SAVITRIBAI PHULE PUNE UNIVERSITY

A PROJECT STAGE I REPORT ON

“ Prediction of Electricity Consumption Based on Geodemographic Factors
Using LSTM Model ”

SUBMITTED TOWARDS THE

PARTIAL FULFILMENT OF THE REQUIREMENTS OF
BACHELOR OF ENGINEERING
(INFORMATION TECHNOLOGY ENGINEERING)
by
Ms. Sanika Patil
Ms. Gayatri Surwade
Ms. Kaveri Shinde
Ms. Yashashri Shinde
Under The Guidance Of
Dr.Jaya R Suryawanshi

DEPARTMENT OF INFORMATION TECHNOLOGY

Maratha Vidya Prasarak Samaj’s

Karmaveer Adv. Baburao Ganpatrao Thakare
College of Engineering, Nashik-13

Academic Year: 2024-25

i
 Department of Information Technology

This is to certify that the Project Entitled

“Prediction of Electricity Consumption Based on

Geodemographic Factors Using LSTM Model”
Submitted by

Ms. Sanika Patil [B400290]

Ms. Gayatri Surwade [B400290400]
Ms. Kaveri Shinde [B400290397]
Ms. Yashashri Shinde [B400290398]

is a record of bonafide work carried out by them under the supervision and guidance of
Dr.Jaya R Suryawanshi in partial fulfilment of the requirement for Bachelor of Engineering
(Information Technology).

Dr.Jaya R Suryawanshi Dr. S.A.Talekar

Project Guide HOD(IT)

Dr. S. R. Devane External Examiner

Principal

MVPS’s KBT College of Engineering

ACKNOWLEDGEMENT

It gives us great pleasure in presenting the project report on “Prediction of Elec-

tricity Consumption Based on Geodemographic Factors Using LSTM Model
”.We would like to take this opportunity to thank our internal guide Dr.Jaya R Suryawan-
shi for giving us all the help and guidance we needed. We are really grateful to them for
their kind support. Their valuable suggestions were very helpful.
We are grateful to Dr. S. A. Talekar, Head of Information Technology Engineer-
ing Department, Maratha Vidya Prasarak Samaj’s Karmaveer Adv. Baburao Ganpatrao
Thakare College of Engineering, Nashik-13 for his indispensable support and suggestions.
We are also grateful to Management of Maratha Vidya Prasarak Samaj, Nashik
and Respected Principal Dr. S. R. Devane for providing all necessary facilities and
supports to complete our project within stipulated period. We are also grateful to Vice-
Principal, Project Coordinator, all teaching and non-teaching staff for their valuable
suggestions and support.

Ms. Sanika Patil [B400290]

Ms. Gayatri Surwade [B400290400]
Ms. Kaveri Shinde [B400290397]
Ms. Yashashri Shinde [B400290398]
(B.E. IT Engg.)

iii
ABSTRACT

The residential sector is a major consumer of electricity, and its demand will rise
by 65 percent by the end of 2050. Therefore, it is essential for the utility companies to
forecast the consumer demand and manage its supply. The electricity consumption of
a household is determined by various factors, e.g. house size, socio-economic status of
the family, size of the family, etc. Previous studies have only identified a limited number
of factors that affect electricity consumption with the ARIMA model. In this project ,
the influence of geodemographic factors in predicting the future power consumption of
a city. The dataset contains the geodemographic factors on electricity consumption for
homes in the City of Mumbai. Geodemographic factors cover a wide array of categories
e.g. social, economic, dwelling, family structure, health, education, finance, occupation,
and transport. Using Spearman correlation, factors that are strongly correlated with
electricity consumption are taken . To study the impact of geodemographic factors in
designing forecasting models. Specifically, build an encoder-decoder LSTM model which
shows improved accuracy with geodemographic factors. The idea will help energy com-
panies design better energy management strategies.

Keywords: Socio-economic factors, geodemographic factors, electricity forecasting, en-

coderdecoder model.

iv
Contents

CERTIFICATE ii

ACKNOWLEDGEMENT iii

ABSTRACT iv

LIST OF FIGURES viii

1 INTRODUCTION 1
1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Aim . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.3 Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.4 Objectives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3

2 LITERATURE REVIEW 4

3 PROBLEM STATEMENT DEFINITION 6

4 SOFTWARE REQUIREMENT SPECIFICATION 7

4.1 Purpose . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
4.2 Scope . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
4.3 Project Perspectives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
4.4 Assumptions and Dependencies . . . . . . . . . . . . . . . . . . . . . . . 8
4.5 Functional Requirements . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
4.6 PROJECT REQUIREMENT . . . . . . . . . . . . . . . . . . . . . . . . 9
4.7 Software and Hardware Requirement Specifications . . . . . . . . . . . . 10
4.7.1 Software Requirement . . . . . . . . . . . . . . . . . . . . . . . . 10
4.8 PyCharm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
4.9 Jupyter Notebook . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11

v
 4.10 Flow Chart . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13

5 PROPOSED SYSTEM ARCHITECTURE 14

5.1 Purposed System Architecture . . . . . . . . . . . . . . . . . . . . . . . 14
5.2 High Level Design Of Project . . . . . . . . . . . . . . . . . . . . . . . . 16
5.2.1 DFD Diagram . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
5.2.2 DFD Diagram . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
5.3 UML Diagram . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
5.3.1 UML Description . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
5.4 Activity Diagram . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22

