Hybrid Intelligent Approaches For Smart Energy Practical Applications Senthil Kumar Mohan Full Chapter

Hybrid Intelligent Approaches for Smart
Energy: Practical Applications Senthil

Kumar Mohan
Visit to download the full and correct content document:
https://ebookmass.com/product/hybrid-intelligent-approaches-for-smart-energy-practi
cal-applications-senthil-kumar-mohan/
Hybrid Intelligent Approaches
for Smart Energy
Scrivener Publishing
100 Cummings Center, Suite 541J
Beverly, MA 01915-6106
Publishers at Scrivener
Martin Scrivener (martin@scrivenerpublishing.com)
Phillip Carmical (pcarmical@scrivenerpublishing.com)
Hybrid Intelligent Approaches
for Smart Energy
Practical Applications
Edited by
John A.
Senthil Kumar Mohan
Sanjeevikumar Padmanaban
and
Yasir Hamid
This edition first published 2022 by John Wiley & Sons, Inc., 111 River Street, Hoboken, NJ 07030, USA
and Scrivener Publishing LLC, 100 Cummings Center, Suite 541J, Beverly, MA 01915, USA
© 2022 Scrivener Publishing LLC
For more information about Scrivener publications please visit www.scrivenerpublishing.com.
All rights reserved. No part of this publication may be reproduced, stored in a retrieval system, or
transmitted, in any form or by any means, electronic, mechanical, photocopying, recording, or other-
wise, except as permitted by law. Advice on how to obtain permission to reuse material from this title
is available at http://www.wiley.com/go/permissions.
Wiley Global Headquarters

111 River Street, Hoboken, NJ 07030, USA
For details of our global editorial offices, customer services, and more information about Wiley prod-
ucts visit us at www.wiley.com.
Limit of Liability/Disclaimer of Warranty

While the publisher and authors have used their best efforts in preparing this work, they make no rep
resentations or warranties with respect to the accuracy or completeness of the contents of this work and
specifically disclaim all warranties, including without limitation any implied warranties of merchant-
ability or fitness for a particular purpose. No warranty may be created or extended by sales representa
tives, written sales materials, or promotional statements for this work. The fact that an organization,
website, or product is referred to in this work as a citation and/or potential source of further informa
tion does not mean that the publisher and authors endorse the information or services the organiza
tion, website, or product may provide or recommendations it may make. This work is sold with the
understanding that the publisher is not engaged in rendering professional services. The advice and
strategies contained herein may not be suitable for your situation. You should consult with a specialist
where appropriate. Neither the publisher nor authors shall be liable for any loss of profit or any other
commercial damages, including but not limited to special, incidental, consequential, or other damages.
Further, readers should be aware that websites listed in this work may have changed or disappeared
between when this work was written and when it is read.
Library of Congress Cataloging-in-Publication Data
ISBN 9781119821243
Cover image: Electrical Tower, Patrick Daxenbichler | Internet of Things, Monthira Yodtiwong |
Dreamstime.com
Cover design by Kris Hackerott
Set in size of 11pt and Minion Pro by Manila Typesetting Company, Makati, Philippines
Printed in the USA
10 9 8 7 6 5 4 3 2 1
Contents
List of Contributors xiii

Preface xv
Acknowledgements xix
1 Review and Analysis of Machine Learning Based Techniques
for Load Forecasting in Smart Grid System 1
Shihabudheen KV and Sheik Mohammed S
1.1 Introduction 2
1.2 Forecasting Methodology 4
1.3 AI-Based Prediction Methods 5
1.3.1 Single Prediction Methods 5
1.3.1.1 Linear Regression 5
1.3.1.2 Artificial Neural Networks (ANN) 7
1.3.1.3 Support Vector Regression (SVR) 8
1.3.1.4 Extreme Learning Machine 9
1.3.1.5 Neuro-Fuzzy Techniques 10
1.3.1.6 Deep Learning Techniques 11
1.3.2 Hybrid Prediction Methods 12
1.3.2.1 Combined AI-Based Prediction Techniques 12
1.3.2.2 Signal Decomposition Based Prediction
Techniques 13
1.3.2.3 EMD Based Decomposition 14
1.3.2.4 Wavelet Based Decomposition 14
1.4 Results and Discussions 15
1.4.1 Description of Dataset 15
1.4.2 Performance Analysis of Single Prediction
Methods for Load Forecasting 16
1.4.2.1 Feature Selection 16
1.4.2.2 Optimal Parameter Selection 17
1.4.2.3 Prediction Results of Single Prediction
Methods 17
v
vi Contents
1.4.3 Performance Analysis of Hybrid Prediction Methods

for Load Forecasting 17
1.4.4 Comparative Analysis 21
1.5 Conclusion 22
References 23
2 Energy Optimized Techniques in Cloud and Fog Computing 27
N.M. Balamurugan, TKS Rathish babu, K Maithili
and M. Adimoolam
2.1 Introduction 28
2.2 Fog Computing and Its Applications 33
2.3 Energy Optimization Techniques in Cloud Computing 38
2.4 Energy Optimization Techniques in Fog Computing 42
2.5 Summary and Conclusions 44
References 45
3 Energy-Efficient Cloud Computing Techniques
for Next Generation: Ways of Establishing and Strategies
for Future Developments 49
Praveen Mishra, M. Sivaram, M. Arvindhan, A. Daniel
and Raju Ranjan
3.1 Introduction 50
3.2 A Layered Model of Cloud Computing 52
3.2.1 System of Architecture 53
3.3 Energy and Cloud Computing 54
3.3.1 Performance of Network 55
3.3.2 Reliability of Servers 55
3.3.3 Forward Challenges 55
3.3.4 Quality of Machinery 56
3.4 Saving Electricity Prices 56
3.4.1 Renewable Energy 57
3.4.2 Cloud Freedom 57
3.5 Energy-Efficient Cloud Usage 58
3.6 Energy-Aware Edge OS 58
3.7 Energy Efficient Edge Computing Based
on Machine Learning 59
3.8 Energy Aware Computing Offloading 61
3.8.1 Energy Usage Calculation and Simulation 63
3.9 Comments and Directions for the Future 63
References 64
Contents vii
4 Energy Optimization Using Silicon Dioxide Composite

and Analysis of Wire Electrical Discharge Machining
Characteristics 67
M.S. Kumaravel, N. Alagumurthi and P. Mathiyalagan
4.1 Introduction 67
4.2 Materials and Methods 69
4.3 Results and Discussion 72
4.3.1 XRD Analysis 72
4.3.2 SEM Analysis 73
4.3.3 Grey Relational Analysis (GRA) 73
4.3.4 Main Effects Graph 76
4.3.5 Analysis of Variance (ANOVA) 77
4.3.6 Confirmatory Test 78
4.4 Conclusion 80
Acknowledgement 80
References 80
5 Optimal Planning of Renewable DG and Reconfiguration
of Distribution Network Considering Multiple Objectives
Using PSO Technique for Different Scenarios 83
Balmukund Kumar and Aashish Kumar Bohre
5.1 Introduction 84
5.2 Literature Review for Recent Development in DG
Planning and Network Reconfiguration 84
5.3 System Performance Parameters and Index 87
5.4 Proposed Method 88
5.4.1 Formulation of Multi-Objective Fitness Function 88
5.4.2 Backward-Forward-Sweep Load Flow Based
on BIBC-BCBV Method 89
5.5 PSO Based Optimization 90
5.6 Test Systems 92
5.7 Results and Discussions 92
5.8 Conclusions 101
References 102
6 Investigation of Energy Optimization for Spectrum
Sensing in Distributed Cooperative IoT Network
Using Deep Learning Techniques 107
M. Pavithra, R. Rajmohan, T. Ananth Kumar, S. Usharani
and P. Manju Bala
viii Contents
6.1 Introduction 108

6.2 IoT Architecture 111
6.3 Cognitive Spectrum Sensing for Distributed
Shared Network 113
6.4 Intelligent Distributed Sensing 115
6.5 Heuristic Search Based Solutions 117
6.6 Selecting IoT Nodes Using Framework 118
6.7 Training With Reinforcement Learning 119
6.8 Model Validation 120
6.9 Performance Evaluations 123
6.10 Conclusion and Future Work 125
References 126
7 Road Network Energy Optimization Using IoT
and Deep Learning 129
N. M. Balamurugan, N. Revathi and R. Gayathri
7.2 Road Network 132
7.2.1 Types of Road 132
7.2.2 Road Structure Representation 134
7.2.3 Intelligent Road Lighting System 135
7.3 Road Anomaly Detection 139
7.4 Role of IoT in Road Network Energy Optimization 141
7.5 Deep Learning of Road Network Traffic 142
7.6 Road Safety and Security 142
7.7 Conclusion 144
References 144
8 Energy Optimization in Smart Homes and Buildings 147
S. Sathya, G. Karthi, A. Suresh Kumar and S. Prakash
8.2 Study of Energy Management 150
8.3 Energy Optimization in Smart Home 150
8.3.1 Power Spent in Smart-Building 153
8.3.2 Hurdles of Execution in Energy Optimization 156
8.3.3 Barriers to Assure SH Technologies 156
8.4 Scope and Study Methodology 157
8.4.1 Power Cost of SH 158
8.5 Conclusion 159
References 159
Contents ix
9 Machine Learning Based Approach for Energy Management

in the Smart City Revolution 161
Deepica S., S. Kalavathi, Angelin Blessy J.
and D. Maria Manuel Vianny
9.1.1 Smart City: What is the Need? 162
9.1.2 Development of Smart City 163
9.2 Need for Energy Optimization 166
9.3 Methods for Energy Effectiveness in Smart City 166
9.3.1 Smart Electricity Grids 166
9.3.2 Smart Transportation and Smart Traffic
Management 169
9.3.3 Natural Ventilation Effect 172
9.4 Role of Machine Learning in Smart City Energy
Optimization 173
9.4.1 Machine Learning: An Overview 173
9.5 Machine Learning Applications in Smart City 175
9.6 Conclusion 177
References 178
10 Design of an Energy Efficient IoT System for Poultry Farm
Management 181
G. Rajakumar, G. Gnana Jenifer, T. Ananth Kumar
and T. S. Arun Samuel
10.2 Literature Survey 183
10.3 Proposed Methodology 187
10.3.1 Monitoring and Control Module 188
10.3.2 Monitoring Temperature 188
10.3.3 Monitoring Humidity 189
10.3.4 Monitoring Air Pollutants 189
10.3.5 Artificial Lightning 190
10.3.6 Monitoring Water Level 190
10.4 Hardware Components 190
10.4.1 Arduino UNO 190
10.4.2 Temperature Sensor 190
10.4.3 Humidity Sensor 191
10.4.4 Gas Sensor 192
10.4.5 Water Level Sensor 192
10.4.6 LDR Sensor 193
x Contents
10.4.7 GSM (Global System for Mobile Communication)

