power loss in
radial distribution systems using PSO. J. Theoret. Appl. Inf. Technol. 13(2), 107–116 (2010)
23. A. Arya, Y. Kumar, M. Dubey, Reconfiguration of electric distribution network using
modified particle swarm optimization. Int. J. Comput. Appl. 34(6) (2011)
24. M.P. Lalitha, N.S. Reddy, V.V. Reddy, Optimal DG placement for maximum loss reduction
in radial distribution system using ABC Algorithm. J. Theoret. Appl. Inf. Technol. (2010)
25. K. Nara, A. Shiose, M. Kitagawa, T. Ishihara, Implementation of genetic algorithm for
distribution systems loss minimum re-configuration. Power Syst. IEEE Trans. 7(3), 1044–
1051 (1992)
26. T. Ackermann, G. Andersson, L. Söder, Distributed generation: a definition. Electr. Power
Syst. Chapter 57(3), 195–204 (2001)
27. A. El-Zonkoly, Optimal placement of multi-distributed generation units including different
load models using Particle Swarm Optimisation. IET Gener. Transm. Distrib. 5(7), 760–771
(2011)
28. J. Mendoza, R. López, D. Morales, E. López, P. Dessante, R. Moraga, Minimal loss
reconfiguration using genetic algorithms with restricted population and addressed operators:
real application. Power Syst. IEEE Trans. 21(2), 948–954 (2006)
29. Y. Del Valle, G.K. Venayagamoorthy, S. Mohagheghi, J.-C. Hernandez, R.G. Harley, Particle
swarm optimization: basic concepts, variants and applications in power systems. Evol.
Comput. IEEE Trans. 12(2), 171–195 (2008)
Chapter 11
Optimal Incentive Plans for
Plug-in Electric Vehicles

Mehdi Rahmani-Andebili, Mahmud Fotuhi Firuzabad

and Moein Moeini-Aghtaie

Abstract This chapter investigates implementation of some parking lots for a

plug-in electric vehicle (PEV) aggregator to participate in energy market. Herein,
behaviors of the PEVs’ drivers regarding their cooperation with the aggregator with
respect to the introduced incentive (value of discount on charging fee of PEVs) are
modeled. The considered incentive includes the value of discount on the charging
fee of PEVs’ batteries. In addition, the capability of parking lots for transacting
electrical energy is modeled based on the hourly arrival/departure time of PEVs to/
from the parking lots and the hourly state of charge (SOC) of PEVs’ batteries. Also,
the degradation of PEVs’ batteries is modeled based on the effective ampere-hours
throughput of the PEVs’ batteries due to vehicle-to-grid (V2G). Moreover, the
economic factors such as inflation and interest rates and the technical factors
including the PEVs’ batteries power limit, the depth of discharge (DOD) constraint
of PEVs’ batteries, the yearly maintenance of parking lot, and the yearly replace-
ment rate of the conventional vehicles with the PEVs are taken into consideration in
the problem over the definite planning horizon. Furthermore, due to variability and
uncertainties involved with the energy market prices and the PEVs’ drivers’
behavior, the planning problem is solved stochastically.

Keywords Modeling capability of parking lot for energy transaction

Modeling behavior of PEVs’ drivers Modeling life loss of PEVs’ batteries


Optimal incentive plans Stochastic optimization

M. Rahmani-Andebili (&)
M. F. Firuzabad

Department of Electrical Engineering, Sharif University of Technology,
Tehran, Iran
e-mail: [email protected]
M. F. Firuzabad
e-mail: [email protected]
M. Moeini-Aghtaie
Sharif Energy Research Institute, Sharif University of Technology,
Tehran, Iran
e-mail: [email protected]

© Springer Nature Singapore Pte Ltd. 2018 299

F. Shahnia et al. (eds.), Electric Distribution Network Planning,
Power Systems, https://doi.org/10.1007/978-981-10-7056-3_11
300 M. Rahmani-Andebili et al.

11.1 Introduction

Replacing internal combustion vehicles with plug-in electric vehicles (PEVs) is a

promising strategy to calm the energy security and environmental issues, since
PEVs can be charged by the electricity generated by renewables or clean energy
resources [1]. Nowadays, governments around the world call for the deployment of
PEVs and hybrid PEVs [2–5]. A recent study demonstrates that almost 27% of total
energy consumption and 33% of greenhouse gas emissions in the world are related
to the transportation sector [6]. Based on the studies presented in [7, 8], PEVs
utilization is being increased rapidly in some developed countries due to the
advancement of battery technology. Recently, the role of energy storage has
become more important with development of smart grids [9]. An individual PEV
has a trifle impact on an electric distribution network; however, aggregation of a
large number of PEVs can noticeably affect the network performance [10, 11].
Through effective coordination and communication technologies, the PEVs can be
considered as the mobile energy storage and play an important role in the smart
grids [12]. Nonetheless, replacing conventional vehicles with electric ones may put
the network at risk and bring about new issues such as system overload and spikes
in energy market prices due to uncontrolled charging of the PEVs’ batteries [13,
14]. Herein, a PEV aggregator can play an effective role to calm the above men-
tioned issues, since it can motivate the PEVs’ drivers (by introducing a variety of
incentives to them) to park their vehicles in the specific locations (parking lots) to
manage and coordinate the charging time of the PEVs’ batteries. By implementing
this strategy, the aggregator can take part in different power markets and provide
benefit for itself, for the PEVs’ drivers, and also for the network.
It has been reported that private vehicles are parked at parking lots in idle state
for more than 90% during a day [15]. Therefore, the PEVs as the energy storage
units have a huge potential for doing energy transactions in power market. Since
every individual PEV’s driver is not able to participate in energy market and
compete with other powerful market players, due to a low power capacity, a PEV
aggregator is introduced to aggregate them [16]. A comprehensive literature
overview regarding the economic and technical management of a PEV aggregator
has been given in [17]. In [18], the methods for optimal charging management of
PEVs have been reviewed. Moreover, advantages and disadvantages, and also
economic and technical characteristics of V2G technology have been discussed in
[10, 19–21].
In [22], the feasibility of utilizing Ontario’s grid for charging PEVs has been
analyzed applying a zonal model of Ontario’s transmission network and base-load
generation capacities for the period of 2009–2025. In [23, 24], real-time load
management strategies for coordinating the charging time of PEVs for minimum
energy losses and voltage control have been proposed. In [25], reliability assess-
ment of network considering PEVs fleet has been studied. In [26], feeder recon-
figuration has been used for coordinating V2G of PEVs in a stochastic framework.
11 Optimal Incentive Plans for Plug-in Electric Vehicles 301

In [27], energy management of one PEV connected to a smart home has been
investigated.
Presence of the PEV aggregator in different power markets has been investigated
in several studies [28–37]. In [28–30], PEVs have been utilized to support smart
grids by offering ancillary service including frequency regulation. In [31, 32],
participation of the PEVs in the spinning reserve market has been studied. In [33],
the PEV charging scheduling by an aggregator in a day-ahead energy market
applying mixed integer linear programming (MILP) has been investigated. In [34],
an optimal bidding strategy of a PEV aggregator participating in day-ahead energy
and regulation markets using stochastic optimization has been presented. The
authors in [35] have presented a method to manage the PEVs charging in real-time
for participation of the PEV aggregators in the energy market. In [36], solar parking
lots have been sized and allocated in an electrical distribution system based on their
optimal power factor applying quantum annealing.
In spite of the numerous studies in the literature about PEVs and their aggre-
gator, the behavior of PEVs’ drivers regarding their cooperation level with the
aggregator with respect to the incentive plans has not been modeled. In this chapter,
in addition to the PEVs’ drivers’ responsiveness level, capability of parking lots for
energy transaction in energy market is modeled. Moreover, model of a battery life
loss presented in [37] is applied for the PEVs’ batteries in the problem simulation.

11.2 Modeling Capability of the Parking Lots for Energy

Transaction

Figure 11.1 illustrates the schematic diagram of a PEV’s battery indicating its
capacity, state of charge (SOC) level, and the defined depth of discharge
(DOD) limit. As can be seen, the value of available grid-to-vehicle (G2V) power
can be determined based on the difference between the PEV’s battery capacity and
its SOC level. Also, the value of available vehicle-to-grid (V2G) power can be
calculated based on the difference between the SOC level of the PEV’s battery and
the given DOD limit for the PEV’s battery. Therefore, in a parking lot, at every

Fig. 11.1 Schematic diagram

of a PEV’s battery indicating
its capacity, SOC level, and
the defined DOD limit
302 M. Rahmani-Andebili et al.

Fig. 11.2 The aggregator as an intermediate agent between the energy market and the PEVs
connected to the distribution network

hour, the total value of available V2G and G2V powers can be computed by
knowing the values of some parameters concerned with the PEVs’ drivers’
behavior. These parameters include the number of PEVs arriving to the parking
lots, the number of PEVs departing from the parking lots, and the SOC level of the
batteries of the PEVs arrived to the parking lots [31]. Figure 11.2 shows the role of
aggregator as an intermediate agent between the energy market and the PEVs
connected to the distribution system through the parking lots. Based on this, the
aggregator can participate in the energy market transactions considering the price
signals and the total available V2G and G2V powers of the parking lots.
The PEVs’ drivers’ behavior and the energy market price are uncertain and they
may have a wide range of variability. Nevertheless, the variability range of every
uncertain parameter can be estimated based on the hourly historical data gathered
by the aggregator. Figure 11.3 shows the upper and lower bands for the possible
value of an uncertain parameter at every hour of the day.
Herein, the data gathered for every uncertain parameter and for every hour of a
day are fitted on a Gaussian distribution function as the most appropriate distri-
bution function, as can be seen in Fig. 11.4 [31]. Then, in order to address the
11 Optimal Incentive Plans for Plug-in Electric Vehicles 303

Fig. 11.3 The upper and lower bands for the possible value of an uncertain parameter at every
hour of the day

Fig. 11.4 Considering five distinct values for each uncertain parameter at every hour of the day

prediction uncertainty, five distinct values are considered for every uncertain
parameter with the probabilities 0.0228, 0.1359, 0.6826, 0.1359, and 0.0228
according to the areas in the related Gaussian distribution function concerned with
the l − 2r, l − r, l, l + r, l + 2r. Figure 11.4 graphically illustrates the above
mentioned approach. After that, in order to investigate the problem stochastically,
15 comprehensive and diverse scenarios are defined for the hourly value of each
uncertain parameter throughout the day, as can be seen in Fig. 11.5. These sce-
narios have been defined arbitrarily; however, it has been tried to design the diverse
and comprehensive scenarios to include the most probable scenarios and eliminate
the similar ones.
304 M. Rahmani-Andebili et al.

Fig. 11.5 The considered scenarios for the values of each uncertain parameter over the day

11.3 Modeling Cooperation Between PEVs’ Drivers

and Aggregator

In this part, cooperation between PEVs’ drivers (n) and the aggregator with respect
to the value of discount on charging fee (w) is modeled applying linear, power,
logarithmic, and exponential functions. As can be seen in (11.1)–(11.4) in
Table 11.1, these models have been designed for 100% cooperation of the PEVs’
drivers under free charging and no cooperation under zero discount on the charging
fee. The cooperation percentage curves related to the linear model, power model
with exponents 0.1, 0.3, 1.5, and 3, logarithmic model, and exponential model
respect to value of discount on charging fee for the range of (0%, 100%) are
illustrated in Fig. 11.6. As can be seen, the considered models are very compre-
hensive, since they cover all the two dimensional space. Therefore, all the possible
linear and nonlinear behaviors of the PEVs’ drivers are taken into consideration.

Table 11.1 Models for cooperation percentage of the PEVs’ drivers with the aggregator as the
function of discount on charging fee
Model Cooperation percentage of PEVs’ drivers (%)
Linear nLin ¼ w (11.1)

n
Power w (11.2)
nPow ¼ 100  100 ; n2R

nExp ¼ 100  eMð Þ; M  1

w
Exponential 1001 (11.3)


Logarithmic w (11.4)
nLog ¼ 100  ln 100  ðexpð1Þ  1Þ þ 1
11 Optimal Incentive Plans for Plug-in Electric Vehicles 305

Fig. 11.6 Curves of

cooperation percentage of the
PEVs’ drivers with the
aggregator respect to value of
the discount on charging fee
assuming various linear and
nonlinear models

11.4 Modeling PEV’s Battery Life Loss Cost Due to V2G

Herein, the value of life loss cost or aging cost of a PEV’s battery is modeled based
on the effective ampere-hours throughput of the PEV’s battery due to the V2G
actions [37], as can be seen in (11.5). In (11.5), Ahb;t is the ampere-hours
throughput of the PEV’s battery due to V2G at hour t, AhTot b is the total cumulative
ampere-hours throughput of the PEV’s battery in its life cycle, PriceBat is the price
of a PEVs’ battery, and k, as the effective weighting factor, is determined using the
model introduced in [37]. As can be seen in Fig. 11.7, in the presented model, the
value of the effective weighting factor has a nonlinear relationship with the SOC
level of the PEV’s battery. For instance, at SOC of 50%, removing 1 A h from the
PEV’s battery is equivalent to removing 1.3 A h from the total cumulative
ampere-hours throughput of the PEV’s battery. However, at SOC of 100%,

Fig. 11.7 Relationship

between effective weighting
factor and SOC level of a
PEV’s battery [37]
306 M. Rahmani-Andebili et al.

removing 1 A h results in only about 0.55 A h reduction. This reality indicates that
the PEVs’ batteries should be operated at high SOC levels to optimize their lifetime.

k  Ahb ðtÞ
CostLL ðtÞ ¼  PriceBat ð11:5Þ
AhTot
b

11.5 Planning Problem Formulation

In this chapter, the PEV aggregator builds and implements some parking lots in a
residential area to participate in the energy market transactions to maximize its
profit over the given planning horizon.

11.5.1 Objective Function

Different terms of the objective function include the income resulted from trans-
actions in energy market, investment cost for structuring and implementing parking
lots and equipping them, yearly maintenance cost of the parking lots, aging cost of
the PEVs’ batteries due to V2G, and cost of considering discount on the charging
fee of the PEVs’ batteries. Herein, some economic factors such as inflation and
interest rates and several technical factors including the PEVs’ batteries’ power
limit, DOD constraint of the PEVs’ batteries, PEVs’ batteries aging due to V2G,
yearly maintenance of the parking lot, and yearly replacement rate of internal
combustion engine based vehicles with electric ones are considered in the problem.
Furthermore, due to variability and uncertainties involved with the energy market
prices and the PEVs’ drivers’ behavior, the planning problem is solved stochasti-
cally considering several comprehensive scenarios for every uncertain parameter.
The uncertain parameters include the energy market price, the number of PEVs
arriving to the parking lot, the number of PEVs departing from the parking lot, and
the SOC level of the PEVs’ batteries arrived to the parking lots. The objective
function of the planning problem is presented in (11.6).
      
OFpp ¼ Max CostTot
Inv
 PWV CostM þ PWV IncomeT  PWV CostBA
 
PWV CostInc
ð11:6Þ

The first term of the objective function is related to the total investment cost for
structuring the parking lots and equipping them and the second term of the
objective function is concerned with the present worth value of maintenance cost of
the parking lots over the planning period.
11 Optimal Incentive Plans for Plug-in Electric Vehicles 307

Table 11.2 The binary ua ub Decision

variables used to code the
decisions of the aggregator 0 0 Idle
1
1 0 G2V
1 V2G

The third term of the objective function is related to the present worth value of
the aggregator’s income over the planning period because of optimal transactions in
the energy market by doing optimal V2G and G2V actions considering all the
defined scenarios for the uncertain parameters. Equation (11.7) presents the
deterministic value of aggregator’s income per year. As can be seen in Table 11.2,
ua and ub as the binary numbers are used to code the decision of the aggregator for
being in idle state or performing one of the V2G or G2V actions. Equation (11.8)
presents the number of PEVs’ drivers who are willing to cooperate with the
aggregator and connect their PEVs to the parking lots. Herein, NevTot indicates
the total number of PEVs in the area. Also, nModel , as the cooperation percentage of
the PEVs’ drivers with the aggregator respect to value of the discount on charging
fee (w), is determined using Table 11.1 or Fig. 11.6. Equations (11.9) and (11.10)
present the stochastic value of aggregator’s income per year and the present worth
value of aggregator’s income over the planning period, respectively.

365 X
X 24
IncomeTy ¼ ua ðtÞ
d¼1 t¼1 !
X
Nev   X
Nev
 u ðt Þ  p ðt Þ
b E
V2Gev;y;d ðtÞ  1  u ðtÞ  pE ðtÞ
b
G2Vev;y;d ðtÞ
ev¼1 ev¼1

ð11:7Þ

Nev ¼ NevTot  nModel ð11:8Þ

 XX X Xn o
Stoch IncomeTy ¼ IncomeTy Prip PrjNarr PrkNdep PrlSOC
i2Sp j2SNarr k2SNdep l2SSOC

ð11:9Þ


 X
pp
  y
1 þ IFR
PWV Stoch IncomeTy ¼ Stoch IncomeTy  ð11:10Þ
y¼1
1 þ ITR

The forth term of the objective function is related to the present worth value of
aging cost of the PEVs’ batteries over the planning period due to V2G actions
considering all the defined scenarios for the uncertain parameters.
Equations (11.11) and (11.13) give the deterministic value of yearly aging cost of
the PEVs’ batteries, the stochastic value of yearly aging cost of the PEVs’ batteries,
308 M. Rahmani-Andebili et al.

and the present worth value of aging cost of the PEVs’ batteries over the planning
period, respectively.

365 X
X 24 X
Nev
CostyBA ¼ ua ðtÞ  ub ðtÞ LL
Costev;y;d ðt Þ ð11:11Þ
d¼1 t¼1 ev¼1

 XX X Xn o
Stoch CostyBA ¼ CostyBA  Prip  PrjNarr  PrkNdep  PrlSOC
i2Sp j2SNarr k2SNdep l2SSOC

ð11:12Þ

  X pp
 1 þ IFR y
PWV CostBA ¼ Stoch CostyBA  ð11:13Þ
y¼1
1 þ ITR

The fifth term of the objective function is related to the present worth value of
discount on charging fee over the planning period considering all the scenarios
defined for the uncertain parameters. Herein, it is assumed that the cooperative
drivers’ PEVs’ batteries will be charged from the initial SOC level to the full charge
considering discount on the charging fee. Equations (11.14) and (11.16) present the
deterministic value of yearly incentive paid to the drivers, the stochastic value of
yearly incentive paid to the drivers, and the present worth value of incentive paid to
the drivers over the planning period, respectively.