6 SYSTEM IMPLEMENTATION 25
6.1 Algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
6.1.1 System Initialization . . . . . . . . . . . . . . . . . . . . . . . . . 25
6.1.2 User Authentication Module . . . . . . . . . . . . . . . . . . . . 25
6.1.3 Admin Workflow . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
6.1.4 Data Preprocessing Workflow . . . . . . . . . . . . . . . . . . . . 26
6.1.5 6.1.5 Model Training Workflow . . . . . . . . . . . . . . . . . . . 26
6.1.6 Prediction and Evaluation . . . . . . . . . . . . . . . . . . . . . . 26
6.1.7 Optimization (Optional) . . . . . . . . . . . . . . . . . . . . . . . 26
6.1.8 Data Synchronization . . . . . . . . . . . . . . . . . . . . . . . . . 26
6.1.9 Reports and Analytics . . . . . . . . . . . . . . . . . . . . . . . . 26
6.1.10 System Termination . . . . . . . . . . . . . . . . . . . . . . . . . 27
6.2 Methodology/Protocols . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
6.2.1 Problem Identification . . . . . . . . . . . . . . . . . . . . . . . . 27
6.2.2 Technological Integration . . . . . . . . . . . . . . . . . . . . . . . 27
6.2.3 Tracking and Feature Verification . . . . . . . . . . . . . . . . . . 27
6.2.4 Cloud Synchronization . . . . . . . . . . . . . . . . . . . . . . . . 27
6.2.5 Database Management . . . . . . . . . . . . . . . . . . . . . . . . 28
6.2.6 6.2.6 User Engagement . . . . . . . . . . . . . . . . . . . . . . . . 28
6.2.7 Outcome-Oriented Design . . . . . . . . . . . . . . . . . . . . . . 28
6.3 Testing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29

7 WORKING MODULE WITH EXPERIMENTAL RESULTS 33

8 PROJECT PLAN 38

vi
9 FUTURE SCOPE 40

10 CONCLUSION 43

11 BIBLIOGRAPHY 44

vii
List of Figures

4.1 Flow chart . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13

5.1 Architecture . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
5.2 DFD-0 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
5.3 DFD-1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
5.4 DFD-2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
5.5 UML Diagram . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
5.6 Activity Diagram . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24

7.1 Download Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33

7.2 Download Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
7.3 Data Preprocessing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
7.4 Data Preprocessing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
7.5 Data Preprocessing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
7.6 Model Building . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
7.7 Model Building . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
7.8 Correlation Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
7.9 Actual vs. Predicted Consumption . . . . . . . . . . . . . . . . . . . . . 37

viii
List of Tables

2.1 Research Papers and Summaries . . . . . . . . . . . . . . . . . . . . . . . 4

2.2 Research Papers and Summaries . . . . . . . . . . . . . . . . . . . . . . . 5

ix
Chapter 1

INTRODUCTION

1.1 Introduction
This chapter briefly focuses on the project idea , it gives us the information about
the technologies that would be used in the project implementation. This chapter focuses
project Aim, Scope and Objectives in detailed The residential sector is the largest sector
for the electricity usage, and its demand is rapidly increased in last few decades. So, it is
necessary for the utility companies to forecast the user demand and manage its supply.
The power consumption is basically determined by the various factors, social,economical,
transport, education, dwelling and digital. This all factors are considered to find elec-
tricity consumption of the consumers and the whole city. The geodemographic factor
info. can help the electricity generation companies build the better power management
strategies. The consumer usage of the electricity is identified by the various factors.,For
example- size of the house, no. of members in the house, various social and economi-
cal factors, etc. Geo-demographic information can help power generation companies to
design better forecasting models that can help to predict the consumer demand more
precisely. This will also ensure that the companies meet their demand supply balance.
To find the impact of geodemographic factors in predicting the future power consump-
tion of a city, we are using Spearman correlation to identify the factors that are strongly
correlated with consumption of electricity.

1
1.2 Aim

1.To find the impact of geodemographic factors in predicting the future power consump-
tion of a city, we are using Spearman correlation to identify the factors that are strongly
correlated with consumption of electricity.
2. To study the impact of geodemographic factors in designing forecasting models. We
will build an encoder-decoder LSTM model which shows improved accuracy with geode-
mographic factors.

1.3 Motivation
The need for efficient electricity management is growing as populations and cities
expand, putting more pressure on energy resources. Geodemographic factors, like where
people live, their income levels, and weather conditions, greatly affect how much electric-
ity is used in different areas. Traditional methods for predicting electricity consumption
often struggle to account for these complex patterns over time. This is where LSTM
(Long Short-Term Memory) models can help. LSTM is a type of machine learning model
that can analyze past electricity usage and predict future demand more accurately by
considering these factors. This project aims to use LSTM to improve electricity con-
sumption predictions, helping utility companies plan better, reduce waste, and ensure a
reliable energy supply for everyone.

2
1.4 Objectives
The Objectives of our project are as follows:

• Understand the prediction method using LSTM Neural Network.

• Create accurate model for companies, to design better energy management strate-
gies.

• Incorporate geodemographic data to identify key factors like population density,

income levels, and weather that influence electricity consumption patterns.

• Enhance the model’s accuracy by testing and tuning hyperparameters, such as

learning rate, batch size, and number of LSTM units.

