Content uploaded by Nadeem Javaid

Author content

All content in this area was uploaded by Nadeem Javaid on Jul 16, 2016

Content may be subject to copyright.

Performance Evaluation of Heuristic Algorithms in

Smart Grids

By

Sahar Rahim

CIIT/SP14-REE-004/ISB

MS Thesis

In

Electrical Engineering

COMSATS Institute of Information Technology

Islamabad – Pakistan

Fall, 2015

Performance Evaluation of Heuristic Algorithms in

Smart Grids

A Thesis Presented to

COMSATS Institute of Information Technology, Islamabad

In partial fulfillment

of the requirement for the degree of

MS (Electrical Engineering)

By

Sahar Rahim

CIIT/SP14-REE-004/ISB

Fall, 2015

ii

Performance Evaluation of Heuristic Algorithms in

Smart Grids

A Graduate Thesis submitted to Department of Electrical Engineering as partial

fulfillment of the requirement for the award of Degree of M.S (Electrical Engineering).

Name

Registration Number

Sahar Rahim

CIIT/SP14-REE-004/ISB

Supervisor

Prof. Dr. Shahid A. Khan,

Professor,

Department of Electrical Engineering,

COMSATS Institute of Information Technology (CIIT),

Islamabad Campus,

January 2016.

Co-Supervisor

Dr. Nadeem Javaid,

Associate Professor,

Department of Computer Science,

COMSATS Institute of Information Technology (CIIT),

Islamabad Campus,

January 2016. iii

Final Approval

This thesis titled

Performance Evaluation of Heuristic Algorithms in

Smart Grids

By

Sahar Rahim

CIIT/Sp14-REE-004/ISB

has been approved

For the COMSATS Institute of Information Technology, Islamabad

External Examiner:

Prof. Dr. Muhammad Sher,

Dean, Faculty of Basic and Applied Sciences

International Islamic University, Islamabad

Supervisor:

Prof. Dr. Shahid A. Khan

Department of Electrical Engineering,

CIIT, Islamabad

Co-Supervisor:

Dr. Nadeem Javaid

Associate Professor, Department of Computer Science,

CIIT, Islamabad

HoD:

Prof. Dr. Shahid A. Khan

Department of Electrical Engineering,

CIIT, Islamabad

iv

Declaration

I Ms. Sahar Rahim, CIIT/SP14-REE-004/ISB, hereby declare that I have

produced the work presented in this thesis, during the scheduled period of

study. I also declare that I have not taken any material from any source

except referred to wherever due that amount of plagiarism is within acceptable

range. If a violation of HEC rules on research has occurred in this thesis, I shall

be liable to punishable action under the plagiarism rules of the HEC.

Date:

Signature of the student:

Sahar Rahim

CIIT/SP14-REE-004/ISB

v

Certificate

It is certified that Sahar Rahim CIIT/SP14-REE-004/ISB has carried out all the

work related to this thesis under my supervision at the Department of Electrical

Engineering, COMSATS Institute of Information Technology, Islamabad and the

work fulfills the requirements for the award of the MS degree.

Date:

Supervisor:

Prof. Dr. Shahid A. Khan,

Professor

Co-Supervisor:

Dr. Nadeem Javaid,

Associate Professor

Head of Department:

Prof. Dr. Shahid A. Khan

Department of Electrical Engineering.

vi

DEDICATION

To Almighty Allah, My Family

&

COMSATS Institute of Information Technology

vii

ACKNOWLEDGMENT

Alhamdulillah. With the Bounty, Mercy and Blessing of ALLAH, this dissertation is completed.

Allah guided me in many ways to successfully finalize my efforts. I have received innumerable

support from various people. I would like to mention few words for adequately capture all my

gratitude. First of all, I would like to express my heartiest appreciation to my supervisor, Prof.

Dr. Shahid A. Khan, for his continuous and inspiring guide and co-supervisor, Dr. Nadeem

Javaid, for his patience, support, encouragement, insightful criticism and guidance from

foundation to concluding level. It is impossible for me to complete this dissertation within due

time without their guidance and utmost efforts. I would also like to appreciate my dissertation

committee, Prof. Dr. Junaid Mughal, Dr. Qadeer Ul Hassan and Dr. Mustafa Shakir for their

critical comments and valuable time.

I want to take it as opportunity to heartily thank my best friend and all my fellows who assist me

in all my harsh time and during precarious phases of this dissertation. Finally, I like to thank my

father, whose support, trust and tremendous care makes me able to work hard and my mother,

whose hands always raised to pray for my success. I would like to extend my thanks to my

sisters and brother for their love, friendly attitude and motivate me with encouraging words.

Sahar Rahim

CIIT/SP14-REE-004/ISB

viii

ABSTRACT

Performance Evaluation of Heuristic Algorithm in

Smart Grids

Smart grid (SG) is evolutionary idea in which all components of conventional power grid are

modernized with the advance integration of information technology, sensors and autonomous

system. The bi-directional flow of information and power in SG with optimal integration of

renewable energy sources encourage customers to participate in energy management schemes

and demand response. Meanwhile, innovation components of grid such as transmission, demand

side management and demand response are develop as modernized applications with lots of

benefits as well as challenges in SG. Therefore, we explore potential solution to the interesting

and challenging problems of SG. In this dissertation, we comparatively evaluate the performance

of home energy management controller which is designed on the basis of heuristic algorithms;

genetic algorithm (GA), binary particle swarm optimization (BPSO) and ant colony optimization

(ACO). In this regard, we introduce a generic architecture for demand side management (DSM)

which integrates residential area domain with smart area domain via wide area network. In

addition, problem formulation is carried via multiple knapsack problems. For energy pricing,

combined model of time of use tariff and inclined block rates is used. Simulation results show

that all designed models for energy management act significantly to achieve our objectives and

proven as a cost-effective solution to increase sustainability of SG. GA based energy

management controller performs more efficiently than BPSO based energy management

controller and ACO based energy management controller in terms of electricity bill reduction,

peak to average ratio minimization and user comfort level maximization and its execution time is

also less than other two models.

ix

TABLE OF CONTENTS

List of Figures x

List of Tables xi

1 Introduction 1

2 Related Work and Motivation 6

2.1 Motivation ................................ 11

3 Proposed Model 13

3.1 Energy consumption model ....................... 15

3.2 Load categorization ........................... 16

3.2.1 Fixed appliances ........................ 17

3.2.2 Shiftable appliances ....................... 17

3.2.3 Elastic appliances ........................ 18

3.3 Energy price model ........................... 19

3.4 Local energy generation ........................ 20

3.5 Energy storage ............................. 21

3.6 Residential users ............................ 22

3.6.1 Passive users .......................... 22

3.6.2 Semi-active users ........................ 22

3.6.3 Active users ........................... 23

3.7 Problem formulation .......................... 23

3.8 PAR ................................... 24

3.9 Waiting time .............................. 25

3.10 Objective function ........................... 27

4 Heuristic algorithms 29

4.1 GA .................................... 30

4.2 BPSO .................................. 32

4.3 ACO ................................... 33

5 Simulations and results 35

5.1 Energy consumption pattern and Electricity bill reduction ..... 39

5.2 PAR ................................... 43

5.3 Waiting time .............................. 44

5.4 Parametric tuning for all models .................... 45

viii

LIST OF FIGURES

1.1 SG .................................... 2

3.1 Components of DSM .......................... 14

3.2 EMC model for residential users .................... 15

3.3 DSM functional diagram ........................ 16

3.4 End users classiﬁcation ......................... 23

3.5 Parameters of appliance ........................ 25

3.6 Range of operation time ........................ 26

3.7 Waiting time .............................. 27

5.1 TOU Tariﬀ Model ........................... 36

5.2 Energy consumption (kWh) ...................... 40

5.3 Electricity bill (cent) .......................... 41

5.4 Electricity bill reduction per day .................... 41

5.5 Electricity bill per day ......................... 42

5.6 Electricity bill (cent/day) for 50 users ................. 42

5.7 PAR curve ................................ 43

5.8 Possible trade oﬀ between electricity cost and waiting time ..... 44

x

LIST OF TABLES

1.1 Brief comparison between traditional grid and SG .......... 3

2.1 Summarized related work ........................ 10

4.1 Key points of GA-EMC ........................ 32

4.2 Key points of BPSO-EMC ....................... 33

4.3 Key points of ACO-EMC ........................ 34

5.1 Parameters of Fixed Appliances .................... 37

5.2 Parameters of Shiftable Appliances .................. 37

5.3 Parameters of Elastic Appliances ................... 37

5.4 GA parametric list ........................... 38

5.5 BPSO parametric list .......................... 38

5.6 ACO parametric list .......................... 38

5.7 Execution time ............................. 39

5.8 Summarized results ........................... 45

5.9 Parameter Evaluation for GA ..................... 47

5.10 Parameter Evaluation for BPSO .................... 48

5.11 Parameter Evaluation for ACO .................... 50

xi

1

INTRODUCTION

1

Chapter 1 Introduction 2

Traditional electrical power system is inadequate to meet modern power grid chal-

lenges such as reliability, stability, robustness, etc. [1]. Thus, a new infrastructure

is needed to smartly meet these challenges and reduce pressure on global envi-

ronment. In this regard, smart grid (SG) integrates communication technologies,

computational abilities, control systems and sensors with existing grid and enables

two way ﬂow of information between utility and end users. SG has modernized all

sections of present electrical system as shown in ﬁg. 1.1.

Figure 1.1: SG

Speciﬁcally, the consumers are now prosumers because they have the ability to sell

their generated electricity to the utility. Thus, renewable energy sources (RESs)

(e.g., solar, wind, etc.) play a vital role in the concept of SG. Utilities are always

interested in increasing their proﬁt and reducing of peak to average load. On the

other hand, prosumers wish to reduce their electricity bills without compromising

their comfort level. Main aims of SG are to enhance eﬃciency, sustainability,

capacity and customer engagement [2]. Some of the major diﬀerences between

traditional power grid and SG are summarized in table. 1.1

Chapter 1 Introduction 3

One of the important aspects of SG is demand side management (DSM) which is

the best way to maintain balance between demand and supply. Two main func-

tions of DSM are load management and demand response (DR). Load management

focuses on the improvement of energy eﬃciency to avoid major distress and black-

out. The beneﬁts of load management are numerous such as reduced number of

peak power plants, eﬃcient energy consumption, electricity bill reduction and im-

proved performance of the power grid in term of reliability and ﬂexibility [3]. DR

is a responsive action taken by a customer against dynamic price models. It oﬀers

many ﬁnancial and operational beneﬁts for electricity utilities, end user, and grid

operations. The highly volatile nature of load may threaten the integrity of grid

within seconds. Therefore, DR is important to tackle these uncertainties, as it

provides ﬂexibility at relatively low rates [4].

The common objectives of SG are electricity bill reduction, minimization of aggre-

gated power consumption and minimization of both electricity bill and aggregated

power. To achieve these objectives, many DSM techniques and algorithms are

Table 1.1: Brief comparison between traditional grid and SG

Infrastructures Traditional grid SG

Power system

Centralized generation

Uni-directional power and

information ﬂow

(utility to consumer)

Low storage capacity

Distributed generation

Bi-directional power and

information ﬂow

(utility to (from) consumer)

Grid energy storage capacity

Information technology

Aged metering system

No monitoring system

Lack of management units

Advanced metering system

Phasor management unit

Information management unit

Communication system Wired technology Wired and wireless technologies

Energy sources system Non- renewable sources

(mainly fossil fuel)

Both non-renewable and

renewable sources

(photovoltaic panels (PV), wind

turbine, plug-in electric

vehicles, etc)

Power losses Wastage of electricity due

to limited power storage

Eﬃcient use of electricity

minimizes power losses

Chapter 1 Introduction 4

proposed in the previous years. For example, in [5-7], integer linear programming,

mixed integer linear programming and mixed integer non-linear programming are

used for electricity cost minimization. Similarly, in [8], authors use convex pro-

gramming for relatively large number of users to reduce their electricity bills. In

[9, 10], integer linear programming and mixed integer linear programming are used

to optimally schedule the appliances to minimize the aggregated power consump-

tion. [11] uses mixed integer linear programming to reduce both electricity bills

and aggregated power consumption. However, these techniques can not tackle

large number of diﬀerent household appliances having unpredictable, non-linear

and complex energy consumption patterns due to randomness in human behavior.