365 X
X 24 XNev  arr
ðtÞ
SOCev;y;d w
CostyInc ¼ 1  Pev   pch ð11:14Þ
d¼1 t¼1 ev¼1
100 100

 XX X Xn o
Stoch CostyInc ¼ CostyInc  Prip  PrjNarr  PrkNdep  PrlSOC
i2Sp j2SNarr k2SNdep l2SSOC

ð11:15Þ

  X pp
 1 þ IFR y
PWV CostInc ¼ Stoch CostyInc  ð11:16Þ
y¼1
1 þ ITR

11.5.2 Constraints

The first constraint of the problem relates to supplying each PEV after daily V2G
and G2V actions. As can be seen in (11.7), the daily energy demand of each PEV
must be supplied considering the daily cumulative values of G2V and V2G done by
the PEV.
11 Optimal Incentive Plans for Plug-in Electric Vehicles 309

X
24   X
24
ua ðtÞ  1  ub ðtÞ  G2Vev;y;d ðtÞ  ua ðtÞ  ub ðtÞ  V2Gev;y;d ðtÞ
t¼1 t¼1
24 
X arr
SOCev;y;d ðt Þ
¼ 1  Pev ð11:17Þ
t¼1
100

8i 2 Sp ; 8j 2 SNarr ; 8k 2 SNdep ; 8l 2 SSOC ; 8ev ¼ 1; . . .; Nev; 8y ¼ 1; . . .; pp; 8d

¼ 1; . . .; 365

The second and third constraints concerned with the allowable injectable power
of every PEV’s battery into the grid and the allowable injectable power of the grid
into every PEV’s battery, respectively. These constraints must be regarded at every
hour of the planning period and in every scenario.

V2Gev;y;d ðtÞ ¼ Pev ð11:18Þ

G2Vev;y;d ðtÞ ¼ Pev ð11:19Þ

8i 2 Sp ; 8j 2 SNarr ; 8k 2 SNdep ; 8l 2 SSOC ; 8ev ¼ 1; . . .; Nev; 8y ¼ 1; . . .; pp; 8d

¼ 1; . . .; 365; 8t ¼ 1; . . .; 24

The forth constraint is related to the obligation of the aggregator respect to the
PEVs’ drivers. In order to prolong the lifetime of PEVs’ batteries, at every hour of
the planning period and in every defined scenario, the battery of every PEV must
not be discharged more than the defined DOD limit. In addition, the SOC level
cannot be considered more than 100%.

DODlimit  SOCev;y;d ðtÞ  100 ð11:20Þ

8i 2 Sp ; 8j 2 SNarr ; 8k 2 SNdep ; 8l 2 SSOC ; 8ev ¼ 1; . . .; Nev; 8y ¼ 1; . . .; pp; 8d

¼ 1; . . .; 365; 8t ¼ 1; . . .; 24

11.6 Proposed Optimization Technique

In this chapter, the problem is solved by applying genetic algorithm (GA) as the
optimization methodology [38]. Other optimization algorithms could be used in this
problem, however capability of GA for parallel optimization and its competence in
complex and nonlinear environments are the main reasons for utilization of GA in
this problem.
Variables of the optimization problem include ua and ub (the indicator of the
aggregator’s decision regarding being in idle state or performing one of the V2G or
G2V actions) at every hour of a day. Based on this, every chromosome in the
310 M. Rahmani-Andebili et al.

Fig. 11.8 Structure of the

defined chromosome

population is defined as the transaction matrix with 24  2 as its dimensions.

Figure 11.8 illustrates the structure of the defined chromosome. Herein, the value of
net profit of the aggregator over the planning period is defined as the value of fitness
of the chromosome. Different steps for applying GA in the problem are presented
and described below.
Step 1: Obtaining primary data
Parameters for applying GA: These parameters includes mutation probability of
the genes ðPMutation Þ and the size of population ðNch Þ.
Parameters of the problem: The values of all the problem parameters and the
initial data are obtained. Moreover, the value of discount on charging fee and the
cooperation model of the PEVs’ drivers with the aggregator are determined.
Initial population: The chromosomes of the population are initialized with
random binary values.
Step 2: Updating the population
Applying crossover operator: Two crossover points are randomly selected for
every pair chromosomes, and then, crossover operator is applied on every two
chromosomes of the population to reproduce two new chromosomes as the off-
spring, as can be seen in Fig. 11.9.
Applying mutation operator: This operator is applied on every gene of every
chromosome of the population with the definite probability PMutation .
Step 3: Selecting new population
Evaluating fitness of the chromosomes: For every chromosome, the problem is
run and if all the constraints are satisfied, the fitness of chromosome is measured.
Applying selection process: As can be seen in (11.21), new chromosomes are
selected through the probabilistic fitness-based selection process, where the fit-
ter chromosomes are more likely to be chosen. The value of selection probability of
every chromosome is calculated using (11.22) which is proportional to the fitness of
chromosome.
11 Optimal Incentive Plans for Plug-in Electric Vehicles 311

Fig. 11.9 Applying

crossover operator on two
chromosomes for reproducing
new chromosomes

1 PSelection [ rch
ach ¼ ch ð11:21Þ
0 PSelection
ch \rch

fch
PSelection ¼ ; S f ¼ ff1 ; . . .; fch ; . . .; fNch g ð11:22Þ
ch
MaxðS f Þ

Step 4: Checking termination criterion

Herein, the convergence status of the optimization procedure is checked. Based
on this, the values of improvements in fitness of the chromosomes of the old and
new populations are measured and if there are no significant improvements (1% of
the fitness of chromosome) in them, the optimization process is finished, otherwise,
the algorithm is continued form Step 2.
Step 5: Introducing the outcomes
The consequences include the best fitted chromosome as the optimal transaction
matrix.
This process is repeated for all possible values of discount on charging fee with a
10% step, and also for every cooperation model of the PEVs’ drivers with the
aggregator. After that, the optimal incentive, the optimal cooperation percentage of
the PEVs’ drivers with the aggregator, and the maximum net profit of the aggre-
gator over the given time horizon are determined.
312 M. Rahmani-Andebili et al.

11.7 Numerical Studies

11.7.1 Primary Data

The initial data and the value of problem parameters are presented in Table 11.3.
Figures 11.10, 11.11, 11.12 and 11.13 illustrate the variability range of the energy
market price and the uncertain parameters of the PEVs’ drivers’ behavior including
the number of PEVs arriving to the parking lot, the number of PEVs departing from
the parking lot, and the SOC level of the PEVs’ batteries arrived to the parking lots

Table 11.3 The initial data and parameters of the problem

Planning period (y) 20 Total cumulative ampere-hours 700,000
throughput of a PEV’s battery
in its life cycle
Inflation rate (%/y) 10 Power of PEV’s battery (kW)a 10
Interest rate (%/y) 15 Capacity of PEV’s battery (kWh)a 50
Investment cost for a 100,000 Charging/discharging voltage 480
parking lot ($) level (volt)a
Maintenance cost for a 1000 DODlimit based on the contract (%) 20
parking lot ($/y)
Size of a parking lot 200 Growth rate of PEVs (%/y) 1
Total number of parking lots 10 Charging fee ($/kWh) 0.043
PEVs’ battery price ($) 10,000 Size of population in GA 100
Mutation probability 0.05
of genes
a
TESLA, level 3 charging

Fig. 11.10 The hourly upper and lower bands for the possible energy market price
11 Optimal Incentive Plans for Plug-in Electric Vehicles 313

Fig. 11.11 The hourly upper

and lower bands for the
possible percentage of PEVs
arriving to the parking lots

Fig. 11.12 The hourly upper

and lower bands for the
possible percentage of PEVs
departing from the parking
lots

Fig. 11.13 The hourly upper

and lower bands for the
possible level of SOC of the
PEVs batteries
314 M. Rahmani-Andebili et al.

Table 11.4 The average value and the standard deviation of Gaussian distribution functions
related to the uncertain parameters
Hour Arriving Departing SOC (%) Energy
PEVs (%) PEVs (%) market price
($/MWh)
l r l r l r l r
1 0 0 0 0 0 0 43 4
2 0 0 0 0 0 0 41 3
3 0 0 0 0 0 0 40 3
4 0 0 0 0 0 0 36 2
5 0 0 0 0 0 0 36 2
6 0 0 5 1 0 0 35 2
7 0 0 55 10 0 0 35 2
8 0 0 40 8 0 0 38 3
9 1 0 0 0 80 20 40 3
10 2 0 0 0 80 20 56 4
11 2 0 0 0 60 10 69 5
12 3 0 0 0 60 10 72 6
13 2 0 0 0 60 10 71 6
14 1 0 0 0 40 5 74 6
15 1 0 0 0 40 5 62 5
16 2 0 0 0 40 5 62 5
17 4 0 0 0 40 5 69 5
18 3 0 0 0 40 5 71 6
19 10 2 0 0 40 5 89 7
20 14 3 0 0 40 5 99 7
21 16 4 0 0 40 5 110 10
22 16 4 0 0 20 0 89 9
23 15 3 0 0 20 0 87 8
24 8 1 0 0 20 0 81 7

at every hour of the day. The average value and the standard deviation of the
Gaussian distribution functions related to the energy market price and the uncertain
parameters of the PEVs’ drivers’ behavior at every hour of the day are presented in
Table 11.4. Also, the scenarios considered in the problem simulation are based on
the details presented in Fig. 11.5.

11.7.2 Results

The curves related to the value of aggregator’s net benefit over the planning period
with respect to the discount value are illustrated in Fig. 11.14 for every cooperation
11 Optimal Incentive Plans for Plug-in Electric Vehicles 315

Fig. 11.14 Aggregator’s net

profit curves versus the value
of discount on charging fee
considering a variety of
cooperation models (the
arrows show the peak point of
the curves)

model. As can be seen, by changing the value of discount on the charging fee, the
profit of aggregator over the planning horizon is changed. However, increasing the
value of incentive for raising motivation of the PEVs’ drivers is not always
effective, since the benefit curves do not have pure ascending trend. In other words,
the curves are nonlinear and there is just one optimal value for the incentive in
every model. Moreover, the optimal value of the incentive is different in every
cooperation model. Thus, it can be concluded that assuming an incidental value of
incentive would not lead to the maximum profit of aggregator and even it may
result in detriment for the aggregator in some models.
The detailed results of the problem simulation including optimal value of
incentive, cooperation percentage of the PEVs’ drivers with the aggregator, the
values of income and cost terms of the objective function, and the value of max-
imum benefit of the aggregator over the given horizon for every model are pre-
sented in Table 11.5. As can be seen, the cooperation between the aggregator and
the PEVs’ drivers with power behavioral model (n = 0.1) results in maximum
benefit for the aggregator. In addition, no profit is achieved for the aggregator due
to cooperation of the aggregator with the PEVs’ drivers with exponential behavioral
model. Therefore, this cooperation is not practical and beneficial.
 316

Table 11.5 Detailed results of the problem simulation

Optimal PEVs cooperation Investment Maintenance Batteries aging Incentive Income of Net
discount (%) percentage (%) cost ($/pp) cost ($/pp) cost ($/pp) cost ($/pp) transaction benefit ($/
($/pp) pp)
Power model 10 79 1,000,000 10,000 7,008,300 1,154,200 17,794,000 8,621,500
with n = 0.1
Power model 20 61 1,000,000 10,000 5,444,000 1,793,100 13,823,000 5,575,900
with n = 0.3
Logarithmic 40 52 1,000,000 10,000 4,615,600 3,040,500 11,719,000 3,052,900
model
Linear model 50 50 1,000,000 10,000 4,411,400 3,632,500 11,201,000 2,147,100
Power model 60 46 1,000,000 10,000 4,100,500 4,051,800 10,411,000 1,248,700
with n = 1.5
Power model 70 34 1,000,000 10,000 3,026,200 3,488,700 7,683,800 158,900
with n = 3
Exponential 0 0 1,000,000 10,000 0 0 0 0
model
M. Rahmani-Andebili et al.
11 Optimal Incentive Plans for Plug-in Electric Vehicles 317

Appendix

The notation used throughout the chapter is listed below:

Problem parameters and Variables
Ahb ð:Þ Ampere-hours throughput of the PEV’s battery due to V2G
AhTot
b Total cumulative ampere-hours throughput of the PEV’s battery in its
life cycle
Cost ð:Þ
LL Battery life loss cost due to V2G
Inv Total investment cost for structuring the parking lots and equipping
CostTot
them
CostM Maintenance cost of the parking lots
CostBA Aging cost of the PEVs’ batteries due to V2G
CostInc Cost of considering discount on the PEVs’ batteries charging fee
IncomeT Income resulted from transactions in energy market
G2V ð:Þ Grid-to-vehicle
IFR; ITR Inflation rate and interest rate
OFpp Objective function of the problem over the given planning period
Pev Nominal input or output power of the PEV
PriceBat Price of a PEV’s battery
Prip Occurrence probability of the ith scenario related to the energy market
price
Narr Occurrence probability of the jth scenario related to number of arriving
Prj
PEVs to the parking lot
PrkNdep Occurrence probability of the kth scenario related to number of
departing PEVs from the parking lot
SOC Occurrence probability of the lth scenario related to SOC level of the
Prl
PEVs’ batteries
SOC ð:Þ State of charge of the PEVs’ batteries
SOC arr ð:Þ State of charge of the PEVs’ batteries arrived to the parking lots
DODlimit Depth of discharge limit based on the contract that must be respected
by the aggregator
ua ð:Þ; ub ð:Þ Controlling parameters for indicating decision of the aggregator for
being in idle state or doing one of the V2G and G2V actions
V2Gð:Þ Vehicle-to-grid
w Value of discount on charging fee
n Cooperation percentage of the PEVs’ drivers with the aggregator
pE Energy market price
p ch PEV’s battery charging fee
l; r Mean and standard deviation of the uncertain parameter
k Effective weighting factor.
318 M. Rahmani-Andebili et al.

GA Parameters
PMutation Mutation probability of the genes
Nch Size of the population
ach Binary variable as the indicator for selection of the chromosome for the
new population
rch Random number in the range of ð0; 1Þ
PSelection
ch
Value of selection probability of a chromosome
fch Value of fitness of a chromosome
Sf Set of chromosomes’ fitness.

References

1. W. Kempton et al., A Test of Vehicle-to-Grid (V2G) for Energy Storage and Frequency
Regulation in the PJM System (University of Delaware, Newark, DE, 2008)
2. D.W. Kurtz, R.R. Levin, EHV systems technology—a look at the principles and current
status. IEEE Trans. Veh. Technol. 32(1), 42–50 (Feb. 1983)
3. Y. Wu, H. Gao, Optimization of fuel cell and super capacitor for fuel-cell electric vehicles.
IEEE Trans. Veh. Technol. 55(6), 1748–1755 (2006)
4. S. Li, S. Sharkh, F. Walsh, C. Zhang, Energy and battery management of a plug-in series
hybrid electric vehicle using fuzzy logic. IEEE Trans. Veh. Technol. 60(8), 3571–3585
(2011)
5. B. Zhao, Y. Shi, X. Dong, Pricing and revenue maximization for battery charging services in
PHEV markets. IEEE Trans. Veh. Technol. 63(4), 1987–1993 (2014)
6. S.F. Tie, C.W. Tan, A review of energy sources and energy management system in electric
vehicles. Renew. Sustain. Energy Rev. 20, 82–102 (2013)
7. Canadian Automobile Association Electric Vehicles: What You Need to Know. http://
electricvehicles.caa.ca/government-incentives/
8. According to BC Hydro’s Draft Integrated Resource Plan, Appendix 2A–2011 Electric
Load Forecast. https://www.bchydro.com/content/dam/hydro/medialib/internet/documents/
environment/EVcharging_infrastructure_guidelines09.pdf
9. M.L. Di Silvestre, G. Graditi, E.R. Sanseverino, A generalized framework for optimal sizing
of distributed energy resources in micro-grids using an indicator-based swarm approach. IEEE
Trans. Ind. Informat. 10(1), 152–162 (2014)
10. W. Su, H. Eichi, W. Zeng, M. Chow, A survey on the electrification of transportation in a
smart grid environment. IEEE Trans. Ind. Informat. 8(1), 1–10 (2012)
11. Z. Darabi, M. Ferdowsi, Anevent-based simulation framework to examine the response of
power grid to the charging demand of plug-in hybrid electric vehicles. IEEE Trans. Ind.
Informat. 10(1), 313–322 (2014)
12. F. Kennel, D. Gorges, S. Liu, Energy management for smart grids with electric vehicles based
on hierarchical MPC. IEEE Trans. Ind. Informat. 9(3), 1528–1537 (2013)
13. A.S. Masoum, S. Deilami, P.S. Moses, M.A.S. Masoum, A. Abu-Siada, Smart load
management of plug-in electric vehicles in distribution and residential networks with charging
stations for peak shaving and loss minimization considering voltage regulation. IET Gener.
Trans. Distrib. 5, 877–888 (2011)
11 Optimal Incentive Plans for Plug-in Electric Vehicles 319