• Compare the performance of the LSTM model with traditional prediction methods,
such as ARIMA or linear regression, to highlight the advantages of using deep
learning for time-series forecasting.

• Analyze the impact of temporal patterns (daily, weekly, seasonal) on electricity

consumption to help predict peaks and valleys in energy demand.

• Provide insights for decision-makers by developing a model that allows utility com-
panies to make data-driven decisions on energy distribution, resource allocation,
and demand-side management.

• To provide fast and efficient service for the customer in hotels.

• To develop a time saving and cost-effective device which will help in providing fast
service in hotels.

3
Chapter 2

LITERATURE REVIEW

Research Pa- Author Summary

per
Machine- Gang Chen, This paper discusses an ML-based framework
Learning-Based Qingchang Hu, to enhance electricity demand forecasting,
Electric Power Jin Wang, Xu emphasizing socio-economic and climatic fac-
Forecasting Wang tors.
Electricity Load Yuyu Zhu This systematic review covers methods in
Forecasting: A electricity load forecasting, comparing data-
Systematic Re- driven AI techniques with traditional models.
view It categorizes studies based on forecast type.
Electricity De- Beullens, W., et The research applied linear regression, sup-
mand Prediction al. port vector regression (SVR), and ensem-
Using Regres- ble learning to predict household energy use.
sion Models Data preprocessing involved handling miss-
ing values and training/testing splits, leading
to improved accuracy when ensemble models
were used.

Table 2.1: Research Papers and Summaries

4
Research Pa- Author Summary
per
The impact of Mohamed This paper uses multiple linear regression
economic and models to predict power consumption in New
demographic Zealand from 1965 to 1999 using GDP, elec-
variables on tricity prices, and population data.
power consump-
tion
A GPRM Akay This study develops a GPRM approach to
approach to predict electricity demand in Turkey. GPRM
predict electric- is computationally simple and effective for
ity demand in volatile data prediction.
Turkey
A Regression Bianco This paper uses a regression model to pre-
model for fore- dict electricity consumption in Italy based
casting electrical on GDP per capita, focusing on high-level
consumption in parameters.
Italy using GDP
per capita
Merkle Signa- Merkle, R MSS is hash-based cryptographic scheme
ture Scheme that employs AES-based hash functions. It
is recognized for its smaller code size and
faster verification process compared to RSA
and ECC.
NTRU (Number Hoffstein, J., et NTRU is a public key cryptosystem based
Theory Research al. on lattice-based cryptography, offering resis-
Unit) tance to quantum attacks such as those posed
by Shor’s algorithm.

Table 2.2: Research Papers and Summaries

5
Chapter 3

PROBLEM STATEMENT
DEFINITION

3.1 Problem Definition

The rise in the electricity consumption and the demand for that will increased so
rapidly. So the electricity companies need to manage the supply of electricity and ful-
filled the consumers demand.To get all of this we study the different factors and applying
algorithm to find the electricity demand. The problem statement of the system is come
up with the simple lines, i.e, ”are the geodemographic factors are important to find the
electricity consumption of city/home? ” To get the answer of above question, we split
the methodology in two partsThe first part of this include the study of correlation be-
tween the factors and the electricity consumption of the home. Then in the second part
we build an encoder-decoder LSTM model to predict the consumption of the city using
demographic factors information.

6
Chapter 4

SOFTWARE REQUIREMENT
SPECIFICATION

4.1 Purpose
The purpose of this project is to develop a software system that forecasts residential
electricity consumption by incorporating a wide range of geodemographic factors. This
system will provide utility companies with accurate demand predictions, enabling better
supply management and more efficient energy planning. The project specifically focuses
on the city of Mumbai and aims to help energy providers design improved management
strategies based on detailed, data-driven insights.

4.2 Scope
The system will analyze and forecast electricity consumption patterns in Mumbai’s res-
idential sector by considering geodemographic factors such as household size, socio-
economic status, dwelling characteristics, family structure, and more. This project will
employ an encoder-decoder LSTM model to improve forecasting accuracy compared to
traditional methods like ARIMA. The system’s outputs will be used by utility compa-
nies to manage supply more effectively, optimize resource allocation, and plan for future
demand increases.

7
4.3 Project Perspectives
The system will operate as a standalone analytical and forecasting tool, integrating with
external datasets provided by city or utility authorities. Users, primarily utility company
analysts and planners, will access a web-based interface to interact with the model’s
outputs. The system will utilize machine learning libraries (e.g., TensorFlow, Scikit-
learn) for model training and evaluation, and visualization tools (e.g., Plotly, Matplotlib)
for interactive dashboards.

4.4 Assumptions and Dependencies

Assumptions: The geodemographic and electricity consumption data are accurate, con-
sistent, and up-to-date. The utility companies have access to historical electricity con-
sumption data. Sufficient computational resources are available for training and running
the LSTM model. Users of the system are familiar with interpreting forecasting outputs
and making data-driven decisions. Dependencies: Access to comprehensive geodemo-
graphic datasets for Mumbai. Machine learning frameworks such as TensorFlow/Keras
for implementing the LSTM model. Data processing and analysis libraries, including
Pandas and NumPy. Visualization libraries like Plotly or Matplotlib for creating dash-
boards and reports. A hosting environment for deploying the web-based dashboard (e.g.,
cloud-based platform).