In this dissertation, heuristic optimization techniques are used due to its ex-

ceptional characteristics: ﬂexibility for speciﬁed constraints, ease of implemen-

tation, low computational complexity and low computational time [12]. Earlier

researchers demonstrated the beneﬁts of diﬀerent heuristic techniques for their de-

signed objectives. [13] uses evolutionary algorithm (a heuristic approach) to mini-

mize electricity cost for all types of sectors (residential, commercial and industrial).

We have chosen three popular heuristic optimization techniques: genetic algorithm

(GA), binary particle swarm optimization (BPSO) and ant colony optimization

(ACO) to evaluate our designed objective function due to their self-organization,

self-optimization, self- protection, self-healing and decentralized control system

[12]. In the literature, various works have been done to enhance the eﬃciency

of DSM using these heuristic optimization techniques. For example, authors in

[14, 15, 16] presents diﬀerent models to reduce utility electricity bills using GA.

Electricity cost reduction for end users is achieved in [17, 18] with particle swarm

optimization (PSO) technique. Signaled PSO (a heuristic approach) is used to

reduce aggregated load and execution time with absolute error for number of con-

sumers in [19]. An eﬃcient self-optimizing system for DSM is proposed in [20] to

optimally schedule load using ACO technique and [21, 22] investigates congestion

problem in SG through DR by applying ACO to minimize cost and maximize user

Chapter 1 Introduction 5

comfort level. Although mentioned work performed well for their designed objec-

tives but some critical issues are ignored.

To attain electricity cost minimization objective, they ignored user comfort level

and their electricity pricing model is not compatible with real scenarios and pro-

vides comprehensive analysis about the eﬀectiveness of diﬀerent heuristic algo-

rithms in home energy management. We ﬁrst design an energy management

controller (EMC) model for single and multiple homes using multiple knapsack

problem (MKP) and then apply three heuristic algorithms (GA, BPSO and ACO)

to get feasible solution for designed objective function. To calculate electricity

bills, we use time of use (TOU) tariﬀ model with inclined block rate (IBR), so

that peak formation is avoided. Performance evaluation for GA, BPSO and ACO

on the basis of designed EMC via simulations is done in terms of energy consump-

tion pattern, electricity bill, peak to average ratio (PAR), user comfort level, and

execution time.

2

RELATED WORK AND

MOTIVATION

6

Chapter 2 Related Work and Motivation 7

Many researchers around the world worked to optimally schedule smart appliances.

In this regard, some of the papers are discussed as follow:

In [23], authors investigate the problem of household appliance scheduling to en-

hance energy eﬃciency of electrical grid and provide beneﬁts to end users. They

proposed a solution that optimally schedules a set of appliances. To minimize

customer electricity bills and maintain energy consumption within a limit, they

use day-ahead variable peak pricing model and map their problem by using MKP.

By limiting the energy demand within certain capacity, problem of load shedding

can be removed. Results show that this model eﬀectively reduces utility electricity

bills while keeping power consumption within pre-deﬁned limits. Another model

of home energy management controller for residential users is proposed in [24].

Objective function is formulated by knapsack problem and dynamic programming

approach is used to solve problem and to set consumer preferences for each ap-

pliance. These priorities were the value of appliances that are used to schedule

the appliance to satisfy their operational time constraints to avoid peak formation

and to reduce electricity cost.

In [13], authors present an eﬃcient model of DSM that reduces PAR and elec-

tricity bills for residential, industrial and commercial users. Scheduling problem

is formulated as a minimization problem and then problem is evaluated by using

heuristic evolutionary approach. Heuristic algorithms show better results because

of their ﬂexible nature that allow the implementation of individual load pattern

in order to minimize inconvenience. Proposed model is beneﬁcial for both utilities

and customers in a way that PAR reduction causes minimization in the number

of peak power plants while incentive based model helps consumer to reduce their

electricity bills. Simulation results show that the proposed DSM strategy achieves

signiﬁcant savings, while reducing the peak load demand of the smart grid.

In [14], authors discuss an eﬃcient architecture for energy management system

by using home area network (HAN) for residential users. They combine real-time

pricing (RTP) tariﬀ model with the IBR because when only the RTP is adopted,

Chapter 2 Related Work and Motivation 8

there is a risk that most of the appliances operate during the hours of lowest elec-

tricity price that cause peak formation. To strengthen the stability of electricity

system, peak formation must be avoided. To solve these issues in an optimized

way, objective function is formulated. As this kind of optimization problem is non-

linear, therefore they use GA to optimize their problem. Simulation results shows

that proposed model is very eﬀective to reduce PAR and electricity cost. Another

DSM model is proposed in [15] for residential users to reduce PAR and electricity

bill minimization. GA is used to get optimal start time of each appliance in each

time slot while satisfying its operational constraints. There is a tradeoﬀ between

electricity cost and waiting time. When waiting time of an appliance is zero, its

electricity cost is increased and vice versa. Combined model of RTP with IBR is

used to avoid peak formation. Simulations are carried out for single and multiple

users. Results show the eﬀectiveness of proposed DSM model for both single and

multiple user scenarios.

An eﬃcient home energy management scheme is proposed in [17] to schedule large

number of interruptible load in the time period of 16 hours. Binary particle swarm

optimization (BPSO), which is an extended version of PSO is used to achieve the

scheduling objective. The objective is to minimize the electricity bills while satis-

fying the operational constraints and minimize the frequency to interruptions. The

eﬀectiveness of proposed approach is improved by dividing the swarm into number

of subswarms. The scheduling technique proven as useful scheme for a relatively

challenging scheduling task, and have potential advantages in scheduling widely

varied and technically complex interruptible loads. In [18], real-time model for op-

timal power usage of household appliances is proposed. BPSO algorithm is used

to solve the formulated problem by encouraging the participation of both utilities

and consumers. By considering the features of the appliances and living habits

of customers, the appliances are divided into three categories. Results show the

signiﬁcant performance of proposed scheme for load shifting, energy saving and

cost reduction.

Chapter 2 Related Work and Motivation 9

An eﬃcient heuristic approach is presented in [25] for scheduling of smart appli-

ances in residential area. The proposed algorithm is evaluated by comparing the

electricity cost and computational time with an exact algorithm. Variable energy

price model is used for scheduling of appliances. Hourly prices for electricity, the

operation start times of set of appliances are optimized to reduce cost of energy

consumption while satisfying the operational and peak power constraints. Re-

sults show that electricity cost obtained by heuristic algorithm is within 5% of the

optimal cost of exact algorithm whereas computational time is reduced by expo-

nential factor. In [26], a home energy management model is designed in which

each appliance is operated according to its schedule within predeﬁned time lim-

its. The objective is to reduce utility electricity bills while satisfying operational

constraints. GA is used to evaluate the objective function and get optimal start

operational time for each appliance to reduce electricity bill and avoid peak for-

mation. It is a comparative study in which GA based energy management model

results are compared with simulated annealing and greed method. Simulation re-

sults show that GA acts eﬃciently to reduce cost while minimizing power usage

at any instant of time than others.

An adaptive energy management model for DSM in residential area is described

in [20]. Authors aim to optimize the use of distributed RESs to reduce utility

electricity bill. They use ACO as an optimization algorithm to schedule shiftable

load and MAPE-K feedback loop as a predictive model to deal with the inter-

mittent nature of RESs. Results show that proposed cost-eﬀective self-optimizing

model is able to adapt sudden changes in environmental conditions and optimize

the power usage of residential users. Authors in [21] proposed an eﬃcient scheme

to manage congestion problem in SG through DR. Their objective is to optimally

schedule diﬀerent generation resources to minimize cost and maximize customer

satisfaction. ACO is used as optimization technique to provide beneﬁt to con-

sumers and fuzzy satisfying technique to choose the most feasible solution from

the set of pareto optimal solution. Results show that proposed scheme is eﬀective

Chapter 2 Related Work and Motivation 10

Table 2.1: Summarized related work

No. Ref. Techniques Objective (s) Achievement (s) Deﬁciency (ies)

1 [23] MKP

Electricity consumption

and bill

reduction

Eﬃciently limits power

consumption to reduce

electricity bills

User comfort level

and integration of

renewable

energy resources

2 [24]

MKP

+

Dynamic

Programming

Reduction in electricity

bills and avoid

peak formation

Sets priorities to satisfy their

operational time constraints

with user preference

Integration of renewable

energy resources

3 [13]

Heuristic

evolutionary

approach

Electricity bills

reduction for

residential, commercial and

industrial user

Beneﬁcial model for both

utilities and customers in

a way to PAR reduction and

peak load minimization

User comfort level

and integration of

renewable

energy resources

4 [14]

GA

+

RTP and IBR

Avoid peak formation

and electricity bill

minimization

Eﬀectively reduce PAR

and electricity cost

Congestion problem, user

comfort level and

integration of renewable

energy resources

5 [15]

Kp

+

GA

+

RTP and IBR

Utility electricity cost

minimization and peak

formation reduction

Eﬀective model for both

single and multiple users

Congestion problem,user

comfort level and

integration of renewable

energy resources

6 [17] BPSO

Minimize electricity

bills and frequency

to improve

Acts potentially to

achieve designed

objectives

Congestion problem,user

comfort level and

integration of renewable

energy resources

7 [18] BPSO

Energy saving and

electricity cost

reduction

Encourage utility and

consumer participation to

maintain balance

between demand and

supply

User comfort level

and integration of renewable

energy resources

8 [25]

Exact

algorithm

+

Heuristic

algorithm

Compare electricity

cost reduction

and computational

time for both algorithms

Electricity cost obtained

from heuristic algorithm is 5%

optimized than exact algorithm

User comfort level

and integration of renewable

energy resources

9 [26]

GA

+

Simulated

annealing

+

Greedy

method

Comparative study for

electricity cost reduction

GA acts eﬀectively

than other to two

algorithm

User comfort level

and integration of renewable

energy resources

10 [20]

ACO

+

MAPE-K

Deals intermittent

nature of RESs to

reduce electricity bills

Cost-eﬀective and

self-optimized model

User comfort level

and peak formation

problem

11 [21]

ACO

+

Fuzzy

techniques

Congestion problem

with electricity cost

minimization

Eﬀective model for

both generation and

demand management

Integration of renewable

energy resources

12 [22] ACO Congestion control with

cost reduction

Improved model to minimize

electricity cost

User comfort level

and integration of renewable

energy resources

both for generation selection and demand management in the most economical

way. In [22], authors investigate congestion cost model in real-time power grid

system. They built a congestion factor to control opening of both generation and

load sides. Non-linear programming is used to formulate real time congestion

problem and cost is minimized on the basis of an optimization algorithm (ACO).

Results show that improved model can signiﬁcantly minimize electricity cost. All

the techniques, objectives, achievement (s) and deﬁciency (ies) mentioned above

is summarized in table. 2.1.

Chapter 2 Related Work and Motivation 11

2.1 Motivation

In SG, optimization of energy consumption schedules and user cost minimization

are two diﬃcult tasks due to randomness in the energy consumption patterns

of end users. In literature, an eﬃcient home energy management controller to

reduce utilities electricity bills and PAR is still an issue. Mostly, user comfort

level is neglected while reducing electricity bills. Typically, the target of DSM is

to eﬃciently manage the energy schedules such that electricity price is minimized

while maximizing user comfort level. In SG, optimization problems are as follow:

•Minimize the electricity bill.

•Minimize the aggregated power consumption.

•Minimize both electricity bill and aggregated power consumption.

•Minimize PAR.

•Maximize user comfort.

•Eﬃcient integration of RESs.