Modem 194
10.5 Results and Discussion 195
10.5.1 Hardware Module 195
10.5.2 Monitoring Temperature 196
10.5.3 Monitoring Gas Content 198
10.5.4 Monitoring Humidity 198
10.5.5 Artificial Lighting 198
10.5.6 Monitoring Water Level 198
10.5.7 Poultry Energy-Efficiency Tips 199
10.6 Conclusion 201
References 203
11 IoT Based Energy Optimization in Smart Farming Using AI 205
N. Padmapriya, T. Ananth Kumar, R. Aswini, R. Rajmohan,
P. Kanimozhi and M. Pavithra
11.2 IoT in Smart Farming 208
11.2.1 Benefits of Using IoT in Agriculture 208
11.2.2 The IoT-Based Smart Farming Cycle 209
11.3 AI in Smart Farming 210
11.3.1 Artificial Intelligence Revolutionises Agriculture 210
11.4 Energy Optimization in Smart Farming 211
11.4.1 Energy Optimization in Smart Farming
Using IoT and AI 212
11.5 Experimental Results 215
11.5.1 Analysis of Network Throughput 216
11.5.2 Analysis of Network Latency 217
11.5.3 Analysis of Energy Consumption 218
11.5.4 Applications of IoT and AI in Smart Farming 219
11.6 Conclusion 220
References 221
12 Smart Energy Management Techniques in Industries 5.0 225
S. Usharani, P. Manju Bala, T. Ananth Kumar, R. Rajmohan
and M. Pavithra
12.2 Related Work 227
12.3 General Smart Grid Architecture 229
12.3.1 Energy Sub-Sectors 230
12.3.1.1 Smart Grid: State-of-the-Art Inside
Energy Sector 230
Contents xi
12.3.2 EV and Power-to-Gas: State-of-the-Art

within Biomass and Transport 231
12.3.3 Constructing Zero Net Energy (CZNE):
State-of-the-Art Inside Field of Buildings 233
12.3.4 Manufacturing Industry: State-of-the-Art 234
12.3.5 Smart Energy Systems 235
12.4 Smart Control of Power 236
12.4.1 Smart Control Thermal System 236
12.4.2 Smart Control Cross-Sector 237
12.5 Subsector Solutions 238
12.6 Smart Energy Management Challenges in Smart Factories 239
12.7 Smart Energy Management Importance 240
12.8 System Design 241
12.9 Smart Energy Management for Smart Grids 241
12.10 Experimental Results 247
12.11 Conclusions 250
References 251
13 Energy Optimization Techniques in Telemedicine
Using Soft Computing 253
R. Indrakumari
13.2 Essential Features of Telemedicine 255
13.3 Issues Related to Telemedicine Networks 256
13.4 Telemedicine Contracts 257
13.5 Energy Efficiency: Policy and Technology Issue 258
13.5.1 Soft Computing 258
13.5.2 Fuzzy Logic 260
13.5.3 Artificial Intelligence 260
13.5.4 Genetic Algorithms 263
13.5.5 Expert System 263
13.5.6 Expert System Based on Fuzzy Logic Rules 264
13.6 Patient Condition Monitoring 266
13.7 Analysis of Physiological Signals and Data Processing 271
13.8 M-Health Monitoring System Architecture 272
13.9 Conclusions 275
References 276
14 Healthcare: Energy Optimization Techniques Using IoT
and Machine Learning 279
G. Vallathan, Senthilkumar Meyyappan and T. Rajani
xii Contents
14.2 Energy Optimization Process 281

14.3 Energy Optimization Techniques in Healthcare 283
14.3.1 Energy Optimization in Building 283
14.3.2 Machine Learning for Energy Optimization 284
14.3.3 Reinforcement Learning for Energy Optimization 286
14.3.4 Energy Optimization of Sustainable Internet of
Things (IoT) 287
14.4 Future Direction of Energy Optimizations 288
14.5 Conclusion 289
References 289
15 Case Study of Energy Optimization: Electric Vehicle Energy
Consumption Minimization Using Genetic Algorithm 291
Pedram Asef
15.2 Vehicle Modelling to Optimisation 295
15.2.1 Vehicle Mathematical Modelling 295
15.2.2 Vehicle Model Optimisation Process: Applied
Genetic Algorithm 298
15.2.3 GA Optimisation Results and Discussion 301
15.3 Conclusion 305
References 305
About the Editors 307
Index 309
List of Contributors
Shihabudheen KV
Sheik Mohammed S
N.M. Balamurugan
TKS Rathish babu
K Maithili
M. Adimoolam
Praveen Mishra
M. Sivaram
A. Daniel
Raju Ranjan
M.S. Kumaravel
N. Alagumurthi
Mathiyalagan
Balmukund Kumar
Aashish Kumar Bohre
M. Pavithra
R. Rajmohan
T. Ananth Kumar
S. Usharani
P. Manju Bala
N. Revathi
R. Gayathri
S. Sathya
G. Karthi
A. Suresh Kumar
S. Prakash
Deepica S.
S. Kalavathi
Angelin Blessy J.
D. Maria Manuel Vianny
G. Rajakumar
xiii
xiv List of Contributors
G. Gnana Jenifer
T. S. Arun Samuel
N. Padmapriya
R. Aswini
P. Kanimozhi
R. Indrakumari
G. Vallathan
Senthilkumar Meyyappan
T. Rajani
Pedram Asef
Preface
This book aims to primarily address issues surrounding the optimization,

consumption, and management of energy resources with the use of hybrid
intelligence techniques. The consumption and optimization of energy
play a crucial role in sustaining the development goals of modern society.
The need to save energy while reducing its overall cost cannot be empha-
sized enough. In recent times, smart computing technologies have slowly
but inevitably replaced the traditional computational methods in energy
optimization and consumption and its optimal scheduling and usage.
Smart computing has permeated almost all areas of technological inno-
vation today. Smart computation techniques such as artificial intelligence,
machine learning, deep learning, and IoT have become indispensable to
designing and building applications that span diverse areas like distrib-
uted environments, healthcare, smart cities, agriculture, and a host of other
functional areas. Hence, it is no surprise that in power usage, these smart
computation techniques have been incorporated along with traditional
computation and scheduling methods to bring about optimal reductions.
This book is predominantly focused on emerging concepts and algorith-
mic approaches in machine learning and artificial intelligence to bring
about enhancements in soft computing techniques in energy optimization.
This site will contribute to research work undertaken by researchers, aca-
demicians, data scientists, and technology developers alike.
This book comprises of fifteen chapters, each chapter elaborating
upon various aspects and techniques related to optimizing and manag-
ing energy. Chapter 1 elaborates upon the review and analysis of machine
learning-based techniques for load forecasting. A load-forecasting algo-
rithm for time-series loads using AI techniques with supervised methods
is presented and discussed. This includes a comparative assessment of load
forecasting based on supervised artificial intelligent algorithms such as
Artificial Neural Networks (ANN), Support Vector Machine (SVM), and
Extreme Learning Machine (ELM), which is performed on smart meter
xv
xvi Preface
data. The results are presented in this chapter, along with a performance
analysis of the selected algorithms.
Chapter 2 elaborates on various energy optimization techniques, exam-
ines how energy can be optimized, and what techniques are available and
used in cloud and fog computing. Chapter 3 is focused on the energy
efficiencies of various next-generation cloud computing techniques. This
chapter delves deeper into the benefits and trade-offs arising from the adop-
tion of various energy conservation measures. It outlines the challenges
and recommendations to be considered in the future. Chapter 4 presents
a method of energy optimization using a Silicon dioxide composite and an
analysis of wire electrical discharge machining characteristics. Here, a stir
casting method to produce the uniform distribution of SiO2 particles in an
Aluminum matrix composite (AMC) is expanded upon, along with confir-
mation by microstructural analysis during which the XRD pattern reveals
that the SiO2 indeed has a monoclinic crystalline structure.
Chapter 5 presents a discussion on optimal planning of renewable DG
and the reconfiguration of the distribution network that considers mul-
tiple objectives using the PSO technique under different scenarios. This
chapter discusses the methods for improving voltage and reliability and
reducing power losses (active and reactive) and proposes an IEEE-69-bus
test distribution network with separate scenarios for optimal sizing-siting
of multi-renewable DGs with the reconfiguration of the particle swarm
optimization (PSO) technique. Chapter 6 propounds upon energy opti-
mization for spectrum sensing in Distributed Cooperative IoT Networks
using deep learning techniques. This chapter further investigates applying
deep learning methods to the Internet of Things (IoT) applications while
focusing on energy optimization mechanisms.
Chapter 7 expands upon the energy optimization for a road network
using IoT and deep learning methods. In this chapter, an overview of the
components of an intelligent road transportation system is first presented.
It is then followed by a discussion on the potential for leveraging the
advancements in deep learning and Internet-of-Things (IoT) in modeling
the short-term traffic states for the prediction of traffic flow and energy
optimization in using fuel and electricity consumption and road lighting.
Chapter 8 explores the means of energy optimization in smart homes and
buildings. In this chapter, information about communication technologies
that enable energy optimization is presented. Such optimizations in smart
homes and buildings are proposed using energy management and optimal
scheduling by utilizing the Internet of Things and smart grids. This chapter
further touches upon methodologies considered and used in the study and
presents recommendations for the future direction of the work.
Preface xvii
Chapter 9 focuses on machine learning-based approaches in energy

management for spurring the smart city revolution. This chapter proposes
a machine algorithm for optimizing the energy sources in a smart city. Such
an approach can benefit future smart city plans and help adequately deal
with existing energy resources. Chapter 10 similarly explores the design
of an energy efficient IoT system in the management of poultry farms. It
proposes a technique for enabling efficient poultry farming that will cause
increased production and profits. It propounds the use of solar technolo-
gies for generating electricity, which, apart from being used in households,
have also been successfully implemented in commercial farms to make use
of energy-saving batteries and the opportunity to sell the power back to
the grid.
Chapter 11 considers IoT-based energy optimization methods that use
artificial intelligence for enabling smart farming. This chapter outlines how
IoT can contribute to the improvement of agricultural productivity for
reaching new sustainability heights. The first part of the chapter introduces
the many ways of using IoT in smart farming and its manifold benefits.
The second part elaborates on revolutionizing the farming process by using
Artificial Intelligence. The third part emphasizes energy optimization in
agriculture, in general, using IoT and Artificial Intelligence. Chapter 12
proposes the use of smart energy management techniques in industries. It
elaborates upon the concept of energy consumption and control to effec-
tively regulate energy demands using smart grids in manufacturing indus-
tries. It also offers a description and details of the statistical method that
can be tailored to the specific needs of manufacturing customers for build-
ing energy-efficient power systems.
Chapter 13 reviews energy optimization techniques using soft comput-
ing in telemedicine. This chapter proposes an expert system based on fuzzy
rules that can be used to calculate patient risk levels. This is followed by
discussing various energy optimization techniques using the Internet of
Things, machine learning, and deep learning techniques in Chapter 14.
These techniques are used in different applications and for various optimi-
zation. The Internet of Things is connected to multiple technologies such
as Cyber-Physical Systems, Big Data Management, Cloud Management,
etc. Using IoT connectivity continuously to monitor energy usage and
optimization in various devices and energy consumption can be reduced
in several industrial and manufacturing fields, leading to many benefits,
including environmental benefits such as reducing CO2 emissions.
Acknowledgements
We are deeply indebted to Almighty God for giving us this opportunity

and it is only possible with the presence of God.
I want to thank the Almighty for giving me enough mental strength and
belief in completing this work successfully. I thank my friends and family
members for their help and support. I express my sincere thanks to the
management of Galgotias University, Greater Noida, Uttar Pradesh, India.
I wish to express my deep sense of gratitude and thanks to Wiley-Scrivener,
publisher, for their valuable suggestions and encouragement.
John A. PhD
I sincerely thank my VIT management for continuous support and encour-

agement. I thank my family members and friends for their timely support
and help. I extend my thanks to Wiley-Scrivener, publisher, for excellent
support and guidance.
Senthilkuamr Mohan, PhD
I express my sincere thanks to the management of CTiF Global Capsule,

Department of Business Development and Technology, Aarhus University,
Birk Centerpark 15, Herning 7400, Denmark. Also, I would like to thank
the Wiley-Scrivener Press for allowing me to edit this book.
P. Sanjeevikumar, PhD
I express my sincere thanks to the management of Abu Dhabi Polytechnic.