4.5 Functional Requirements

1: Data Collection and Preprocessing - The system shall collect and preprocess geodemo-
graphic and electricity consumption data, including handling missing values and outliers.
2: Feature Selection - The system shall perform feature selection using Spearman
correlation to identify factors with strong correlations to electricity consumption.
3: Model Training and Forecasting - The system shall implement an encoder-decoder
LSTM model to predict future electricity consumption based on selected features. The
system shall allow model retraining with updated data to maintain prediction accuracy
over time.
4: Model Evaluation - The system shall evaluate model accuracy using metrics like
MAE, RMSE, and R². The system shall compare LSTM forecasting results with those

8
from traditional models like ARIMA for benchmarking.
5: Data Visualization - The system shall provide a dashboard with visualizations
of historical trends, forecasted electricity demand, and key influencing factors. The
dashboard shall allow users to view predictions over various time horizons (e.g., monthly,
yearly).
6: Report Generation - The system shall generate reports on electricity consumption
trends and forecasted demand for utility company stakeholders.
7: User Interface - The system shall provide a web-based interface that allows users
to input data, run forecasts, and view visualizations easily. The system shall allow users
to download generated reports and graphs.

4.6 PROJECT REQUIREMENT

In the field of energy management, predicting electricity consumption is crucial for effi-
cient resource allocation and sustainability. Accurate forecasting can aid utility compa-
nies in optimizing power generation, distribution, and consumption, thus reducing waste
and promoting energy efficiency. Traditionally, predictive models have relied on limited
factors such as GDP or electricity prices, often ignoring the influence of geodemographic
factors. However, these factors, such as population density, income levels, and weather
conditions, play a significant role in determining electricity demand.
To overcome the limitations of conventional models and incorporate a broader range
of influencing factors, we propose a solution that leverages advanced machine learning
techniques. Our project involves developing a predictive model using Long Short-Term
Memory (LSTM) networks, which are well-suited for handling time-series data. By incor-
porating geodemographic factors, our model aims to provide more accurate and dynamic
electricity consumption forecasts. This will empower utility companies to design better
energy management strategies and optimize resources, ultimately contributing to a more
sustainable and energy-efficient future.

9
4.7 Software and Hardware Requirement Specifica-
tions

4.7.1 Software Requirement

Functional Requirements
• User can upload the Dataset.

• User can view the co-related factors

• User can view the prediction of electricity consumption of month.

• User can observe the difference between the results of consumption of electricity
with and without Geo-demographic factors.

Non-Functional Requirements
• The prediction system shall be easier to use and smoother user experience.

• Even if there are any updates system need to perform efficiently i.e., the system
should be scalable .

• The system needs to work on every device.

4.8 PyCharm

PyCharm is a powerful Integrated Development Environment (IDE) designed specif-

ically for programming in Python. An IDE is a software application that provides
comprehensive facilities to programmers for software development. PyCharm, de-
veloped by the Czech company JetBrains, is widely used by both beginners and
professional developers due to its user-friendly interface and a rich set of features.
One of the standout features of PyCharm is its excellent code completion. This
means that as you start typing, PyCharm suggests possible completions for your
code. This not only speeds up coding but also helps prevent errors by guiding
you toward correct syntax and function names. Additionally, PyCharm includes
smart code inspection tools that automatically check your code for potential issues

10
and suggest improvements. Debugging is an essential part of programming, and
PyCharm provides a robust graphical debugger that allows you to step through
your code line by line. This visual debugging feature helps you identify and fix
bugs more efficiently. You can set breakpoints to pause the execution of your
code at specific points, making it easier to examine variables and the flow of your
program. Another significant feature of PyCharm is its integrated unit testing sup-
port. Unit tests are essential for ensuring that individual parts of your code work as
intended. PyCharm simplifies the process of writing and running these tests, mak-
ing it easier to maintain code quality. PyCharm also integrates seamlessly with
version control systems like Git, allowing you to manage changes to your code-
base effectively. This integration is crucial for collaboration in team environments,
where multiple developers work on the same project. You can easily track changes,
create branches, and merge code without leaving the IDE.For those interested in
web development, PyCharm supports popular web frameworks, including Django,
Flask, and FastAPI. This makes it an excellent choice for developers building web
applications in Python. The IDE provides tools for HTML, CSS, and JavaScript
development, along with features tailored specifically for web frameworks.

4.9 Jupyter Notebook

Jupyter Notebook is a powerful tool that helps people work on data projects in
one convenient place. It allows users to combine different elements like data, code,
and visualizations into a single document. This makes it easier to showcase the
entire process of a project to others.In a Jupyter Notebook, you can write and
run computer code, such as Python, which is a popular programming language for
data analysis. Besides code, you can also include text to explain your work, use
markdown for formatting, and add figures and links for better understanding.The
Jupyter Notebook runs as a web application, which means you access it through
your web browser. When you open it, you see a Dashboard or control panel that
shows all your local files. This dashboard allows you to open notebook documents
and run code snippets directly within the notebook.As you run the code, the out-
puts are displayed right below the code cells in a neat and organized way. This
makes it easy to see the results of your code without switching between different
programs or windows. You can also edit the notebook easily, allowing for quick ad-

11
justments and updates as needed.One of the best things about Jupyter Notebook is
that it is widely used and well-documented. This means there are many resources
available for learning how to use it effectively, making it accessible for beginners
and experienced users alike.Additionally, Jupyter Notebooks can be shared with
others, making it a great way to collaborate on projects. You can create interactive
stories that show your analysis step-by-step, helping others understand your work
better.Overall, Jupyter Notebook provides a user-friendly interface for creating,
editing, and running data projects, making it a valuable tool for anyone working
with data. Whether you’re a student, a researcher, or a data professional, Jupyter
Notebook can help you organize your thoughts and present your findings clearly.