14. L.P. Fernandez, T.G. San Roman, R. Cossent, C.M. Domingo, P. Frias, Assessment of the
impact of plug-in electric vehicles on distribution networks. IEEE Trans. Power Syst. 26,
206–213 (2011)
15. W. Kempton, Vehicle to grid power, FERC (2007)
16. W. Kempton, J. Tomic, S. Letendre, A. Brooks, T. Lipman, Vehicle to grid power: battery,
hybrid, and fuel cell vehicles as resources for distributed electric power in California.
University of California Davis Institute for Transportation Studies, Rep. ECD-ITS-RR-01-03
(2001)
17. R.J. Bessa, M.A. Matos, Economic and technical management of an aggregation agent for
electric vehicles: a literature survey. Eur. Trans. Elect. Power (2011) [Online]. Available:
http://onlinelibrary.wiley.com/doi/10.1002/etep.565/abstract
18. S.G. Wirasingha, A. Emadi, Classification and review of control strategies for plug-in hybrid
electric vehicles. IEEE Trans. Veh. Technol. 60(1), 111–122 (2011)
19. J.C. Ferreira et al., Vehicle-to-anything application (V2Anything App) for electric vehicles.
IEEE Trans. Ind. Informat. 10(3), 1927–1937 (2014)
20. M. Bertoluzzo, G. Buja, Development of electric propulsion systems for light electric
vehicles. IEEE Trans. Ind. Informat. 7(3), 428–435 (2011)
21. J.H. Zhao, F. Wen, Z.Y. Dong, Y. Xue, K. Wong, Optimal dispatch of electric vehicles and
wind power using enhanced particle swarm optimization. IEEE Trans. Ind. Informat. 8(4),
889–899 (2012)
22. A. Hajimiragha, C.A. Caizares, M.W. Fowler, A. Elkamel, Optimal transition to plug-in
hybrid electric vehicles in Ontario, Canada, considering the electricity-grid limitations. IEEE
Trans. Indust. Electron. 57, 690–701 (2010)
23. D.Q. Oliveira, A.C. Zambroni de Souza, L.F.N. Delboni, Optimal plug-in hybrid electric
vehicles recharge in distribution power systems. Elect. Power Syst. Res. 98, 77–85 (2013)
24. Z. Liu, F. Wen, G. Ledwich, Optimal planning of electric-vehicle charging stations in
distribution systems. IEEE Trans. Power Del. 28, 102–110 (2013)
25. C. Chen, S. Duan, Optimal integration of plug-in hybrid electric vehicles in microgrids. IEEE
Trans. Ind. Informat. 10(3), 1917–1926 (2014)
26. A. Kavousi-Fard, M.A. Rostami, T. Niknam, Reliability-oriented reconfiguration of
vehicle-to-grid networks. IEEE Trans. Ind. Informat 11(3), 682–691 (2015)
27. X. Wu, X. Hu, S. Moura, X. Yin, V. Pickert, Stochastic control of smart home energy
management with plug-in electric vehicle battery energy storage and photovoltaic array.
J. Power Sources 333, 203–212 (2016)
28. S. Han, S. Han, K. Sezaki, Development of an optimal vehicle-to-grid aggregator for
frequency regulation. IEEE Trans. Smart Grid, pp. 65–72 (2010)
29. E. Sortomme, M.A. El-Sharkawi, Optimal charging strategies for unidirectional
vehicle-to-grid. IEEE Trans. Smart Grid 2, 131–138 (2011)
30. J.R. Pillai, B. Bak-Jensen, Integration of vehicle-to-grid in the Western Danish power system.
IEEE Trans. Sustain. Energy 2, 12–19 (2011)
31. M. Rahmani-andebili, Spinning reserve supply with presence of plug-in electric vehicles
aggregator considering compromise between cost and reliability. IET Gener. Trans. Distrib. 7,
1442–1452 (2013)
32. R.J. Bessa, M.A. Matos, F.J. Soares, J.A. Peças Lopes, Optimized bidding of a EV
aggregation agent in the electricity market. IEEE Trans. Smart Grid 3(1), 443–452 (2012)
33. C. Jin, J. Tang, P. Ghosh, Optimizing electric vehicle charging with energy storage in the
electricity market. IEEE Trans. Smart Grid 4(1), 311–320 (2013)
34. S.I. Vagropoulos, A.G. Bakirtzis, Optimal bidding strategy for electric vehicle aggregators in
electricity markets. IEEE Trans. Power Syst. 28(4), 4031–4041 (2013)
35. F.J. Soares, P.M. Rocha Almeida, J.A. Pecas Lopes, Quasi-real-time management of Electric
Vehicles charging. Elect. Power Syst. Res. 108, 293–303 (2014)
320 M. Rahmani-Andebili et al.

36. M. Rahmani-andebili, Optimal power factor for optimally located and sized solar parking lots
applying quantum annealing. IET Gener. Transm. Distrib. 10, 2538–2547 (2016)
37. D.P. Jenkins J. Fletcher, D. Kane, Lifetime prediction and sizing of lead–acid batteries for
micro generation storage applications. IET Renew. Power Gener. 2(3), 191–200 (Sept. 2008)
38. L. Zhang, Z. Wang, X. Hu, F. Sun, D.G. Dorrell, A comparative study of equivalent circuit
models of ultracapacitors for electric vehicles. J. Power Sources 274, 899–906 (2015)
Chapter 12
Optimal Allocation of Compensators

Mohamed Ebeed, Salah Kamel, Shady H. E. Abdel Aleem

and Almoataz Y. Abdelaziz

Abstract Electric distribution networks mainly deliver the electric power from the
high-voltage transmission system to the consumers. In these networks, the R/X ratio
is significantly high compared to transmission systems hence power loss is high
(about 10–13% of the generated power). Moreover, poor quality of power including
the voltage profile and voltage stability issues may arise. The inclusion of shunt
capacitors and distributed Flexible ac transmission system (D-FACTS) devices can
significantly enhance the performance of distribution networks by providing the
required reactive power. D-FACTS include different members such as; distributed
static compensator (DSTATCOM), Distribution Static Var Compensator (D-SVC)
and unified power quality conditioner (UPQC). Optimal allocation of these con-
trollers in the distribution networks is an important task for researchers for power
loss minimizing, voltage profile improvement, voltage stability enhancement,
reducing the overall system costs and maximizing the system load ability and
reliability. Several analytical and optimization methods have been presented to find
the optimal siting and sizing of capacitors and shunt compensators in electric
distribution networks. This chapter presents a survey of new optimization tech-
niques which are used to find the optimal sizes and locations of such devices. This
chapter also presents an application of new optimization technique called

M. Ebeed
S. Kamel
Electrical Engineering Department, Aswan Faculty of Engineering,
Aswan University, Aswan, Egypt
e-mail: [email protected]
S. Kamel
e-mail: [email protected]
S. H. E. Abdel Aleem
Mathematical, Physical and Engineering Sciences Department,
15th of May Higher Institute of Engineering, Cairo, Egypt
e-mail: [email protected]
A. Y. Abdelaziz (&)
Electric Power and Machines Department, Faculty of Engineering,
Ain Shams University, Cairo, Egypt
e-mail: [email protected]

© Springer Nature Singapore Pte Ltd. 2018 321

F. Shahnia et al. (eds.), Electric Distribution Network Planning,
Power Systems, https://doi.org/10.1007/978-981-10-7056-3_12
322 M. Ebeed et al.

Grasshopper Optimization Algorithm (GOA) to determine the optimal locations and

sizes of capacitor banks and DSTATCOMs. The obtained results are compared with
different algorithms such as; Grey Wolf Optimizer (GWO), Sine Cosine Algorithm
(SCA).

Keywords D-FACTS
UPQC
Capacitor
DSTATCOM
Optimization

12.1 Introduction

Reactive power compensation can be used for enhancing system power quality,
reducing power loss, improving voltage profile, increasing power factor and net-
work capacity and reliability, reducing power flow in feeder lines, and enhancing
the network’s loadability and stability, as well as minimizing energy cost.
The most conventional devices that have been applied for reactive power
compensation are capacitor banks which include the switched and fixed types, in
addition to phase shifters and shunt reactor. D-FACTS devices have been incor-
porated in the distribution network for reactive power compensation. The main
advantages of D-FACTS devices are fast response, fine controllable and continuous
adjustment compared to conventional devices. Several types of D-FACTS devices
have been presented for enhancing the performance of distribution networks such as
DSTATCOM [1], UPQC [2] and Distribution Static Synchronous Series
Compensator (DSSSC) [3].
Optimal allocation of such compensation devices is an important issue to
maximize the benefits of these devices. Several techniques have been presented for
solving the optimal allocation problem of compensation devices in distribution
networks such as analytical techniques, numerical programming techniques,
heuristic techniques and artificial intelligence techniques [4]. The analytical
methods are based on calculus analytical approaches to determine the maximum of
a certain objective function, and the shortage of these methods is the obtained
capacitor sizes aren’t matched with the standard sizes hence the solution is rounded
up to standard capacitor sizes which may lead to overvoltage or less loss saving
[5–7]. The numerical programming techniques are iterative optimization approach
that can be applied to determine the optimal size and locations of compensation
devices [8–11]. It should point out that the obtained results using these methods are
more accurate compared to the analytical methods, but these techniques could be
trapped in local optimal solution. Heuristic techniques are applied for minimizing
the search space of optimization techniques where heuristic techniques are based on
determining the most candidate nodes for reactive power compensation using
sensitivity analysis [12]. Recently, artificial intelligence (AI) techniques are widely
used for solving the allocation problem of compensation devices in distribution
networks. Most of AI techniques are inspired from the natural phenomena behav-
iors. The AI methods can be applied to the nonlinear and complex problems.
12 Optimal Allocation of Compensators 323

This chapter introduces an application of Grasshopper Optimization Algorithm

(GOA) for solving problem allocation of compensators in distribution network
where GOA is employed to determine the optimal placement of shunt capacitor
banks for minimizing the total cost (energy loss cost along with capacitor cost)
moreover GOA is applied for assigning the optimal location and size of
DSTATCOM for minimizing the total loss, improving the voltage profile and
enhancing the voltage stability simultaneously.

12.2 Operation Principles of Distributed Compensators

The fixed and switched capacitor types are the most common devices that have been
incorporated for reactive power compensation. Different FACTS devices are imple-
mented for changing the parameters of network such as; transmission line impedance,
the bus voltage, the active and reactive power through networks for enhancing the
performance of electric systems [13, 14]. FACTS devices can be classified as:
(a) series members such as Thyristor Controlled Series Capacitor (TCSC) and Static
Synchronous Series Compensator (SSSC) (b) Shunt connected devices include
Static VAR Compensator (SVC), Static Synchronous Compensator (STATCOM) and
(c) Combined shunt-series controllers like Interline Power Flow Controller (IPFC) and
Generalized Unified Power Flow Controller (GUPFC) [15–18].

12.2.1 Shunt Capacitor

The power flow equations of distribution system can be obtained from Fig. 12.1 as
!
P2n þ jQ2n
Pn þ 1 ¼ Pn  PL;n þ 1  Rn ð12:1Þ
jVn j2

V0+1 Vn+1 Vk
V0 Vn
P0+jQ0
Pn+jQn Rn+jXn Pn+1+jQn+1

PL,0+1+jQL,0+1 P +jQ PL,n+1+jQL,n+1 Pk+jQk

L,n L,n

Fig. 12.1 Single line diagram of a radial distribution network

324 M. Ebeed et al.

!
P2n þ jQ2n
Qn þ 1 ¼ Qn  QLn þ 1  Xn ð12:2Þ
jVn j2
!
  P2n þ jQ2n
Vn2þ 1 ¼ Vn2  2ðRn Pn þ Xn Qn Þ þ R2n þ Xn2 ð12:3Þ
jVn j2

where
Pn ; Qn Real and reactive power flows into the receiving end of branch n + 1
connecting bus n and node n + 1.
Rn ; Xn Resistance and reactance of the line section between buses n and n + 1.
Vn The bus voltage magnitude at bus n
The active and reactive power loss of the nth line between buses n and n + 1 are
given as
!
P2n þ jQ2n
Plossðn;n þ 1Þ ¼ Rn ð12:4Þ
jVn j2
!
P2n þ jQ2n
Qlossðn;n þ 1Þ ¼ Xn ð12:5Þ
jVn j2

The system security level can be realized using the voltage stability index [19] as

VSIðn þ 1Þ ¼ jVn j4 4ðPn þ 1 Xn  Qn þ 1 Rn Þ2 4ðPn þ 1 Xn þ Qn þ 1 Rn ÞjVn j2 ð12:6Þ

where VSIðn þ 1Þ is the voltage stability index at bus n + 1. Enhancing the voltage
profile depends upon minimizing the voltage deviations as

X
k  2
VD ¼ Vn  Vref ð12:7Þ
n¼1

where k is a number of buses and Vref is the reference voltage that commonly equals
to 1 pu.
The capacitor banks are included in distribution systems for enhancing the
power quality and minimizing the total cost by injecting reactive power into the
systems. Figure 12.2 illustrates a shunt capacitor that is incorporated at bus n + 1
and the reactive power through the transmission line is given as
12 Optimal Allocation of Compensators 325

V0+1 Vn+1 Vk
V0 Vn
Pn+jQn Rn+jXn
Pn+1+jQn+1

Qc

PL,n+jQL,n PL,n+1+jQL,n+1

Fig. 12.2 Radial distribution system with a shunt capacitor

!
P2n þ jQ2n
Qn þ 1 ¼ Qn  QLn þ 1  Xn þ QC;n þ 1 ð12:8Þ
jVn j2

12.2.2 Distributed Static Compensator (DSTATCOM)

New members of FACTS controllers have been emerged due to progress of power
electronic devices. DSTATCOM is a developed controller based on voltage source
converter (VSC). DSTATCOM can inject or absorb both active and reactive power
at a point of coupling connection (PCC) by injecting a variable magnitude and
phase angle voltage at PCC. DSTATCOM is incorporated in electric systems for
enhancing the power quality, improving the power factor, balancing the loading,
mitigating the harmonic, reactive power compensation, reducing the power fluc-
tuations of photovoltaic units minimizing the voltage sag, mitigating the flicker in
the electric system and minimizing the power loss [20–23].
DSTATCOM consists of voltage source converter, dc bus capacitor, ripple filter
and coupling transformer as shown in Fig. 12.3. VSC is constructed by using
insulated gate bipolar transistors (IGBT) and MOSFET where the switching of
component is based on pulse-width modulation (PWM) sequences. The coupling
transformer is utilized for matching the inverter voltage with the bus voltage.
The DSTATCOM topologies are categorized based on three-phase three-wire
(3P3 W) and three-phase four-wire (3P4 W) as illustrated in [24].
DSTATCOM has an ability to exchange active and reactive current with the
network. A steady state modeling DSTATCOM has been presented in [25].
Figure 12.4 shows DSTATCOM controller which included in the radial distri-
bution system at bus n + 1 where DSTATCOM inject or absorb ID at this bus. By
applying KVL, the voltage at bus n + 1 can be obtained as
326 M. Ebeed et al.

PCC

Ripple Filter

Coupling Transformer

P Q

Controller
VSC

V ref Q ref
P

Energy source
(OpƟonal)

Fig. 12.3 Schematic diagram of DSTATCOM device

V0 V0+1 Vn+1 Vk
Vn
Pn+jQn Rn+jXn In Pn+1+jQn+1

ID

PL,n+jQL,n PL,n+1+jQL,n+1

D-STATCOM

Fig. 12.4 Radial distribution system with DSTATCOM

12 Optimal Allocation of Compensators 327

  p
Vn þ 1 \hn þ 1 ¼ Vn \hn  ðRn þ jXn Þ In \d þ ID \ hn þ 1 þ ð12:9Þ
2

where
Vn þ 1 \hn þ 1 Voltage of bus n + 1 after inclusion DSTATCOM.
ID The injected current by DSTATCOM.
In The line current after inclusion of DSTATCOM
Equation (12.9) represents the essential idea for modeling DSTATCOM which
can be solved by separating it to real and imaginary terms as
 p
Vn þ 1 Cosðhn þ 1 Þ ¼ ReðVn \hn Þ  ReðIn \dðRn þ jXn ÞÞ þ Xn ID Sin hn þ 1 þ
 p
2
 Rn ID Cos hn þ 1 þ
2
ð12:10Þ
 p
Vn þ 1 Sinðhn þ 1 Þ ¼ ImðVn \hn Þ  ImðIn \dðRn þ jXn ÞÞ  Xn ID Cos hn þ 1 þ
 p
2
 Rn ID Sin hn þ 1 þ
2
ð12:11Þ

Equations (12.10) and (12.11) can be simplified as

aCosx2 ¼ k1  b1 x1 Sinx2  b2 x1 Cosx2 ð12:12Þ

aSinx2 ¼ k2  b2 x1 Sinx2 þ b1 x1 Cosx2 ð12:13Þ

where

k1 ¼ ReðVn \hn Þ  ReðIn \dðRn þ jXn ÞÞ

k2 ¼ ImðVn \hn Þ  ImðIn \dðRn þ jXn ÞÞ

a ¼ Vn þ 1

b1 ¼ Rn

b2 ¼ Xn

x1 ¼ I D

x 2 ¼ hn þ 1
328 M. Ebeed et al.

Equations (12.12) and (12.13) can be rewritten as

aCosx2  k1
x1 ¼ ð12:14Þ
b1 Sinx2  b2 Cosx2

aSinx2  k2
x1 ¼ ð12:15Þ
b2 Sinx2 þ b1 Cosx2

Solving (12.14) and (12.15) yields

ðk1 b2  k2 b1 ÞSinx2 þ ðk1 b1  k2 b2 ÞCosx2 þ ab1 ¼ 0 ð12:16Þ

The previous equation can be simplified as

   
d12 þ d22 x2 þ ð2d1 ab1 Þx þ a2 b21  d22 ¼ 0 ð12:17Þ

where
x ¼ Sinðx2 Þ

d 1 ¼ ð k 1 b2  k 2 b1 Þ

d2 ¼ ðk1 b1  k2 b2 Þ

Hence, (12.17) can be solved as

pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
B  B2  4AC
x¼ ð12:18Þ
2A

where
 
A ¼ d12 þ d22

B ¼ ð2d1 ab1 Þ
 
C ¼ a2 b21  d22

Hence

hn þ 1 ¼ Sin1 ð xÞ ð12:19Þ

The value of ID can be obtained from (12.14) or (12.15). The voltage at PCC, the
DSTATCOM current and injected reactive power by DSTATCOM can be found as

!
Vn þ 1 ¼ Vn þ 1 \hn þ 1 ð12:20Þ
12 Optimal Allocation of Compensators 329

!  p
I D ¼ I D \ hn þ 1 þ ð12:21Þ
2
   p   
QD ¼ Im Vn þ 1 \hn þ 1 ID \ hn þ 1 þ ð12:22Þ
2

12.2.3 Unified Power Quality Conditioner (UPQC)

UPQC is a powerful controller that has applied for enhancing the power quality of
the electric system where it has the ability to minimize the voltage sags, balance the
system, mitigate the existed harmonics and minimizing the power loss, etc.
UPQC consists of two inverters on of these inverters is connected in series with a
certain transmission line while the other converter is connected in shunt to the
common bus. These inverters are combined thought dc linked bus. The inverters are
connected to the network by coupling transformers as shown in as shown in
Fig. 12.5 [26–28]. The main purpose of the series inverter is injecting an ac series
voltage to system to mitigate the supply voltage flickers or imbalance from the load
and forces the shunt branch to absorb harmonics generated by the nonlinear loads.
The shunt converter is employed for delivering the reactive power compensation for
improving the power factor correction in addition the shunt converter is used to
mitigate of current distortions and adjusting the dc bus voltage. In other words, the
series converter regulates the load voltage to be balanced and sinusoidal while the
shunt converter ensures the balancing of system current and become sinusoidal
(harmonic free). Several types of UPQC have produced which can be classified
based on the converter topology or the supply system or UPQC configuration [28].
Figure 12.5 shows the UPQC controller which included in the radial distribution
system where the series controller is included between buses n, n + 1 while the
shunt converter is connected at bus n + 1. It should highlight that the series injected