Also, I would like to thank the Wiley-Scrivener Press for allowing me to
edit this book.
Yasir Hamid, PhD
xix
1
Review and Analysis of Machine
Learning Based Techniques for Load
Forecasting in Smart Grid System
Shihabudheen KV1 and Sheik Mohammed S2*
1
Electrical Engineering Department, National Institute of Technology, Calicut,
Kerala, India
2
Electrical and Electronic Engineering Programme Area, Faculty of Engineering,
Universiti Teknologi Brunei, Gadong, Brunei Darussalam
Abstract
Electrical load forecasting is an important process that can improve the efficiency
and economy of the utility grid especially in the smart grid environment. Load
forecasting plays a significant role in making decisions such as planning, gener-
ation scheduling, operation, pricing customer satisfaction, and system security.
However, load forecasting is a tedious and difficult task due to the intermittent
nature of Renewable Energy Systems (RES) that varies depending on the seasons
and parameters such as change in temperature and humidity. Moreover, the con-
nect loads are also complex in nature as they vary from season to season. Artificial
Intelligent (AI) techniques are a promising approach for better load forecasting
having chaotic and random variation of both load and generation. In this chap-
ter, a load-forecasting algorithm for time series loads using AI techniques with
supervised methods is presented and discussed. A comparative assessment of load
forecasting based on supervised artificial intelligent algorithms, such as Artificial
Neural Networks (ANN), Support Vector Machine (SVM), and Extreme Learning
Machine (ELM), is performed on smart meter data. The results are presented and
performance of the selected algorithms are analysed.
Keywords: Electricity load forecasting, Artificial Intelligence (AI), Artificial
Neural Networks (ANN), Support Vector Machine (SVM), extreme learning
machine (ELM), smart grid, smart meter
*Corresponding author: ssheikmd@yahoo.co.in
John A., Senthil Kumar Mohan, Sanjeevikumar Padmanaban and Yasir Hamid (eds.) Hybrid Intelligent
Approaches for Smart Energy: Practical Applications, (1–26) © 2022 Scrivener Publishing LLC
1
2 Hybrid Intelligent Approaches for Smart Energy
1.1 Introduction
Traditionally, electric grids are a network of electric power generation,
transmission, and distribution systems controlled by a centralized sys-
tem. The conventional electric grid system has one way power flow and a
communication approach from the generating station until the end cus-
tomer. However, the power generation, transmission, and distribution
systems had a noteworthy revolution in recent times. The progress and
development in power generation using renewable energy sources is one
of the important reasons behind this transformation. The restructured
power system has made distribution systems bi-directional, controllable
grids called Smart Grids. The Smart Grid consists of a number of RES
OPERATIONS SERVICE
PROVIDER
MARKETS CUSTOMER
FOUNDATIONAL
SUPPORT SYSTEMS
TRANSMISSION DISTRIBUTION
GENERATION
BULK NON-BULK
GENERATION GENERATION
Figure 1.1 Structure of smart grid [1].

Machine Learning Based Load Forecasting of Smart Grid 3
with loads, ESS, sensors, and communications networks connected in

a well-arranged fashion so that it has the potential to improve overall
system performance. The coordinated and controlled operation of this
integrated structure makes the grid smarter by managing generation,
distribution, customers, and the market in both an efficient and effective
manner. Figure 1.1 shows the different domains and stakeholders of the
Smart Grid.
Electricity load forecasting is a projection of the load demand that elec-
tricity users are expected to have in the future. Load forecasts enable the
utilities to manage supply and demand and also ensure the stability of
power grids. Load forecasting is the key element for smart grid operation,
as it plays a vital role in decision-making such as planning, scheduling,
operation, capacity addition, pricing, generation planning, and system
security. Another major advantage of load forecasting is that it helps both
the utility and consumers to optimize their energy usage.
Load forecasting is classified based on horizon and scale. Scale level is
the unit size at which the forecasting is performed. Scale level forecast-
ing ranges from individual forecasting (meter-level) in homes and build-
ing levels (multi-meters level) to region, district, state, and up to country
(integrated) level load forecasting. On the other hand, horizon defines
the time range of load forecasting. Horizon level forecasting is classified
as very short-term load forecasting (VSTLF), short-term load forecasting
(STLF), medium-term load forecasting (MTLF) and long-term load fore-
casting (LTLF). In VSTLF, minutes to hour ahead prediction is carried out
and SLTF is day ahead to weekly forecasting. MTLF deals with one week
to three years prediction and more than three years prediction is known as
LTLF. Each type of forecasting serves different purposes in the power sys-
tem for scheduling, economic dispatch, operation planning, maintenance,
capacity expansion planning, fuel economy, sales, etc. [2–4].
Traditionally, expected demand is forecast using the information about
past use and other related data with the aid of charts and graphs by apply-
ing an engineering approach. Later, data driven approaches are predomi-
nantly used to build prediction algorithms to improve the efficiency and
accuracy of forecasting. Statistical based techniques and intelligent com-
puting are the two main categories of data driven approaches applied for
electricity load forecasting [5, 6]. Statistical approaches use historical data
to compare energy consumption with the most relevant variables as inputs.
More high-quality historical data therefore plays a vital role in the efficacy
of statistical models. Conventional approaches like Regression Models,
Conditional Demand Analysis (CDA), Auto Regressive Moving Average

(ARMA), and ARIMA are the most commonly adopted statistical methods
for time series prediction. However, many researchers have investigated
forecasting using AI based techniques and deep learning and those tech-
niques have become the widely accepted technology over the past decade
[7]. In addition to that, machine learning techniques like Classification
and Regression Trees (CART), as well as Support Vector Machine
(SVM) techniques are also used for time series prediction. Fuzzy Logic
Systems, Artificial Neural Networks (ANN), Evolutionary Programming,
and expert systems are some of the AI based approaches. Among them,
ANN has widespread acceptance for time series forecasting [8–12]. Many
attempts are made to solve the load-forecasting problems using AI based
hybrid approaches. A comprehensive review on all such types of forecast-
ing techniques are discussed in [13–16]. A review based on different cat-
egorization of the various forecasting, including the hybrid method, is less
attempted in most of the literature.
In this book chapter, an extensive review of different supervised AI
based load forecasting methodology is discussed. The review includes dif-
ferent categories of prediction such as single prediction and hybrid pre-
diction methods. The details of hybrid prediction such as combined AI
based prediction and signal decomposition based prediction techniques
are included. Moreover, a comparative simulation study is performed on
smart meter data. Methodology of forecasting is described in Section 1.2.
A comprehensive review of various AI based prediction strategies applied
for load forecasting is presented in Section 1.3. Comparative assessment
of single and hybrid predictions is performed on smart meter data and
the results are presented and discussed in Section 1.4. AI techniques such
as Back Propagation Based Neural Network (BPNN), Support Vector
Machine (SVM), and Extreme Learning Machine (ELM) are used for sin-
gle prediction. For hybrid prediction, Empirical Model Decomposition
(EMD) based BPNN and SVM (EMD-BPNN and EMD-SVM) prediction
are performed. The chapter is concluded in Section 1.5.
1.2 Forecasting Methodology

The forecasting methodology for prediction of time series loads consists of
two steps, as shown in Figure 1.2. The first step of methodology is feature
extraction. Initially, a sufficient quantity of features is extracted form load
time series data. Feature extraction procures the features which aid in the
prediction of time series data. Transferring the collected features into a
Load time Feature Predicted

Predictor
series data extraction result
Figure 1.2 Overall steps for load time series forecasting methodology.
more informative analysis domain helps in sensing the hidden character-

istics of future loads. The second step is the implementation of a predictor
for accurate forecasting.
Let y(t) represent a load time series data. The prediction equation can be
mathematically represented by
yˆ(t + k ) = f ( g (t )) (1.1)
where g(t) indicates the extracted features, f represents the predictive func-
tion to be approximated by predictor and yˆ(t + k ) is k step ahead of pre-
dicted values of y(t).
1.3 AI-Based Prediction Methods

Many AI-based prediction methods are proposed in literatures for time
series based load forecasting. The classification of AI based prediction
techniques is shown in Figure 1.3. An overview some of commonly used
prediction methods are outlined in this section.
1.3.1 Single Prediction Methods

Single prediction indicates the prediction of time series, which is formed
using a single AI technique. Some of the AI techniques used for single
prediction of time series data are Linear Regression, Artificial Neural
Networks, Support Vector Regression, Extreme Learning Machine, and
Neuro-Fuzzy Methods.
1.3.1.1 Linear Regression

This approach is used to predict the dependent variable using several inde-
pendent variables or features. It uses the assumption that a linear relation
may exist between the features and output signals. A linear regression
model is represented as
yˆ(k ) = a0 + ∑a x (k) k = 1,2,..., N

i =1
i i (1.2)
where
yˆ(k ) is estimated load
N total signal length.
xi (k ) is input feature variables
ai is parameters to be calculated
λ number of span.
The unknown parameters of the regression model are calculated by the
least square estimation method [17]. A short term load forecasting with
multiple linear regression was implemented in [18]. A medium term load
Linear
Regression
ANN
Single
prediction SVM
methods
ELM
AI based
Neuro-fuzzy
prediction
Deep learning
Hybrid
prediction
methods Combined AI
models
Signal
decomposition
Figure 1.3 Classification of AI-based prediction.

forecasting strategy using multivariable linear regression is proposed in

[19]. A lesser number of tuning parameters are required for linear regres-
sion forecasting problems, but handling the nonlinearities and uncertain-
ties in the prediction is a tedious job [20].
1.3.1.2 Artificial Neural Networks (ANN)

Artificial Neural Network (ANN) is a stochastic based prediction tech-
nique inspired by the concept of biological neurons. ANN is a parallel
computational algorithm where neurons are connected parallel through
a definite pattern. There are three common ANN structures, Single-Layer
Feed-Forward NN, Multi-Layer Feed-Forward NN and Recurrent NN.
The Multi-Layer Feed-Forward NN, which consist of one input layer, one
output layer, and one or more hidden layers between the input and output
layers is the most commonly preferred ANN structure [21]. The number
of neurons in the input and output layer is determined by the application,
whereas the hidden layer neurons are fixed often based on trial and error
methods [22]. Data provided to the input neuron is multiplied by an arbi-
trary connection weight and is summed to derive the net input to the neu-
ron in the next layer. Once the combined weight from preceding neurons
Input Hidden Output

layer layer layer
1 W b1
β
1
1
2 b2
2
xj 2 yj
3 bj
j
bs m
4
s
Figure 1.4 Feed forward neural network with single hidden layer.
arrives at a particular neuron, an activation function at this particular neu-

ron approximates the received input to known output. The architecture of
a single hidden layered feed forward neural network is shown in Figure 1.4
and its input/output relation can be written as
y (x) = wT σ (vTx + θv) + θw (1.3)
where Ih is the number of neurons of the hidden layer, w ∈ R Ih and

v ∈ Rn × Ih are the threshold of the output θw ∈ R and threshold of the hid-
den layer vector θ v ∈R Ih .
The training of ANN can be performed by a well-known algorithm
called the back propagation based gradient descent rule [23]. In [24], Park
et al. proposed a multi-layered perceptron (MLP) based ANN model for
efficient time series load forecasting. The Generalized Delta Rule (GDR)
[25] based error back propagation algorithm is used for the training of
parameters. A Hybrid Quantized Elman Neural Network (HQENN) is
applied for short-term load forecasting using an extended quantum learn-
ing algorithm [26]. Most of the above works use a gradient rule to optimize
parameters. A gradient based back propagation algorithm produces slow
converges and suffers from local minima trapping.
1.3.1.3 Support Vector Regression (SVR)