12
4.10 Flow Chart

Figure 4.1: Flow chart

13
Chapter 5

PROPOSED SYSTEM
ARCHITECTURE

5.1 Purposed System Architecture

14
Figure 5.1: Architecture

15
5.2 High Level Design Of Project

5.2.1 DFD Diagram

Figure 5.2: DFD-0

16
Figure 5.3: DFD-1

17
Figure 5.4: DFD-2

18
5.2.2 DFD Diagram

The Data Flow Diagram (DFD) for the electricity consumption prediction project
provides a detailed overview of how data moves within the system and how various
processes interact with one another. The DFD is divided into different levels to give
both a high-level and more detailed understanding of the system’s architecture.
Level 0 DFD : At the highest level, the system consists of three main entities:
the User, the Electricity Consumption Prediction System, and an External Data
Source. The User uploads a dataset containing electricity consumption records
and geodemographic factors like population density, weather conditions, etc. The
External Data Source provides additional geodemographic data that can be inte-
grated into the prediction model. The Electricity Consumption Prediction System
processes these inputs and provides the predicted electricity consumption results
back to the User.
The Electricity Consumption Prediction System from Level 0 is further broken
down into sub-processes to show the detailed flow of data within the system.

– Data Preprocessing: The first step in the system is preprocessing the uploaded
dataset from the User. This involves cleaning the data, handling missing
values, and formatting it for further analysis.

– Data Integration: In the second step, the cleaned dataset is integrated with
geodemographic data from the External Data Source. This step is crucial
for combining electricity consumption data with factors such as population
density, climate conditions, and other environmental factors that could impact
electricity usage.

– Model Training and Prediction: Once the data is integrated, the system moves
to the most critical phase: model training and prediction. Here, the LSTM
(Long Short-Term Memory) model is used to analyze historical electricity
consumption patterns, considering the external factors, and predict future
consumption trends.

– Result Generation: After the prediction is complete, the system prepares the
results in a user-friendly format. The predicted electricity consumption data
is then sent back to the User as the final output of the system.

19
5.3 UML Diagram

Figure 5.5: UML Diagram

5.3.1 UML Description

The UML Class Diagram presented above illustrates the structure of a system
designed for electricity consumption forecasting using an LSTM model based on
geodemographic factors. It includes several key classes: User, Geodemographic-
Data, HistoricalData, ForecastModel, and ForecastOutput. Each class contains
attributes representing the data it holds, such as user information, data IDs, and
forecast results, as well as methods that define the operations the class can perform.
For example, the User class has methods for inputting data and viewing forecasts,
while the ForecastModel class includes methods for training the model and gener-
ating forecasts based on the provided data.
The relationships between the classes highlight how they interact within the sys-
tem. The User class has a one-to-many association with the ForecastOutput class,
indicating that a single user can access multiple forecast outputs. Additionally,

20
the ForecastModel class has dependencies on both GeodemographicData and His-
toricalData, which means it requires these data types to function effectively. This
diagram effectively encapsulates the relationships and functionalities necessary for
the electricity consumption forecasting process, providing a clear overview of how
different components work together to achieve the system’s objectives.

21
5.4 Activity Diagram

The diagram represents the workflow of a project focused on predicting electric-

ity consumption using geodemographic factors with an LSTM (Long Short-Term
Memory) model. The project follows a series of structured steps, ensuring data is
collected, processed, and fed into the LSTM algorithm to predict future electricity
usage effectively.
The process begins with user interaction, where the user uploads the dataset
that contains relevant time-series data. This dataset likely includes various fea-
tures such as past electricity consumption records, geodemographic information
(like population size, income levels, housing characteristics), and environmental
factors (such as temperature, seasons, or weather conditions). Once the dataset is
uploaded, the system enters a data loading phase, where the raw data is prepared
for further processing. This may involve handling missing data, normalizing values,
or integrating additional external datasets to provide more context. Data fetching
may also happen here to bring in information from external sources, like weather
databases, economic reports, or regional development data.
After the dataset is loaded and ready, the user triggers the application of the
LSTM algorithm. The Long Short-Term Memory (LSTM) model is a specialized
type of neural network that excels at time-series data analysis. In this context,
the LSTM will learn the patterns of electricity consumption over time, considering
how different factors (like weather, population, and time of year) impact usage.
Encoder : The first component within the LSTM architecture is the encoder. The
encoder processes the input time-series data (e.g., electricity consumption records
and geodemographic factors) and captures long-term dependencies and patterns.
For example, it might recognize that electricity consumption spikes in summer
months or weekends based on historical data. The encoder learns which patterns
are important to keep in memory and which to forget, ensuring that relevant data
is passed forward for future predictions.
Decoder : After encoding the time-series data, the decoder takes this encoded
information and translates it into predictions. The decoder’s job is to interpret the
patterns it learned from the encoder and apply them to the data, predicting future
electricity consumption for different regions. For example, if the model knows that
electricity consumption rises with temperature, it can predict higher consumption

22
for an area based on upcoming weather forecasts.
Dense Layer : Following the decoder, the dense layer refines the prediction
further. This layer is a fully connected neural network layer, which means it pro-
cesses the outputs of the decoder and makes final adjustments to the predicted
values. This step ensures that the final output is as accurate as possible, using the
learned relationships between features (like geodemographics and electricity use)
to create a reliable forecast.
Prediction Result : The final output is the result, which represents the pre-
dicted electricity consumption for the future time periods. This result is based on
all the patterns the model has learned from historical data, such as consumption
trends, the influence of geodemographic factors, and environmental variables. The
prediction can help energy providers, municipalities, or businesses better manage
electricity resources, optimize energy supply, and anticipate demand.