V0 V0+1 Vn Vn+1 Vk
Rn+jXn
Pn+jQn Ise In Pn+1+jQn+1
Vse
T se Ish
Filter
Series Shunt
Lsh
inverter inverter

PL,n+jQL,n PL,n+1+jQL,n+1
Vsh
T sh

UPQC

Fig. 12.5 Schematic diagram of UPQC controller

330 M. Ebeed et al.

voltage is kept in quadrature with current flow. In other words, the series and shunt
current are kept in quadrature with the voltage of bus n + 1 [29]. Referring to
Fig. 12.5, the voltage at bus n + 1 can be given as
  p
Vn þ 1 \hn þ 1 ¼ Vn \hn  ðRn þ jXn Þ In \d þ Ish \ hn þ 1 þ þ Vse \hse
2
ð12:23Þ

where
Vse The magnitude of the series injected voltage.
hse The phase angle of the injected voltage.
In   The current flow through the transmission line.
Ish \ hn þ 1 þ p2 The injected current of the shunt converter
The injected current of the series converter can be found as
! ! !
Ise ¼ In þ Ish ð12:24Þ

However, two equations are obtained by separating the real and imaginary part
of (12.23). Three quantities are unknown ðVse ; hn þ 1 ; Ish Þ. For solving this problem,
it is assumed that the reactive shunt power by shunt converter is represented as the
negative reactive load at bus n + 1 as shown in Fig. 12.6 [28].
Referring to Fig. 12.6, the injected series voltage can be found as
 
Vse \hse ¼ Vn þ 1 \hn þ 1 þ Zn
In \

d  Vn \hn ð12:25Þ

where
p
hse ¼
d þ
d  0 ð12:26Þ
2
p
hse ¼
d 
d [ 0 ð12:27Þ
2

V n+1 Vk
V0 V 0+1 Vn
R n+jX n Ise
P n+jQ n P n+1 +jQ n+1
Vse
T se

Series
inverter
P L,n +jQ L,n P L,n+1 +jQ L,n+1 - jQ sh

Series compansator

Fig. 12.6 Representation of UPQC in a distribution system

12 Optimal Allocation of Compensators 331

By separating the real and imaginary terms of (12.25) as

  
Vse cosðhse Þ ¼ Vn þ 1 cosðhn þ 1 Þ þ Re Zn
In \

d  ReðVn \hn Þ ð12:28Þ
  
Vse sinðhse Þ ¼ Vn þ 1 sinðhn þ 1 Þ þ Im Zn
In \

d  ImðVn \hn Þ ð12:29Þ

Equations (12.28) and (12.29) can be simplified as

Vse K1 ¼ b3 cosðhn þ 1 Þ þ b1 ð12:30Þ

Vse K2 ¼ b3 sinðhn þ 1 Þ þ b2 ð12:31Þ

where

x1 ¼ Vse

x 2 ¼ hn þ 1

K1 ¼ cos ðhse Þ

K2 ¼ sin ðhse Þ
  
b1 ¼ Re Zn
In \
d  ReðVn \hn Þ
  
b2 ¼ Im Zn
In \
d  ImðVn \hn Þ

b3 ¼ Vn þ 1

Solving (12.30) and (12.31), the value of Vse can be given as

pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
B  B2  4AC
Vse ¼ ð12:32Þ
2A

where

k12 þ k22
A¼
b3

K1 b1 þ K2 b2
B ¼ 2 
b3

b21 þ b22
C¼
b3
332 M. Ebeed et al.

The value of hn þ 1 can be obtained from (12.30) or (12.31) as


K 1 x 1  b1
hn þ 1 ¼ cos1 ð12:33Þ
b3

K 2 x1  b 2
hn þ 1 ¼ sin1 ð12:34Þ
b3

The reactive power of series compensator can be found as

   
Qseries ¼ Im Vn þ 1 \hn þ 1
In \

d ð12:35Þ

12.3 Optimization Techniques

Recently, the several optimization techniques are widely applied to determine the
optimal sizes and locations of compensation device in distribution networks.
Variety of optimization techniques have been proposed based on nature-swarm
inspired methods, human-inspired methods, physics inspired methods and evolu-
tionary inspired algorithms. In this section, a survey including the previous tech-
niques for solving the allocation problem of compensation devices is presented.
Table 12.1 shows an overview of application the optimization techniques in radial
distribution systems.

12.4 Problem Formulation

12.4.1 Capacitor Allocation Problem Formulation

The objective of optimal capacitor placement problem of the radial distribution

system is minimizing the total cost including the energy loss cost along with
capacitor cost. The objective function can be formulated as

X
nc
Minimize Cost ¼ Kp Ploss þ Kc;i Qc;i ð12:36Þ
i¼1
Table 12.1 Summary of the literature review regarding compensation devices placement problem
12

Algorithm Objective function Controller Sensitivity Year Refs.

using?
Nature inspired Ant Colony Ploss Reconfiguration and No 2008 [30]
capacitor
Bacterial Foraging Ploss and Cost Capacitor Yes 2015 [31]
Particle Swarm Optimization Cost Capacitor No 2010 [32]
Cuckoo Search Algorithm Cost Capacitor Yes 2013 [33]
Artificial Bee Colony Algorithm Cost and VSI Capacitor Yes 2014 [34]
Ant Colony Optimization Ploss and Cost and VD Capacitor Yes 2016 [35]
Crow Search Algorithm Ploss and Cost Capacitor No 2016 [36]
Flower Pollination Algorithm Capacitor 2016 [37]
Optimal Allocation of Compensators

Bat and Cuckoo Search Algorithms Ploss and Cost Capacitor No 2015 [38]
Oppositional Krill Herd Algorithm Ploss Reconfiguration and No 2016 [39]
capacitor
Monkey Search Optimization Ploss and Cost and VD Capacitor No 2016 [40]
and emission
MOPSO Ploss and VSI and VD Capacitor and DG No 2015 [41]
and current balancing
Firefly Algorithm SAIFI and SAIDI and Capacitor No 2014 [42]
AENS and Cost
Particle Swarm Optimization Cost Capacitor Yes 2014 [43]
Genetic Algorithm Loadability and total DSTATCOM and DG No 2016 [44]
cost
Fuzzy system and Expanded Ploss and VSI and VD DSTATCOM and DG No 2016 [45]
Invasive Weed Optimization
Particle Swarm Optimization Ploss and VD DSTATCOM No 2014 [46]
No 2015 [47]
333

(continued)
Table 12.1 (continued)
334

Algorithm Objective function Controller Sensitivity Year Refs.

using?
Fuzzy Multiobjective Approach Ploss and VD and Reconfiguration and
and Ant Colony Optimization feeder load balancing DSTATCOM and Photovoltc
arry
Bacterial Foraging Optimization Ploss and VD and Cost DSTATCOM and DG 2016 [48]
Algorithm
Bat algorithm Ploss and VSI and Cost DSTATCOM Yes 2015 [49]
Improved Cat Swarm Optimization Ploss and VD DSTATCOM and DG No
Cuckoo Optimization Algorithm Ploss and VD and Cost UPQC No 2016 [50]
and harmonic
Particle Swarm Optimization Ploss and VD and UPQC No 2014 [51]
UPQC rating
Plant Growth Simulation Algorithm Ploss and Cost Capacitor Yes 2011 [52]
Human-inspired Tabu Search Cost Capacitor Yes 1996 [53]
methods Improved Harmony Search Cost Capacitor 2016 [54]
Harmony Search and the Particle Ploss and VD DG and Capacitor Yes 2016 [55]
Artificial Bee Colony
Modified Cultural Algorithm Cost Capacitor No 2013 [56]
Imperialist Competitive Algorithm Ploss and VSI and VD DG and Capacitor No 2014 [57]
and Genetic Algorithm and load balancing
Teaching Learning Based Ploss and Cost Capacitor No 2014 [58]
Optimization
Harmony Search (HS) Algorithm Ploss and VD UPQC No 2015 [59]
(continued)
M. Ebeed et al.
Table 12.1 (continued)
12

Algorithm Objective function Controller Sensitivity Year Refs.

using?
Physics inspired Gravitational Search Algorithm Cost Capacitor yes 2015 [60]
methods Simulated Annealing and Greedy Cost Capacitor No 1995 [61]
Search
Biogeography Based optimization Ploss Capacitor and DG No 2016 [62]
(BBO)
Big bang-big Crunch Optimization Ploss Capacitor No 2011 [63]
Simulated Annealing Cost Capacitor No 1990 [64]
Evolutionary Integrated Evolutionary Algorithms Cost Capacitor yes 2013 [65]
inspired Penalty Free Genetic Algorithm Cost Capacitor No 2016 [66]
Optimal Allocation of Compensators

algorithms
Differential Evolutionary Cost Capacitor No 2015 [67]
Fuzzy GA Genetic Algorithm VSI and VD Capacitor Yes 2014 [68]
Genetic Algorithms Cost Capacitor and voltage No 2012 [69]
regulators
Genetic Algorithms Cost Capacitor Yes 1994 [61]
Immune Algorithm Cost DSTATCOM No 2014 [1]
Differential Evolution Ploss and VD DSTATCOM No 2011 [70]
Differential Evolution Ploss DSTATCOM No 2016 [71]
SAIFI System average interruption frequency index
SAIDI System average interruption duration index
AENS Average energy not supplied index
335
336 M. Ebeed et al.

where
Cost The total cost
Ploss The total active power loss (kW)
Qc The capacitor reactive power (kVar)
Kp The annual cost of energy losses
Kc The cost of capacitor per kVar

12.4.2 DSTATCOM Allocation Problem Formulation

The objective of optimal placement problem of DSTATCOM in the radial distri-

bution system is minimizing the total loss, improving the voltage profile and
enhancing the voltage stability index simultaneously as
Pnl
i¼1 ðPloss ðiÞÞafter DSTATCOM
f1 ¼ Pnl ð12:37Þ
i¼1 ðPloss ðiÞÞbefore DSTATCOM
Pnb  
i¼1 V ðiÞ  Vref after DSTATCOM
f2 ¼ Pnl   ð12:38Þ
i¼1 V ðiÞ  Vref before DSTATCOM

1
f3 ¼ Pnb ð12:39Þ
i¼1 ðjVSI ðiÞjÞafter DSTATCOM

where nl is the number of branches in electric distribution network while nb is the

number of buses in the network.

12.4.3 System Constraints

The required objective functions are subjected to equality and inequality constraints
related to electric distribution network which can be represented as
– Equality constraints
The equality constraints of the system are the active and reactive power flow
constraints which can be obtained as

X
n X
nb
Pslack ¼ PL ðiÞ þ Ploss ð jÞ ð12:40Þ
i¼1 j¼1
12 Optimal Allocation of Compensators 337

X
nc X
n X
nb
Qslack þ Q c ði Þ ¼ QL ðiÞ þ Qloss ð jÞ ð12:41Þ
i¼1 i¼1 j¼1

where Pslack and Qslack are active power and reactive powers supplied from the slack
bus, respectively. PL and QL are the active and reactive load demands respectively.
nb is the number of branches in the network while nc is the number of compen-
sation units.
– Inequality constraints
I. Bus voltage constraints

Vmin  Vi  Vmax ð12:42Þ

where Vmin and Vmax are the minimum and the maximum allowable bus voltage
limit.
II. Total reactive power constraint
Practically, the total injected reactive power using compensation devices is equal
to or less than the reactive load demand.

X
nc X
n
Qc ðiÞ  QL ðiÞ ð12:43Þ
i¼1 i¼1

where QL is the reactive load at a certain bus and Qc is compensator reactive power.
III. Thermal limit
The current flow through network branches must be within their allowable
limits as

In;i  Imax;i i ¼ 1; 2; 3. . .Nb ð12:44Þ

Nb is the number of branches in the distribution system.

12.5 Overview of Grasshopper Optimization Algorithm

(GOA)

GOA is a new optimization technique that is inspired form the movement and
migration of grasshopper in natural. The adult insects of grasshopper travel together
over long distance which mimics exploration of optimization technique while the
338 M. Ebeed et al.

Fig. 12.7 The life cycle of grasshopper

nymphs have no wings, so it move in small area which mimics the exploitation of
optimization technique [72].
Grasshoppers are harmful insects that can destroy a wide area of the agriculture
and crops where the grasshoppers swarm consist of million members which can
cover a wide area up to 1000 km. The life cycle of Grasshopper consists of three
stages as depicted in Fig. 12.7. The grasshopper can be found in two phases. In the
first phase the individual of grasshoppers avoids interaction together (solitary
phase) while in the other phase (gregarious phase), grasshoppers became sociable
and form a swarm. The swarm became a flying swarm depends upon environmental
consideration such as air temperature, sunshine and wind speed [73].
The swarm of grasshopper moves in rolling motion where groups are formed in
ground firstly by a collection of individuals of insects which move in the ground or
locally and short flight then these groups became coordinated together, and the
insects share a common spatial orientation. The behavior of grasshopper swarm can
be summarized as
(1) The swarm flies with downwind.
(2) The grasshoppers in front of swarm settle on the ground.
(3) The settled insects start eating and resting.
(4) The swarm starts taking of gain to altitude.
The grasshopper swarm navigation behavior aligned the wind is depicted in
Fig. 12.8.
12 Optimal Allocation of Compensators 339

Fig. 12.8 Motion of grasshopper swarm aligned with wind

The grasshopper swarm behavior depends upon the social interaction between a
grasshopper, the gravity force and the downwind advection. Hence mathematical
behavior can be represented as [74]:

Xi ¼ r1 Ai þ r2 Bi þ r3 Ci ð12:45Þ

where
Xi The position of ith grasshopper
Ai The social interaction
Bi The gravity force on the ith grasshopper
Ci Wind advection
r1 ; r2 ; r3 Random numbers
A social forced between two grasshoppers is established biologically where the
repulsion forces are existed in order to prevent collisions over a short length scale
and attraction force is existed for aggregation. The social interaction between
grasshoppers can define as

X 
N   xi  xj
Ai ¼ s Disij ð12:46Þ
Disij
j¼1
i 6¼ j

where Disij is the distance between i and j grasshoppers that equals to Disij ¼
xi  xj and the s function represents the social forces which can be represented as

  Disij
s Disij ¼ Fe l  eDisij ð12:47Þ

where F is the intensity of attractive force and l is the attractive length scale. The
swarm motion is directly affected by the gravity force which can be found as
340 M. Ebeed et al.

Bi ¼ ge!g ð12:48Þ

where g is the gravitational constant and !

eg the unity vector towards the center of
earth. The wind advection effect on the motion swarm

Ci ¼ ue!
w ð12:49Þ

Substituting the value of Ai , Bi and Ci from (12.46), (12.48) and (12.49) in

(12.45) yields

X 
N   xi  xj
Xi ¼ s Disij  ge! !
g þ uew ð12:50Þ
Disij
j¼1
i 6¼ j

The previous equation is modified to be implemented for optimization problems

and for enhancing the capability global searching of the algorithm it can be mod-
ified as
0 1
B X   C
B N Upper ðmÞ  Lower ðmÞ   xi  xj C
Xim B
¼ CB C s Disij C þ Pm ð12:51Þ
C best
@j ¼ 1 2 Disij A
i 6¼ j

where Upper ð:Þ and Lower ð:Þ are the upper and lower limits of the control variable,
respectively. Pm
best is the best position (the target position). C is an adaptive coef-
ficient that decrease linearly for enhancing the search capability of GOA which can
be represented as

Cmax  Cmin
C ¼ Cmax  T ð12:52Þ
Tmax

where Cmax , Cmin are the maximum and the minimum values of C, respectivly. T
and Tmax are the current iterations and the maximum iteration, respectively.
Step 1: Determine the input data of GOA algorithm including number of the search
agents (N), maximum number of iterations,Cmin , Cmax , F, L and the upper and lower
boundaries of control variables.
Step 2: Initialize the population of GOA algorithm as

i ¼ Lower ði; mÞ þ rand * ðUpper ði; mÞ  Lower ði; mÞÞ

Pm ð12:53Þ
12 Optimal Allocation of Compensators 341

Step 3: Calculate the fitness functions for each search agent.

Step 4: Determine the best position (target position) in term of the best fitness
function.
Step 5: Update the position of search agent according to (12.51).
Step 6: Check the boundaries of the updated agents and bring the violated variable
to accepted limit.
Step 7: Calculate the fitness function for the updated positions and determine the
target position.
Step 8: Repeat steps form (12.5) to (12.7) until the stopping criterion is achieved
(current iteration equals to maximum iteration).
Step 9: Obtain the optimal solution by capture the target position and the related
fitness function.