In machine learning, Support Vector Machines (SVM) are the most pop-
ular algorithm based on kernel methods. SVM is gaining popularity com-
pared to other soft computing techniques such as AN, as it overcomes
the bottleneck issues of ANN, like over fitting and local minima. Support
Vector Regression (SVR) [27] is an extended version of SVM that can be
effectively used for regression problems. The architecture of SVR is shown
in Figure 1.5. All the input combinations are transferred to a higher dimen-
sional plane using a special function called kernel functions [28]. The
input/output relation of SVR is written as
yˆ = w TTϕ ( x ) + b (1.4)
where w is the weight connected between input and output, b indicates the
bias in the output node, and ϕ (x) is the transferred function in a higher
dimensional space. The optimum value of weight and base can be obtained
using a quadratic optimization technique. The output equation after sim-
plification can be written as
x1 κ(x1, x)
(α1-α1*)
x2 κ(x2, x)
(α2-α2*)
ŷ
κ(x3, x) (α3-α3*)
x3
(αN-αN*)
1
xN κ(xN , x)
Figure 1.5 Support vector regression architecture.
yˆ = ∑α K ( x , x ) + b
i
i i (1.5)
where αi is the Lagrangian function to be calculated by quadratic optimiza-

tion and K (xi, x) indicates kernel function. Commonly used kernel func-
tions are the polynomial function and RBF functions. The polynomial and
RBF kernel functions are expressed as
Polynomial Kernel: K ( xi , x ) = (1 + γ xiT x )

d
(1.6)
||xi − x||2
RBF Kernel: K ( xi , x ) = e 2σ 2 (1.7)
1.3.1.4 Extreme Learning Machine

An Extreme Learning Machine (ELM) [29] is a single layer feedforward
neural network in which hidden layers are not tuned. According to ELM
theory, the weight between input layer and hidden layer is found randomly
and the weight between hidden layer and output layer are calculated by
pseudo-inverse [30]. Hence, ELM provides less computational complexity
along with better generalization capability when compared to back prop-

agation based ANN. The input/output relation of ELM can be represented
in matrix form, as given below
Hθ = T (1.8)
where H is the hidden layer output matrix with input weights, θ is output
parameter matrix, and T is the target vector output for the dataset with N
samples (Xi, Ti). Matrix H and vectors θ and T are represented as
 g (W1 X1 + b1 )  g (WNˆ X1 + bNˆ ) 

 
H=    ,
 g (W1 X N + b1 )  g (WNˆ X N + bNˆ ) 
 
 θT   T1T 
 1   
θ=   and T =    t (1.9)
 θ Tˆ   TT 
 N  Nˆ ×m  N  N ×m

N̂ is the number of hidden nodes with activation function g(x) and
Wi, θi, and bi are the weight between input node and hidden node, weight
between hidden node and output node, and hidden bias, respectively. The
output linear layer weights are computed by following a simple three step
algorithm given below:
Step 1: Randomly assign Wi and bi.

Step 2: Calculate hidden layer output matrix H.
Step 3: Calculate the output linear layer weights using Moore–
Penrose generalized inverse. θ = H⊗ * T where H⊗
indicates the Moore–Penrose generalized inverse.
1.3.1.5 Neuro-Fuzzy Techniques

The knowledge and reasoning capacity of the ANN can be improved by
integrating ANN with fuzzy logic. The integration of ANN with fuzzy
logic provides better learning capacity and robustness and can operate
with uncertain environments [31]. The Adaptive Neuro-Fuzzy Inference
System (ANFIS) is the one of the promising neuro-fuzzy systems and it
is widely used for modelling and prediction applications [32], but ANFIS
with hybrid learning may result in more computational complexity.
1.3.1.6 Deep Learning Techniques

The conventional neural network will not provide a satisfactory result if
the load forecasting data is big. This is due to a lesser number of adaptive
parameters. A Deep Neural Network is a strategy that uses a larger num-
ber of hidden layers in a conventional ANN. The deep learning technique
is more capable of handling the complexity and the model is more accu-
rate when compared to conventional ANN. The basic architecture of deep
learning techniques is called a Recurrent Neural Network (RNN) [33] and
it is shown in Figure 1.6. RNN allows the signal to travel both forward and
backward, which creates loops in hidden nodes.
RNN computes the output
Y = [y1 y2 … yT] through the hidden layer
H = [h1 h2 … hT] for the input series
X = [x1 x2 … xT] using the following iterative equations
ht = F (Whx xt + Whh ht – 1 + bh) (1.10)
yt = G (Wyh ht + by) (1.11)
where Whx, Whh, and Wyh are the weights between input to hidden node,
hidden to hidden node, and hidden to output node, respectively, and bh
and by represent bias for the hidden layer and output layer, respectively.
F (∙) and G (∙) are the activation functions used in the hidden layer and
output layer, respectively.
Despite the advantages, the long term information storing capacity is
not available in RNN [34].
YT yT–n yt yT–1
W hy W hy W hy
Whh Whh
RNN = hT–n ht hT–1
W xh W xh W xh
XT xT–n xt xT–1
Figure 1.6 RNN architecture.

ht
ct
× +
ct-1 it
th
ƒt σ × ~
Ct
σ ×
σ
ht-1 th Ot
ht
xt
Figure 1.7 LSTM network.
Long Short Term Memory (LSTM) is another promising deep learning

approach for accurate time series forecasting [35]. LSTM provides the idea
of gate and memory cells in each hidden layer. Figure 1.7 shows the rep-
resentation of an LSTM network where x(t) and h(t) are the current input
and output of the LSTM layer. Sigmod is used for recollecting or disre-
membering the data in the hidden layers. Therefore, by providing memory
elements in the hidden layer, LSTM is able to learn what information to
store in long term memory and what to avoid.
1.3.2 Hybrid Prediction Methods

In many cases, particularly for electricity load, data contains lots of nonlinear
and uncertain information and prediction of such data using a single model
may not provide satisfactory results. Hence, combined models are often used
for prediction. In the hybrid prediction method, two or more prediction/
feature extraction techniques are adaptively combined to extract the advan-
tages of individual methods. Hybrid prediction methods can be classified as
(i) combined AI-based prediction and

(ii) signal decomposition based prediction techniques.
1.3.2.1 Combined AI-Based Prediction Techniques

In this method, two or more AI based single prediction models are
combined for accurate load forecasting. An adaptive learning strategy,
consisting of unsupervised and supervised methods, is used for short term

load forecasting in [36]. Initially, a Self-Organizing Map (SOM) is used to
cluster the input load data into a number of subsets. Then, 24 SVMs are
used for hourly prediction in a day. A hybrid short term prediction strat-
egy using Fuzzy Inductive Reasoning (FIR) with a Simulated Rebounding
Algorithm (SRA) was proposed in [37]. SRA is used to determine the non-
linear relationship between the variables and the new set of input vari-
ables is calculated. In [38], day ahead forecasting using ARIMA and SVR
with a Cuckoo Search Algorithm (CSA) was proposed. Initially, a properly
designed ARIMA model is utilised to find linear and residual behaviour
of the data. Then, in the second stage, SVR is used to forecast the non-
linear portion of the system. CSA is implemented to find out the hyper
parameter of SVR. A better performance has been achieved in this hybrid
system compared to individual models. It is to be noted that implementa-
tion of such feature extraction and optimization techniques is complex and
tedious.
A two stage load forecasting method combining a double seasonal Holt–
Winters–Taylor technique (HWT) and other conventional ML techniques
has been discussed [39]. Fuzzy C-means (FCM) Clustering, k-Nearest
Neighbours (k-NN), Wavelet Packet Decomposition (WPD), and ANFIS
algorithms are used in secondary prediction. The proposed method is
compared with the individual prediction model and higher accuracy is
obtained in terms of mean absolute percentage error.
1.3.2.2 Signal Decomposition Based Prediction Techniques

The block schematic of signal decomposition based prediction is shown
in Figure 1.8. The actual load time series data is decomposed into a
number of simple adaptive series using signal decomposition. Empirical
Model Decomposition (EMD) [40] and Wavelet Transform (WT) [41]
based decomposition are the two commonly available signal decompo-
sition methods. After decomposition, feature extraction of each signal is
performed to get the optimum training input. In certain cases, wavelet
transform is used to perform feature extraction along with decomposition.
Load time Signal Feature Learning Predicted

series data Aggregation result
decomposition extraction model
Figure 1.8 Signal decomposition based load-forecasting strategy.

The extracted features of each decomposed signal are modelled using any
learning models. Finally, aggregations of outputs of all learning models are
performed to find the final prediction result through certain combination
schemes.
1.3.2.3 EMD Based Decomposition

EMD is a signal decomposition method which converts the nonstation-
ary non-linear time series into simple and adaptive series called Intrinsic
Mode Functions (IMF) [40]. EMD can convert any complicated series into
simple adaptive series, which eliminates the need for spurious harmon-
ics to represent nonlinear and non-stationary signals. After this, IMF are
extracted and each IMF is modelled by simple learning techniques and the
final prediction result is found by summing the output of each prediction
model.
Load forecasting using hybrid EMD and SVR is proposed in [42].
Thirteen (13) IMFs and one residue are extracted for the EUNITE Dataset
using EMD algorithms and SVM is used to train each subseries. The EMD-
SVM technique provides better accuracy compared to individual ARMA
and SVR models. In [43], differential EMD is used for signal decomposi-
tion and corresponding IMF and residue are predicted by SVR and Auto
Regression (AR), respectively. Short term load forecasting using EMD and
deep learning techniques is proposed in [44]. In this work, a Deep Belief
Network (DBN) is used for modelling and an Auto Correlation Function
is used for feature extraction. The proposed combination of algorithms for
load forecasting improves the overall accuracy of the prediction.
All of the above studies show that the implementation of signal decom-
position methods like EMD for time series load forecasting improves the
overall accuracy of prediction compared with individual methods.
1.3.2.4 Wavelet Based Decomposition

In this type of prediction, initially, time series data is decomposed into dif-
ferent high frequency and low frequency components using wavelet trans-
forms. Then, single prediction is applied on this decomposed data. Pandey
et al. [45] proposed neural network based load forecasting. In this work,
load data is decomposed using wavelet transform. An ensemble forecast-
ing method using wavelet transform and ELM is utilized for load fore-
casting in [46]. A Modified Artificial Bee Colony (MABC) based ELM is
applied on wavelet decomposed signals for better prediction accuracy [47].
1.4 Results and Discussions

In this chapter, very short term forecasting with hour ahead prediction is
carried out using the selected techniques. Performance comparison of sin-
gle prediction and hybrid prediction techniques are carried out. For analy-
sis of single prediction methods, Back Propagation Based Neural Network,
Support Vector Machine, and Extreme Learning Machine are selected. For
hybrid prediction, Empirical Model Decomposition based BPNN (EMD-
BPNN) and Empirical Model Decomposition based SVM (EMD-SVM) are
considered. In BPNN, a tansig activation function is used and a Gaussian
kernel activation function is used for SVM and ELM. Details of the data
used for analysis and prediction accuracy of the selected methods are dis-
cussed in the below subsections.
1.4.1 Description of Dataset

The original dataset of smart meter data collected from the household
of the London for UK Power Networks led Low Carbon London Project
is used in this work [48]. The dataset consists of household energy con-
sumption in KWh. The sampling rate of data is 30 minutes and the data is
available for a period of 10 months from January to October in 2013. The
dataset from January, April, July, and September (Figure 1.9) is used for
the analysis of prediction performance. The dataset for the first 21 days in
January April
1 1
KWh
KWh
0.5 0.5
0 0
0 500 1000 1500 0 500 1000 1500
Half hour Half hour
July September
1 1
KWh
KWh
0.5 0.5
0 0
0 500 1000 1500 0 500 1000 1500
Half hour Half hour
Figure 1.9 Dataset collected for months of January, April, July, and September 2013.
each month is used for training and the dataset for the remaining days is
used for testing.
1.4.2 Performance Analysis of Single Prediction Methods

for Load Forecasting
Single prediction methods such as BPNN, SVM, and ELM are discussed in
the following subsection. These methods include optimal feature extraction,
parameter identification, and prediction.
1.4.2.1 Feature Selection

The optimum number of features for prediction should be carefully selected.
Partial Autocorrelation Function (PACF) is used for feature selection [22].
The PACF graph with respect to the lag for the dataset of January, April,
July, and September are shown in Figure 1.10. Based on results shown in
the figure, it can be realized that the optimum number of lag for all cases
can be selected as 2.
January April
1 1
0.5 0.5
PACF
PACF
0 0
-0.5 -0.5
0 5 10 15 20 0 5 10 15 20
Lag Lag
July September
1 1
0.5 0.5
PACF
PACF
0 0
-0.5 -0.5
0 5 10 15 20 0 5 10 15 20
Lag Lag
Figure 1.10 PACF graph of dataset for January, April, July, and September.
Table 1.1 Optimum parameters selected for single prediction methods.