Many strategies have been proposed in the previous years to eﬀectively tackle

these mentioned problems. Authors in [5-7], present three diﬀerent techniques:

integer linear programming, mixed integer linear programming and integer non-

linear programming for electricity bill reduction. However, integration of RES,

user comfort and power consumption minimization problems are ignored in these

models. Similarly, [8] uses convex programming to deal with large number of

consumer for electricity bill reduction. Results show that this technique gives ef-

fective solution however, at the cost of increased computational time. The linear

programming based scheme in [9] is eﬀective for residential areas. However, lack

of RES integration and non-adaptability with dynamic pricing model are its ma-

jor drawbacks. In [11], both electricity bill minimization and aggregated power

minimization problems are investigated using mixed integer linear programming

for dual optimization functions. Proposed scheme is implemented for single home

Chapter 2 Related Work and Motivation 12

but it does not deal with the ﬂexibility of power usage patterns and human be-

haviors. Thus, to resolve these issues in previously proposed schemes for DSM,

we proposed an eﬃcient model using MKP to formulate an objective function for

residential area. To evaluate our objective function, heuristic optimization tech-

niques are used due to its ability to deal with large and complex scenarios within

less computational time and less computational complexity [12].

We have considered three heuristic techniques: GA, BPSO and ACO to achieve

our objectives and compare their results. Despite of great eﬀorts in literature for

DSM strategy using these heuristic algorithms, there is still a room for improve-

ment to make system compatible with growing demand of power. As in [14], GA

based DSM model is presented to reduce electricity bills for residential area while

ignoring user comfort level. Authors used RTP tariﬀ model to calculate electricity

bills which is a major drawback of this model because real time data transmis-

sion causes great chance for data loss that cause discomfort both for utility and

customer. Authors in [17], proposed a model for DSM using PSO techniques to

get objective of electricity bill reduction without considering user comfort max-

imization. Whereas, in [20], authors used ACO and feedback prediction model

to enhance eﬃciency of DSM. They proposed an economical model that optimize

power usage without considering user satisfaction parameters.

We apply these heuristic optimization techniques in a novel way to enhance the

eﬃciency of DSM. In our focused scenario, household appliances are classiﬁed into

three categories and problem is formulated by using MKP. TOU tariﬀ model is

used to calculate electricity bills for end users and to get feasible solution for de-

signed objective function; we used GA, BPSO and ACO. Also our proposed model

signiﬁcantly integrates RES energy with grid power to deals with issues keeping

in view the interest of both players (utilities and consumers).

3

PROPOSED MODEL

13

Chapter 3 Proposed Model 14

In SG, DSM enables more eﬃcient and reliable grid operations. Its two main

functions are energy management and demand side control activities for end users.

In residential area, every smart home is equipped with EMCs and smart meters

to make stable and reliable bi-directional communication between utilities and

customers. All elements, such as electrical appliances, sensors, local generation

and energy storage systems (ESSs) give their information to EMC through HAN

and EMC controls scheduling of appliances. After collecting all information, EMC

sends it to SG domain through WAN. There are various wireless solutions for

communication links between the smart meters and the EMCs such as ZigBee,

Z-Wave, Wi-Fi, or a wired (HomePlug) protocol [1]. Simple architecture of DSM

is shown in ﬁg. 3.1.

Sensors

Distributed RESs

Smart devices

Residential area

domain

ESSs

EMCs

HAN

SG domain

Distribution

Operation

Market

Service provider

Customers

WAN

Two way communication

One way communication

Figure 3.1: Components of DSM

In residential area based DSM, we consider Nsmart homes and Msmart appli-

ances as shown in ﬁg. 3.2. In this model, all smart homes have smart metering

system and EMC. End users change their energy usage according to incentive

based schemes oﬀered by utilities.

Conceptual DSM diagram of our proposed scheme is shown in ﬁg. 3.3. In each

home, consumer inputs diﬀerent parameters of appliances to appliances scheduler

and then appliance manager gives signal to various appliances about their on/oﬀ

Chapter 3 Proposed Model 15

Smart

meter EMC

Smart

meter EMC Smart

meter EMC

Smart

meter EMC

Power distribu on

ulity

User N+1

User N+2 User N

User 1

Two way communica on

One way communica on

Figure 3.2: EMC model for residential users

status. For electricity pricing model, TOU tariﬀ is used to calculate electricity bill

against the energy consumption cost per day.

3.1 Energy consumption model

Let A={a1, a2, a3, . . . , am}be the set of appliances such that a1,a2,a3,···,amare

number of appliances that belong to each category. If t∈T={1,2,3,··· ,24 }

denotes the scheduling horizon, then hourly energy consumption demand of a

appliance is given as,

Ea(t) = {Ea

t1+Ea

t2+Ea

t3+. . . +Ea

t24 }(3.1)

where, Ea

t1,Ea

t2,Ea

t3,···,Ea

t24 denotes energy consumption demand of each appliance

in the respective time slots. The per day total energy consumption demand for all

Chapter 3 Proposed Model 16

Appliance 1

Appliance 2

Appliance M

User

Appliance

manager

Appliance

scheduler

Master control

TOU pricing

genera on

Aggrega on

process

Network

interface

Hourly projected

energy demand

Next day TOU

pricing

User energy

uliza on

New data pricing

Appliance

status

Appliance

schedule

Heuris c

techniques

System se ng

and parameters

Database manager

Historical appliance data

a a

a

a

a

a

User Interface

Fixed device

Shi able device

Elas c device

HAN

a

a

a

Two way communica on

One way communica on

Figure 3.3: DSM functional diagram

appliances is calculated as follows,

ET=

24

X

t=1 A

X

a=1

E(i,t)(3.2)

In order to design the optimization model for home energy management, we have

categorized the load according to the characteristic of appliances and life style of

end users as discussed in the following section.

3.2 Load categorization

We classify appliances into three categories; ﬁxed, shiftable and elastic appliances

according to their power consumption pattern and time of use [27]. Detail of all

these categories is given as follow:

Chapter 3 Proposed Model 17

3.2.1 Fixed appliances

These are also called regular appliances because their usage or length of operation

can not be modiﬁed. For example, lights, fans, clothes iron, microwave oven,

toaster, tv, etc. We represent ﬁxed appliances by Fed and its power consumption

as ν. If each ﬁxed appliance fed ∈Fed has power rating ρfed , then total power

consumption in each time slot is calculated as,

ν(t) = X

fed∈Fed

24

X

t=1

ρfed (t)×χfed (t)(3.3)

where, χfed (t) is the state of each ﬁxed appliance in particular time slot and it is

given as,

χfed (t) =

1 if appliance is ON

0 if appliance is OFF

(3.4)

3.2.2 Shiftable appliances

These are also called burst load because these are manageable and can be shifted

in time without altering their load proﬁle. For example, washing machine, dish

washer, clothes dyer, etc. We denote shiftable appliances by Sed and their power

consumption by ∆. Each shiftable appliance is characterized by its length of

operation which is denoted as τsed and it is pre-deﬁned by end users each day.

Consumers set start time and end time for each shiftable appliance as,

αsed ≤τsed ≤βsed (3.5)

where, αsed and βsed are the start and end times of a shiftable appliance that are

set by end consumer. If each shiftable appliance sed ∈Sed has power rating factor

Chapter 3 Proposed Model 18

ρsed , then the total power consumption is calculated as,

∆(t) = X

sed∈Sed

24

X

t=1

ρsed (t)×χsed (t)(3.6)

where, χsed (t) is the state of each shiftable appliance in particular time slot and it

is given as,

χsed (t) =

1 if appliance is ON

0 if appliance is OFF

(3.7)

3.2.3 Elastic appliances

These are also called interruptible appliances because these are fully controllable

in terms of both usage time and power consumption proﬁle. For example, air

conditioner, refrigerator, water heater, space heater, etc. We represent elastic ap-

pliances by Eed and its power consumption is denoted by κ. Each elastic appliance

eed ∈Eed has power rating ρeed, power quantity factor λeed, length of operation

τeed , start time αeed and end time βeed . These attributes are set by the consumer,

such that,

αeed ≤τeed ≤βeed (3.8)

Power consumption of each elastic appliance ζeed is calculated as follows,

ζeed =λeed ×ρeed ∀eed ∈Eed (3.9)

The total power consumption is calculated as,

κ(t) = X

eed∈Eed

24

X

t=1

ρeed (t)×χeed (t)(3.10)

Chapter 3 Proposed Model 19

where, λis used to vary the power quantity use in the predeﬁned range and χsed (t)

is the state of each elastic appliance in particular time slot given as,

χeed (t) =

1 if appliance is ON

0 if appliance is OFF

(3.11)

3.3 Energy price model

A number of tariﬀ models are available to deﬁne electric energy prices for a day or

for short time duration. Among these, TOU tariﬀ model is deﬁned for electricity

prices depend on the time of day and are pre-deﬁned in advance. Critical peak

pricing (CPP) is a variant of TOU in which price is considerably raised in case of

emergency situations (e.g. high demand). RTP based electricity prices can change

as often as hourly, reﬂecting the utility cost of supplying energy to customers at

that speciﬁc time. In our model, we use TOU with power dependent tariﬀ known as

inclined block tariﬀ or IBR. The energy price at time tis an increasing, piecewise,

linear function of the total energy demand. As E(t) is the total power consumption

of all appliances in a home at each time slot tand it is calculated as,

E(t) =

24

X

t=1 ν(t) + ∆(t) + κ(t)(3.12)

To calculate electricity bills, energy price for each unit consumed in each time slot

is represented by C(t) and according to IBR model, it is designed as,

C(t) =

C1(t) 0 ≤E(t)≤Eth1(t)

C2(t)Eth1(t)≤E(t)≤Eth2(t)

C3(t)Eth2(t)< E(t)

(3.13)

Chapter 3 Proposed Model 20

where, Eth1(t) and Eth2(t) are two power consumption thresholds and C1,C2and

C3are costs for these particular cases.

3.4 Local energy generation

The residential users can also use distributed RESs such as PV panels, wind tur-

bines, electric vehicles, etc. The distributed RESs are used to fulﬁll the energy

demand locally or to charge the storage devices. Assume that each home is ﬁtted

with PV panel that is capable of generating 50% of total grid power. In this case,

consumer becomes prosumer because he/she can generate its own energy. They

can also sell the generated RESs energy back to the grid depending upon their

agreement with the utility. PV panel generates solar power depending on solar

radiations and total estimated radiation varies for every month. Solar power out-

put depends on radiation amount, direction of panels and transfer eﬃciency. The

generated energy in each time slot is characterized as Ψr

tand it is calculated by

the following expression [28].

Ψr(t) = 10 ×1

√2Πσexp −(t−µ)2

2σ2(3.14)

Where, µis the mean of distribution and σis the variance. The hourly RESs

energy must be greater than zero during day time. The daily energy supply from

renewable system (PV panel) installed by the users is denoted by Θ and calculated

as,

Θ(t) =

24

X

t=1

Ψr(t) (3.15)

Let Θmax be the maximum generation capacity of PV panel then available energy

must be in following range,

0≤Ψr(t)≤Θmax ∀t∈T(3.16)

Chapter 3 Proposed Model 21

If locally generated power is greater than total power demand for appliances in each

time slot, then the total power is in negative sign which means that the generated

power can be sold back to grid or it can be stored to reduce energy usage during

peak hours. In order to be eligible for participation in some agreement with grid to

sell negative power back to grid, user must oblige to meet a speciﬁc power capacity

Ψrmin ;

Θmax(t)≥Ψrmin (t)∀t∈T(3.17)

3.5 Energy storage

When the energy generation exceeds consumption, it is stored. The stored energy

is used at high peak hours or it can be used in night, when solar energy is not

present. We model ESS with a set Band for each battery b∈B, we introduce a

binary variable χbnthat shows charging and discharging status of all batteries. A

binary variable χbnis deﬁned for each time slot,

χbn(t) =

1 Charging

0 Discharging

(3.18)

Charging and discharging rates of a battery are represented by non-negative vari-

ables as rc

bnand rd

bn, respectively. Such variables are bounded by the following

constraints [27],

rc

bn< rc,max

b×χbn∀b∈B(3.19)

rd

bn< rd,max

b×(1 −χbn)∀b∈B(3.20)

where, rc,max

band rd,max

bare maximum capacity of charging and discharging rates,

respectively. There are energy losses during charging and discharging in each

battery and its eﬃciency rate is between 0 <ℵc<1 and 0 <ℵd<1 for charging

Chapter 3 Proposed Model 22

and discharging. A binary variables shows that both these operations cannot

occur at the same time. Despite the beneﬁts of ESSs, their cost may limit their

applicability in real scenarios.