23
Figure 5.6: Activity Diagram

24
Chapter 6

SYSTEM IMPLEMENTATION

6.1 Algorithm

6.1.1 System Initialization

Load electricity consumption and geodemographic datasets into the environment.

Initialize required libraries and data processing pipelines (Pandas, NumPy, Tensor-
Flow). Load pre-trained models or prepare the LSTM architecture for training.

6.1.2 User Authentication Module

User (utility admin, researcher) logs in with credentials. Authenticate via backend
service or cloud identity provider. Grant access to dashboards and forecasting tools
upon verification.

6.1.3 Admin Workflow

Upload, view, and manage datasets. Select geodemographic features using correla-
tion analysis. Train the model, visualize results, and download forecasts. Monitor
model accuracy and retrain with new data as needed.

25
6.1.4 Data Preprocessing Workflow

Clean null and duplicate values from the dataset. Normalize and reshape the time-
series data for LSTM compatibility. Apply Spearman correlation to rank features
by relevance. Prepare train-test datasets.

6.1.5 6.1.5 Model Training Workflow

Initialize encoder-decoder LSTM model with defined layers (e.g., 200 units). Train
the model on selected features and past consumption values. Validate using test
set and calculate metrics (MSE, MAE). Save trained model for future use.

6.1.6 Prediction and Evaluation

Load the trained model. Input geodemographic features for a given time window.
Generate future electricity consumption predictions. Display or store results and
performance graphs.

6.1.7 Optimization (Optional)

Tune hyperparameters: number of LSTM units, learning rate, batch size. Use K-
fold cross-validation for model robustness. Implement dropout or early stopping to
avoid overfitting.

6.1.8 Data Synchronization

Forecast results are stored in a database (e.g., SQLite or Firebase). Prediction

dashboards update dynamically. Logs and previous predictions remain accessible
for version tracking.

6.1.9 Reports and Analytics

Generate and visualize: Predicted vs actual consumption. Influence of top geode-

mographic features. Forecast accuracy over different time periods. Export data in
Excel or PDF format.

26
6.1.10 System Termination

Shut down model training and save final weights. Export processed datasets and
evaluation reports. Log out user sessions and close active dashboards.

6.2 Methodology/Protocols

6.2.1 Problem Identification

The residential electricity demand is growing rapidly, yet most existing models fail
to incorporate geodemographic variability. Traditional methods like ARIMA lack
the ability to process multiple features and capture temporal non-linearity. There
is a need for intelligent forecasting systems that adapt to real-world city dynamics.

6.2.2 Technological Integration

Python (TensorFlow/Keras): Used to implement the encoder-decoder LSTM ar-

chitecture. Jupyter/VS Code: Environment for development and experimentation.
Data Visualization Libraries: Matplotlib, Seaborn for trend analysis and results
display.

6.2.3 Tracking and Feature Verification

Spearman Correlation: Applied to analyze relationships between geodemographic

variables and electricity consumption. Rank-based Filtering: Ensures only statis-
tically relevant features are used.

6.2.4 Cloud Synchronization

Forecasts and visualizations can be hosted using cloud dashboards (e.g., Firebase
or Streamlit). Ensures access across teams and devices with real-time updates.

27
6.2.5 Database Management

Input datasets and forecast results can be stored using MySQL or SQLite. Helps
maintain long-term records, user logs, and feature-impact analysis.

6.2.6 6.2.6 User Engagement

Energy companies or policymakers can interact via a frontend to upload data and
view predictions. Results can assist in strategy planning and identifying high-
consumption localities based on demography.

6.2.7 Outcome-Oriented Design

The system is built to provide accurate, scalable, and interpretable energy forecasts.
Helps reduce power shortages, optimize supply distribution, and promote efficient
urban planning through data-driven decision-making.

28
6.3 Testing

Testing Types

1. Accessibility Testing Accessibility testing for data integrity proofs focuses on

ensuring that the system is accessible to all users, including those with disabilities
or impairments.
2 Acceptance Testing Acceptance testing ensures that the end-user (customers)
can achieve the goals set in the business requirements, which determines whether
the software is acceptable for delivery or not. It is also known as user acceptance
testing (UAT). In UAT we checked that all the seven principal emotions are recog-
nized.
3 Blackbox Testing Black box testing involves testing against a system where
the code and paths are invisible. Giving the system to the user and how verified
how feasible the system is for the user.
4 End To End Testing End-to-end testing is a technique that tests the ap-
plication’s workflow from beginning to end to make sure everything functions as
expected. Here check the complete flow of the system.
5 Functional Testing Functional testing checks an application, website, or sys-
tem to ensure it’s doing exactly what it’s supposed to be doing. A website which
we develop is working properly
6 Interactive Testing Also known as manual testing, interactive testing enables
testers to create and facilitate manual tests for those who do not use automation
and collect results from external tests.
7 Integration Testing Integration testing ensures that an entire, integrated sys-
tem meets a set of requirements. It is performed in an integrated Hardware and
software environment to ensure that the entire system functions properly. Here we
verifed that system is capable of taking the image from camera and extract features
from it.
8 Non Functional Testing Non-functional testing verifies the readiness of a sys-
tem according to non functional parameters (performance, accessibility etc.) which
are never addressed by functional testing. So now here we checked the performance
of the system i.e how fast it is able to recognize the emotions of individual or group.
9 Performance Testing Performance testing examines the speed, stability, reli-
ability, scalability, and resource usage of a software application under a specified