Dataset BPNN SVM ELM
Number of Hidden σ C σ C
Neurons
January 8 2-3 21 24 2
April 10 25 24 22 24
July 7 22 2-1 2-2 23
October 11 22 23 20 2-1
1.4.2.2 Optimal Parameter Selection

The optimum parameters of prediction methods are calculated by cross
validation. In BPNN, the number of neurons in the hidden layers is varied
from 3 to 20 and the optimum number is selected by cross validation. The
kernel parameter (σ) and regularization parameter (C) of SVM and ELM
are selected in the logarithmic range between 2-10 and 215. The optimum
value of σ and C are found by cross validation. The optimum parameters of
the selected single prediction methods are listed in Table 1.1.
1.4.2.3 Prediction Results of Single Prediction Methods

The prediction results obtained using BPNN, SVM, and ELM for the data-
set January, April, July, and October are shown in Figures 1.11a through
1.11d along with the actual data. From the presented results, it can be seen
that ELM based prediction results are closer to the actual data, i.e., the
prediction accuracy of ELM is higher when compared to BPNN and SVM.
1.4.3 Performance Analysis of Hybrid Prediction Methods

for Load Forecasting
Hybrid prediction of load data using EMD-BPNN and EMD-SVM are
performed. Initially, the data is converted to a simple adaptive series
using EMD. Ten adaptive series (IMF) are generated using EMD decom-
position of the dataset for January. The IMF generated after decompo-
sition of the dataset for January, April, July, and October are shown in
Figures 1.12a–1.12d. In hybrid prediction, each IMF is modelled by single
(a)
Actual data
1 NN
ELM
0.8 SVM
KWh
0.6
0.4
0.2
0
0 50 100 150 200 250 300 350 400 450
Samples
(b) 1 Actual data
ELM
0.8 NN
SVM
0.6
KWh
0.4
0.2
0
0 50 100 150 200 250 300 350 400
Samples
(c)
0.8 Actual data
BPNN
ELM
0.6 SVM
KWh
0.4
0.2
0
0 50 100 150 200 250 300 350 400 450
Samples
(d)
Actual data
1.2 BPNN
ELM
1 SVM
KWh
0.8
0.6
0.4
0.2
0
0 50 100 150 200 250 300 350 400
Samples
Figure 1.11 (a) Prediction results of single prediction methods for January. (b) Prediction
results of single prediction methods for April. (c) Prediction results of single prediction
methods for July. (d) Prediction results of single prediction methods for September.
(a) IMF1 IMF2 IMF3 IMF4 IMF5

0.5 0.5 0.5 0.5 0.2
0 0 0 0 0
-0.5 -0.5 -0.5 -0.5 -0.2

0 500 1000 0 500 1000 0 500 1000 0 500 1000 0 500 1000
IMF6 IMF7 IMF8 IMF9 IMF10

0.1 0.1 0.02 0.02 0.26
0 0.24
0 0 0
-0.02 0.22
-0.1 -0.04 -0.02 0.2

-0.1
0 500 1000 0 500 1000 0 500 1000 0 500 1000 0 500 1000
(b) IMF1 IMF2 IMF3 IMF4 IMF5

0.4 1 0.4 0.2 0.2
0.2 0.2 0.1 0.1
0.5
0 0 0 0
0 -0.2
-0.2 -0.1 -0.1
-0.4 -0.5 -0.4 -0.2 -0.2
0 500 1000 1500 0 500 1000 1500 0 500 1000 1500 0 500 1000 1500 0 500 1000 1500
IMF6 IMF7 IMF8 IMF9 IMF10
0.1 0.1 0.04 0.04 0.25
0.05 0.05 0.02 0.02
0 0 0 0 0.2
-0.05 -0.05 -0.02 -0.02
-0.1 -0.1 -0.04 -0.04 0.15
0 500 1000 1500 0 500 1000 1500 0 500 1000 1500 0 500 1000 1500 0 500 1000 1500
(c) IMF1 IMF2 IMF3 IMF4 IMF5

0.5 0.5 0.5 0.2 0.2
0.1
0 0 0 0
0
-0.5 -0.2 -0.1
-0.5 -0.5
0 500 1000 1500 0 500 1000 1500 0 500 1000 1500 0 500 1000 1500 0 500 1000 1500
IMF6 IMF7 IMF8 IMF9 x 10-3 IMF10
0.1 0.1 0.5 0.02 5
0 0 0 0 0
-0.1 -0.1 -0.5 -0.02 -5

0 500 1000 1500 0 500 1000 1500 0 500 1000 1500 0 500 1000 1500 0 500 1000 1500
IMF11
0.22
0.2
0.18
0 1000 2000
(d) IMF1 IMF2 IMF3 IMF4 IMF5

0.4 0.5 0.4 0.2 0.2
0.2 0.2 0.1 0.1
0 0 0 0 0
-0.2 -0.2 -0.1 -0.1
-0.4 -0.5 -0.4 -0.2 -0.2

0 500 1000 1500 0 500 1000 1500 0 500 1000 1500 0 500 1000 1500 0 500 1000 1500
IMF6 IMF7 IMF8 IMF9

0.2 0.1 0.04 0.22
0.1 0.05 0.02
0.21
0 0 0
0.2
-0.1 -0.05 -0.02
-0.2 -0.1 -0.04 0.19

0 500 1000 1500 0 500 1000 1500 0 500 1000 1500 0 500 1000 1500
Figure 1.12 (a) EMD decompositions of dataset (January). (b) EMD decompositions of
dataset (April). (c) EMD decompositions of dataset (July). (d) EMD decompositions of
dataset (September).
(a) January
Actual data
1 EMD-BPNN
EMD-SVM
0.8
0.6
KWh
0.4
0.2
0
0 50 100 150 200 250 300 350 400 450
Samples
(b) 1
Actual data
EMD-BPNN
0.8 EMD-SVM
0.6
KWh
0.4
0.2
0
0 50 100 150 200 250 300 350 400
Samples
(c)
0.8
Actual data
EMD-BPNN
EMD-SVM
0.6
KWh
0.4
0.2
0
0 50 100 150 200 250 300 350 400 450
Samples
(d)
1.5
Actual data
EMD-BPNN
EMD-SVM
1
KWh
0.5
0
0 50 100 150 200 250 300 350 400
Samples
Figure 1.13 (a) Prediction results using hybrid prediction method for dataset (January).
(b) Prediction results using hybrid prediction method for dataset (April). (c) Prediction
results using hybrid prediction method for dataset (July). (d) Prediction results using
hybrid prediction method for dataset (September).
prediction techniques such as BPNN and SVM. The prediction results for
different datasets are shown in Figures 1.13a–1.13d. From the results, it
can be seen that the detection of load spikes is better in EMD based hybrid
methods when compared to single prediction techniques.
1.4.4 Comparative Analysis

Performance comparison of the single prediction and hybrid prediction
method are carried out in terms of Root Means Square Error (RMSE) and
Mean Average Percentage Error (MAPE). RMSE and MAPE of different
prediction algorithms for the datasets January, April, July, and September
are listed in Table 1.2. Figure 1.14a shows the RMSE of the selected pre-
dictions techniques for different datasets. Similarly, MAPE of the selected
predictions techniques for different datasets is shown in Figure 1.14b.
The obtained results show the variability between datasets. For the
selected datasets, overall prediction performance of ELM is good among
the single prediction models. On the other hand, the EMD-SVM hybrid
model provides better prediction results when compared to EMD-BPNN
in terms of error accuracy.
It is clear from the obtained results that the prediction accuracy of
hybrid prediction methods is higher than the single prediction methods in
terms of error accuracy and error percentage.
Table 1.2 Performance evaluation of different prediction methods.