3.6 Residential users

We design our model for three types of users in residential area such as passive,

semi-active and active users. We deﬁne these categories as,

3.6.1 Passive users

They only consume electrical energy of the grid and does not generate or store

electrical energy. They can only shift there load from high peak to low peak and

reduce their electricity bills. The set of passive users is represented by P. The

energy consumption proﬁle for each user is calculated by the following equation:

Ei∈p(t) =

24

X

t=1

Ei(t) (3.21)

where, i∈Pconsumes electrical energy in time slot t.

3.6.2 Semi-active users

They have RESs such as solar panels and wind turbines. They consume energy

both from power grid and RES to reduce their electricity bills. The set of semi-

active users is represented by S. The energy consumption proﬁle for t∈Tis

calculated as,

Es∈S(t) =

24

X

t=1 Es(t)−Θs(t)(3.22)

where, s∈Sbelongs to set of semi-active users and Θs(t)is the solar panel

generated energy.

Chapter 3 Proposed Model 23

3.6.3 Active users

They take energy from RES and store it in storage devices such as batteries as

well as also take electrical energy from grid to fulﬁll their need. The set of active

users is represented by A. The energy consumption proﬁle for t∈Tis calculated

by the following equation:

Ea∈A(t) =

24

X

t=1 Ea(t)−Θa(t)±Ba(t)(3.23)

where, a∈Abelongs to set of active users and Θa(t) is the solar panel generated

energy and Ba(t) is the energy stored in batteries. The batteries are charged from

RES (not from grid). If Bais positive, it means battery is charging and if negative,

battery is discharging. The conceptual diagram of all types of users is shown in

ﬁg. 3.4.

SG

Smart

meter EMC

Grid power

+

RES

Grid power

Grid power

+ RES +

ESS

Passive user

Semi-

active

user

Active user

Two way communication

One way communication

Figure 3.4: End users classiﬁcation

3.7 Problem formulation

In this work, main objectives are to reduce consumer cost by optimizing the energy

consumption patterns of appliances to maximize the comfort level of end user

and to maintain balance between demand and supply. Here, we formulate our

Chapter 3 Proposed Model 24

scheduling problem by using MKP. MKP is a resource allocation problem that

consists of Mresources (capacities) and set of Nobjects [29]. We take jnumber

of knapsacks, and map our scheduling problem in MKP as follows:

•We consider jnumber of knapsacks as power capacities in each time slot.

•Number of appliances as number of objects.

•The weight of each object as the energy consumed by appliances in each

time slot. Note that it is independent of t.

•The value of object in a speciﬁc time slot is the cost of power consumption

of the appliance in that time slot.

•The value of binary variable χcan be 0 or 1 depending on the state of

electrical appliance.

With the help of this model customer’s electricity bill can be controlled, and for

utility side, it is also beneﬁcial because it ensures that the grid is not over stressed.

As the total power consumption for all types of appliances should not exceed

maximum power capacity in each hour denoted as γ(t), we introduce constraint

which limits the power consumption and depends on load proﬁle and its states.

Constraints show that power consumption is predeﬁned,

24

X

t=1 E(t)×χ(t)≤γ(t)(3.24)

Here, γ(t) is the power capacity in each hour that is available from grid and

χ(t)∈[0,1] denotes the states of appliances. Total power consumption in each

hour must be limited to this capacity factor.

3.8 PAR

It is beneﬁcial for the utility and consumer to reduce PAR so that power supply

and demand balance can be maintained. We have deﬁned PAR for single user as

the ratio of peak load and average load in each time slot. It is represented as φ

Chapter 3 Proposed Model 25

and its mathematical form is as follow,

φ(t) = maxa∈A(E(t))

1

TP24

t=1(E(t)(3.25)

Therefore, for nnumber of users, the PAR is written as,

φn(t) = maxa∈A(E(t, n))

1

TPN

n=1(P24

t=1 E(t, n)) (3.26)

3.9 Waiting time

It is necessary for residents to set some parameters for each shiftable and elastic

appliance. In scheduling problem, we omit ﬁxed appliances because these appli-

ances do not play any role in energy management system and must run with ﬁrst

priority. We assume start time αaand end time βafor each schedulable appli-

ance such that αa< βa. The operation time interval (OTI) for each appliance is

the time in which it performs its functionality. Let τabe the length of operation

(LOT) of an appliances that is required to complete the task. These parameters

are needed to set by the resident via user interface and then this information is

sent to EMC. Assumed that βa−αamust be greater than or equal to τa, we deﬁne

operation start time by ηa. As, we already know αa,βa,τaand χafor each ap-

pliance but ηais unknown. Once we get ηa, we can calculate power consumption

pattern. Relationship between all these parameters is shown in ﬁg. 3.5. Now for

α

β

τ

ƞ

Figure 3.5: Parameters of appliance

each appliance, there exists a group of parameters comprising the OTI, LOT, and

power consumption values per unit time. ηamust be greater than or equal to αa

Chapter 3 Proposed Model 26

and less than or equal to βa−αa. Therefore, range of ηais given by,

ηa∈[αa, βa−τa] (3.27)

The range of ηais shown in ﬁg. 3.6. Usually, residents want to ﬁnish their work

α

β

α

β

τ

τ

Range of ƞa

ƞa=βa-τa

Figure 3.6: Range of operation time

as soon as possible. Therefore user comfort depends upon waiting time reduction

and cost minimization. There is a trade oﬀ between cost and waiting time. When

we minimize cost, customer compromises on waiting time and when waiting time

reduces customer pays huge cost. Mathematically, waiting time is represented as

ϕand for each schedulable appliance and is given as,

ϕa=ηa−αa

βa−τa−αa

(3.28)

If an appliance operates at a later time, the later the appliance operates the larger

the waiting time will be. The smallest and the largest values of ϕare set between

0 and 1. Assume that for washing machine, a resident sets the parameter OTI as

[αw, βw] and LOT as τw. If it starts working at its starting time that is αwthen

its ϕwis zero and if it start working at latest time such that βw−lwthen its ϕw

would be one, as shown in ﬁg. 3.7.

Chapter 3 Proposed Model 27

α

β

α

β

τ

τ

ƞ=βa-τa

ƞa=αa

αa

βa

τa

βa-τa-αa

ƞa=αa

Figure 3.7: Waiting time

3.10 Objective function

The overall objective function of our scheduling problem is to minimize electricity

bill with optimal use of power from grid and to minimize waiting time (to avoid

frustration of end users). Additionally, optimal integration of RESs is also a key

point to reduce green house gas (GHG) emission. We formulate our objective

function as an optimization function and is modeled as,

min

24

X

t=1 a1

A

X

a=1

(∆a(t)×Υa(t))+a2ϕa(t) (3.29)

s.t:

αssd ≤τssd ≤βssd (3.29a)

αsed ≤τsed ≤βsed (3.29b)

ηa∈[αa, βa−la] (3.29c)

ϕa≤5 (3.29d)

0≤Ψr

t≤Θmax ∀t∈T(3.29e)

Chapter 3 Proposed Model 28

rc

bn< rc,max

b×χbn∀b∈B(3.29f)

rd

bn< rd,max

b×χbn∀b∈B(3.29g)

24

X

t=1

(∆a(t)×χa(t)≤γa(t)) (3.29h)

χa(t)∈[0,1] (3.29i)

where, Υais the electricity cost in each time slot that must be minimized while

keeping waiting time of shiftable appliances minimized. a1and a2are weights of

two parts of objective function and their values are a1, a2∈[0,1] or a1+a2= 1.

It shows that either a1or a2would be 0 or 1. In this work, our major concern is

with electricity cost reduction with maximizing comfort level of end users. For this

purposed model, we assume waiting time of each shiftable appliance not greater

than 5, if operation start time of an appliance is greater than our assumption then

utility pays penalty.

4

HEURISTIC ALGORITHMS

29

Chapter 4 Heuristic algorithms 30

Due to highly volatile load behavior of residential users and intermittent nature

of RESs, we consider our deﬁned problem as non-linear optimization function and

traditional optimization techniques in [5-11] can not handle the complexity of our

proposed model due to their non-ﬂexible nature. Therefore, we apply heuristic al-

gorithms (GA, BPSO and ACO) to solve our designed MKP. These algorithms are

similar due to population based search methods. They move from one population

to another population in number of iterations with improvement using a combina-

tion of deterministic and probabilistic rules. In the following sections, we discuss

some of the latest research works of GA, BPSO and ACO as an optimization

solutions.

4.1 GA

It is most suitable for complex non-linear models where location of the global opti-

mum is a diﬃcult task. Due to its probabilistic nature for development of solution,

GA does not guarantee optimality even when it may be reached [30]. As in [16],

a combined model of RESs and ESSs is proposed to minimize electricity bills for

residential, commercial and industrial areas. Authors use probabilistic model to

design their optimization function and optimal solution is obtained from GA tech-

nique. RTP is used for electricity pricing model. Results show favorable eﬀects

for designed model but in our proposed model, we used GA in more promising

manner to achieve our objectives: electricity cost reduction, PAR minimization,

maximizing user comfort and optimal integration of RESs. In comparison to [16],

we used MKP to balance the demand and supply capacity model and to optimize

our objective function, GA is used. Furthermore, combined TOU with IBR pric-

ing model is used instead of RTP because in real time pricing model, chances of

data loss are increased due to congestion problem. Another model to improve the

eﬃciency of DSM is proposed in [14]. Authors investigate electricity cost minimiza-

tion and peak formation problem to make DSM eﬃcient. They deﬁned appliance

Chapter 4 Heuristic algorithms 31

scheduling problem for power consumption as optimization problem. GA is used

to get optimal solution subject to cost minimization and PAR reduction and for

electricity pricing model, they used RTP tariﬀ model with IBR. In comparison to

[14], our proposed solution is more eﬀective due to its unique implementation. In

our model, grid capacity is predeﬁned by using MKP to maintain balance between

generation and demand curve. To give compromising results for user satisfac-

tion, we used GA in more appropriate manner. However, they use RTP with IBR

which is not suitable for electricity bill calculation due to increased information

loss chances during high data transfer rates and we use TOU tariﬀ model with

IBR in which date loss problem is diminished. Whereas, we used PV panel to deal

with greenhouse gas emission problem that they totally ignored in their proposed

scheme. Authors in [15], proposed GA based home energy management controller

for single home in residential area. RTP is used for electricity bill calculation.

Results show eﬀectiveness of this model but compared to our model, they ignore

some parameters of DSM model. Detailed GA based energy management con-

troller (GA-EMC) model is shown in algorithm 1 (refer to appendix (7.1)), which

is improved form of algorithm in [15]. Objective function (refer eq. 29) and its

constraints (refer eq. 29a to eq. 29i) are used to ﬁnd feasible solution. Users input

initial parameters (αa,βa,τaand ρa) for all appliances whereas we treat ηas vari-

able quantity. GA creates a random population initially that consists of certain

number of chromosomes represented by binary string as ON/OFF status of each

appliance. Each chromosome is evaluated using eq. 29. TOU with IBR is embed-

ded as electricity pricing scheme and PV panels model is used as distributed RES

to achieve our objectives. Key modiﬁcations that we have implemented in GA al-

gorithm [15] to achieve our objectives and its expected outcome (s) are mentioned

in table. 4.1.