12.6 Numerical Examples

In this section the grasshopper optimization technique is employed to determine the

optimal locations and sizes of shunt capacitors and DSTATCOM in the 69-bus
radial distribution network. The line diagram of the network is shown in Fig. 12.9.
The network data are given in [75] which are also tabulated in Table 12.7.
A program code for optimal allocation of compensators is written using MATLAB
2009a and run on a PC with core i5 processor, 2.50 GHz and 4 GB RAM. The
selected parameters of GOA technique are listed in Table 12.2. The parameters
required for implementation of the proposed algorithm are adjusted by 50 times
running of this algorithm. The obtained results using the GOA algorithm are
compared with compared with other well-known optimization algorithms such as;
Grey Wolf Optimizer (GWO) [76], Sine Cosine Algorithm (SCA) [77] and other
meta-heuristics techniques. The studied cases are presented as

Fig. 12.9 The line diagram of the 69-bus system

342 M. Ebeed et al.

Table 12.2 The selected parameters of GOA

Parameter Tmax Search agents NO. Cmax Cmin F L
Value 100 30 1 0.00001 0.5 1.5

12.6.1 Case 1

The GOA technique is applied for optimal allocation of the capacitor in the 69-bus
network to minimize the total cost as described in (12.36). The sizes of capacitors
are selected to be standard with the available industrial market. The available sizes
and costs of capacitors are listed in Table 12.3. The Total active and reactive load
demands are 3801.89 kW and 2694.1 kVar respectively. The substation voltage is
12.66 kV and the single line diagram. The system power loss without inclusion
compensation devices equal to 225 kW and the total cost for the system without
any capacitor is found to be 37,800.0 $. The optimal size of capacitors, their
locations and the impact of optimal placement and sizing of capacitors on the
energy loss cost, capacitor cost and total cost of the system by 50 run trials are
given in Table 12.4. Moreover, the best, worst and mean obtained results by GOA
also are listed in Table 12.4. The power loss decreased to 145.405 MW with
incorporating capacitor banks optimally using GOA. Moreover, the value of total
cost is enhanced to 24,820.84 $. From Table 12.5 it can also be found that the
objective value found by the GOA technique is better than those obtained by the
CSA [33], DSA [78], TLBO [58], GSA [2], GWO and SCA. This demonstrates that
the GOA successfully achieves better simulation results than other techniques. The
voltage profiles of all system buses are enhanced significantly with incorporating
capacitor banks optimally using GOA as shown in Fig. 12.10. The average com-
putational time taken by the GOA technique and the other techniques are reported
in Table 12.4. It can be obvious that GOA needs less computational time compared
with other reported techniques. The convergence characteristic of the GOA, GWO

Table 12.3 Available capacitor size and related cost ($/kVar)

Size (kVar) 150 300 450 600 750 900 1050 1200 1350
Cost ($/kVar) 0.5 0.35 0.253 0.22 0.276 0.183 0.228 0.170 0.207
Size (kVar) 1500 1650 1800 1950 2100 2250 2400 2550 2700
Cost $/kVar 0.201 0.193 0.187 0.211 0.176 0.197 0.170 0.189 0.187
Size (kVar) 2850 3000 3150 3300 3450 3600 3750 3900 4050
Cost ($/kVar) 0.183 0.180 0.195 0.174 0.188 0.170 0.183 0.182 0.179
 12

Table 12.4 Obtained results for the 69-bus test system using different optimization techniques
Base case CSA [33] DSA [78] TLBO [58] GSA [2] GWO SCA GOA
Vmin ðp:uÞ 0.9092 0.930 0.9318 0.9321 0.9519 0.93079 0.93145 0.93079
Ploss ðkWÞ 225.00 147.95 147 146.35 145.9 145.569 145.440 145.405
Capacitor location (kVar) – 21 (250) 61 (900) 22 (300) 26 (150) 61 (1200) 61 (1200) 61 (1200)
62 (1200) 15 (450) 61 (1050) 13 (150) 12 (450) 9 (450) 12 (450)
Optimal Allocation of Compensators

60 (450) 62 (300) 15 (1050) 26 (150) 17 (350) 21 (150)

Total (kVar) – 1450 1800 1650 1350 1800 2000 1800
Capacitors cost ($) – 291.5 392.4 446.4 451.5 392.85 440.35 392.85
Energy cost ($) 37,800.0 24,855.6 24,696 24,586.8 24,511.2 24,455.51 24,433.98 24,427.99
Total annual cost ($) 37,800.0 25,147.1 25,088.4 25,033.2 24,962.7 24,848.36 24,874.33 24,820.84
Net savings ($) – 12,652.9 12,711.6 12,766.8 12,837.3 12,951.6 12,925.7 12,979.1
Best annual cost ($) 37,800.0 25,147.1 25,088.4 25,033.2 24,962.7 24,848.36 24,874.33 24,820.84
Worst annual cost ($) NA NA NA NA 25,093.56 25,040.17 25,040.17
Mean annual cost ($) NA NA NA NA 24,938.10 24,930.81 24,930.81
Average CPU time (S) 125.80 NA 36.87 NA 23.66 24.15 21.30
343
344 M. Ebeed et al.

and SCA are depicted in Fig. 12.11. From the convergence graph, it may be
observed that the objective value (total cost) converges and smoothly rapidly at the
15th iteration compared to GWO and SCA. This confirms the convergence relia-
bility of the proposed GWO algorithm.

12.6.2 Case2

In this case, GOA technique is employed to determine the optimal locations and
sizes of DSTATCOMs in the 69-bus network for minimizing the total loss,
improving the voltage profile and enhancing the voltage stability index simulta-
neously as described in (12.37), (12.38) and (12.39). Hence, in this case, the
objective function is a multi-objective function which can be formulated as

ft ¼ w1 f1 þ w2 f2 þ w3 f3 ð12:54Þ

where w1 , w2 and w3 are weighting factors. The value of any weighting factor is
selected based on the relative important on its related objective function with others
objective functions. The sum of the absolute values of the weight factors in (12.54)
assigned to all impacts should add up to one as [79]

jw1 j þ jw2 j þ jw3 j ¼ 1 ð12:55Þ

In this chapter, w1 is set as 0.5 while w2 and w3 equal 0.25.It should point out
that the constraint of injected reactive power of DSTATCOM is restricted as [1]

0  QSTATCOM  10; 000KVAR ð12:56Þ

X
nc X
n
QSTATCOM ðiÞ  QL ðiÞ ð12:57Þ
i¼1 i¼1

In this case, three DSTATCOM devices are included in the 69-bus system. The
optimal locations and sizes of DSTATCOMs that have been determined using
GOA, GWO and SCA, are listed in Table 12.5. It is obvious that the power loss is
reduced to 145.146 and the summation of voltage deviations is also reduced from
1.8374 to 1.3872 p.u with incorporating of the DSTATCOMs optimally using
GOA. Moreover, the voltage stability is also enhanced to 62.7759 p.u with inclu-
sion of DSTATCOMs. From Table 12.6, it is clear that the obtained results by
GOA are better than those obtained by GWO and SCA.
12 Optimal Allocation of Compensators 345

Table 12.5 Simulation results of the 69-bus system at different loadings

Loading Base case GWO SCA GOA
100% Minimum 0.9092 0.93079 0.93145 0.93079
voltage
Total active 225.00 145.569 145.440 145.405
loss kW
Annual cost 37,800.0 24,848.36 24,874.33 24,820.84
$/year
Location – 61 (1200) 61 (1200) 61 (1200)
and size 12 (450) 9 (450) 12 (450)
26 (150) 17 (350) 21 (150)
75% Minimum 0.93353 0.94874 0.94873 0.94874
voltage
Total active 121.030 79.971 81.383 79.971
loss kW
Annual cost 20,333.04 13,722.35 13,959.48 13,722.35
$/year
Location – 61 (900) 61 (900) 61 (900)
and size 12 (350) 9 (350) 12 (350)
50% Minimum 0.95668 0.96569 0.96569 0.96569
voltage
Total active 51.606 35.757 35.757 35.757
loss kW
Annual cost 8669.808 6139.1694 6139.1694 6139.1694
$/year
Location – 61 (600) 61 (600) 61 (600)
and size
Net Fixed 600 at Fixed 600 at Fixed 600 at
injected bus 61 bus 61 bus 61
kVar Switched 600 Switched 600 Switched 600
at bus 61 at bus 61 at bus 61
Switched 450 Switched 450 Switched 450
at bus 12 at bus 9 at bus 12
Switched 350 Switched 350 Switched 350
at bus 26 at bus 17 at bus 21
346 M. Ebeed et al.

1
4 3 21 69 68 67
6 5 6665
7 0.98 64
8 63
9 62
10 0.96 61
11 60
12 0.94 59
13 58
14 0.92 57
15 0.9 56
16 55
17 0.88 54
18 0.86 53
19 52
20 51
21 50
22 49
23 48
24 47
25 46
26 45
27 44
28 43
29 42
3031 4041
32 33 38 39
34 35 36 37

Before Compensation After Compensation

Fig. 12.10 Effect of compensation on system voltages for the 69-bus system

4
x 10
2.6 GOA
GWO
2.58
Total Cost($)

SCA
2.56
2.54
2.52
2.5
2.48
10 20 30 40 50 60 70 80 90 100
Iteration

Fig. 12.11 Change of total cost with iterations for the 69-bus using GOA, GWO and SCA
12 Optimal Allocation of Compensators 347

Table 12.6 Obtained results for optimal allocation of DSTATCOM using different optimization
techniques
Base case GWO SCA GOA
Vmin ðp:uÞ 0.90919 0.93093 0.93132 0.93121
Vmax ðp:uÞ 0.99997 0.9999 0.99998 0.99998
VSImin ðp:uÞ 0.6833 0.7511 0.7523 0.7520
VSImax ðp:uÞ 0.9999 1.0000 0.9999 0.9999
P
VSI 61.2181 62.6904 62.7154 62.7759
Ploss ðKW Þ 225.00 146.453 145.840 145.146
VDðp:uÞ 1.8374 1.4105 1.4046 1.3872
Optimal locations and size – 61 (1264.5) 12 (548.01) 11 (374.71)
of DSTATCOM (kVar) 17 (346.9973) 61 (1245.6) 61 (1224.21)
36 (687.7078) 49 (562.84) 18 (242.430)

Appendix

See Table 12.7.

Table 12.7 Data of the 69-bus test systems

S. NO. From bus To bus R (X) X (X) PL (kW) QL (kVar)
1 1 2 0.0005 0.0012 0 0
2 2 3 0.0005 0.0012 0 0
3 3 4 0.0015 0.0036 0 0
4 4 5 0.0251 0.0294 0 0
5 5 6 0.366 0.1864 2.60 2.20
6 6 7 0.3811 0.1941 40.40 30
7 7 8 0.0922 0.0470 75 54
8 8 9 0.0493 0.0251 30 22
9 9 10 0.819 0.2707 28 19
10 10 11 0.1872 0.0619 145 104
11 11 12 0.7114 0.2350 145 104
12 12 13 1.0300 0.3400 8 5
13 13 14 1.0440 0.3450 8 5.50
14 14 15 1.0580 0.3496 0 0
15 15 16 0.1966 0.0650 45.50 30
16 16 17 0.3744 0.1238 60 35
17 17 18 0.0047 0.0016 60 35
18 18 19 0.3276 0.1083 0 0
(continued)
348 M. Ebeed et al.

Table 12.7 (continued)

S. NO. From bus To bus R (X) X (X) PL (kW) QL (kVar)
19 19 20 0.2106 0.0690 1 0.60
20 20 21 0.3416 0.1129 114 81
21 21 22 0.0140 0.0046 5 3.50
22 22 23 0.1591 0.0526 0 0
23 23 24 0.3463 0.1145 28 20
24 24 25 0.7488 0.2475 0 0
25 25 26 0.3089 0.1021 14 10
26 26 27 0.1732 0.0572 14 10
27 27 28 0.0044 0.0108 26 18.60
28 28 29 0.0640 0.15650 26 18.60
29 29 30 0.3978 0.1315 0 0
30 30 31 0.0702 0.0232 0 0
31 31 32 0.3510 0.1160 0 0
32 32 33 0.8390 0.2816 14 10
33 33 34 1.7080 0.5646 9.50 14
34 34 35 1.4740 0.4873 6 4
35 35 6 0.0044 0.0108 26 18.55
36 36 37 0.0640 0.1565 26 18.55
37 37 38 0.1053 0.1230 0 0
38 38 39 0.0304 0.0355 24 17
39 39 40 0.0018 0.0021 24 17
40 40 41 0.7283 0.8509 1.20 1
41 41 42 0.3100 0.3623 0 0
42 42 43 0.0410 0.0478 6 4.30
43 43 44 0.0092 0.0116 0 0
44 44 45 0.1089 0.1373 39.22 26.30
45 45 46 0.0009 0.0012 39.22 26.30
46 4 47 0.0034 0.0084 0 0
47 47 48 0.0851 0.2083 79 56.40
48 48 49 0.2898 0.7091 384.70 274.50
49 49 50 0.0822 0.2011 384.70 274.50
50 8 51 0.0928 0.0473 40.50 28.30
51 51 52 0.3319 0.1114 3.60 2.70
52 9 53 0.1740 0.0886 4.35 3.50
53 53 54 0.2030 0.1034 26.40 19
54 54 55 0.2842 0.1447 24 17.20
55 55 56 0.2813 0.1433 0 0
56 56 57 1.5900 0.5337 0 0
57 57 58 0.7837 0.2630 0 0
58 58 59 0.3042 0.1006 100 72
(continued)
12 Optimal Allocation of Compensators 349

Table 12.7 (continued)

S. NO. From bus To bus R (X) X (X) PL (kW) QL (kVar)
59 59 60 0.3861 0.1172 0 0
60 60 61 0.5075 0.2585 1244 888
61 61 62 0.0974 0.0496 32 23
62 62 63 0.1450 0.0738 0 0
63 63 64 0.7105 0.3619 227 162
64 64 65 1.0410 0.5302 59 42
65 11 66 0.2012 0.0611 18 13
66 66 67 0.0047 0.0014 18 13
67 12 68 0.7394 0.2444 28 20
68 68 69 0.0047 0.0016 28 20
Tie lines
69 11 43 0.5 0.5 6.0 4.30
70 13 21 0.5 0.5 5.00 3.50
71 15 46 1.0 1.0 39.22 26.30
72 50 59 2.0 2.0 100.0 72
73 27 65 1.0 1.0 59.0 42.0

References

1. S.A. Taher, S.A. Afsari, Optimal location and sizing of DSTATCOM in distribution systems
by immune algorithm. Int. J. Electr. Power Energy Syst. 60, 34–44 (2014)
2. S. Ganguly, Impact of unified power-quality conditioner allocation on line loading, losses,
and voltage stability of radial distribution systems. IEEE Trans. Power Delivery 29, 1859–
1867 (2014)
3. S. Devi, M. Geethanjali, Optimal location and sizing of distribution static synchronous series
compensator using particle swarm optimization. Int. J. Electr. Power Energy Syst. 62, 646–
653 (2014)
4. H. Ng, M. Salama, A. Chikhani, Classification of capacitor allocation techniques. IEEE Trans.
Power Delivery 15, 387–392 (2000)
5. J. Schmill, Optimum size and location of shunt capacitors on distribution feeders. IEEE Trans.
Power Appar. Syst. 84, 825–832 (1965)
6. N. Neagle, D. Samson, Loss reduction from capacitors installed on primary feeders. Trans.
Am. Inst. Electr. Eng. Part III: Power Appar. Syst. 75, 950–959 (1956)
7. Y. Bae, Analytical method of capacitor allocation on distribution primary feeders. IEEE
Trans. Power Appar. Syst. 1232–1238 (1978)
8. T.H. Fawzi, S.M. El-Sobki, M.A. Abdel-halim, New approach for the application of shunt
capacitors to the primary distribution feeders. IEEE Trans. Power Appar. Syst. 10–13 (1983)
9. H. Dura, Optimum number, location, and size of shunt capacitors in radial distribution feeders
a dynamic programming approach. IEEE Trans. Power Appar. Syst. 1769–1774 (1968)
10. M. Baran, F.F. Wu, Optimal sizing of capacitors placed on a radial distribution system. IEEE
Trans. Power Delivery 4, 735–743 (1989)
350 M. Ebeed et al.

11. M. Ponnavsikko, K.P. Rao, Optimal choice of fixed and switched shunt capacitors on radial
distributors by the method of local variations. IEEE Trans. Power Appar. Syst. 1607–1615
(1983)
12. S. Lee, J. Grainger, Optimum placement of fixed and switched capacitors on primary
distribution feeders. IEEE Trans. Power Appar. Syst. 345–352 (1981)
13. K. Padiyar, FACTS Controllers in Power Transmission and Distribution (New Age
International, 2007)
14. N.G. Hingorani, L. Gyugyi, Understanding Facts (IEEE press, 2000)
15. S. Kamel, F. Jurado, D. Vera, A simple implementation of power mismatch STATCOM
model into current injection Newton-Raphson power-flow method. Electr. Eng. 96, 135–144
(2014)
16. S. Kamel, F. Jurado, Z. Chen, M. Abdel-Akher, M. Ebeed, Developed generalised unified
power flow controller model in the Newton-Raphson power-flow analysis using combined
mismatches method. IET Gener. Transm. Distrib. 10, 2177–2184 (2016)
17. S. Abd el-sattar, S. Kamel, M. Ebeed, Enhancing security of power systems including SSSC
using moth-flame optimization algorithm, in Power Systems Conference (MEPCON), 2016
Eighteenth International Middle East (2016), pp. 797–802
18. M. Ebeed, S. Kamel, F. Jurado, Determination of IPFC operating constraints in power flow
analysis. Int. J. Electr. Power Energy Syst. 81, 299–307 (2016)
19. M. Chakravorty, D. Das, Voltage stability analysis of radial distribution networks. Int.
J. Electr. Power Energy Syst. 23, 129–135 (2001)
20. A. Elnady, M.M. Salama, Unified approach for mitigating voltage sag and voltage flicker
using the DSTATCOM. IEEE Trans. Power Delivery 20, 992–1000 (2005)
21. Z. Shuai, A. Luo, Z.J. Shen, W. Zhu, Z. Lv, C. Wu, A dynamic hybrid var compensator and a
two-level collaborative optimization compensation method. IEEE Trans. Power Electron. 24,
2091–2100 (2009)
22. R. Majumder, Reactive power compensation in single-phase operation of microgrid. IEEE
Trans. Industr. Electron. 60, 1403–1416 (2013)
23. R. Yan, B. Marais, T.K. Saha, Impacts of residential photovoltaic power fluctuation on
on-load tap changer operation and a solution using DSTATCOM. Electr. Power Syst. Res.
111, 185–193 (2014)
24. O.P. Mahela, A.G. Shaik, A review of distribution static compensator. Renew. Sustain.
Energy Rev. 50, 531–546 (2015)
25. M. Hosseini, H.A. Shayanfar, Modeling of series and shunt distribution FACTS devices in
distribution systems load flow. J. Electr. Syst. 4, 1–12 (2008)
26. A. Ghosh, G. Ledwich, Power Quality Enhancement Using Custom Power Devices (Springer
Science & Business Media, 2012)
27. M.-C. Wong, C.-J. Zhan, Y.-D. Han, L.-B. Zhao, A unified approach for distribution system
conditioning: distribution system unified conditioner (DS-UniCon), in Power Engineering
Society Winter Meeting, 2000. IEEE (2000), pp. 2757–2762
28. V. Khadkikar, Enhancing electric power quality using UPQC: a comprehensive overview.
IEEE Trans. Power Electron. 27, 2284–2297 (2012)
29. M. Hosseini, H. Shayanfar, M. Fotuhi-Firuzabad, Modeling of unified power quality
conditioner (UPQC) in distribution systems load flow. Energy Convers. Manag. 50, 1578–
1585 (2009)
30. C.-F. Chang, Reconfiguration and capacitor placement for loss reduction of distribution
systems by ant colony search algorithm. IEEE Trans. Power Syst. 23, 1747–1755 (2008)
31. K. Devabalaji, K. Ravi, D. Kothari, Optimal location and sizing of capacitor placement in
radial distribution system using bacterial foraging optimization algorithm. Int. J. Electr. Power
Energy Syst. 71, 383–390 (2015)
32. A.A. Eajal, M. El-Hawary, Optimal capacitor placement and sizing in unbalanced distribution
systems with harmonics consideration using particle swarm optimization. IEEE Trans. Power
Delivery 25, 1734–1741 (2010)
12 Optimal Allocation of Compensators 351