29
workload.
10 Unit Testing Unit testing is the process of checking small pieces of code to
ensure that the individual parts of a program work properly on their own, speeding
up testing strategies and reducing wasted tests.
.

Test Case 1: Data Upload and Preprocessing

1. Test Case Name: Data Upload and Preprocessing
2. Test Description: Verify that the user can upload electricity consumption
and geodemographic datasets and that the system processes the data correctly.
3. Test Steps:
– Open the Electricity Forecasting application or Jupyter Notebook.
– Click or execute the data upload cell.
– Select valid CSV files for electricity consumption and geodemographic
data.
– Click “Upload” or run the preprocessing cell.
– Verify that null values are handled and features are scaled properly.
– Confirm that the processed data is displayed or saved.

Test Case 2: Feature Selection using Spearman

Correlation
1. Test Case Name: Feature Selection using Spearman Correlation
2. Test Description: Verify that the system correctly identifies and selects the
most relevant geodemographic features for forecasting.
3. Test Steps:
– Load the cleaned dataset.
– Run the feature selection module.
– Check the correlation scores for each feature.

30
 – Verify that only features with high absolute correlation are selected.
– Confirm that selected features are stored for model training.

Test Case 3: Model Training with LSTM

1. Test Case Name: Model Training with LSTM
2. Test Description: Verify that the encoder-decoder LSTM model trains suc-
cessfully with selected features and outputs performance metrics.
3. Test Steps:
– Load the selected features and consumption data.
– Initialize the LSTM model architecture.
– Start model training.
– Monitor the training loss and accuracy.
– Verify that the model completes training and saves the weights.
– Confirm that performance metrics such as MSE are displayed.

Test Case 4: Electricity Consumption Prediction

Test Case Name: Electricity Consumption Prediction Test Description:Verify

that the model generates accurate consumption forecasts for future time periods.
Test Steps:

– Load the trained model.

– Provide a test input (e.g., recent geodemographic and consumption data).
– Click “Predict” or run the prediction cell.
– Verify that output values for future consumption are generated.
– Compare predicted values with actual data to evaluate accuracy.

31
Test Case 5: Visualization of Results
1. Test Case Name: Visualization of Results
2. Test Description: Verify that the application displays prediction results and
feature correlations through graphs.
3. Test Steps:
– Run the result visualization module.
– Generate line plots for actual vs predicted electricity consumption.
– Generate heatmaps or bar charts for feature correlations.
– Confirm that the plots are clear, labeled, and match expected trends.

Test Case 6: Forecast Report Export

1. Test Case Name: Forecast Report Export
2. Test Description: Verify that the user can export prediction results and
model evaluation metrics as a report.
3. Test Steps:
– After prediction and evaluation, click on “Export Report” or run the
export cell.
– Choose the export format (CSV, Excel, PDF).
– Confirm that the file downloads successfully.
– Open the file and verify that it contains prediction results, metrics, and
feature details.

32
Chapter 7

WORKING MODULE WITH

EXPERIMENTAL RESULTS

Figure 7.1: Download Data

33
 Figure 7.2: Download Data

Figure 7.3: Data Preprocessing

34
Figure 7.4: Data Preprocessing

Figure 7.5: Data Preprocessing

35
Figure 7.6: Model Building

Figure 7.7: Model Building

36
 Figure 7.8: Correlation Analysis

Figure 7.9: Actual vs. Predicted Consumption

37
Chapter 8

PROJECT PLAN

The Electricity Consumption Prediction System is designed to assist utility

companies and city planners in accurately forecasting residential electricity
demand by incorporating deep learning and geodemographic analytics. The
system utilizes an encoder-decoder LSTM model and Spearman correlation to
understand consumption patterns and the influence of various social, economic,
and household-related factors. The primary goal is to enhance decision-making
for energy supply management and urban energy planning through reliable and
data-driven forecasting.

The project will be implemented through the following key modules:

– Data Management Module: Handles collection, preprocessing, and nor-

malization of electricity usage and geodemographic datasets. Ensures all data
is cleaned, formatted, and ready for analysis and modeling.
– Feature Selection Module:Applies Spearman correlation to identify and
extract the most relevant geodemographic features that impact electricity con-
sumption, thereby improving model efficiency and interpretability.
– Forecasting Module: Implements an encoder-decoder LSTM model that
takes selected features as input to generate future electricity consumption
predictions. This module handles model training, validation, and prediction
output.
– Visualization and Evaluation Module: Provides visual outputs such as
actual vs predicted consumption graphs and correlation heatmaps. Evaluates

38
 model accuracy using metrics like MSE and MAE to ensure reliability.
– Report Generation Module:Generates downloadable reports summarizing
forecast results, feature influences, and performance statistics. Helps stake-
holders review insights and make informed decisions.
– Database Module:Manages and stores datasets, model configurations, pre-
diction logs, and reports using databases like MySQL or SQLite. Enables
structured access and long-term data retention.