Chapter 4 Heuristic algorithms 32

Table 4.1: Key points of GA-EMC

Enhancement mode Expected Outcome (s)

MKP capacity factor

(refer eq. 29a to 29i)

Limit energy consumption

within certain range

Combined model of

TOU and IBR

(steps 8, 9, ... ,14)

Reduce electricity bills

PAR reduction

Integration of RES

energy model

(steps 22, 23,..., 27)

Reduce peak power plants

Reduction in green house gas emission

Maximize consumer participation

Further minimized electricity cost

4.2 BPSO

BPSO becomes the prominent evolutionary approach to solve global optimization

problems due to its ability to handle non-diﬀerential, non-linear multimodal func-

tion, parallel behavior, ease of implementation and good convergence properties

[31]. Recently, many researches have proposed to make eﬃcient energy manage-

ment controller using BPSO. In this regard, a real-time appliance usage model is

proposed in [17]. Authors use BPSO technique to achieve their objectives; electric-

ity bill reduction, peak shaving, valley ﬁlling and demand curve smoothing. They

categorized domestic appliances on the basis of characteristics of appliances and

living habits of end users. TOU tariﬀ pricing model is considered as a billing model

for electricity cost calculation. In our work, we categorized household appliances

on the basis of time of usage and appliance power consumption patterns. Com-

bined pricing model, TOU with IBR is used to avoid peak formation and overﬂow

problem. To make system more eﬃcient, we used PV panels to reduce electricity

bills and avoid environmental pollution that is totally ignored in [17]. Similarly, in

[32], another model is proposed for energy management in DSM based on BPSO.

Main objective of this work is to minimize electricity cost for residential area by

scheduling shiftable load. They ignore user comfort level while investigating DR

program and use TOU pricing model to calculate electricity bills for end users.

Chapter 4 Heuristic algorithms 33

However, in our model, we formulate our objection function by MKP techniques

and BPSO is used to evaluate our designed optimization function. We used TOU

with IBR to avoid peak formation. Thus, our proposed model gives more signif-

icant solution for electricity bill minimization, PAR reduction and user comfort

with optimal integration of RESs. All steps of our proposed model is shown in al-

gorithm. 2 (refer appendix(7.2)). Compared to [31], we modiﬁed BPSO according

to customer needs. Feasible operation time ηis calculated by evaluating objec-

tive function (refer eq. 29) and its constraints (refer eq. 29a to eq. 29i). Each

particle in the generation is represented by a binary string denoted as states of

appliances. These particles are updated by individual velocity and particle posi-

tion as in [31]. Our proposed model is applicable for single and multiple homes in

residential areas. In table. 4.2, enhancement points and its expected outcomes for

BPSO algorithms are mentioned.

Table 4.2: Key points of BPSO-EMC

Enhancement mode Expected Outcome (s)

MKP capacity factor

(refer eq. 29a to eq. 29i)

Balance between demand

and supply

Combined model of TOU and IBR

(steps 26 to 37)

Reduce electricity bills

PAR reduction

Integration of RES energy model

(steps 39 to 44)

Reduce peak power plants

Reduction in green house gas emission

Maximize consumer participation

Further minimized electricity cost

4.3 ACO

ACO is a meta-heuristic optimization approach that is used to solve discrete com-

binatorial optimization problems. It has unique properties of self-healing, self-

protection and self-organization [20]. In literature, ACO is used for DSM in many

ways. For-example, authors in [22], investigate congestion management and cost

minimization problems. They formulate their focused problem as a non-linear

Chapter 4 Heuristic algorithms 34

Table 4.3: Key points of ACO-EMC

Enhancement mode Expected Outcome (s)

MKP capacity factor

(refer eq. 29a to eq. 29i)

Balance between demand

and supply

Model of TOU and IBR

(steps 9 to 15)

Reduce electricity bills

and PAR reduction

Integration of RES model

(steps 25 to 30)

Reduce peak power plants

Reduction in greenhouse gas emission

Maximize consumer participation

Minimized electricity cost

programming problem and electricity bill minimization is achieved using ACO. To

our knowledge, ACO implementation in residential area is not done before. In our

work, we use ACO to evaluate the designed optimization function to get optimized

schedules for home appliances. Our scheme gives novel idea to implement ACO as

optimization tool for DSM in residential area. In [33], linear programming is used

to designed the optimization function. Refer to [34], we modiﬁed its algorithm for

our designed scenario. Algorithm. 3 (refer appendix(7.3)) gives detailed view of

ACO based EMC (ACO-EMC) model. ACO is used to evaluate objective function

(refer eq. 29) and its constraints (refer eq. 29a to eq. 29i) to get feasible oper-

ational time for all appliances. Our proposed model is applicable for single and

multiple homes in residential areas. Major modiﬁcations and possible outcomes

in ACO algorithm in contrast to [34] are given in table. 4.3

5

SIMULATIONS AND RESULTS

35

Chapter 5 Simulations and results 36

To evaluate diﬀerent performance metrics of EMC, we conduct extensive simu-

lations in MATLAB. In these simulations, we compare our objectives: electricity

bill reduction, energy consumption pattern, PAR reduction, user comfort level and

optimal integration of RESs (PV panels) by using diﬀerent heuristic algorithms;

GA, BPSO and ACO. Subject to fair comparison, we used TOU tariﬀ model of

Jemena Electricity Networks (VIC) Ltd [35, 36] for residential area with IBR. Ac-

cording to this model, time horizon of 24 hours is divided into three periods as

shown in ﬁg. 5.1. Peak hours are from 3 PM to 9 PM in local time weekdays;

shoulder hours are 7 AM to 3 PM and 9 PM to 10 PM in local time weekdays

and 7 AM to 10 PM in local time weekends while oﬀ peak hours are 10 PM to 7

AM local time all days. Price rate for the peak hours, shoulder peak hours and oﬀ

peak hours are 15 cent/kWh, 9 cent/kWh and 4 cent/kWh. The purpose of using

dynamic pricing model instead of ﬁxed pricing schemes is to enable customers to

make informed decisions that can be beneﬁcial for them in terms both of electric-

ity bill reduction and comfort level. These TOU tariﬀ model prices are readily

available to customers having advance metering infrastructure.

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24

2

4

6

8

10

12

14

16

Time (hours)

Cost (cent/kWh)

TOU prices

Figure 5.1: TOU Tariﬀ Model

For simulations, we design a model for residential area in which each home is

equipped with 10 smart appliances and 4 end users. These appliances are further

characterized into ﬁxed, shiftable and elastic appliances. Diﬀerent parameters of

all appliances can be deﬁned through end users directly or can be obtained from

Chapter 5 Simulations and results 37

learning algorithms. In this work, diﬀerent parameters are obtained from users in

advance. Appliances with their parametric values that are used in simulations are

shown in table. 5.1, table. 5.2, and table. 5.3, respectively. In table. 5.1, ﬁxed ap-

pliance has only ρaparameter measured in kWh because these are non-manageable

appliances and do not play any role in load scheduling problem. Whereas, other

Table 5.1: Parameters of Fixed Appliances

Appliances ρa(kWh)

Lighting 0.6

Fans 0.75

Clothes iron 1.5

Microwave oven 1.18

Toaster 0.5

Coﬀee maker 0.8

two categories of appliances; shiftable and elastic appliances are known as schedu-

lable appliances. As, in table. 5.2, the parameters for shiftable appliances are αa,

βa,ξa,ϕaand ρaare kWh. ϕais the unique parameter in shiftable appliance

because these appliances can be interruptible during its length of use. For elastic

appliances, the parameters are αa,βaand ρain kWh are shown in table. 5.3.

Table 5.2: Parameters of Shiftable Appliances

Appliances αa

(hours)

βa

(hours)

ϕa

(hours)

ρa

(kWh)

Washing machine 8 16 5 0.78

Dish washer 7 12 5 3.60

Clothes dyer 6 18 5 4.40

Table 5.3: Parameters of Elastic Appliances

Appliances αa

(hours)

βa

(hours)

ρa

(kWh)

Air conditioner 6 24 1.44

Refrigerator 6 24 0.73

Water heater 6 24 4.45

Space heater 6 24 1.50

Chapter 5 Simulations and results 38

To evaluate the performance of GA-EMC, BPSO-EMC and ACO-EMC, it is re-

quired to set important parameters of these heuristic algorithms. Parameters of

GA-EMC, BPSO-EMC and ACO-EMC are given in table. 5.4, table. 5.5 and ta-

ble. 5.6, respectively.

Table 5.4: GA parametric list

Parameters Values

Population size 10

Selection Roulette wheel

Elite count 2

Crossover 0.8%

Mutation 0.2%

Stopping criteria Max. generation

Max. generation 800

Table 5.5: BPSO parametric list

Parameters Values

Swarm size 200

Max. velocity 4 m/s

Min. velocity -4 m/s

Local pull (c1) 2 N

Global pull (c2) 2 N

Initial momentum weight 1.0 Ns

Final momentum weight 0.4 Ns

Stopping criteria Max. iteration

Max. iteration 600

When we apply these algorithms on our designed objective function, execution

time is diﬀerent depending on some characteristics. Execution time of an algorithm

Table 5.6: ACO parametric list

Parameters Values

Ant quantity 10

Pheromone intensity factor 2

Visibility intensity factor 6

Evaporation rate 5

Trail decay factor 0.5

Stopping criteria Max. iteration

Max. iteration 600

Chapter 5 Simulations and results 39

is the time in which an algorithm completes its functionality. BPSO executes in

more time than GA and unscheduled EMC and ACO takes more time to complete

its functionality than BPSO, GA and unscheduled EMCs. Execution time for all

models is shown in table. 5.7.

Table 5.7: Execution time

Execution time Values (seconds)

Without EMC 0.0983

GA-EMC 1.0191

PSO-EMC 24.1933

ACO-EMC 77.7434

5.1 Energy consumption pattern and Electricity

bill reduction

Let the knapsack capacity of power grid is 20 kWh for each time slot per day. To

integrate distributed energy sources, we have considered PV panel as a source of

renewable energy and batteries as storage system. We use solar panel for power

generation which meets 50% of the total load demand. Each smart house is

equipped with 1-kW PV arrays. This translates 500 W to 10 kW energy gen-

eration capacities. The purpose of integrating RESs with GA-EMC, BPSO-EMC

and ACO-EMC is to reducing greenhouse gas and to give further advantage to

end users by minimizing their electricity bills. Energy consumption pattern using

GA-EMC, BPSO-EMC and ACO-EMC without or with RES is shown in ﬁg. 5.2.

It is shown in ﬁg. 5.2 that maximum energy consumption value are limited to

19.4250 kWh, 18.6750 kWh, 19.4250 kWh, 19.4250 kWh, 18.6450 kWh, 18.8250

kWh and 18.2450 kWh for without EMC, GA-EMC, BPSO-EMC, ACO-EMC,

GA-EMC (RES), BPSO-EMC (RES) and ACO-EMC (RES), respectively. It is

concluded that energy consumption pattern of all models are under predeﬁned

knapsack capacity of grid. It is important to notice that GA-EMC acts slightly

Chapter 5 Simulations and results 40

better than BPSO-EMC and ACO-EMC whereas, ACO-EMC (RES) performed

well than others by reducing maximum energy consumption value. During high

energy consumption hours, consumers use energy from RES and ESS to further

minimize utility electricity bills. Results show that electricity consumption and

loses can be further optimally reduced when consumers smartly handle their elec-

tricity usage and accomplish their energy needs by using energy in an intelligent

way. The maximum value of electricity bill in unscheduled model is 266 cent as

2 4 6 8 10 12 14 16 18 20 22 24

0

5

10

15

20

Time (hours)

Energy consumption (kWh)

Without EMC

With GA−EMC

With BPSO−EMC

With ACO−EMC

With GA−EMC(RES)

With BPSO−EMC(RES)

With ACO−EMC(RES)

Figure 5.2: Energy consumption (kWh)

shown in ﬁg. 5.3. It is reduced to 81 cent in the case of GA-EMC while it is

reduced from 266 cents to 98 cent in BPSO-EMC and to 114 cent in ACO-EMC.