33. A.A. El-Fergany, A.Y. Abdelaziz, Cuckoo search-based algorithm for optimal shunt
capacitors allocations in distribution networks. Electric Power Components and Systems
41, 1567–1581 (2013)
34. A.A. El-Fergany, Involvement of cost savings and voltage stability indices in optimal
capacitor allocation in radial distribution networks using artificial bee colony algorithm. Int.
J. Electr. Power Energy Syst. 62, 608–616 (2014)
35. A.A.A. El-Ela, R.A. El-Sehiemy, A.-M. Kinawy, M.T. Mouwafi, Optimal capacitor
placement in distribution systems for power loss reduction and voltage profile improvement.
IET Gener. Transm. Distrib. 10, 1209–1221 (2016)
36. A. Askarzadeh, Capacitor placement in distribution systems for power loss reduction and
voltage improvement: a new methodology. IET Gener. Transm. Distrib. 10, 3631–3638
(2016)
37. A. Abdelaziz, E. Ali, S.A. Elazim, Flower pollination algorithm and loss sensitivity factors for
optimal sizing and placement of capacitors in radial distribution systems. Int. J. Electr. Power
Energy Syst. 78, 207–214 (2016)
38. S.K. Injeti, V.K. Thunuguntla, M. Shareef, Optimal allocation of capacitor banks in radial
distribution systems for minimization of real power loss and maximization of network savings
using bio-inspired optimization algorithms. Int. J. Electr. Power Energy Syst. 69, 441–455
(2015)
39. S. Sultana, P.K. Roy, Oppositional krill herd algorithm for optimal location of capacitor with
reconfiguration in radial distribution system. Int. J. Electr. Power Energy Syst. 74, 78–90
(2016)
40. F.G. Duque, L.W. de Oliveira, E.J. de Oliveira, An approach for optimal allocation of fixed
and switched capacitor banks in distribution systems based on the monkey search
optimization method. J. Control Autom. Electr. Syst. 27, 212–227 (2016)
41. A. Zeinalzadeh, Y. Mohammadi, M.H. Moradi, Optimal multi objective placement and sizing
of multiple DGs and shunt capacitor banks simultaneously considering load uncertainty via
MOPSO approach. Int. J. Electr. Power Energy Syst. 67, 336–349 (2015)
42. A.K. Fard, T. Niknam, Optimal stochastic capacitor placement problem from the reliability
and cost views using firefly algorithm. IET Sci. Meas. Technol. 8, 260–269 (2014)
43. A. Elsheikh, Y. Helmy, Y. Abouelseoud, A. Elsherif, Optimal capacitor placement and sizing
in radial electric power systems. Alex. Eng. J. 53, 809–816 (2014)
44. H. Karami, B. Zaker, B. Vahidi, G.B. Gharehpetian, Optimal multi-objective number,
locating, and sizing of distributed generations and distributed static compensators considering
loadability using the genetic algorithm. Electr. Power Compon. Syst. 44, 2161–2171 (2016)
45. H. Bagheri Tolabi, A. Lashkar Ara, and R. Hosseini, A fuzzy-ExIWO method for optimal
placement of multiple DSTATCOM/DG and tuning the DSTATCM’s controller, COMPEL:
Int. J. Comput. Math. Electr. Electron. Eng. 35, 1014–1033 (2016)
46. S. Devi, M. Geethanjali, Placement and sizing of D-STATCOM using particle swarm
optimization, in Power Electronics and Renewable Energy Systems (Springer, 2015),
pp. 941–951
47. H.B. Tolabi, M.H. Ali, M. Rizwan, Simultaneous reconfiguration, optimal placement of
DSTATCOM, and photovoltaic array in a distribution system based on fuzzy-ACO approach.
IEEE Trans. Sustain. Energy 6, 210–218 (2015)
48. K. Devabalaji, K. Ravi, Optimal size and siting of multiple DG and DSTATCOM in radial
distribution system using bacterial foraging optimization algorithm. Ain Shams Eng. J. 7,
959–971 (2016)
49. T. Yuvaraj, K. Ravi, K. Devabalaji, DSTATCOM allocation in distribution networks
considering load variations using bat algorithm. Ain Shams Eng. J. (2015)
50. J. Sarker, S. Goswami, Optimal location of unified power quality conditioner in distribution
system for power quality improvement. Int. J. Electr. Power Energy Syst. 83, 309–324 (2016)
51. S. Ganguly, Multi-objective planning for reactive power compensation of radial distribution
networks with unified power quality conditioner allocation using particle swarm optimization.
IEEE Trans. Power Syst. 29, 1801–1810 (2014)
352 M. Ebeed et al.

52. R.S. Rao, S. Narasimham, M. Ramalingaraju, Optimal capacitor placement in a radial

distribution system using plant growth simulation algorithm. Int. J. Electr. Power Energy Syst.
33, 1133–1139 (2011)
53. Y.-C. Huang, H.-T. Yang, C.-L. Huang, Solving the capacitor placement problem in a radial
distribution system using tabu search approach. IEEE Trans. Power Syst. 11, 1868–1873
(1996)
54. E. Ali, S.A. Elazim, A. Abdelaziz, Improved harmony algorithm and power loss index for
optimal locations and sizing of capacitors in radial distribution systems. Int. J. Electr. Power
Energy Syst. 80, 252–263 (2016)
55. K. Muthukumar, S. Jayalalitha, Optimal placement and sizing of distributed generators and
shunt capacitors for power loss minimization in radial distribution networks using hybrid
heuristic search optimization technique. Int. J. Electr. Power Energy Syst. 78, 299–319 (2016)
56. V. Haldar, N. Chakraborty, Power loss minimization by optimal capacitor placement in radial
distribution system using modified cultural algorithm. Int. Trans. Electr. Energy Syst. 25, 54–
71 (2015)
57. M.H. Moradi, A. Zeinalzadeh, Y. Mohammadi, M. Abedini, An efficient hybrid method for
solving the optimal sitting and sizing problem of DG and shunt capacitor banks
simultaneously based on imperialist competitive algorithm and genetic algorithm. Int.
J. Electr. Power Energy Syst. 54, 101–111 (2014)
58. S. Sultana, P.K. Roy, Optimal capacitor placement in radial distribution systems using
teaching learning based optimization. Int. J. Electr. Power Energy Syst. 54, 387–398 (2014)
59. V. Renu, S. Jeyadevi, Optimal design of UPQC devices in radial distribution network for
voltage stability enhancement. Int. J. Appl. Eng. Res. 10 (2015)
60. Y.M. Shuaib, M.S. Kalavathi, C.C.A. Rajan, Optimal capacitor placement in radial
distribution system using gravitational search algorithm. Int. J. Electr. Power Energy Syst.
64, 384–397 (2015)
61. S. Sundhararajan, A. Pahwa, Optimal selection of capacitors for radial distribution systems
using a genetic algorithm. IEEE Trans. Power Syst. 9, 1499–1507 (1994)
62. H. Sadeghi, N. Ghaffarzadeh, A simultaneous biogeography based optimal placement of DG
units and capacitor banks in distribution systems with nonlinear loads. J. Electr. Eng. 67, 351–
357 (2016)
63. M. Sedighizadeh, D. Arzaghi-Haris, Optimal allocation and sizing of capacitors to minimize
the distribution line loss and to improve the voltage profile using big bang-big crunch
optimization. Int. Rev. Electr. Eng. 6 (2011)
64. H.-D. Chiang, J.-C. Wang, O. Cockings, H.-D. Shin, Optimal capacitor placements in
distribution systems. II. Solution algorithms and numerical results. IEEE Trans. Power
Delivery 5, 643–649 (1990)
65. A.A. El-Fergany, Optimal capacitor allocations using evolutionary algorithms. IET Gener.
Transm. Distrib. 7, 593–601 (2013)
66. J. Vuletić, M. Todorovski, Optimal capacitor placement in distorted distribution networks
with different load models using penalty free genetic algorithm. Int. J. Electr. Power Energy
Syst. 78, 174–182 (2016)
67. R. Hosseinzadehdehkordi, H. Shayeghi, M. Karimi, P. Farhadi, Optimal sizing and siting of
shunt capacitor banks by a new improved differential evolutionary algorithm. Int. Trans.
Electr. Energy Syst. 24, 1089–1102 (2014)
68. A.R. Abul’Wafa, Optimal capacitor placement for enhancing voltage stability in distribution
systems using analytical algorithm and Fuzzy-Real Coded GA. Int. J. Electr. Power Energy
Syst. 55, 246–252 (2014)
69. I. Szuvovivski, T. Fernandes, A. Aoki, Simultaneous allocation of capacitors and voltage
regulators at distribution networks using genetic algorithms and optimal power flow. Int.
J. Electr. Power Energy Syst. 40, 62–69 (2012)
70. S. Jazebi, S. Hosseinian, B. Vahidi, DSTATCOM allocation in distribution networks
considering reconfiguration using differential evolution algorithm. Energy Convers. Manag.
52, 2777–2783 (2011)
12 Optimal Allocation of Compensators 353

71. J. Sanam, A. Panda, S. Ganguly, Optimal phase angle injection for reactive power
compensation of distribution systems with the allocation of multiple distribution STATCOM.
Arab. J. Sci. Eng. 1–9 (2016)
72. S. Saremi, S. Mirjalili, A. Lewis, Grasshopper optimisation algorithm: theory and application.
Adv. Eng. Softw. 105, 30–47 (2017)
73. B. Uvarov, Grasshoppers and locusts.in A Handbook of General Acridology Vol. 2.
Behaviour, Ecology, Biogeography, Population Dynamics (Centre for Overseas Pest
Research, 1977)
74. C.M. Topaz, A.J. Bernoff, S. Logan, W. Toolson, A model for rolling swarms of locusts. Eur.
Phys. J.-Spec. Top. 157, 93–109 (2008)
75. S. Chandramohan, N. Atturulu, R.K. Devi, B. Venkatesh, Operating cost minimization of a
radial distribution system in a deregulated electricity market through reconfiguration using
NSGA method. Int. J. Electr. Power Energy Syst. 32, 126–132 (2010)
76. S. Mirjalili, S.M. Mirjalili, A. Lewis, Grey wolf optimizer. Adv. Eng. Softw. 69, 46–61
(2014)
77. S. Mirjalili, SCA: a sine cosine algorithm for solving optimization problems. Knowl.-Based
Syst. 96, 120–133 (2016)
78. M.R. Raju, K.R. Murthy, K. Ravindra, Direct search algorithm for capacitive compensation in
radial distribution systems. Int. J. Electr. Power Energy Syst. 42, 24–30 (2012)
79. E. Ali, S.A. Elazim, A. Abdelaziz, Ant lion optimization algorithm for renewable distributed
generations. Energy 116, 445–458 (2016)
Chapter 13
Optimal Allocation of Automatic
Reclosers

Carlos Frederico Meschini Almeida, Gabriel Albieri Quiroga,

Henrique Kagan and Nelson Kagan

Abstract This chapter presents a methodology for the allocation of Automatic

Reclosers (AR) in medium voltage electric distribution networks. The methodology
defines strategic positions for installing Normally Closed (NC) and Normally
Opened (NO) reclosers to improve the system’s performance in terms of quality of
power supply. The restriction relies on the budget available for investing in pur-
chasing and installing AR. The methodology supports power distribution planning
activities, as it focusses on defining the optimal positions for installing reclosers in a
large network. Due to the size of the electric distribution networks considered
during planning activities, hundreds different positions for installing Normally-
Opened Automatic Reclosers (NO-AR) and Normally-Closed Automatic Reclosers
(NC-AR) must be assessed. To deal with the size of the problem, covering all states
the network may assume and assuring the positions for installing AR were optimum
ones, the proposed methodology divides this problem into three states. Through this
approach, the planning engineer need to carry out several simulations in just a few
minutes, evaluating the technical benefits achieved from different investment levels.
Similar approaches could not be found in the current literature. The methodology
was assessed considering two substations of a Brazilian electric distribution com-
pany, corresponding to twenty-five medium voltage feeders. Two analyses were
carried out: the brown field analysis, where the positions of thirty new automatic
reclosers were determined; and the green field analysis, where forty-five existing
automatic reclosers were reallocated. The results indicate significant improvements
in quality of service indices, which may reach over 30% reduction level.

C. F. M. Almeida (&)
G. A. Quiroga
N. Kagan

Escola Politecnica, Universidade de Sao Paulo, Sao Paulo, Brazil
e-mail: [email protected]
G. A. Quiroga
e-mail: [email protected]
N. Kagan
e-mail: [email protected]
H. Kagan
Sinapsis Inovacao em Energia, Sao Paulo, Brazil
e-mail: [email protected]

© Springer Nature Singapore Pte Ltd. 2018 355

F. Shahnia et al. (eds.), Electric Distribution Network Planning,
Power Systems, https://doi.org/10.1007/978-981-10-7056-3_13
356 C. F. M. Almeida et al.

Keywords Electric distribution planning Genetic algorithms
Electric distribu-



tion reliability Automatic reclosers Power quality

13.1 Introduction

The allocation of AR is an interesting alternative to improve the quality of supply

provided by electric distribution utilities. Besides the operational flexibility pro-
vided by such equipment, these investments usually come fully in the regulatory
asset base in Brazil. Other actions, such as tree pruning or change of network
patterns, are alternatives that represent an increase in operational costs and may
represent a regulatory risk of not being classified as prudent investment. Several
Brazilian distribution companies have invested for the significant increase in the
presence of AR on their networks. Such investments correspond to installation of
hundreds of devices per year. Thus, a new problem arose in distribution planning:
how to determine the optimal locations for the installation of hundreds of equip-
ment in the concession area of an electric distribution company?
Allocating an AR is a very difficult task. Firstly, planning engineers obtain the
data regarding the interruption occurrences recorded in the Outage Management
System (OMS) database from the previous years to define where most of fault
problems occur. Then, the fault occurrence information is crossed with the customer
distribution throughout the electric distribution network, obtained from the
Georeferenced Information System (GIS). In a very simplified way, the planning
engineers try to define the strategic positions for installing AR that would not allow
the regions where most of the faults occur affect the regions where there are most of
the customers are located. Thus, just to locate one single AR may take up to a few
hours according to the traditional approach.
Today, planning engineers need to assess different configurations the electric
distribution network may assume during its operation. For each configuration, one
also needs to assess the level of improvement that different sets of AR may
introduce to the network performance in terms of quality of supply indices.
Depending on the level of improvement of each set of AR, one can determine the
investment level required to deliver the quality of supply required by the customers.
It is important to clarify that these analyses cannot take much longer to be per-
formed, because the investment plan on utility companies is normally executed in a
few months of the year. So, a methodology that proposes the strategic positions for
allocating AR in the networks becomes an interesting alternative.
There are several references dealing with the problem of allocating switching
devices. Some references simply provide conceptual approaches for allocating
switching resources [4, 6]. Others, for the sake of simplicity, focus on allocation NC
switching devices only [5, 10, 11, 14, 15, 17, 18, 20, 23] or focus on the problem of
allocating NO switches only [9]. Normally, the methodologies proposed in current
references deal with simplified networks [2, 3, 5, 7, 8, 10–12, 14–16, 18, 23, 25].
Moreover, some methodologies address the problem of allocation of switching
13 Optimal Allocation of Automatic Reclosers 357

devices in networks by testing the installation of such equipment in every bus [7, 8],
others consider every line segment as candidate position for assessing the
improvement of installing a switching device [12, 25], and others have the candi-
date positions for the installation of the switching devices previously selected [16,
18, 21, 22]. Due to the combinatorial nature of the problem, and the vast possible
candidate position available for testing the allocation of a switching device these
approaches are impractical for planning investments across an entire concession
area.
When dealing with electric distribution networks of considerable size, the
methodologies present in the current literature tend to simplify the approach in
the decision-making problem for defining the number of switching devices to be
allocated. In [13], authors have defined a methodology that distributes AR in
medium voltage feeders regarding their importance. Thus, no reliability simulation
regarding a detailed modelling approach towards the network topology configura-
tion and the spatial distribution of fault occurrences throughout the network is
considered. Besides, the specific position where the AR should be installed is not
provided either. In [20], the methodology defines the optimal position for installing
AR one at a time. Thus, the optimal position for the next AR depends on the
position defined for the previous one. Such approach does not allow de possibility
of defining an optimum when one is trying to allocate a set of AR simultaneously.
In [21, 22, 24] authors propose an approach for allocation NC-AR and NO-AR
in electric distribution networks of considerable size. These references only con-
sider the allocation of new AR. In [21, 22] the authors proposed a methodology that
simultaneously allocates a NC-AR and a NO-AR, so it would be possible to assure
that the load blocks may be supplied during a contingency from another source.
That approach is not always required when installing new AR in network, as the
power demand may significantly vary throughout the day. In [24] the proposed
methodology considers average failure rates and average failure duration times for
the entire network. Such approach does not provide a realistic behavior of the
network, as one region may be more prone for failure than others, due to the
existence of trees, for example.
In this chapter, the methodology proposed for the optimal allocation of AR aims
its application at real distribution networks, with several thousands of buses. It
defines optimal alternatives for the installation of new NC-AR and NO-AR
simultaneously, indicating the positions where the devices should be installed to
improve the quality of electricity supply and respecting the total amount of
equipment to be allocated and the total investment budget available. The method-
ology also considers interruption occurrences, extracted from the utility company’s
OMS, and the real topology, extracted from the utility company’s GIS. Thus,
reliability calculation is performed in a more realistic way.
An important innovation of the proposed methodology relies on the three-stage
approach proposed. At the first stage, the possible states for each medium voltage
feeder, depending on the allocation of NO-AR are enumerated. At the second stage,
the optimum solutions for each set of NC-AR to be allocated are defined for each
possible state and stored. At last, the optimum solution for each medium voltage
358 C. F. M. Almeida et al.

feeder is chosen, maximizing the improvement in quality of supply regarding the

level of budget available. As the set of optimum solutions are stored, the planning
engineer may carry out several different simulations, considering different budget
levels to evaluate the investment level required to deliver the proper quality of
supply. The methodology also allows one to ignore the existing AR and reallocate
them to adequate the electric distribution network towards the current condition it is
subjected to.
As the proposed methodology indicates the strategic position for installing AR
with respect to the spatial fault occurrence distribution and spatial customer dis-
tribution, no coordination between the protective devices is assure. Several other
references that deal with the switch allocation problem do not guarantee the
coordination and selectivity of the protection device either. Nevertheless, it is
important to highlight that the coordination may be achieved by altering the pro-
tection parameters of the existing devices of the network and that defining the
strategic positions is the more arduous task when one is trying to install new AR.