EMD-
BPNN SVM ELM EMD-BPNN SVM
January RMSE 0.2151 0.1841 0.1995 0.1471 0.1349
MAPE 0.5982 0.5567 0.6770 0.5768 0.5205
April RMSE 0.0798 0.0793 0.0741 0.0615 0.0661
MAPE 0.5814 0.5783 0.5679 0.5241 0.4754
July RMSE 0.0642 0.0795 0.0768 0.0791 0.0758
MAPE 0.4793 0.4067 0.3791 0.3935 0.4166
September RMSE 0.1895 0.1491 0.1530 0.1380 0.1328
MAPE 0.5986 0.3312 0.3741 0.1281 0.3970
(a)
0.25
0.2 BPNN
SVM
0.15 ELM
RMSE
EMD-BPNN
EMD-SVM
0.1
0.05
0
January April July September
Dataset
(b)
0.8
BPNN
SVM
0.6 ELM
EMD-BPNN
EMD-SVM
MAPE
0.4
0.2
0
January April July September
Dataset
Figure 1.14 (a) RMSE of selected predictions techniques for different datasets. (b) MAPE
of selected predictions techniques for different datasets.
1.5 Conclusion
A comprehensive review of different supervised AI-based load forecasting
techniques is presented and discussed. A comprehensive overview of sin-
gle and hybrid prediction strategies for load forecasting is presented and a
comparative assessment of single and hybrid predictions is also performed
on smart meter data for load forecasting. To carry out the performance
analysis, BPNN, SVM, and ELM are selected from single prediction tech-
niques and EMD-BPNN or EMD-SVM are selected among the hybrid
approaches. The analysis is carried out by performing simulation using
the smart meter datasets for January, April, July, and September, which
are collected from Low Carbon London Project database. The results are
analysed in detail in terms of prediction accuracy, RMSE, and MAPE. It
is shown that the hybrid prediction method produces better accuracy in
terms of magnitude of error and error percentage. Among the different
prediction methods selected for analysis, the EMD-SVM method has sig-
nificant improvement in terms of error accuracy.
References
1. S. M. Sulaiman, P. Aruna Jeyanthy, D. Devaraj, S. S. Mohammed, and K.
V Shihabudheen, “Smart Meter Data Analytics for Load Prediction using
Extreme Learning Machines and Artificial Neural Networks,” in 2019 IEEE
International Conference on Clean Energy and Energy Efficient Electronics
Circuit for Sustainable Development (INCCES), Dec. 2019, pp. 1–4, doi:
10.1109/INCCES47820.2019.9167736.
2. W. Kong, Z. Y. Dong, D. J. Hill, F. Luo, and Y. Xu, “Short-Term Residential
Load Forecasting Based on Resident Behaviour Learning,” IEEE Trans. Power
Syst., vol. 33, no. 1, pp. 1087–1088, 2018, doi: 10.1109/TPWRS.2017.2688178.
3. E. Eskandarnia, S. Abdul Kareem, and H. Al-Ammal, “A Review of Smart
Meter Load Forecasting Techniques: Scale and Horizon,” 2018.
4. K. Amasyali and N. M. El-Gohary, “A review of data-driven building energy
consumption prediction studies,” Renew. Sustain. Energy Rev., vol. 81, pp.
1192–1205, 2018, doi: https://doi.org/10.1016/j.rser.2017.04.095.
5. H. Zhao and F. Magoulès, “A review on the prediction of building energy
consumption,” Renew. Sustain. Energy Rev., vol. 16, no. 6, pp. 3586–3592,
2012, doi: https://doi.org/10.1016/j.rser.2012.02.049.
6. L. Pérez-Lombard, J. Ortiz, and C. Pout, “A review on buildings energy con-
sumption information,” Energy Build., vol. 40, no. 3, pp. 394–398, 2008, doi:
https://doi.org/10.1016/j.enbuild.2007.03.007.
7. W. Kong, Z. Y. Dong, Y. Jia, D. J. Hill, Y. Xu, and Y. Zhang, “Short-Term
Residential Load Forecasting Based on LSTM Recurrent Neural Network,”
IEEE Trans. Smart Grid, vol. 10, no. 1, pp. 841–851, 2019, doi: 10.1109/
TSG.2017.2753802.
8. D. C. Park, M. A. El-Sharkawi, R. J. Marks, L. E. Atlas, and M. J. Damborg,
“Electric load forecasting using an artificial neural network,” IEEE Trans.
Power Syst., vol. 6, no. 2, pp. 442–449, 1991, doi: 10.1109/59.76685.
9. K. L. Ho, Y.-. Hsu, and C.-. Yang, “Short term load forecasting using a mul-
tilayer neural network with an adaptive learning algorithm,” IEEE Trans.
Power Syst., vol. 7, no. 1, pp. 141–149, 1992, doi: 10.1109/59.141697.
10. Z. Yun, Z. Quan, S. Caixin, L. Shaolan, L. Yuming, and S. Yang, “RBF Neural
Network and ANFIS-Based Short-Term Load Forecasting Approach in Real-
Time Price Environment,” IEEE Trans. Power Syst., vol. 23, no. 3, pp. 853–
858, 2008, doi: 10.1109/TPWRS.2008.922249.
11. P. Li, Y. Li, Q. Xiong, Y. Chai, and Y. Zhang, “Application of a hybrid quan-
tized Elman neural network in short-term load forecasting,” Int. J. Electr.
Power Energy Syst., vol. 55, pp. 749–759, 2014, doi: https://doi.org/10.1016/j.
ijepes.2013.10.020.
12. D. L. Marino, K. Amarasinghe, and M. Manic, “Building energy load forecast-
ing using Deep Neural Networks,” in IECON 2016 - 42nd Annual Conference
of the IEEE Industrial Electronics Society, 2016, pp. 7046–7051, doi: 10.1109/
IECON.2016.7793413.
Another random document with
no related content on Scribd:
mamma, who was rather glad to be spared that satisfaction.)
A single day was now all that remained of the visit,—a day which
dawned finer than ever, making it so hard to quit the hills, and the
loch, and all the charms of this beautiful place. Not a cloud on the
sky, not a ripple on the waters, blackberries saying “come gather
me,” by hundreds from every bramble, ferns of rare sort growing on
dikes, and banks, and roots of trees. This whole morning must be
spent on the hillside by Sunny and her mamma, combining business
with pleasure, if possible.
So they took a kitchen knife as an extempore spade; a basket, filled
with provisions, but meant afterwards to carry roots, and the well-
known horn cup, which was familiar with so many burns. Sunny used
it for all sorts of purposes besides drinking; filled it with pebbles,
blackberries, and lastly with some doubtful vegetables, which she
called “ferns,” and dug up, and brought to her mamma to take home
“very carefully.”
Ere long she was left to mamma’s charge entirely, for this was the
last day, and Lizzie had never climbed a mountain, which she was
most anxious to do, having the common delusion that to climb a
mountain is the easiest thing in the world,—as it looks, from the
bottom.
Off she started, saying she should be back again directly, leaving
mamma and the child to watch her from the latest point where there
was a direct path,—the cottage where the old woman had come out
and gone to the burn at sunrise. Behind it was a large boulder,
sunshiny and warm to sit on, sheltered by a hayrick, on the top of
which was gambolling a pussy-cat. Sunny, with her usual love for
animals, pursued it with relentless affection, and at last caught it in
her lap, where it remained about one minute, and then darted away.
Sunny wept bitterly, but was consoled by a glass of milk kindly
brought by the old woman; with which she tried to allure pussy back
again, but in vain.
So there was nothing for it but to sit on her mamma’s lap and watch
her Lizzie climbing up the mountain, in sight all the way, but
gradually diminishing to the size of a calf, a sheep, a rabbit; finally of
a black speck, which a sharp eye could distinguish moving about on
the green hillside, creeping from bush to bush, and from boulder to
boulder, till at last it came to the foot of a perpendicular rock.
“She’ll no climb that,” observed the old woman, who had watched the
proceeding with much interest. “Naebody ever does it: she’d better
come down. Cry on her to come down.”
“Will she hear?”
“Oh, yes.”
And in the intense stillness, also from the law of sound ascending, it
was curious how far one could hear. To mamma’s great relief, the
black dot stopped in its progress.
“Lizzie, come down,” she called again, slowly and distinctly, and in a
higher key, aware that musical notes will reach far beyond the
speaking-voice. “You’ve lost the path. Come down!”
“I’m coming,” was the faint answer, and in course of time Lizzie
came, very tired, and just a little frightened. She had begun to climb
cheerfully and rapidly at first, for the hillside looked in the distance
nearly as smooth as an English field. When she got there, she found
it was rather different,—that heather bushes, boulders, mosses, and
bogs were not the pleasantest walking. Then she had to scramble on
all-fours, afraid to look downward, lest her head should turn dizzy,
and she might lose her hold, begin rolling and rolling, and never stop
till she came to the bottom. Still, she went on resolutely, her stout
English heart not liking to be beaten even by a Scotch mountain;
clinging from bush to bush,—at this point a small wood had grown
up,—until she reached a spot where the rock was perpendicular,
nay, overhanging, as it formed the shoulder of the hill.
“I might as well have climbed up the side of a house,” said poor
Lizzie, forlornly; and looked up at it, vexed at being conquered but
evidently thankful that she had got down alive. “Another time,—or if I
have somebody with me,—I do believe I could do it.”
Bravo, Lizzie! Half the doings in the world are done in this spirit.
Never say die! Try again. Better luck next time.
Meanwhile she drank the glass of milk offered by the sympathising
old Highland woman, who evidently approved of the adventurous
English girl, then sat down to rest beside Little Sunny.
But Sunny had no idea of resting. She never has, unless in bed and
asleep. Now she was bent upon also climbing a mountain,—a
granite boulder about three feet high.
“Look, mamma, look at Sunny! Sunny’s going to climb a mountain,
like Lizzie.”
Up she scrambled, with both arms and legs,—catching at the edges
of the boulder, but tumbling back again and again. Still she was not
daunted.
“Don’t help me!—don’t help me!” she kept saying. “Sunny wants to
climb a mountain all by her own self.”
Which feat she accomplished at last, and succeeded in standing
upright on the top of the boulder, very hot, very tired, but triumphant.
“Look, mamma! Look at Sunny! Here she is!”
Mamma looked; in fact had been looking out of the corner of her eye
the whole time, though not assisting at all in the courageous effort.
“Yes, I see. Sunny has climbed a mountain. Clever little girl! Mamma
is so pleased!”
How many “mountains” will she climb in her life, that brave little soul!
Mamma wonders often, but knows not. Nobody knows.
In the meantime, success was won. She, her mamma, and her
Lizzie, had each “climbed a mountain.” But they all agreed that,
though pleasant enough in its way, such a performance was a thing
not to be attempted every day.
CHAPTER XI.
The last day came,—the last hour. Sunny, her mamma, and her
Lizzie, had to turn their ways homeward,—a long, long journey of
several hundred miles. To begin it at four in the morning, with a child,
too, was decided as impracticable; so it was arranged that they
should leave overnight, and sleep at the only available place, an inn
which English superiority scornfully termed a “public-house,” but
which here in the Highlands was called the “hotel,” where
“gentlemen could be accommodated with excellent shooting
quarters.” Therefore, it was supposed to be able to accommodate a
lady and a child,—for one night, at least.
Fortunately, the shooting gentlemen did not avail themselves of it; for
the hotel contained only two guest-rooms. These being engaged,
and the exact time of the boat next morning learned,—which was not
so easy, as everybody in the neighbourhood gave different advice
and a different opinion,—the departure was settled.
Lovelier than ever looked the hills and the loch when the carriage
came around to the door. All the little boys crowded around it, with
vociferous farewell,—which they evidently thought great fun,—Sunny
likewise.
“Good-bye! good-bye!” cried she, as cheerfully as if it had been “how
d’ye do,” and obstinately refused to be kissed by anybody. Indeed,
this little girl does not like kisses, unless she offers them of her own
accord.
One only grief she had, but that was a sharp one. Maurice’s papa,
who had her in his arms, suddenly proposed that they should “send
mamma away and keep Sunny;” and the scream of agony she gave,
and the frantic way she clung to her mamma, and would not look at
anybody for fear of being kept prisoner, was quite pathetic.
At last the good-byes were over. For Little Sunshine these are as yet
meaningless; life to her is a series of delights,—the new ones
coming as the old ones go. The felicity of kissing her hand and
driving away was soon followed by the amusement of standing on
her mamma’s lap, where she could see everything along the road,
which she had passed a fortnight before in dark night.
Now it was golden twilight,—such a twilight! A year or two hence
Sunny would have been in ecstasy at the mountains, standing range
behind range, literally transfigured in light, with the young moon
floating like a “silver boat” (only turned the wrong way uppermost)
over their tops. As it was, the large, distant world interested her less
than the small, near one,—the trees that swept her face as she
drove along the narrow road, and the numerous cows and calves
that fed on either side of it.
There was also a salt-water loch, with fishing-boats drawn up on the
beach, and long fishing-nets hanging on poles; but not a living
creature in sight, except a heron or two. These stood on one leg,
solemnly, as herons do, and then flew off, flapping their large wings
with a noise that made Little Sunshine, as she expressed it, “nearly
jump.” Several times, indeed, she “nearly jumped” out of the carriage
at the curious things she saw: such funny houses, such little
windows,—“only one pane, mamma,”—and, above all, the girls and
boys barefooted, shock-headed, that hung about staring at the
carriage as it passed.
“Have those little children got no Lizzie to comb their hair?” she
anxiously inquired; and mamma was obliged to confess that
probably they had not, at which Sunny looked much surprised.
It was a long, long drive, even with all these entertainments; and
before it ended, the twilight had faded, the moon crept higher over
the hill, and Sunshine asked in a whisper for “Maymie’s apron.” The
little “Maymie’s apron,” which had long lain in abeyance, was
produced, and she soon snuggled down in her mamma’s arms and
fell fast asleep.
When she woke up the “hotel” was reached. Such a queer hotel! You
entered by a low doorway, which opened into the kitchen below, and
a narrow staircase leading to the guest-rooms above. From the
kitchen Sunny heard a baby cry. She suddenly stopped, and would
not go a step till mamma had promised she should see the baby,—a
very little baby, only a week old. Then she mounted with dignity up
the rickety stairs, and began to examine her new apartments.
They were only two, and as homely as they well could be. Beside the
sitting-room was a tiny bedroom, with a “hole in the wall,” where
Lizzie was to sleep. This “hole in the wall” immediately attracted
Sunny; she jumped in it, and began crawling about it, and tried to
stand upright under it, which, being such a very little person, she was
just able to do. Finally, she wanted to go to sleep in it, till, hearing
she was to sleep with mamma, a much grander thing, she went up to
the bed, and investigated it with great interest likewise. Also the
preparations for her bath, which was to be in a washing-tub in front
of the parlour fire,—a peat fire. It had a delicious, aromatic smell,
and it brightened up the whole room, which was very clean and tidy,
after all.
So was the baby, which shortly appeared in its mother’s arms. She
was a pale, delicate woman, speaking English with the slow
precision of a Highlander, and having the self-composed, courteous
manner that all Highlanders have. She looked much pleased when
her baby was admired,—though not by Sunny, who, never having
seen so young a baby before, did not much approve of it, and
especially disapproved of seeing it taken into her own mamma’s
arms. So presently it and its mother disappeared, and Sunny and her
mamma were left to eat their supper of milk, bread and butter, and
eggs; which they did with great content. Sunny was not quite so
content to go to bed, but cried a little, till her mamma set the parlour
door half open, that the firelight might shine in. Very soon she also
crept in beside her little girl; who was then not afraid of anything.
But when they woke, in the dim dawn, it was under rather
“frightening” circumstances. There was a noise below, of a most
extraordinary kind, shouting, singing, dancing,—yes, evidently
dancing, though at that early hour of the morning. It could not have
been continued from overnight, mamma having distinctly heard all
the family go to bed, the children tramping loudly up the stairs at nine
o’clock, after which the inn was quite quiet. No, these must be new
guests, and very noisy guests, too. They stamped, they beat with
their feet, they cried “whoop!” or “hech!” or some other perfectly
unspellable word, at regular intervals. Going to sleep again was
impossible; especially as Sunny, unaccustomed to such a racket,
began to cry, and would have fallen into a downright sobbing fit, but
for the amusement of going to the “hole in the wall,” to wake her
Lizzie. Upon which everybody rose, the peat fire was rekindled, and
the new day began.
The good folk below stairs must have begun it rather early. They
were a marriage party, who had walked over the hills several miles,
to see the bride and bridegroom off by the boat.
“Sunny wants to look at them,” said the child, who listens to
everything, and wants to have a finger in every pie.
So, as soon as dressed, she was taken down, and stood at the door
in her mamma’s arms to see the fun.
Very curious “fun” it was. About a dozen young men and women,
very respectable-looking, and wonderfully dressed, though the
women had their muslin skirts pretty well draggled,—not surprising,
considering the miles they had trudged over mountain and bog, in
the damp dawn of the morning,—were dancing with all their might
and main, the lassies with their feet, the lads with feet, heads, hands,
tongues, snapping their fingers and crying “hech!” or whatever it
was, in the most exciting manner. It was only excitement of dancing,
however; none of them seemed the least drunk. They stopped a
minute, at sight of the lady and child, and then went on again,
dancing most determinedly, and as solemnly as if it were to save
their lives, for the next quarter of an hour.
English Lizzie, who had never seen a Highland reel before, looked
on with as much astonishment as Sunny herself. That small person,
elevated in her mamma’s arms, gazed on the scene without a single
smile; there being no music, the dance was to her merely a noise
and a scuffle. Presently she said, gravely, “Now Sunny will go away.”
They went away, and after drinking a glass of milk,—oh, what
delicious milk those Highland cows give!—they soon heard the
distant paddles of the boat, as she steamed in between the many
islands of which this sea is full.
Then mounting an extraordinary vehicle, which in the bill was called
a “carridge,” they headed a procession, consisting of the wedding
party walking sedately two and two, a young man and young woman
arm in arm, down to the pier.
The married couple were put on board the boat (together with Sunny,
her mamma, and her Lizzie, who all felt very small, and of no
consequence whatever), then there was a great shouting and waving
of handkerchiefs, and a spluttering and splattering of Gaelic good
wishes, and the vessel sailed away.
By this time it was broad daylight, though no sun was visible. Indeed,
the glorious sunrises seemed ended now; it was a gray, cheerless
morning, and so misty that no mountains could be seen to take
farewell of. The delicious Highland life was all gone by like a dream.
This homeward journey was over the same route that Sunny had
travelled a fortnight before, and she went through it in much the
same fashion.
She ran about the boat, and made friends with half a dozen people,
for no kindly face is long a strange face to Little Sunshine. She was
noticed even by the grim, weather-beaten captain (he had a lot of
little people of his own, he said), only when he told her she was “a
bonnie wee lassie,” she once more indignantly repelled the
accusation.
“I’m not a bonnie wee lassie. I’m Sunny, mamma’s little Sunny,”
repeated she, and would not look at him for at least two minutes.
She bore the various changes from sea-boat to canal-boat, etc., with
her usual equanimity. At one place there was a great crush, and they
got so squeezed up in a crowd that her mamma did not like it at all,
but Sunny was perfectly composed, mamma’s arms being
considered protection against anything. And when the nine locks
came, she cheerfully disembarked, and walked along the towing-
path for half a mile in the bravest manner. Gradually, as amusement
began to fail her, she found several playfellows on board, a little dog
tied by a string, and a pussy-cat shut up in a hamper, which formed
part of the luggage of an unfortunate gentleman travelling to London
with five daughters, six servants, and about fifty boxes,—for he was
overheard counting them. In the long, weary transit between the
canal-boat and the sea, Sunny followed this imprisoned cat, which
mewed piteously; and in its sorrows she forgot her own.
But she was growing very tired, poor child! and the sunshine, which
always has a curious effect upon her temper and spirits, had now
altogether disappeared. A white, dull, chill mist hung over the water,
fortunately not thick enough to stop traffic, as had happened two
days before, but still enough to make the river very dreary.
Little Sunshine, too, went under a cloud; she turned naughty, and
insisted on doing whatever she was bid not to do; climbing in the
most dangerous places, leaning over the boat’s side to look at the
waves: misbehaviour which required a strong hand and watchful
eyes to prevent serious consequences. But mamma was more sorry
than angry, for it was hard for the little woman; and she was
especially touched when, being obliged to forbid some stale,
unwholesome fruit and doubtful “sweeties,” over which Sunny
lingered and longed, by saying “they belonged to the captain,” the
child answered, sweetly:
“But if the kind captain were to give Sunny some, then she might
have them?”
The kind captain not appearing, alas! she passed the basket with a
sigh, and went down to the engines. To see the gigantic machinery
turning and turning, never frightened, but only delighted her. And
mamma was so thankful to find anything to break the tedium of the
fourteen hours’ journey, that though her little girl went down to the
engine-room neat and clean in a white pelisse, and came up again
looking just like a little sweep, she did not mind it at all!
Daylight faded; the boat emptied gradually of its passengers,
including the gentleman with the large family and the fifty boxes; and
on deck it began to grow very cold. Sunny had made excursions
down below for breakfast, dinner, and tea, at all of which meals she
conducted herself with the utmost propriety, but now she took up her
quarters permanently in the comfortable saloon.
Not to sleep, alack! though her mamma settled down in a corner, and
would have given anything for “just one little minute,” as Sunny says,
of quiet slumber, but the child was now preternaturally wide awake,
and as lively as a cricket. So was a little boy, named Willie, with
whom she had made friends, and was on such terms of intimacy that
they sat on the floor and shared their food together, and then jumped
about, playing at all sorts of games, and screaming with laughter, so
that even the few tired passengers who remained in the boat, as she
steamed up the narrow, foggy river, could not help laughing too.
This went on for the space of two hours more, and even then Sunny,
who was quite good now, was with difficulty caught and dressed, in
preparation for the stopping of the boat, when she was promised she
should see papa. But she will endure any martyrdom of bonnet-tying
or boot-buttoning if only she thinks she is going to meet her papa.
Unluckily there had been some mistake as to hours, and when she
was carried on deck, in the sudden darkness, broken only by the
glimmer of the line of lights along the wharf, and plunged into the
midst of a dreadful confusion,—porters leaping on board and
screaming to passengers, and passengers searching wildly for their
luggage,—no papa was there. To double her grief, she also lost her
mamma, who of course had to see to things at once herself. Through
the noise and whirl she heard the voice of the child, “Mamma!
mamma!” It was a cry not merely of distress,—but agony, with a
“grown-up” tone in it of actual despair. No doubt the careless jest of
Maurice’s papa had rankled in her little mind, and she thought
mamma was torn from her in real truth, and for ever.
When at last mamma came back, the grasp with which the poor little
girl clung to her neck was absolutely frantic.
“Mamma went away and left Sunny,—Sunny lost mamma,” and
mamma could feel the little frame shaking with terror and anguish.
Poor lamb! there was nothing to be done but to take her and hold her
tight, and stagger with her somehow across the gangway to the cab.
But even there she never loosened her clasp for a minute till she got
safe into a bright, warm house, where she found her own papa. Then
the little woman was content.
She had still another journey before her, and without her papa too. A
night journey, which promised to be easy and comfortable, but
turned out quite the contrary. A journey in which Sunny’s powers of
endurance were taxed to the utmost, so that it will be years before
she forgets the wind-up of her holiday.
Her papa put his family safe in a carriage all to themselves, and
under special charge of the guard. Then he left them, just settling
down to sleep; Sunny being disposed of in a snug corner, with an air-
cushion for a pillow, and furry shawls wrapped about her, almost as
cosy as in her own little crib, in which, after her various changes and
vicissitudes, she was soon to repose once more.
She fell asleep in five minutes, and her mamma, who was very tired,
soon dozed also, until roused by a sharp cry of fright. There was the
poor little girl, lying at the bottom of the carriage, having been thrown
there by its violent rocking. It rocked still, and rocked for many, many
miles, in the most dreadful manner. When it stopped the guard was
appealed to, who said it was “the coupling-chains too slack,” and
promised to put all right. So the travellers went to sleep again, this
time Sunny in her mamma’s arms, which she refused to quit.
Again more jolting, and another catastrophe; mamma and the child
finding themselves lying both together on the floor. This time Sunny
was much frightened, and screamed violently, repulsing even her
mamma.
“I thought you were not my own mamma; I thought you were
somebody else,” said she, afterward, and it was a long time before
she came to her right self and cuddled down; the oscillation of the
carriage continuing so bad that it was as much as her mamma could
do, by wrapping her own arms around her, to protect the poor child
from being hurt and bruised.
The guard, again appealed to, declared there was no danger, and
that he would find a more comfortable carriage at the next stopping-
place: but in vain. It was a full train, and the only two seats vacant
were in a carriage full of gentlemen, who might object to a poor,
sleepy, crying child. The little party went hopelessly back.
“Perhaps those gentlemen might talk so loud they might waken
Sunny,” said the child, sagely, evidently remembering her
experiences of five weeks ago. At any rate, nobody wished to try the
experiment.
Since there was no actual danger, the only remedy was endurance.
Mamma settled herself as firmly as she could, making a cradle of her
arms. There, at length, the poor child, who had long ceased crying,
and only gave an occasional weary moan, fell into a doze, which
ended in quiet sleep. She was very heavy, and the hours seemed
very long, but still they slipped away somehow. Nothing is absolutely
unbearable when one feels that, being inevitable, it must be borne.
Of course nobody slept, except the child, until near daybreak, when
a new and more benevolent guard came to the rescue, had the
coupling-chains fastened (which, they found, had never been done
at all till now), and lessened the shaking of the carriage. Then tired
Lizzie dropped asleep too, and the gray morning dawned upon a
silent carriage, sweeping rapidly across the level English country, so
different from that left behind. No more lochs, no more mountains.
No more sunshine either, as it appeared; for there was no sign of
sunrise, and the day broke amidst pelting rain, which kept drip, drip,
upon the top of the carriage, till it seemed as if a deluge would soon
be added to the troubles of the journey.
But these were not so bad now. Very soon the little girl woke up,
neither frightened nor cross, but the same sunshiny child as ever.
“Mamma!” she said, and smiled her own beaming smile, and sat up
and looked about her. “It’s daylight. Sunny wants to get up.”
That getting up was a most amusing affair. It lasted as long as
mamma’s ingenuity could possibly make it last, without any
assistance from poor, worn-out Lizzie, who was left to sleep her fill.
First, Sunny’s face and hands had to be washed with a damp
sponge, and wiped with mamma’s pocket-handkerchief. Then her
hair was combed and brushed, with a brush that had a looking-glass
on the back of it; in which she contemplated herself from time to
time, laughing with exceeding merriment. Lastly, there was breakfast
to be got ready and eaten.
A most original breakfast! Beginning with a large pear, out of a
basketful which a kind old gentleman had made up as a special
present to Sunny; then some ham sandwiches,—from which the ham
was carefully extracted; then a good drink of milk. To uncork the
bottle in which this milk had been carried, and pour it into the horn
cup without spilling, required an amount of skill and care which
occupied both mamma and Sunny for ever so long. In fact, they
spent over their dressing and breakfasting nearly an hour; and by
this time they were both in the best of spirits, and benignly
compassionate to Lizzie, who slept on, and wanted no breakfast.
And when the sun at last came out, a watery and rather melancholy
orb, not at all like the sun of the Highlands, the child was as bright
and merry as if she had not travelled at all, and played about in the
railway-carriage just as if it were her own nursery.
This was well, for several weary hours had still to be passed; the
train was far behind its time; and what poor mamma would have
done without the unfailing good temper of her “sunshiny child,” she
could not tell. When London was reached, and the benevolent guard
once more put his head into the carriage, with “Here we are at last. I
should think you’d had enough of it, ma’am,” even he could not help
giving a smile to the “little Missy” who was so merry and so good.
In London was an hour or two more of weary delay; but it was under
a kindly roof, and Sunny had a second beautiful breakfast, all proper,
with tea-cups and a table-cloth; which she did not seem to find half
so amusing as the irregular one in the railway-carriage. But she was
very happy, and continued happy, telling all her adventures in
Scotland to a dear old Scotchwoman whom she loves exceedingly,
and who loves her back again. And being happy, she remained
perfectly good, until once more put into a “puff-puff,” to be landed at
her own safe home.
Home. Even the child understood the joy of going home. She began
talking of “Sunny’s nursery;” “Sunny’s white pussy;” “Sunny’s little
dog Rose;” and recalling all the servants by name, showing she
forgot nothing and nobody, though she had been absent so long.
She chattered all the way down, till some ladies who were in the
carriage could hardly believe she had been travelling all night. And
when the train stopped, she was the first to look out of the window
and call out, “There’s godmamma!”
So it was! Sunny’s own, kind godmamma, come unexpectedly to
meet her and her tired mamma at the station; and oh, they were both
so glad!
“Glad” was a small word to express the perfect and entire felicity of
getting home,—of finding the house looked just as usual; that the
servants’ cheerful faces beamed welcome; that even the doggie
Rose barked, and white pussy purred, as if both were glad Little
Sunshine was back again. She marched up-stairs, lifting her short
legs deliberately one after the other, and refusing to be carried; then
ran into her nursery just as if she had left it only yesterday. And she
“allowed” her mamma to have dinner with her there, sitting at table,
as grand as if she were giving a dinner-party; and chattering like a
little magpie to the very end of the meal. But after that she collapsed.
So did her mamma. So did her Lizzie. They were all so dreadfully
tired that human nature could endure no more. Though it was still
broad daylight, and with all the delights of home around them, they
went to bed, and slept straight on,—mamma “all around the clock,”
and the child and her Lizzie for fourteen hours!
Thus ended Little Sunshine’s Holiday. It is told just as it happened, to
amuse other little people, who no doubt are as fond as she is of
hearing “stories.” Only this is not a story, but the real truth. Not the
whole truth, of course, for that would be breaking in upon what
grown-up people term “the sanctities of private life.” But there is no
single word in it which is not true. I hope you will like it, little people,
simple as it is. And so, good-bye!
Finis
L. C. Page & Company’s
Cosy Corner Series
OF
Charming Juveniles
Each one volume, 16mo, cloth, Illustrated, 50 cents
Ole Mammy’s Torment. By Annie Fellows-Johnston.