During peak hours (16-22), suﬃcient electricity cost reduction is shown for all

designed models (GA-EMC, BPSO-EMC and ACO-EMC). GA-EMC acts more

eﬀectively than BPSO-EMC and ACO-EMC in achieving our designed objective

of electricity cost reduction due to its unique parameters (crossover and mutation)

and BPSO-EMC acts slightly better than ACO-EMC due to its characteristics of

local and global exploration. When we integrate RES with these models electricity

bills is further reduced due to consumer participation. GA-EMC, BPSO-EMC and

ACO-EMC with RES models show maximum values at 75 cent, 90 cent and 98

cent as in ﬁg. 5.3. Now, total electricity bill reduction per day of a single home for

all models is shown in ﬁg. 5.4. Electricity bill reduction in the case of GA-EMC,

Chapter 5 Simulations and results 41

2 4 6 8 10 12 14 16 18 20 22 24

0

50

100

150

200

250

300

Time (hours)

Electricity bill (cent)

Without EMC

With GA−EMC

With BPSO−EMC

With ACO−EMC

With GA−EMC(RES)

With BPSO−EMC(RES)

With ACO−EMC(RES)

Figure 5.3: Electricity bill (cent)

BPSO-EMC ans ACO-EMC is 48.79%, 40.43% and 28.26% respectively. This

shows that GA-EMC is more cost-eﬃcient than BPSO-EMC and ACO-EMC. The

Figure 5.4: Electricity bill reduction per day

total electricity bill reduction per day for all designed models with RES is shown

in ﬁg. 5.5. Here, it is clear that GA-EMC performs more eﬀectively than BPSO-

EMC and ACO-EMC with the integrated RES model. In the case of GA-EMC

with RES, the per day electricity bill is 752 cents, whereas, in case of BPSO-EMC

with RES, it is 940 cents and in ACO-EMC, it is 1046 cents. Therefore, electricity

bill reduction using GA-EMC with RES is 65%, BPSO-EMC with RES is 57%

and ACO-EMC with RES is 52%. All above results are for single home but what

Chapter 5 Simulations and results 42

Figure 5.5: Electricity bill per day

happens if we increase number of homes. To see the eﬀect, we have considered 50

homes for which energy consumption and electricity cost reduction are measured

in a particular day. From ﬁg. 5.6, it is veriﬁed that our designed models achieved

signiﬁcant results. As these controllers designed to optimize starting time of all

appliances while satisfying constraints of objective function in 24 hours so that

residential users gets beneﬁc by reducing their electricity bills and utilities get

advantage by keeping demand under power capacity of power grid.

5 10 15 20 25 30 35 40 45 50

500

1000

1500

2000

2500

Number of homes

Electricity bill (cent/day)

Without EMC

With GA−EMC

With BPSO−EMC

With ACO−EMC

Figure 5.6: Electricity bill (cent/day) for 50 users

Chapter 5 Simulations and results 43

5.2 PAR

Performance of all the designed models (GA-EMC, BPSO-EMC and ACO-EMC)

with respect to PAR reduction is shown in ﬁg. 5.7. It shows that PAR is signif-

icantly reduced for GA-EMC, BPSO-EMC and ACO-EMC as compared to the

unscheduled case because these are designed to avoid peak formation in any hour

of a day. Results prove that our proposed models eﬀectively tackle the peak for-

mation problem. PAR curves for GA-EMC, BPSO-EMC and ACO-EMC describe

that power consumption of appliances are optimally distributed in 24 hours with-

out creating peak in peak hours (16-22) of a day. BPSO-EMC has high PAR than

ACO-EMC and GA-EMC and GA-EMC is more eﬀective in PAR reduction due

to its ability to generate new population of more feasible solution using crossover

and mutation. Peak formation is a major drawback in traditional electric power

system as it causes customer to pay high electricity bills and utility suﬀers high

demand that causes blackout or load shedding. We have used combined model of

TOU and IBR for electricity billing to avoid peak formation via giving information

to consumers. The performance of these algorithms in our scenario is improved

due to power capacity factor that cause utilities to fulﬁll the demand of customers

and gives chance end user to reduce electricity bills.

2 4 6 8 10 12 14 16 18 20 22 24

0

50

100

150

200

250

Time (hours)

PAR curve

Without EMC

With GA−EMC

With BPSO−EMC

With ACO−EMC

Figure 5.7: PAR curve

Chapter 5 Simulations and results 44

5.3 Waiting time

User comfort is related to both electricity bill and waiting time of an appliance. In

order to achieve lower electricity bills, smart users must operate their appliances

according to optimal schedule of EMC. During scheduling horizon of shiftable

appliances, operational time is not ﬁxed due to price variation in dynamic pricing

models. Generally, it is observed that electricity cost reduction and waiting time

show inverse relationship. By applying waiting time constraints (refer eqs. 29c and

29d) on the objective function (refer eq. 29), we have enhanced the performance

of EMC in terms of user comfort and electricity bill reduction. In ﬁg. 5.8, it

is shown that electricity bill is high if rate of waiting time is zero and it is low

with increase in rate of waiting time for all models. Performance of GA-EMC

is much better than other due to minimize eﬀect of tradeoﬀ. The purpose of

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

5

10

15

20

25

30

35

Waiting time rate

Electricity cost (cent)

Without EMC

With GA−EMC

With BPSO−EMC

With ACO−EMC

Figure 5.8: Possible trade oﬀ between electricity cost and waiting time

scheduling algorithm to delay the operation of any appliance is to optimize the

system according to the designed objective function. When energy consumption

of an appliance is more than the power capacity of a particular hour or during

high peak hour, appliance scheduler shifts the appliance to another time slot. By

ignoring waiting time factor, optimized scheduling can not be achieved. On the

other hand, our proposed scheme gives an eﬀective solution. Results of all models

are signiﬁcant to achieve our goals, however, GA-EMC based model is proven

Chapter 5 Simulations and results 45

more eﬀective than the BPSO-EMC and ACO-EMC models. All the results are

summarized in table. 5.8.

Table 5.8: Summarized results

Cases Total load (kWh/day) Total cost (cent/day) PAR reduction Cost reduction (%)

without RESs

Cost reduction (%)

with RESs

Without EMC 258 2201 244.6747 - -

GA-EMC 258 1127 81.8808 48.79 65

BPSO-EMC 258 1311 95.2281 40.43 57

ACO-EMC 258 1579 127.5380 28.26 52

5.4 Parametric tuning for all models

In this section, tuning eﬀects for diﬀerent parameters of three heuristic techniques

(GA, BPSO and ACO) on our designed model is discussed in detail. Results are

summarized in table. 5.9, table. 5.10 and table. 5.11.

Table. 5.9, shows how eﬀectively power consumption, electricity bill with RES,

PAR reduction, electricity bill with RES and execution time values are changed

when parameters of GA altered. In our work, population size, maximum gener-

ation, crossover and mutation parameters are analyzed at diﬀerent values while

keeping others constant as deﬁned in table IV. We consider three values of popu-

lation size (200, 1000 and 2000), four values for maximum generation (2000, 1500,

1000 and 800) and three values for each crossover (1, 0.8 and 0.6) and mutation

(0, 0.2 and 0.4). After evaluating our designed objective function for each value

of considered parameters, we get results that are summarized here. Maximum

value of scheduled load is in the range of 12.7480 kWh to 18.6750 kWh that is less

than 19.4550 kWh which is maximum value unscheduled load and also in limits of

assumed power grid capacity of 20 kWh whereas, maximum value of electricity bill

Chapter 5 Simulations and results 46

is in the range of 55 cent to 84 cent that is optimized results than unscheduled cost

that is 266 cent. For PAR reduction, its values are change between 38.4030 and

215.6530. In the way, electricity cost per day without using RES is 1.125×103cent

that is reduced cost than 2.201×103cent and when we integrate RES with our de-

signed model than electricity cost reduction is between 5.57×102cent and 1.05×102

cent. Execution time is greatly eﬀected when population size and maximum gener-

ation is changed. As, it slightly increase with increase in number of population size

and generation. It is clearly notice from our tables that performance of GA-EMC

is more signiﬁcant than other two models due to its evolutionary nature. As in

GA-EMC, population is randomly generated depending upon nature of problem

whereas, selection is done using designed objective function and roulette wheel cri-

teria, moreover, crossover and mutation plays key role to generate new population

which is ﬁtter than the older one. Thus, GA-EMC gives more optimized results

than other two model within less time. Performance evaluation for diﬀerent

parameters of BPSO-EMC is summarized in table. 5.10. For analy-

sis, we consider three diﬀerent values for swarm size (10, 20 and 40),

three values for maximum iteration (1800, 1500 and 600) and four

values for each local pull factor c1(0, 1, 2 and 3) and global pull

factor c2(4, 3, 2 and 1) and remaining parameters are mentioned in

table V. It is proven from results that BPSO-EMC performs well but

not as good as GA-EMC. Maximum optimized energy consumption

values for BPSO-EMC model are between 15.1543 kWh and 19.3106

kWh that obey power capacity of grid whereas, maximum value for

Chapter 5 Simulations and results 47

Table 5.9: Parameter Evaluation for GA

Pop.

size Max. gen Cros.

(%)

Mut.

(%)

Max. energy

(kWh)

Max. cost

(cent) Max. PAR Cost (cent/day)

without RES

Cost (cent/day)

with RES

Exe. time

(sec.)

200

2000

1500

1000

800

1

0.8

0.6

1

0.8

0.6

1

0.8

0.6

1

0.8

0.6

0

0.2

0.4

0

0.2

0.4

0

0.2

0.4

0

0.2

0.4

13.0650

17.1450

18.6450

17.1450

18.6450

17.1450

17.1450

17.1450

18.6450

12.7450

18.6750

18.6450

57

74

81

74

81

74

74

74

81

55

82

81

44.9548

85.9937

104.1863

81.5619

81.7315

115.0993

58.9935

71.0060

178.6347

38.4030

81.8808

81.4787

1.127 ×103

1.127 ×103

1.127 ×103

1.127 ×103

1.127 ×103

1.127 ×103

1.127 ×103

1.127 ×103

1.127 ×103

1.127 ×103

1.127 ×103

1.127 ×103

1.051 ×103

7.52 ×102

8.02 ×102

6.03 ×102

9.65 ×102

6.05 ×102

7.529 ×103

7.529 ×102

8.83 ×102

6.26 ×102

7.529 ×102

7.20 ×102

2.4326

2.3215

2.2022

1.8583

1.7596

2.0181

1.2520

1.2366

1.2059

1.0403

1.0191

1.0019

1000

2000

1500

1000

800

1

0.8

0.6

1

0.8

0.6

1

0.8

0.6

1

0.8

0.6

0

0.2

0.4

0

0.2

0.4

0

0.2

0.4

0

0.2

0.4

17.1450

18.6450

18.6450

17.1450

18.6450

18.6450

18.6450

18.6450

18.6450

17.1450

18.6450

18.6450

75

84

81

81

75

81

81

81

81

75

81

81

82.0499

215.6530

158.4000

66.4213

136.2192

124.1276

81.7315

110.8996

158.4000

75.1562

94.7048

158.4000

1.127 ×103

1.127 ×103

1.127 ×103

1.127 ×103

1.127 ×103

1.127 ×103

1.127 ×103

1.127 ×103

1.127 ×103

1.127 ×103

1.127 ×103

1.127 ×103

7.53 ×102

6.28 ×102

7.17 ×102

7.20 ×102

6.78 ×102

5.57 ×102

8.83 ×102

6.39 ×102

7.06 ×102

9.03 ×102

9.65 ×102

8.87 ×102

3.7997

3.6981

3.0158

2.2721

2.1938

2.3219

1.5643

1.5487

1.4883

1.3744

1.3384

1.3195

2000

2000

1500

1000

800

1

0.8

0.6

1

0.8

0.6

1

0.8

0.6

1

0.8

0.6

0

0.2

0.4

0

0.2

0.4

0

0.2

0.4

0

0.2

0.4

18.6450

17.9250

18.6450

17.9250

18.6450

18.6450

17.1450

18.6450

18.6450

17.1450

18.8450

18.6450

81

78

81

78

81

81

74

81

81

75

81.48

81.48

81.7315

120.3357

156.7356

90.4732

137.4787

212.5796

94.4845

110.8936

136.2192

74.3415

93.5172

127.3058

1.127 ×103

1.127 ×103

1.127 ×103

1.127 ×103

1.127 ×103

1.127 ×103

1.127 ×103

1.127 ×103

1.127 ×103

1.127 ×103

1.127 ×103

1.127 ×103

7.20 ×102

7.29 ×102

6.35 ×102

8.14 ×102

8.02 ×102

8.02 ×102

7.53 ×102

6.39 ×102

6.25 ×102

6.78 ×102

8.84 ×102

5.62 ×102

3.2209

3.1092

3.0691

2.6474

2.6172

3.0554

2.1044

2.0605

2.0118

1.8535

1.8292

1.7825

cost reduction is in the range of 95 cent to 113 cent for diﬀerent values

of parameters. These variations are due to local pull and global pull

parameter that directly aﬀects optimization phenomena to evaluate

objective function. Similarly, in the case of PAR reduction in which

values are reduced as compared to unscheduled model. Electricity

cost reduction for a particular day without RESs is 1.311 ×103cent

and with RES it is furthered. Execution time is highly increases with

increase in the number of swarm particles due to execution of step in

BPSO for each individual. As, GA is diﬀerent from BPSO and ACO

Chapter 5 Simulations and results 48

Table 5.10: Parameter Evaluation for BPSO

Swarm.