39
Chapter 9

FUTURE SCOPE

1. Smart Grid Integration

The model developed in this project can be integrated with smart grid systems to
enhance energy management. Smart grids use digital technology to monitor and
manage electricity flow more efficiently. By predicting electricity demand based
on geodemographic and environmental data, this project can help energy providers
distribute electricity dynamically, adjust supply in real-time, and reduce power
outages or overloading. The system could also enable demand-side management,
allowing consumers to adjust their energy usage based on real-time forecasts and
price signals, improving energy efficiency.

2. Renewable Energy Optimization

With the growing emphasis on transitioning to renewable energy sources like solar
and wind, predicting electricity consumption will be crucial. Renewable energy
sources are intermittent, meaning their availability can fluctuate (e.g., solar power
depends on sunlight). The LSTM model can predict electricity demand for different
regions and help energy companies plan how to balance supply from renewables
with traditional sources. Additionally, the project could be expanded to predict
renewable energy generation (based on weather conditions), optimizing the mix of
energy sources and improving the integration of green energy into the grid.

40
3.Urban Planning and Policy Making

The project can provide valuable insights for urban planners and policymakers. By
analyzing geodemographic factors like population growth, migration patterns, and
economic development, cities can use the predictions to design more energy-efficient
infrastructures, including buildings, transportation systems, and utilities.

4. Scalability and Regional Customization

The current project can be expanded to cover larger geographic areas and more
complex regions. For example, it could be scaled to predict electricity consumption
for an entire country, accounting for regional differences in energy use.

5. Real-Time Predictions and IoT Integration

With advancements in Internet of Things (IoT) technologies, it is possible to col-

lect real-time data from smart meters, sensors, and other connected devices. By
integrating this real-time data into the LSTM model, the project could provide
up-to-the-minute predictions for electricity consumption. This would allow energy
providers to respond more quickly to changing demand, helping balance load during
peak hours or even forecast short-term fluctuations due to sudden weather changes.

6. Energy Cost Savings and Efficiency

For both consumers and providers, accurate electricity consumption predictions can
lead to significant cost savings. Providers can optimize their operations, reducing
wastage of electricity, and minimizing the need for expensive standby power gen-
eration. Consumers can benefit from more accurate billing, better management of
their electricity consumption, and the possibility of differential pricing based on
predicted demand. The model could also be used by businesses to reduce their
carbon footprint by planning electricity usage more efficiently, contributing to sus-
tainability goals.

41
7. Climate Change Impact and Resilience Planning

As climate change accelerates, the relationship between electricity consumption

and environmental factors (like temperature fluctuations, extreme weather events,
etc.) will become even more important. The project could be expanded to factor
in climate change models, predicting how electricity demand might evolve in the
future due to changing weather patterns.

8. Collaboration with Energy Departments and Industries

The model has the potential to be used by energy departments, governmental bod-
ies, and private sector industries for decision-making. By improving the accuracy
of electricity consumption forecasts, the project could play a role in setting up
power plants, designing energy-efficient communities, and reducing the overall car-
bon footprint. Industries that rely heavily on energy (like manufacturing, data
centers, etc.) could also use this model to optimize their energy usage, aligning it
with grid conditions to minimize energy costs and environmental impact.

42
Chapter 10

CONCLUSION

In conclusion,this work explores the significance of geodemographic factors .Firstly,

it identifies factors that impact the power consumption of a home. Secondly, it
suggests an energy forecasting model using Geo demographic factors. The ap-
proach could benefit energy companies.Geo demographic factors are important for
understanding and forecasting electricity consumption. Electricity companies can
incorporate the Geo demographic factors to implement better energy management
strategies. We have also shown that Geo demographic factors can be used to
improve the accuracy of a fore casting model. This will help the utility compa-
nies to fulfill the consumer demand in more precise manner. Companies can use
Geo-demographic factors for the purpose of forecasting user demand and get the
load-supply balance.

43
 Chapter 11

BIBLIOGRAPHY

• J. Contreras, R. Espinola, F. J. Nogales, and A. J. Conejo, ARIMA models to

predict next-day electricity prices, IEEE Trans. Power Syst., vol. 18, no. 3, pp.
10141020, Aug. 2003.

• K. Kavaklioglu, Modeling and prediction of Turkeys electricity consumption using

support vector regression, Appl. Energy, vol. 88, no. 1, pp. 368375, 2011.

• B. R. Szkuta, L. A. Sanabria, and T. S. Dillon, Electricity price shortterm fore-

casting using artificial neural networks, IEEE Trans. Power Syst., vol. 14, no. 3,
pp. 851857, Aug. 1999.

• R. V. Jones, A. Fuertes, and K. J. Lomas, The socio-economic, dwelling and appli-

ance related factors affecting electricity consumption in domestic buildings, Renew.
Sustain. Energy Rev., vol. 43, pp. 901917, Mar. 2015.

• Acorn User Guide. Accessed: 2020. [Online]. Available: https://acorn.caci.

co.U.K./downloads/Acorn-User-guide.pdf

44