Author of “The Little Colonel,” etc.
The Little Colonel. By Annie Fellows-Johnston.
Author of “Big Brother.”
Big Brother. By Annie Fellows-Johnston.
The Gate of the Giant Scissors. By Annie Fellows-
Johnston.
Two Little Knights of Kentucky, who were “The Little
Colonel’s” neighbors. By Annie Fellows-Johnston.
A sequel to “The Little Colonel.”
The Story of Dago. By Annie Fellows-Johnston.
Farmer Brown and the Birds. By Frances Margaret
Fox.
A little story which teaches children that the birds are
man’s best friends.
Story of a Short Life. By Juliana Horatia Ewing.
This beautiful and pathetic story is a part of the
world’s literature and will never die.
Jackanapes. By Juliana Horatia Ewing.
A new edition, with new illustrations, of this exquisite
and touching story, dear alike to young and old.
The Little Lame Prince. By Miss Mulock.
A delightful story of a little boy who has many
adventures by means of the magic gifts of his fairy
godmother.
The Adventures of a Brownie. By Miss Mulock.
The story of a household elf who torments the cook
and gardener, but is a constant joy and delight to
the children.
His Little Mother. By Miss Mulock.
Miss Mulock’s short stories for children are a constant
source of delight to them, and “His Little Mother,” in
this new and attractive dress, will be welcomed by
hosts of readers.
Little Sunshine’s Holiday. By Miss Mulock.
“Little Sunshine” is another of those beautiful child-
characters for which Miss Mulock is so justly
famous.
Wee Dorothy. By Laura Updegraff.
A story of two orphan children, the tender devotion of
the eldest, a boy, for his sister being its theme.
Rab and His Friends. By Dr. John Brown.
Doctor Brown’s little masterpiece is too well known to
need description.
The Water People. By Charles Lee Sleight.
Relating the further adventures of “Harry,” the little
hero of “The Prince of the Pin Elves.”
The Prince of the Pin Elves. By Chas. Lee Sleight.
A fascinating story of the underground adventures of
a sturdy, reliant American boy among the elves and
gnomes.
Helena’s Wonderworld. By Frances Hodges White.
A delightful tale of the adventures of a little girl in the
mysterious regions beneath the sea.
The Adventures of Beatrice and Jessie. By Richard
Mansfield.
A bright and amusing story of the strange adventures
of two little girls in the “realms of unreality.”
A Child’s Garden of Verses. By R. L. Stevenson.
This little classic is undoubtedly the best of all
volumes of poetry for children.
Little King Davie. By Nellie Hellis.
It is sufficient to say of this book that it has sold over
110,000 copies in England, and consequently
should well be worthy of a place in “The Cosy
Corner Series.”
Little Peterkin Vandike. By Charles Stuart Pratt.
The author’s dedication furnishes a key to this
charming story.
“I dedicate this book, made for the amusement of the
boys who may read it, to the memory of one boy,
who would have enjoyed as much as Peterkin the
plays of the Poetry Party.”
The Making of Zimri Bunker. A Tale of Nantucket. By W.
J. Long.
The story deals with a sturdy American fisher lad
during the war of 1812.
The Fortunes of the Fellow. By Will Allen Dromgoole.
A sequel to “The Farrier’s Dog and His Fellow.”
The Farrier’s Dog and His Fellow. By Will Allen
Dromgoole.
This story, written by the gifted young Southern
woman, will appeal to all that is best in the natures
of her many admirers.
The Sleeping Beauty. A Modern Version. By Martha B.
Dunn.
A charming story of a little fishermaid of Maine,
intellectually “asleep,” until she meets the “Fairy
Prince.”
The Young Archer. By Charles E. Brimblecom.
A strong and wholesome story of a boy who
accompanied Columbus on his voyage to the New
World.
For His Country. By Marshall Saunders.
A beautiful story of a patriotic little American lad.
A Little Puritan’s First Christmas. By Edith Robinson.
A Little Daughter of Liberty. By Edith Robinson.
Author of “A Loyal Little Maid,” “A Little Puritan
Rebel,” etc.
A true story of the Revolution.
A Little Puritan Rebel. By Edith Robinson.
An historical tale of a real girl, during the time when
the gallant Sir Harry Vane was governor of
Massachusetts.
A Loyal Little Maid. By Edith Robinson.