size

Max.

iter

Loc. pull

(c1)

glob. pull

(c2)

Max. energy

(kWh)

Max. cost

(cent) Max. PAR Cost (cent/day)

without RES

Cost (cent/day)

with RES

Exe. time

(sec.)

10

1800

1200

600

0

1

2

3

0

1

2

3

0

1

2

3

4

3

2

1

4

3

2

1

4

3

2

1

18.5684

17.2285

17.2783

17.9422

17.5230

19.3106

15.1543

18.4918

18.8250

19.1240

18.6250

18.8250

105

105

103

105

101

99

96

101

105

101

99

105

95.6426

95.1082

135.1126

110.824

136.4045

158.1255

180.9130

105.5971

133.3874

201.4372

95.2281

186.9246

1.311 ×103

1.311 ×103

1.311 ×103

1.311 ×103

1.311 ×103

1.311 ×103

1.311 ×103

1.311 ×103

1.311 ×103

1.311 ×103

1.311 ×103

1.311 ×103

8.45 ×102

9.40 ×102

9.41 ×102

8.67 ×102

9.39 ×102

8.52 ×102

1.03 ×102

9.40 ×102

8.56 ×102

1.037 ×103

9.40 ×102

9.33 ×102

47.6986

47.8157

48.4773

48.1125

47.1692

47.1188

57.7714

48.0297

24.0676

27.5235

24.1933

24.1033

20

1800

1200

600

0

1

2

3

0

1

2

3

0

1

2

3

4

3

2

1

4

3

2

1

4

3

2

1

17.4665

18.6450

17.7227

16.8228

18.6257

19.1387

16.8228

16.4068

18.4999

18.7275

17.2525

17.3655

101

101

96

96

104

111

101

101

101

100

96

95

120.5838

209.2746

190.8179

181.8585

203.6886

149.9150

122.7205

216.3967

131.4721

205.5694

182.0045

131.0050

1.311 ×103

1.311 ×103

1.311 ×103

1.311 ×103

1.311 ×103

1.311 ×103

1.311 ×103

1.311 ×103

1.311 ×103

1.311 ×103

1.311 ×103

1.311 ×103

9.41 ×102

8.55 ×102

9.42 ×102

9.39 ×102

7.77 ×102

7.72 ×102

8.36 ×102

9.35 ×102

8.62 ×102

7.77 ×102

8.57 ×102

9.36 ×102

142.0157

144.9287

143.9084

143.2447

96.9877

96.5602

97.2115

98.1963

45.6568

44.9692

45.8085

45.1651

40

1800

1200

600

0

1

2

3

0

1

2

3

0

1

2

3

4

3

2

1

4

3

2

1

4

3

2

1

17.4665

18.3676

17.7227

18.9652

17.6810

19.1658

17.5402

18.3967

18.4999

18.7275

17.2525

16.3655

111

102

104

113

111

111

113

102

96

111

113

111

156.9369

99.0604

166.8758

154.7958

100.4732

97.4787

100.5796

186.1581

150.8515

114.8936

156.8815

186.3058

1.311 ×103

1.311 ×103

1.311 ×103

1.311 ×103

1.311 ×103

1.311 ×103

1.311 ×103

1.311 ×103

1.311 ×103

1.311 ×103

1.311 ×103

1.311 ×103

7.41 ×102

6.10 ×102

6.08 ×102

8.54 ×102

6.10 ×102

6.43 ×102

6.57 ×102

8.50 ×102

6.06 ×102

5.24 ×102

6.89 ×102

8.57 ×102

287.0919

306.9691

286.2626

472.8304

289.2429

439.8685

286.8129

196.6268

95.8205

95.7764

95.2590

95.5248

due to its unique parameters (crossover and mutation) due to which

it complete its functionality within less time than BPSO and ACO.

While in BPSO, local and global exploration enables the algorithm

(refer appendix (b)) to give feasible solution while obeying deﬁned

constraints (eq. 29a to eq. 29i). Both global exploration and local

exploration is aﬀected by updating particle’s velocity and position.

From table. 5.11, simulation results for tuning of diﬀerent parame-

ters are shown. In the case of ACO-EMC model, we have considered

two values of ant population (10 and 20), three number of iterations

(2000, 1500 and 600), three values for visibility intensity factor (6,

Chapter 5 Simulations and results 49

10 and 15) and two trial decay values (0.5 and 1). ACO-EMC acts

better than unscheduled models but its performance is not as good

as other two models (GA-EMC and BPSO-EMC). Maximum opti-

mized value after evaluating our objective function is in the range

of 17.5064 kWh to 19.4250 kWh that is under predeﬁned power grid

capacity and electricity bill reduction is between 111 cent and 127

cent that is much less than 266.3492 cent for the cost of unscheduled

model. PAR reduction is in the range of 95.5807 to 215.6530. In the

case of ACO-EMC, electricity bill reduction is 1.58 ×103cent and

with RES it is more reduced to 1.00 ×102cent. Execution time of

ACO-EMC is very high than others due to its pheromone update for

each ant and number of step for designed algorithm (refer appendix

(c)) is repeated till convergence to feasible solution.

Chapter 5 Simulations and results 50

Table 5.11: Parameter Evaluation for ACO

Ant

pop.

Max.

iter.

Vis. int.

(β1)

Tra. dec.

factor

Max. energy

(kWh)

Max. cost

(kWh) Max. PAR Cost (cent/day)

without RES

Cost (cent/day)

with RES

Exe. time

(sec.)

10

2000

1500

600

6

10

15

6

10

15

6

10

15

0.5

1

0.5

1

0.5

1

0.5

1

0.5

1

0.5

1

0.5

1

0.5

1

0.5

1

19.4250

18.6450

19.4250

18.6450

18.3640

18.6450

17.9250

19.4250

18.6450

19.4250

17.9850

19.4250

18.9750

18.2450

18.2450

19.4250

18.2450

18.6750

127

114

119

122

114

114

117

127

122

127

118

127

114

115

112

119

106

115

122.3898

137.3747

132.6316

127.6530

95.5807

127.5206

122.3898

132.6316

215.6530

132.6316

122.7994

132.6316

127.5380

124.5747

127.5253

132.6312

215.6530

127.5655

1.578 ×103

1.578 ×103

1.578 ×103

1.578 ×103

1.578 ×103

1.578 ×103

1.578 ×103

1.578 ×103

1.578 ×103

1.578 ×103

1.578 ×103

1.578 ×103

1.578 ×103

1.578 ×103

1.578 ×103

1.578 ×103

1.578 ×103

1.578 ×103

1.414 ×103

1.465 ×103

9.64 ×102

9.99 ×102

1.464 ×103

9.90 ×102

9.26 ×102

6.67 ×102

9.13 ×102

1.025 ×103

8.266 ×102

8.99 ×102

1.175 ×103

6.62 ×102

1.046 ×103

7.01 ×102

6.59 ×102

9.44 ×102

363.8474

282.1364

311.5912

253.6215

292.9011

283.5336

208.0962

201.0578

211.9933

203.0963

220.3764

211.6243

77.4729

111.5901

78.5933

75.3730

77.7434

76.6911

20

2000

1500

600

6

10

15

6

10

15

6

10

15

0.5

1

0.5

1

0.5

1

0.5

1

0.5

1

0.5

1

0.5

1

0.5

1

0.5

1

19.4250

19.4250

18.6450

17.5046

18.6450

19.4250

19.4250

18.6450

19.4250

17.5046

18.6450

19.4250

18.6450

19.4250

18.6450

17.5064

18.6450

19.4250

119

127

122

130

128

127

119

122

111

132

127

111

122

119

104

132

122

132

132.6316

132.6316

215.6341

127.3058

211.0265

132.6316

132.6316

127.3018

132.6316

127.3065

211.0265

132.6316

161.8373

132.6316

211.6541

205.1154

127.5107

211.6541

1.579 ×103

1.579 ×103

1.579 ×103

1.579 ×103

1.579 ×103

1.579 ×103

1.579 ×103

1.579 ×103

1.579 ×103

1.579 ×103

1.579 ×103

1.579 ×103

1.579 ×103

1.579 ×103

1.579 ×103

1.579 ×103

1.579 ×103

1.579 ×103

6.10 ×102

5.58 ×102

6.58 ×102

7.65 ×102

7.07 ×102

6.45 ×102

7.00 ×102

7.82 ×102

1.14 ×102

7.65 ×102

7.65 ×102

7.85 ×102

6.64 ×102

6.72 ×102

9.49 ×102

9.55 ×102

7.38 ×102

6.64 ×102

1582.6952

1434.3018

1525.2417

1404.2934

1621.2112

1455.0140

1125.8606

1387.2139

1185.5512

1125.3564

1303.6541

1103.6541

622.5746

521.5444

419.7896

411.8112

417.8833

401.9631

6

CONCLUSION AND FUTURE

WORK

51

Chapter 6 Conclusion and Future Work 52

In this dissertation, we have presented an eﬃcient DSM model for

residential energy management system in order to avoid peak for-

mations while decreasing the utilities electricity bill by preserving

user comfort level within acceptable limits. We evaluated our de-

signed objective function by using three heuristic algorithms (GA,

BPSO and ACO) and used combine pricing models, TOU tariﬀ and

IBR model for electricity bill calculation. From the results, it is

clearly justiﬁed that our proposed model works more eﬃciently with

GA-EMC than BPSO-EMC and ACO-EMC in term of electricity

bill reduction, minimizing PAR while considering user satisfaction.

GA-EMC executes with less execution time than others as GA-

EMC>BPSO-EMC>ACO-EMC. Additionally, results of extended

model for multiple users are shown to validate our work in terms of

relative scalability.

In future, we will focus on human behavior to achieve comfort level

of consumer and to minimize frustration cost and improve security

and privacy issues between end user and utility. We will also work on

the diﬀerent optimization methods so that more accurate data trans-

formation is achieved with in less execution time and computational

complexity.

Appendix

Appendix (a): GA-EMC

Algorithm 1 GA-EMC algorithm

1: Initialize all parameters (αa,βa,τa,ρa)

2: For all users n ∈N do

3: For all appliances a ∈A do

4: For all time slots t ∈T do

5: Randomly generate population represents the patterns of appliances.