13.2 Methodology

The methodology proposed was divided into three stages. The first stage consists of
enumerating the possible states a medium voltage network may assume. A state is
defined by the positions for installing Normally-Opened Automatic Reclosers
(NO-AR). The second stage consists of evaluating configurations for the installation
of Normally-Closed Automatic Reclosers (NC-AR). Several optimization problems
are solved to determine the optimal positions for installing a specific number
NC-AR for each medium voltage feeder. Each set of optimization problems is
related to a specific state that the medium voltage feeder may assume. The third and
final stage consists on selecting a set of NC-AR and NO-AR to be installed in each
of the medium voltage feeder. In this stage, the objective is to maximize the
improvement in terms of quality of service that includes, in a weighted way, col-
lective indices of quality of service. The maximum number of AR to be allocated is
the main restriction of the optimization problem in this stage. It addresses the issue
of budget limitations. One may carry out different simulations, considering different
budget values, to evaluate if the estimated improvements would support the
required investment level. Figure 13.1 shows the detailed flowchart of the
methodology. Further details regarding each of the stages of the methodology are
provided in the following sections.
13 Optimal Allocation of Automatic Reclosers 359

Begin

i=0

Network ConƟngency
Topology Data
Select Feeder

j=0 Reliability
Parameters
CalculaƟon
Allocate NO

k=0
Reliability
Parameters

Allocate NC

Indicator
K++ K > #NC total CalculaƟon

j++ j > #NO total

AllocaƟon
END
SelecƟon
i++ i > #Fe. total

AllocaƟon by
Feeder

Fig. 13.1 Simplified methodology flowchart

13.2.1 Stage #1—State Enumeration—NO Recloser

Allocation

As stated before, the first stage consists of enumerating configurations for the
installation of AR-NO between the medium voltage feeders that belongs to the
electric distribution network under analysis. At first, all positions where there is
some kind of NO switch are candidate positions for testing the installation of a new
NO-AR. One may also provide extra candidate position for testing NO-AR by
adding a regular NO switch between two medium voltage feeders. The combination
of possible positions for testing the installation of NO-AR defines a state for a
specific medium voltage feeder.
360 C. F. M. Almeida et al.

For example, consider the medium voltage feeder that already has two NC-AR
installed, as shown in Fig. 13.2a. Also consider that one has indicated three pos-
sibilities for installing NO reclosers, as shown in Fig. 13.2b. Therefore, the pos-
sibility of installing these three NO-AR leads eight possible configurations for the
electric distribution network, as shown in Fig. 13.2c. If there were four NO-AR, the
network would present sixteen possible configurations; if there were five NO-AR,
the network would present thirty-two possible configurations; and so on. These
possible configurations are called states of network in the proposed methodology.
Each state provides a specific behavior of the network towards the set of contin-
gencies the network is subjected to. In other words, the number of NO-AR and their
installation positions determine the resource available for transferring loads during a
contingency, which will lead to a specific behavior in terms of the quality of
service.

13.2.2 Stage #2—NC Recloser Allocation

The second stage consists of determining configurations for installing NC-AR in

each of the medium voltage feeder from the electric distribution network under
evaluation. The second stage of the methodology is illustrated in the blue section
from Fig. 13.1. Firstly, the methodology selects a specific feeder. Then, for each of
the possible states the selected feeder may assume, the methodology determines the
optimal positions for installing a set of NC-AR.
Initially, the methodology determines the optimum position for installing of only
one NC-AR by testing it in every possible candidate position. All positions where
there is some kind NC switch (such as regular switched, fuses, etc.) are candidate
positions for testing the installation of a new NC-AR. However, due to high number
of possible candidates present in real networks, some heuristics rules have been
developed to determine the candidate positions for testing the effect of a new
NC-AR. Some of the heuristics rules considered are:
• it is not possible to install AR after a fuse;
• it is not possible to install more than three NC-AR in series;
• it is not possible to install NC-AR next to the substation (less than 50 m, for
example).
After determining the optimum position for installing only one NC-AR in the
selected feeder, the methodology solves several optimization problems to determine
the optimum positions for installing higher numbers of NC-AR (two NC-AR
installed simultaneously, three NC-AR installed simultaneously, and so on). The
process is repeated for every possible state the selected feeder may assume to
determine the optimal solutions for installing a specific set of NC-AR. After con-
cluding the optimization processes for one feeder, the methodology repeats the
whole process for the next feeder that belongs to the electric distribution network
13 Optimal Allocation of Automatic Reclosers 361

Fig. 13.2 Possible installation alternatives for NO devices

362 C. F. M. Almeida et al.

Fig. 13.3 Stage #2—NC recloser allocation methodology process

Fig. 13.4 Detail of the process of assessing the NC-AR allocation impact on the quality indices

under evaluation. It is important to clarify that the methodology starts by solving

the optimization problem for installing two NC-AR and the number of NC-AR is
increased until the cost of the set of NC-AR exceeds the amount of avoided ENS.
To compare the investment with the avoided ENS he methodology considered a US
$20,000.00 cost for the purchasing and installing a AR, 60.00 US$/MWh/year
interruption cost, a 7.5% return rate and 20-year life time for the AR. Figure 13.3
illustrates the automated process regarding the second stage of the methodology.
The process of assessing the NC-AR allocation impact on the quality of supply
indices is detailed at Fig. 13.4. From Fig. 13.3 one may observe that the optimal
results for every feeder, regarding every possible state and every possible number of
NC-AR are stored in specific files.
As stated before, when the methodology is determining the optimal positions for
installing two or more NC-AR, the optimum positions are determined by solving
several optimization problems. As there are several candidate positions for
13 Optimal Allocation of Automatic Reclosers 363

installing NC-AR, and all possible states each medium voltage feeder may assume
should be covered, several optimization problems must be solved. Using an
approach based on GA becomes convenient, as, when one defines the maximum
number of individual and generations to be considered, it also determines the
maximum time spent for solving each optimization problem. In fact, any other
metaheuristic approach could also be considered to solve the same problem. But,
the focus should not be on the solution method considered, but on the methodology
itself, regarding the allocation of multiple NO and NC devices, simultaneously, and
on real electric distribution networks, composed by thousands of buses.

13.2.2.1 Genetic Algorithms

Genetic Algorithms (GA) optimization technique was introduced by Holland in

1975 [1] with the objective of mathematically formalizing and rigorously explain
adaptation processes in natural systems and developing artificial systems that retain
the original mechanisms found in natural systems. GA start from a string, an
element that must have an explicit relation with the parameters of the problem.
A string can be understood, from genetics, as a chromosome that presents genes (or
bits), at different locus of the chromosome (positions in the string), representing
several characteristics of an individual. The value of each gene, which corresponds
to a specific characteristic, matches to one allele. Also, the genetic package, usually
called a genotype, may have its correspondence in GA by the data structures that
define the string. The interaction of this genetic package with the environment,
which defines the characteristics of the individual is called a phenotype, which
corresponds, in GA, in the decoding of the structure to form a possible alternative
solution or possible set of solution parameters of the problem. Once the basic
element of GA is established, that is a string, and its relation to the real problem,
that is its coding, the mechanism of a GA is relatively simple.

String Coding Approach

The string coding approach for the Stage #2 considers integer vectors. The number
of positions is defined by the number of NC-AR to be allocated. The integer inside
the position is defined by the candidate position where the NC-AR should be
installed. Figure 13.5 shows an example the string coding approach. Figure 13.5a
shows a feeder F, with a circuit breaker CB, two NO switches and five NC switches
that are the candidate position for assessing the NC-AR allocation. At its side, there
is respective general string representation for the allocation of two NC-AR. As there
are five candidate positions, the values in each position of the vector may vary from
0 to 4. Obviously, during the string coding process, the methodology does not allow
repeated values in the string positions. Figure 13.3b indicates that the feeder per-
formance should be assessed considering the installation of two NC-AR in the
position of the switches NC2 and NC4. Figure 13.3c indicates that the feeder
364 C. F. M. Almeida et al.

(a) Example of possible installation alternatives for two AR-NC.

(b) Example of the installation of two AR-NC.

(c) Example of the installation of three AR-NC.

Fig. 13.5 String coding approach for second stage GA-based optimization problems

performance should be assessed considering the installation of three NC-AR in the

position of the switches NC1, NC2 and NC4.
At this point of the methodology, the basic problem lies in determining the
“best” individual or the “best fit”, which is measured by the value of an evaluation
function applied to every individual.

Evaluation Function

For the second stage of the methodology, the evaluation function aims at measuring
the improvement of the quality of supply of the electric distribution network, by
13 Optimal Allocation of Automatic Reclosers 365

reducing the average frequency and duration of contingencies and improving rev-
enue saving by reducing the amount of energy not supplied. Thus, for the modeling
of the evaluation function, three indices are considered:
• Interrupted Customers (IC): total number of customers interrupted in a certain
period;
• Hours of Interrupted Customers (HIC): total number of hours that customers
were interrupted in a certain period;
• Energy Not Supplied (ENS): total energy not supplied due to occurrences of
interruptions in a certain period.
The evaluation function considered during the several optimization problems is
illustrated in Eq. (13.1).

max½KIC
MIIC þ KHIC
MIHIC þ KENS
MIENS  ð13:1Þ

where
• MIIC : is the merit index for IC reduction;
• MIHIC : is the merit index for HIC reduction;
• MIENS : is the merit index for ENS reduction;
• KIC : is the weighting factor for IC;
• KHIC : is the weighting factor for HIC;
• KENS : is the weighting factor for ENS.
The Merit Index (MI) formulation was based on the difference between any
evaluation grade for the initial state of the medium voltage feeder (without
installation of the AR-Gradewithout ) and the same grade for the state determined by
an alternative allocation (with the installation of the AR-Gradewith ). To allow
comparison with impacts on different grades, the difference is then normalized by
Gradewithout . Equation (13.2) illustrates the generic behavior for the proposed MI
formulation.

Gradewithout  Gradewith Gradewith

MI ¼ ¼1 ð13:2Þ
Gradewithout Gradewithout

Thus, (13.3–13.5) illustrates the specific formulation for MIIC , MIHIC , and
MIENS :

ICwith
MIIC ¼ 1  ð13:3Þ
ICwithout

HICwith
MIHIC ¼ 1  ð13:4Þ
HICwithout

ENSwith
MIENS ¼ 1  ð13:5Þ
ENSwithout
366 C. F. M. Almeida et al.

The improvements in terms of each of these indices were determined for every
state of every feeder that belongs to the electric distribution network under eval-
uation. The IC index is related to the SAIFI index. The HIC index is related to the
SAIDI index. IC and HIC were considered instead of the original SAIFI and SAIDI
because the methodology aims for a global improvement over the whole concession
area. Using the values for SAIFI and SAIDI determined for each feeder may not
lead to allocation that minimizes the quality of service indices (i.e. provides the
maximum improvement in terms of quality of supply). The calculation of the
indices is performed during the evaluation of every possible individual and follows
the ‘A Priori’ Reliability Calculation Approach [19].

‘A Priori’ Reliability Calculation Approach

The “A Priori” Reliability Calculation Approach estimates the reliability indicators for
an electric distribution network, based on the use of average reliability parameters
obtained through interruptions historic data. The indicators calculated are the number
of interrupted customers (IC), the number of hours of interrupted customers (HIC), and
the energy not supplied (ENS) due to the interruptions. For this, it is necessary to
calculate the reliability parameters for each load block and for each protection block.
A load block is a set of contiguous line sections delimited by sectioning equipment.
A protection block is a set of load blocks delimited only by protective equipment.
In this approach, a load block is defined by four parameters, which are the
average failure rate, the average failure duration, the number of consumers and its
demand, as illustrated in Fig. 13.6. In the schematic feeder presented in Fig. 13.6,
there are four load blocks and two protection blocks.
The average failure rate ðf avg Þ of each load block is calculated from interruptions
historic data, which may be obtained from the OMS database of electric distribution
companies. Thus, the interruptions occurrences are grouped according to the pro-
tection equipment responsible for the interruption, and the number of total inter-
ruptions is divided by the total length from the corresponding protection block.
Thus, the f avg from Load Block #1 is the same from Load Block #2, because Load
Block #1 and Load Block #2 are in the same protection block.

Fig. 13.6 Example of load blocks and its parameters

13 Optimal Allocation of Automatic Reclosers 367

The average failure duration ðdur avg Þ is also calculated from interruptions his-
toric data. In the same way, the interruptions occurrences are grouped according to
the protection equipment responsible for the interruption, and dur avg is the arith-
metic average of duration of each interruption occurrence. Thus, the dur avg from
Load Block #1 is the same from Load Block #2, because Load Block #1 and Load
Block #2 are in the same protection block.
Only the number of customers (NC) and average demand ðDavg Þ are specific for
each load block. The (NC) is calculated from the sum of the customers that are
inside the block. Typically, such information may be obtained from GIS database of
electric distribution companies. The Davg is calculated through the energy con-
sumed by the customers connected to the load block. Typically, such information
may be obtained from ERP (Enterprise Resource Planning) system database of
electric distribution companies. With these four parameters, one may calculate the
quality of supply indices as presented in the following sections.

Total Number of Customers Interrupted

The (IC) is calculated as illustrated in Eq. (13.6):

X
n
IC ¼ ðNCi
fiavg Þ ð13:6Þ
i

where
• n: refers to the total number load blocks present the electric distribution system
under analysis;
• i: refers to one specific load block;
• fiavg : refers to the average failure rate of load block i;
• NCi : which is the number of customers that had their supply interrupted by a
contingency in load block i.

Total Number Hours of Interrupted Customers

The HIC is calculated as illustrated in Eq. (13.7):

X
n
HIC ¼ ðNCi
fiavg
duriavg Þ ð13:7Þ
i

where duriavg is the average failure duration for a contingency in load block i (which
is usually measured in hours, and considers, only interruptions that last more than
3 min, according to Brazilian regulation).
368 C. F. M. Almeida et al.

Energy Not Supplied

The ENS can be calculated as illustrated in Eq. (13.8):

X
n
ENS ¼ ðDavg
i
fiavg
Duri Þ ð13:8Þ
i

where Davg
i : is the average power demand of load block i.
The average demand for each load block is obtained by the relation between the
monthly consumption of each customers and the average number of hours in a
month, as seen in Eq. (13.9):

emonthly
Davg
i ¼ i
ð13:9Þ
730

where
• i: refers to one specific load block;
• emonthly
i : is total amount of energy consumed by the customers connected to load
block i (which is usually measured in kWh).
For evaluating the possibility of load transference, one needs to check if capacity
and voltage limits conditions would be violated. To guarantee a good computational
performance and to allow a conservative approach, only one power flow calculation
is carried out for the original state of the network.
Regarding loading constraints, the transference capacity for each NO switch is
established through the capacity limits of the line segments that connect the NO
switch to the source. Thus, the line segment with the lowest capacity located
between the NO switch and the source defines the load amount that the medium
voltage feeder may receive during a contingency.
Regarding voltage limit constraints, the transference capacity for each NO
switch is established through the total impedance from the line segments that
connect the NO switch to the source. Thus, the amount of load transferred cannot
increase a voltage drop at the bus where the NO switch to values that would violate
the voltage limits.
Through this approach, there is no need to perform a power flow calculation for
every state the electric network may assume, during the ‘A Priori Reliability
Calculation Approach’. Thus, the time spent of the calculation process does not
compromises the performance for the whole methodology.
It is important to clarify that the ‘A Priori Reliability Calculation Approach’ only
considers the transference of load blocks from one feeder to another, when the load
blocks are located downstream from the load block where a fault is being con-
sidered. Thus, by opening the switches to isolate the faulty load block and closing
the NO switch to reestablish the supply to the downstream load blocks the radiality
13 Optimal Allocation of Automatic Reclosers 369

condition of the electric distribution network is not violated. Thus, there is no need
to perform a radiality checking as commonly is done on reconfiguration problems.
In GA, the initial population of individuals, that is, an initial set of strings, is
usually established randomly. Populations then evolve into generations, basically
through three operators, reproduction, crossing-over and mutation. Reproduction
corresponds to a process in which individuals are copied to the future generation
according to their evaluation function. Crossing-over corresponds to an operator
acting on a randomly chosen pair of strings. And mutation corresponds to an
operator that can modify, with certain probability, the values of genes (alleles) of
the strings. Based on the population from the previous generation, such operators
are applied to create a new one which corresponds to the population from the next
generation. In this new population, new phenotypes are then introduced, which may
lead to new results in terms of the “best” individual. Figure 13.7 shows how these
stages are related to each other, which illustrates the GA considered. Further details
regarding GA may be found in [1]. For the second stage of the methodology, the
selection operator applied was the Tournament Selection with as evaluation with
three individuals, as illustrated in Fig. 13.8.
The mutation operator considered a 1% probability rate and is illustrated in
Fig. 13.9. The crossing-over operator considered a 75% probability rate and is
illustrated in Fig. 13.10.

13.2.3 Stage #3—Global Optimization

The third state of the methodology consists on determining which of the optimal
alternatives listed at the second stage for each medium voltage feeder should be
applied, to maximize the improvement in the overall quality of power supply indices of
the electric distribution network. Thus, the third stage determines which state will be
considered, by indicating how many NO-AR should be installed, and which alter-
native will be considered, by indicating how many NC-AR should be installed, for
each medium voltage feeder. The main restriction in this stage relies on the budget
available for investing in AR allocation, or the maximum number of AR to be installed.
To clarify the approach at this stage of the methodology, Eq. (13.10) illustrates
the optimization process from a linear programming perspective. The evaluation of
each state and alternative combination considers the benefits achieved in terms
of reduction of the quality of supply indices and in terms of the maximum number
of reclosers considered.

Xn
benefi
maximize
i¼1
num devi
subject to : ð13:10Þ
budget
num devi  max num dev max num dev 
unit cost
370 C. F. M. Almeida et al.

Begin

Randomly select
first population

Apply crossing - Apply mutation Evaluate each

over operator operator individual

Was the maximum

Apply selection
operator
No number of generation
reached?