Hybrid Intelligent Approaches For Smart Energy Practical Applications Senthil Kumar Mohan Full Chapter

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Hybrid Intelligent Approaches For Smart Energy Practical Applications Senthil Kumar Mohan Full Chapter

Uploaded by

Copyright:

Available Formats

Hybrid Intelligent Approaches for Smart

Energy: Practical Applications Senthil

Wiley Global Headquarters

Limit of Liability/Disclaimer of Warranty

Library of Congress Cataloging-in-Publication Data

Printed in the USA

List of Contributors xiii

1.4.3 Performance Analysis of Hybrid Prediction Methods

4 Energy Optimization Using Silicon Dioxide Composite

6.1 Introduction 108

9 Machine Learning Based Approach for Energy Management

10.4.7 GSM (Global System for Mobile Communication)

12.3.2 EV and Power-to-Gas: State-of-the-Art

14.2 Energy Optimization Process 281

This book aims to primarily address issues surrounding the optimization,

Chapter 9 focuses on machine learning-based approaches in energy

We are deeply indebted to Almighty God for giving us this opportunity

I sincerely thank my VIT management for continuous support and encour-

Senthilkuamr Mohan, PhD

I express my sincere thanks to the management of CTiF Global Capsule,

I express my sincere thanks to the management of Abu Dhabi Polytechnic.

Yasir Hamid, PhD

*Corresponding author: ssheikmd@yahoo.co.in

Figure 1.1 Structure of smart grid [1].

with loads, ESS, sensors, and communications networks connected in

Conditional Demand Analysis (CDA), Auto Regressive Moving Average

1.2 Forecasting Methodology

Load time Feature Predicted

more informative analysis domain helps in sensing the hidden character-

1.3 AI-Based Prediction Methods

1.3.1 Single Prediction Methods

1.3.1.1 Linear Regression

yˆ(k ) = a0 + ∑a x (k) k = 1,2,..., N

N total signal length.

xi (k ) is input feature variables

Figure 1.3 Classification of AI-based prediction.

forecasting strategy using multivariable linear regression is proposed in

1.3.1.2 Artificial Neural Networks (ANN)

Input Hidden Output

arrives at a particular neuron, an activation function at this particular neu-

y (x) = wT σ (vTx + θv) + θw (1.3)

where Ih is the number of neurons of the hidden layer, w ∈ R Ih and

1.3.1.3 Support Vector Regression (SVR)

Figure 1.5 Support vector regression architecture.

where αi is the Lagrangian function to be calculated by quadratic optimiza-

Polynomial Kernel: K ( xi , x ) = (1 + γ xiT x )

1.3.1.4 Extreme Learning Machine

along with better generalization capability when compared to back prop-

 g (W1 X1 + b1 )  g (WNˆ X1 + bNˆ ) 

Step 1: Randomly assign Wi and bi.

1.3.1.5 Neuro-Fuzzy Techniques

1.3.1.6 Deep Learning Techniques

ht = F (Whx xt + Whh ht – 1 + bh) (1.10)

yt = G (Wyh ht + by) (1.11)

Figure 1.6 RNN architecture.

Figure 1.7 LSTM network.

Long Short Term Memory (LSTM) is another promising deep learning

1.3.2 Hybrid Prediction Methods

(i) combined AI-based prediction and

1.3.2.1 Combined AI-Based Prediction Techniques

consisting of unsupervised and supervised methods, is used for short term

1.3.2.2 Signal Decomposition Based Prediction Techniques

Load time Signal Feature Learning Predicted