6: for p=1:popsize do

7: Each individual evaluate the ﬁtness function using eq. 29

8: F=fitnessf unction

9: if (F(p)< F (p−1))&&(E(t)< γ(t)) then

10: F(p) = F(p)

11: if ϕausing(28) ≥5&&(η≤τa)then

12: start an appliance else wait till low peak hours

13: else

14: F(p) = F(p−1)

15: Pattern(1,:)=popnew(1,p)

16: if AppP atterna== 1 then

17: τa=τa−1

Generate new population Select pair a, b by Roulette selection criteria

18: if Pc> rand then

19: crossover(a, b)

20:

21: if Pm> rand then

22: mutate(a, b)

23: popnew(popsize,N) Repeat until stopping criteria

24: if E(t) is high then

25: Θ(t) energy used

26: else

27: E(t)

28: Return best individuals

Appendix

Appendix (b): BPSO-EMC

Algorithm 2 BPSO-EMC algorithm

1: Initialize all parameters (αa,βa,τa,ρa)

2: For all users n ∈N do

3: For all appliances a ∈A do

4: Initialize particles velocities to 0 and individual best to current population

5: for S=1:swarmsize do

6: Each particle evaluate the ﬁtness function using eq. 3.29 F=fitnessf unction

7: if (F(S)< F (S−1))&&(E(t)< γ(t)) then

8: Fbest =F(S)

9: if ϕausing(28) ≥5&&(η≤τa)then

10: start an appliance

11: else

12: wait till low peak hours

13: else

14: F(S) = F(S−1)

15: Pattern1(1,:)=swarmpop(1,S)

16: if AppP atterni== 1 then

17: τa=τa−1

18: Initialize global best to Fbest

19: Each individual update velocity and position of each particle refer [31]

20: For Each bit

21: if rand < 1

1+e−vthen

22: bit ←1

23: else

24: bit ←0

Repeat until maximum iteration reached

25: if E(t)is high then

26: Θ(t) energy is used

27: else

28: E(t) used

Appendix

Appendix (c): ACO-EMC

Algorithm 3 ACO-EMC algorithm

1: Initialize all parameters (αa,βa,τa,ρa) For all users n ∈N do For all appliances

a∈A do For all time slots t ∈T do

2: for An=1:antpop do

3: Each ant evaluate the ﬁtness function using eq. 29

4: F=fitnessf unction

5: if (F(An)< F (An −1))&&(E(t)< γ(t)) then

6: Fbest =F(An)

7: if ϕausing(28) ≥5&&(η≤τa)then

8: start an appliance

9: else

10: wait till low peak hours

11: else

12: F(An) = F(An −1)

13: Pattern2(1,:)=antpop(1,An)

14: if P attern2An== 1 then

15: τa=τa−1

16: Local update pheromone for each ant refer [34]

17: Choose best solution so far

18: Global update pheromone for each ant refer [34]

19: Repeat until maximum iteration reached

20:

21: if E(t) is high then

22: Θ(t) energy is used

23: else

24: E(t) used

BIBLIOGRAPHY

[1] V. Gungor, D. Sahin, T. Kocak, S. Ergut, C. buccella, C. Cecati

and G. Hancke, “Smart Grid Technologies: Communication Tech-

nologies and Standards”, Industrial Informatics, IEEE Transac-

tion, Vol. 7, No. 4, pp. 529-539, Nov. 2011.

[2] M. Hashmi, S. Hanninen, and K. Maki, “Survey of Smart

Grid Concepts, Architectures and Technological Demonstrations

Worldwide”, IEEE PES Conference, Medellin, Oct. 2011.

[3] L. Gelazanskas, and K. A. A. Gamage, “Demand Side Manage-

ment in Smart Grid: A Eeview and Proposals Forfuture Direc-

tion”, Sustainable Cities and Society, Vol. 11, pp. 22–30, Feb.

2014.

[4] P. Siano, “Demand Response and Smart Grids — A Survey”, Re-

newable and Sustainable Energy Reviews, Vol. 30, pp. 461–478,

Feb. 2014.

[5] A. Molderink, V. Bakker, M. G. Bosman, J. L. Hurink, and G.

56

BIBLIOGRAPHY 57

J. Smit, “Domestic Energy Management Methodology for Opti-

mizing Eﬃciency in Smart Grids”, IEEE PowerTech, Bucharest,

pp. 1-7, 28 Jun(2008) - 2 Jul (2009)

[6] T. Sousa, H. Morais, Z. Vale, P. Faria, and J. Soares, “Intelligent

Energy Resource Management Considering Vehicle-to-Grid: A

Simulated Annealing Approach”, IEEE Transaction Smart Grid,

Vol. 3, No. 1, pp. 535-542, Mar. 2012.

[7] J. Soares, T. Sousa, H. Morais, Z. Vale, and P. Faria, “An Op-

timal Scheduling Problem in Distribution Networks Considering

V2G”, in Proc. IEEE CIASG, Paris, 11-15 Apr. 2011.

[8] K. M. Tsui and S. C. Chan, “Demand Response Optimization

for Smart Home Scheduling Under Real-Time Pricing”, IEEE

Transaction Smart Grid, Vol. 3, No. 4, pp. 1812-1821, Dec. 2012.

[9] Z. Zhu, J. Tang, S. Lambotharan, W. H. Chin, and Z. Fan, “An

Integer Linear Programming Based Optimization for Home De-

mand Side Management in Smart Grid”, in Proc. IEEE PES

ISGT, Washington, DC, 16-20 Jan. 2012.

[10] P. O. Kriett and M. Salani, “Optimal Control of a Residential

Microgrid”,8th World Energy System Conference, Vol. 82, No. 1,

pp. 321-330, Jun. 2012.

BIBLIOGRAPHY 58

[11] J. Wang, Z. Sun, Y. Zhou, and J. Dai, “Optimal Dispatching

Model of Smart Home Energy Management System”, in Proc.

IEEE PES ISGT ASIA, Tianjin, 21-24 May 2012.

[12] D. Maringer, “Portfolio Management with Heuristic Optimiza-

tion”, Springer, 2004.

[13] T. Logenthiran, D. Srinivasan, and T. Z. Shun, “Demand Side

Management in Smart Grid using Heuristic Optimization”, IEEE

Transaction Smart Grid, Vol. 3, No. 3, pp. 1244-1252, Sept. 2012.

[14] Z. Zhao, W. C. Lee, Y. Shin and K. Song, “An Optimal Power

Scheduling Method for Demand Response in Home Energy Man-

agement System”, IEEE Transaction Smart Grid, Vol. 4, No. 3,

pp. 1390-1400, Sept 2013.

[15] M. A. Khan, N. Javaid, A. Mahmood, Z. A. Khan and N. Al-

rajeh, “A Generic Demand-Side Management Model for Smart

Grid”, International Journal of Energy Research, Vol. 39, No. 7,

pp. 954–964, Jun. 2015.

[16] A. Arabali, M. Ghofrani, M. Etezadi-Amoli, M. S. Fadali, and

Y. Baghzouz, “Genetic-Algorithm-Based Optimization Approach

for Energy Management”,IEEE Transaction on Power Delivery,

Vol. 28, No. 1, Jan 2013.

BIBLIOGRAPHY 59

[17] Y. Zhou, Y. Chen, G. Xu and Q. Zhang, “Home Energy Man-

agement with PSO in Smart Grid”,Industrial Electronics (ISIE),

IEEE 23rd International Symposium on, Istanbul, pp. 1666 -

1670, Jun. 2014.

[18] M. A. A. Pedrasa, T. D. Spooner and I. F. MacGill, “Scheduling

of Demand Side Resources Using Binary Particle Swarm Opti-

mization”, IEEE Transactions on Power Systems, Vol. 24, No. 3,

pp. 1173 - 1181, Aug. 2009.

[19] J. Soares, M. Silva, T. Sousa, Z. Vale, and H. Morais, “Dis-

tributed Energy Resource Short-Term Scheduling using Signaled

Particle Swarm Optimization”, 8th World Energy System Con-

ference, Vol. 42, No. 1, pp. 466-476, Jun. 2012.

[20] T. Dethlefs, T. Preisler and W. Renz, “Ant-Colony based

Self-Optimization for Demand-Side-Management”, Conference:

SmartER Europe, Essen, Feb. 2015.

[21] J. Hazra, K. Das and D. P. Seetharam, “Smart Grid Congestion

Management through Demand Response”, 2012 IEEE Third In-

ternational Conference on Smart Grid Communications (Smart-

GridComm), Tainan, pp. 109 - 114, 5 - 8 Nov. 2012.

[22] B. Liu, J. Kang, N. Jiang and Y. Jing, “Cost Control of

the Transmission Congestion Management in Electricity Systems

BIBLIOGRAPHY 60

based on Ant Colony Algorithm”, Energy and Power Engineer-

ing, Vol. 3, No. 1, pp. 17 - 23, Feb. 2011.

[23] N. Kumaraguruparan, H. Sivaramakrishnan and S. Sapatnekar,

“Residential task scheduling under dynamic pricing using the

multiple knapsack method”, IEEE Conference, Washington, DC,

pp. 1-6, Jan. 2012.

[24] O. A. Sianaki, O. Hussian and A.R. Tabesh, “A Knapsack Prob-

lem Approach for Achieving Eﬃcient Energy Consumption in

Smart Grid for End-user Life Style”,IEEE conference, Waltham,

MA, Sept. 2010.

[25] C. Ogwumike, M. Short and M. Denai, “Near-Optimal Schedul-

ing of Residential Smart Home Appliances using Heuristic Ap-

proach”, IEEE International Conference on Industrial Technol-

ogy, Seville, Spain, 2015.

[26] H. Miao, X. Huang and G. Chen, “A Genetic Evolutionary Task

Scheduling Method for Energy Eﬃciency in Smart Homes”, In-

ternational Journal of Electrical Engineering, Vol. 7, No. 5, pp.

1827-6660, 2012.

[27] A. Barbato and A. Capone, “Optimization Models and Methods

for Demand-Side Management of Residential Users: A Survey”,

Sept. 2014

BIBLIOGRAPHY 61

[28] C. Monteiro, T. Santos, L. A. F. Jimenez, I. J. R. Rosado and

M. S. T. Olarte, “Short-Term Power Forecasting Model for Pho-

tovoltaic Plants based on Historical Similarity”, Energies, May

2013.

[29] H. Kellerer, U. Pferschy, and D. Pisinger, “Knapsack Problems”.

Springer, 2004.

[30] S. Mardle and S. Pascoe, “An Overview of Genetic Algorithms

for the Solution of Optimisation Problems”, CHEER, Vol. 13,

1999.

[31] H. M. Lugo-Cordero, A. Fuentes-Rivera, R. K. Guha, and E.

I. Ortiz-Rivera, “Particle Swarm Optimization for Load Balanc-

ing in Green Smart Homes”, Evolutionary Computation (CEC),

IEEE Congress , pp. 715 - 720, 2011

[32] T. Remani, E.A. Jasmin and I. T. P. Ahamed, “Load Scheduling

with Maximum Demand Using Binary Particle Swarm Optimiza-

tion”, IEEE International Conference on Technological Advance-

ments in Power and Energy, Kollam, 24-26 Jun. 2015.

[33] I. K. Yu and Y. H song, “A Novel Short-term Generation

Scheduling Technique of Thermal Units using Ant Colony Search

Algorithms”, International Journal of Electrical Power and En-

ergy Systems, Vol. 23, No. 6, pp. 471–479, Aug. 2001.

BIBLIOGRAPHY 62

[34] S. Rahim, S. A. Khan, N. Javaid, N. Shaheen, Z. Iqbal and

G. Rehman, “Towards Multiple Knapsack Problem Approach

for Home Energy Management in Smart Grid”, 8th IEEE In-

ternational Conference on Network-Based Information Systems,

Taipei, Taiwan, 2015.

[35] “Jemena Electricity Networks (VIC) Ltd - Network Tariﬀs For

The 2015 Calendar Year (Exclusive of GST)”