Yes

Select fittest
individual

END

Fig. 13.7 Simplified GA flowchart

where
• i: refers to a specific medium voltage feeder;
• n: refers to the maximum number of medium voltage feeders present in the
network under evaluation;
• benefi : refers to the reduction on the quality of supply indices in feeder i;
• num devi : refers to the number of reclosers to be allocated in feeder i;
• max num dev: is the maximum number of reclosers to be allocated in the
network under evaluation;
• budget: the value of the budget available;
• unit cost: the unit cost of purchasing and installing one recloser.
13 Optimal Allocation of Automatic Reclosers 371

Fig. 13.8 Tournament selection

1 4 0
[Possible values: 0, 1, 2, 3] [Possible values: 0, 1, 2, 3, 4, 5] [Possible values: 0, 1, 2]
(a) Original String (third position selected for mutation)

1 4 2
[Possible values: 0, 1, 2, 3] [Possible values: 0, 1, 2, 3, 4, 5] [Possible values: 0, 1, 2]
(b) New String (after mutation execution)

Fig. 13.9 Mutation operator

Any other optimization technique could also be considered to solve the opti-
mization problem described through Eq. (13.10). Due to its ease of implementation,
GA were also considered to solve the optimization problem regarding this stage of
the methodology.
372 C. F. M. Almeida et al.

Position selected for crossing-over

0 3 1

1 4 0
(a) Original Strings (second position selected for crossing-over)

0 4 0

1 3 1
(b) New Strings (after crossing-over execution)

Fig. 13.10 Crossing-over operator

13.2.3.1 Genetic Algorithms

The GA process considered at this stage of the methodology was very similar to the
one considered for the second stage of the methodology. At this stage, the selection,
mutation and crossing-over operator were the same as the ones considered in the
second stage. The probability rates for the mutation and crossing-over operators
were also 1 and 75%, respectively.
The main difference from the approach considered for the second stage is that,
instead of solving multiple optimization problems, only one optimization problem
is now solved for the whole electric distribution network. As the NO-AR deter-
mines the state for the network, one should be able to identify the corresponding
state for each feeder, to locate the correct optimal solutions regarding the allocation
of NC-AR. Such problem was addressed through the string coding approach
considered.

String Coding Approach

The coding for the proposed solution alternatives is a string with two types posi-
tions. The binary positions at the beginning of the string correspond to the NO-AR.
They determine the installation (unit value) or not (null value) of an NO-AR at a
specific candidate position at the electric network. The integer positions in the string
correspond to a specific medium voltage feeder present in the network under
evaluation. Figure 13.11 shows an example of the string coding approach.
Figure 13.11 shows four feeders and the candidate positions for installing AR,
defined by NC and NO switches.
One may or may not allocate an NO-AR in each NO switch present in
Fig. 13.11. In that way, its coding is represented by a binary position. Since that are
four NO candidate switches in the example, the string code must have four binary
positions.
13 Optimal Allocation of Automatic Reclosers 373

Fig. 13.11 Feeders and its candidate positions for installing AR

For each medium voltage feeder, there are many optimal alternatives for allo-
cating NC-AR. Each of these alternatives correspond to a specific state of the
electric distribution network and was previously determined at the second stage of
the methodology. Thus, there are several sets of optimal solution for each feeder.
Each set of optimal solutions lists the optimal solutions for allocating one NC-AR,
two NC-AR, three NC-AR and so on, until the ENS avoided by the installation of
the NC-AR does not exceeds the unit cost of the AR. Each set of optimal solutions
also corresponds to a specific state of the medium voltage feeder. And the state of
the medium voltage feeder is determined by the NO-AR considered. For example,
in Fig. 13.11, F1 has two candidate positions for installing NC-AR, which leads to
four possible alternatives for NC-AR installations: the allocation of no NC-AR, the
allocation of an NC-AR at the position of switch NC1, the allocation of an NC-AR
at the position of switch NC2 and the allocation of two NC-AR, one at the position
of switch NC1 and another one at the position of switch NC2. In this way, an
integer coding is convenient. Since there are four medium voltage feeders in the
example, the string code must have four integer positions, one for each medium
voltage feeder. The generic string coding for the network illustrated in Fig. 13.11 is
detailed in Fig. 13.12.
To illustrate it better, Fig. 13.13 shows a coded string for a viable solution.
By assessing the string shown in Fig. 13.13, one can identify that this solution
indicates the allocation of two NO-AR at the positions of switches NO1 and NO2,
the allocation of two NC-AR in feeder F1, at the position of switches NC1 and
NC2, the allocation of one NC-AR in feeder F3, at the position of switch NC6,
and the allocation of two NC-AR in feeder F4, at the positions of switches NC8 and
NC9. The solution represented by this coding can be seen in Fig. 13.14.
374 C. F. M. Almeida et al.

Fig. 13.12 String considered to code all possible AR allocation positions

Fig. 13.13 Example of viable solution defined by a string

Fig. 13.14 Example of a possible solution

13.2.3.2 Evaluation Function

For the third stage of the methodology, the evaluation function also aims at mea-
suring the improvement of quality of supply of electric distribution network, by
reducing the average frequency and duration of contingencies and improving rev-
enue saving by reducing the amount of energy not supplied. Thus, for the modeling
of the evaluation function, IC, HIC and ENS indices were also considered.
The evaluation function for evaluating every possible solution during the several
optimization problems is illustrated in Eq. (13.11).
13 Optimal Allocation of Automatic Reclosers 375

max½KIC
fIC þ KHIC
fHIC þ KENS
fENS  ð13:11Þ

where fIC : is a function that provides the overall reduction in terms of the IC for the
whole network; fHIC : is a function that provides the overall reduction in terms of the
HIC for the whole network; fENS : is a function that provides the overall reduction in
terms of the ENS for the whole network, which are formulated as
Pn
ICwith ðfeederi ; NO1 ; . . .; NOk ; Alternative#Þ
fIC ¼ 1  i Pn ð13:12Þ
i ICwithout ðfeederi Þ

where
• feederi : refers to a specific medium voltage feeder present in the network under
evaluation;
• NO1 ; . . .; NOk : refer to the candidate positions for installing NO-AR, which
defines the state feederi is subjected to;
• Alternative#: refers to the combination of NC-AR to be installed in feederi ;
• ICwith ðfeederi ; NO1 ; . . .; NOk ; Alternative#Þ: is a function that returns the cor-
responding value of IC for feederi , depending on the state defined by the
combination of values on NO1 ; . . .; NOk , and on the combination of NC-AR to
be installed;
• ICwithout ðfeederi Þ: is a function that returns the original value of IC for feederi ,
without the installation of the AR.

Pn
HICwith ðfeederi ; NO1 ; . . .; NOk ; Alternative#Þ
fHIC ¼ 1  i Pn ð13:13Þ
i HICwithout ðfeederi Þ

where
• HICwith ðfeederi ; NO1 ; . . .; NOk ; Alternative#Þ: is a function that returns the
corresponding value of HIC for feederi , depending on the state defined by the
combination of values on NO1 ; . . .; NOk , and on the combination of NC-AR to
be installed;
• HICwithout ðfeederi Þ: is a function that returns the original value of HIC for
feederi , without the installation of the AR.

Pn
ENSwith ðfeederi ; NO1 ; . . .; NOk ; Alternative#Þ
fENS ¼ 1  i Pn ð13:14Þ
i ENSwithout ðfeederi Þ
376 C. F. M. Almeida et al.

where
• ENSwith ðfeederi ; NO1 ; . . .; NOk ; Alternative#Þ: is a function that returns the
corresponding value of ENS for feederi , depending on the state defined by the
combination of values on NO1 ; . . .; NOk , and on the combination of NC-AR to
be installed;
• ENSwithout ðfeederi Þ: is a function that returns the original value of ENS for
feederi , without the installation of the AR.

13.3 Results

The simulations were performed considering interruptions occurrences referring to

the years of 2012, 2013, 2014 and part of 2015 (until February 19, 2015) together,
corresponding to 327,472 records for the whole utility concession area. Such
occurrences were extracted from the OMS database of the Brazilian utility com-
pany. Each occurrence describes the protection equipment that have operated, the
time the interruption started and the time the service was restored. With this
information, it is possible to calculate the average duration time for repairing fault at
each load block and the average failure rate for eat load block.
For the second stage, all GA-based optimizations considered 50 generations with
100 individuals each. For the third stage, all GA-based optimizations considered
200 generations with 500 individuals each. The simulation time spent in both
analysis was around 15 min. All simulations were executed in a virtual machine,
considering 2 processors and 8 GB of RAM memory.
Another relevant premise for the studies carried out was the consideration of AR
already previously installed or not. That is, the studies can be divided into:
• Brown field: considers the effects of existing reclosers and intends to carry out
the installation of new AR;
• Green field: when the existing reclosers are disregarded and it is intended to
reallocate existing AR.
Figure 13.15 illustrates the pilot area where the proposed methodology was
applied. The pilot area for application of the methodology is composed by two real
substations (SED) from a Brazilian utility. Some of the characteristics of the net-
work under evaluation are:
• The first SED is composed of nine feeders, one of which is a distress feeder;
• The second SED consists of sixteen feeders, 2 of which are distress feeders;
• There are 14,513 buses;
• There are 13,803 segments of line;
• There are 17 AR installed in the feeders served by first SED, of which 4 are NO
type;
13 Optimal Allocation of Automatic Reclosers 377

Fig. 13.15 Pilot Area

• There are 28 AR installed in the feeders served by second SED, of which 8 are
NO type;
• There are 765 switches installed (among regular switches and fuses);
• There are 99,946 customers;
• The original HIC value is 779,727.13 h/year;
• The original IC value is 641,862.52 customers/year;
• The original ENS value is 304.65 MWh/year.

13.3.1 Brown Field Analysis

The second stage of the methodology determine 261 possible positions for instal-
ling NC reclosers, and 35 possible positions for installing NO reclosers. Table 13.1
details the results found for the brown field analysis.
As one can observe, regardless of the limitations imposed, the most significant
reduction in terms of CIH is achieved by installing 20 new NO reclosers and 48
378 C. F. M. Almeida et al.

Table 13.1 Brown field simulation

Best Best IC Best Allocation of Allocation of Allocation of
HIC ENS 10 new 20 new 30 new
reclosers reclosers reclosers
# of NO 20 15 20 2 5 8
reclosers
# of NC 48 34 47 8 15 22
reclosers
Benefit in 19.23% – – 10.16% 15.00% 16.44%
HIC
reduction
Benefit in – 19.25% – 10.38% 16.24% 17.84%
IC
reduction
Benefit – – 25.61% 9.52% 16.18% 21.38%
ENS
reduction

new NC reclosers. In terms of CI reduction, the best result is achieved by installing

15 new NO reclosers and 34 new NC reclosers. And in terms of ENS reduction, the
best result is achieved by installing 20 new NO reclosers and 47 new NC reclosers.
These results are obtained by combining the most significant reduction achieved for
each feeder at the second stage of the methodology. Thus, the GA approach for
global optimization had not been used yet.
The global optimization simulations consider constraints for the allocation of
ten, twenty and thirty reclosers. In addition, the analysis considers the best results
for each collective index, regardless of the limitations imposed. Thus, for the
simulation for allocation ten reclosers, the methodology proposed the installation of
2 NO reclosers and 8 NC reclosers. For the simulation for allocation twenty
reclosers, the methodology proposed the installation of 5 NO reclosers and 15 NC
reclosers. And for the simulation for allocation thirty reclosers, the methodology
proposed the installation of 8 NO reclosers and 22 NC reclosers.

13.3.2 Green Field Analysis

The second stage of the methodology determine 286 possible positions for instal-
ling NC reclosers, and 38 possible positions for installing NO reclosers. Table 13.2
details the results found for the green field analysis.
As one can observe, regardless of the limitations imposed, the most significant
reduction in terms of HIC is achieved by installing 26 new NO reclosers and 61
new NC reclosers. In terms of IC reduction, the best result is achieved by installing
13 Optimal Allocation of Automatic Reclosers 379

Table 13.2 Green field simulation

Best Best IC Best Reallocation of 45 existing
HIC ENS reclosers
# of NO reclosers 26 25 28 9
# of NC reclosers 61 56 61 36
Benefit in HIC 37.46% – – 34.44%
reduction
Benefit in IC – 36.12% – 33.29%
reduction
Benefit in ENS – – 41.83% 32.73%
reduction

25 new NO reclosers and 56 new NC reclosers. And in terms of ENS reduction, the
best result is achieved by installing 28 new NO reclosers and 61 new NC reclosers.
The global optimization simulations consider constraints for the reallocation of
forty-five reclosers. Thus, the simulation for allocation ten reclosers, the method-
ology proposed the installation of 9 NO reclosers and 36 NC reclosers.

13.3.3 General Remarks

A comparison of the results obtained from both analysis was presented in

Fig. 13.16. Observing Fig. 13.16 it is possible to identify that the allocation of new
reclosers improves the performance indices of service quality. It is also noticed that
the improvement tends to be less intense with the increase of the number of devices
allocated, leading to a saturation of the benefit coming from the allocation of the
reclosers.

Fig. 13.16 Histogram comparing both analysis

380 C. F. M. Almeida et al.

Another interesting analysis is comparing the results obtained from the brown
field analysis for allocating 20 new reclosers, which results the results obtained with
the analysis for allocating 45 existing reclosers in total and the green field simu-
lation for 45 reclosers, which also results in 45 reclosers in total. Observing these
results, one can see the relevance of the methodology, since the relocation results
are superior to the benefits of installing new reclosers. Thus, it is assumed that the
methodology can present superior results when compared to the current method-
ologies of allocation used by most of the utility companies.

References

1. D.E. Goldberg, Genetic Algorithms in Search, Optimization, and Machine Learning

(Addison-Wesley, 1989)
2. G. Levitin, S. Mazal-Tov, D. Elmakis, Genetic algorithm for optimal sectionalizing in radial
distribution systems with alternative supply. Electr. Power Syst. Res. 35(3) (1995)
3. R. Billinton, S. Jonnavithula, Optimal switching device placement in radial distribution
systems. IEEE Trans. Power Deliv. 11(3) (1996)
4. F. Soudi, K. Tomsovic, Optimal distribution protection design: quality of solution and
computational analysis. Int. J. Electr. Power Energy Syst. 21(5) (1999)
5. G. Celli, F. Pilo, Optimal sectionalizing switches allocation in distribution networks. IEEE
Trans. Power Deliv. 14(3) (1999)
6. F. Soudi, K. Tomsovic, Optimal trade-offs in distribution protection design. IEEE Trans.
Power Deliv. 16(2) (2001)
7. J.H. Teng, C.N. Lu, Feeder-switch relocation for customer interruption cost minimization.
IEEE Trans. Power Deliv. 17(1) (2002)
8. J.H. Teng, Y.H. Liu, A novel ACS-based optimum switch relocation method. IEEE Trans.
Power Syst. 18(1) (2003)
9. M.R. Haghifam, Optimal allocation of tie points in radial distribution systems using a genetic
algorithm. Eur. Trans. Electr. Energy Syst. (2004)
10. L.G.W. Silva, R.A.F. Pereira, J.R.S. Mantovani, Allocation of protective devices in
distribution circuits using nonlinear programming models and genetic algorithms. Electr.
Power Syst. Res. 69(1) (2004)
11. D.H. Popovic, J.A. Greatbanks, M. Begovic, A. Pregelj, Placement of distributed generators
and reclosers for distribution network security and reliability. Int. J. Electr. Power Energy
Syst. 27(5–6) (2005)
12. C.S. Chen, C.H. Lin, H.J. Chuang, C.S. Li, M.Y. Huang, C.W. Huang, Optimal placement of
line switches for distribution automation systems using immune algorithm. IEEE Trans.
Power Syst. 21(3) (2006)
13. V.C. Zamborlini, D.R. Trindade, E. Zambon, B.B. Garcia, E.F Azeredo, Otimização da
Alocação de Religadores em Larga Escala; II CBEE - Congresso Brasileiro de Eficiência
Energética (Vitória/ES, Brazil, 2007)
14. A. Moradi, M.F. Firuzabad, Optimal switch placement in distribution systems using trinary
particle swarm optimization algorithm. IEEE Trans. Power Deliv. 23(1) (2008)
15. L.G.W. Silva, R.A.F. Pereira, J.R. Abbad, J.R.S. Mantovani, Optimised placement of control
and protective devices in electric distribution systems through reactive Tabu search algorithm.
Electr. Power Syst. Res. 78(3) (2008)
16. A. Helseth, A.T. Holen, Impact of energy end use and customer interruption cost on optimal
allocation of switchgear in constrained distribution networks. IEEE Trans. Power Deliv. 23(3)
(2008)
13 Optimal Allocation of Automatic Reclosers 381

17. H. Falaghi, M.R. Haghifam, C. Singh, Ant colony optimization-based method for placement
of sectionalizing switches in distribution networks using a fuzzy multiobjective approach.
IEEE Trans. Power Deliv. 24(1) (2009)
18. W. Tippachon, D. Rerkpreedapong, Multiobjective optimal placement of switches and
protective devices in electric power distribution systems using ant colony optimization. Electr.
Power Syst. Res. 79(7) (2009)
19. N. Kagan, C.C.B. Oliveira, E.J. Robba, Introdução aos Sistemas de Distribuição de Energia
Elétrica. 2ª Edição (Editora Edgard Blucher, 2010)
20. C.C.B. Oliveira, D. Takahata, M. Maia, Metodologia de Alocação Otimizada de Dispositivos
de Proteção em Alimentadores Baseada no Desempenho Máximo do Alimentador (DMA).
IX CBQEE - Conferência Brasileira sobre Qualidade da Energia Elétrica (Cuiabá/MT,
Brazil, 2011)
21. D.P. Bernardon, M. Sperandio, V.J. Garcia, J. Russia, L.N. Canhab, A.R. Abaideb, E.F.B.
Daza, Methodology for allocation of remotely controlled switches in distribution networks
based on a fuzzy multi-criteria decision-making algorithm. Electr. Power Syst. Res. 81(2)
(2011)
22. D.P. Bernardon, M. Sperandio, V.J. Garcia, L.N. Canha, A.R. Abaide, E.F.B. Daza, AHP
decision-making algorithm to allocate remotely controlled switches in distribution networks.
IEEE Trans. Power Deliv. 26(3) (2011)
23. A.A. Jahromi, M.F. Firuzabad, M. Parvania, M. Mosleh, Optimized sectionalizing switch
placement strategy in distribution systems. IEEE Trans. Power Deliv. 27(1) (2012)
24. L.S. Assis, J.F.V. González, F.L. Usberti, C. Lyra, C. Cavellucci, F.J. Von Zuben, Switch
allocation problems in power distribution systems. IEEE Trans. Power Syst. 30(1) (2015)
25. J.C. López, J.F. Franco, M.J. Rider, Optimisation-based switch allocation to improve energy
losses and service restoration in radial electrical distribution systems. IET Gener. Transm.
Distrib. 10(11) (2016)