Edith Cowan University

Copyright Warning

You may print or download ONE copy of this document for the purpose
of your own research or study.

The University does not authorize you to copy, communicate or

otherwise make available electronically to any other person any
copyright material contained on this site.

You are reminded of the following:

 Copyright owners are entitled to take legal action against persons

who infringe their copyright.
 A reproduction of material that is protected by copyright may be a
copyright infringement.
 A court may impose penalties and award damages in relation to
offences and infringements relating to copyright material. Higher
penalties may apply, and higher damages may be awarded, for
offences and infringements involving the conversion of material
into digital or electronic form.
Optimal Power Generation in Microgrids using

Agent-Based Technology

by
Thair S MAHMOUD
BSc(Control Systems Engineering); MSc(Control and Automation Engineering)

This thesis is presented in fullment of the requirements for the degree of

Doctor of Philosophy

SCHOOL OF ENGINEERING
FACULTY OF COMPUTING, HEALTH AND SCIENCE
EDITH COWAN UNIVERSITY

August 9, 2013
2
USE OF THESIS

This copy is the property of Edith Cowan University. However, the literary rights of the
author must also be respected. If any passage from this thesis is quoted or closely para-
phrased in a paper or written work prepared by the user, the source of the passage must be
acknowledged in the work. If the user desires to publish a paper or written work containing
passages copied or closely paraphrased from this thesis, which passages would in total con-
stitute an infringing copy for the purpose of the Copyright Act, he or she must rst obtain
the written permission of the author to do so.

i
Abstract

The existing power grids that form the basis of the respective electrical power infrastruc-

tures for various states and nations around the world, are expected to undergo a period of

rapid change in the near future. The key element driving this change is the emergence of

the Smartgrid. The Smartgrid paradigm represents a transition towards an intelligent,

digitally enhanced, two-way power delivery grid. The aim of the Smartgrid is to promote

and enhance the ecient management and operation of the power generation and delivery

facilities, by incorporating advanced communications, information technology, automation,

and control methodologies into the power grid proper. Smartgrid's are currently an active
topic for research, where the research is strongly focused on developing new technologies

such as: demand response, power generation management, pricing modelling and energy

markets participation, power quality, and self-healing scenarios. In recent times, in both the

United States of America and Europe, many new projects have begun which are specically

directed towards developing  Smartgrid technologies. In Australia, the Federal Govern-

ment has recently initiated funding plans to promote the commercialisation of renewable

energy. In order to exploit these developments, Edith Cowan University (ECU); which is

a High Voltage (HV) customer for the major utility network of Western Australia, and

which owns its own transformers and Low Voltage (LV) network; is planning to integrate

renewable energy suppliers within its LV network.

The aim of this research is to introduce a smart decision making system, which can

manage the operation of disparate power generation sources installed on a LV network (mi-

crogrid); such as that owned by ECU on its campuses. The proposed energy management

system is to gather data in real-time, and it must be capable of anticipating and optimis-

ing energy needs for each operational scenario that the microgrid might be expected to

experience. The system must take into account risk levels, while systematically favouring

low economic and environmental costs. A management system application, based on au-

tonomous and distributed controllers, is investigated in a virtual environment. The virtual

environment being a full-scale simulation of ECU's microgrid; with solar panels, wind tur-

ii
bines, storage devices, gas gen-sets, and utility supply. Hence the simulation studies were

conducted on the basis of realistic demand trends and weather conditions data.

The major factors for reducing the cost of generation in the case study, were identied

as being: 1) demand forecasting; 2) generation scheduling; 3) markets participation; and

4) autonomous strategies conguration, which is required to cope with the unpredictable

operation scenarios in LV networks. Due to the high uncertainty inherent within the oper-

ational scenarios; an Articial Intelligence (AI) deployment for managing the distributed

sub-systems was identied as being an ideal mechanism for achieving the above mentioned

objectives. Consequently it is proposed that Multi-Agent System (MAS) technology be

deployed, to enable the system to respond dynamically to the unpredictable operational

conditions by updating the method of analysis. The proposed system is to behave in

a strategic manner when dealing with the expected operational scenarios, by aiming to

achieve the lowest possible cost of power generation for the microgrid. The simulated

system is based on realistic operational scenarios, which have been scaled to suit the size

and type of load in the case study. The distributed intelligent modules have proven to

be successful in achieving the potential benets of the dynamic operational conditions, by

minimising the cost of power generation.

The distributed intelligent modules, which form the basis of the proposed management

systems, have been designed to perform the following functions:

1. Provide accurate demand forecasts through the utilisation of an AI-based adaptive

demand forecasting model. The novel demand-forecast modelling technique, which was

introduced to model demand in the case study, has been utilised to supply reasonably

accurate demand forecasts to other stages of processing in the management system. The

forecasts are generated from this model, by monitoring and controlling the forecasting error

to ensure consistent and satisfactory forecasts.

2. Make optimum decisions concerning the operation of the power generators by con-

sidering the economic and the environmental costs. In order to deal with the complexity

of the operational conditions, a smart and adaptive generation scheduling method was

implemented for the case study. The method was primarily applied to control the charg-

ing/discharging process of the Storage Devices (SDs) among the other generators. The

proposed method aims at controlling the resources, and extracting the benet of having

an hourly based variable generation cost.

3. Integrate the microgrid into the electricity market, in order to enable the microgrid

to oer its spinning and non-spinning power generation reserve as Ancillary Services (AS)

iii
to the grid. To this end, studying the operational mechanisms of the Australian market

was essential prior to building the proposed market participation rules which form an in-

tegral part of the proposed management system. As a result we used the market data, by

approaching the market operators to create a semi-realistic competitive market environ-

ment for our simulations. Consequently, a smart and adaptive pricing mechanism, that

adapts the AS prices to the amount of electricity on oer, and the level of demand in the

market has been presented.

The motivation for introducing the proposed management system, is to achieve a tran-

sition plan for current microgrids, so that they can have a commercial connection to the

future Smartgrid. The results obtained in this work show that there is a signicant eco-

nomic and environmental advantage to be gained from utilising intelligence when manag-

ing electricity generation within a power grid. As a consequence, selecting the appropriate

management strategy is fundamental to the success of the proposed management system.

In conclusion, modelling of the proposed strategies using MAS technology has proven to

be a successful approach, and one that is able to reect the human attitude; in making

critical decisions and in reducing the cost of generation.

iv
DECLARATION

I certify that this thesis does not, to the best of my knowledge and belief:
( i) incorporate without acknowledgement any material previously submitted for a degree or
diploma in any institution of higher education;
(ii) contain any material previously published or written by another person except where
due reference is made in the text; or
(iii) contain any defamatory material.
I also grant permission for the Library at Edith Cowan University to make duplicate copies
of my thesis as required.

v
Contents

Use of Thesis i

Abstract ii

DECLARATION v

1 INTRODUCTION 4
1.1 Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5

1.2 Problem statement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8

1.3 Research questions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9

1.4 Aims of this thesis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10

1.5 Thesis contributions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11

1.6 Publications from this research . . . . . . . . . . . . . . . . . . . . . . . . . 12

1.7 Thesis outline . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13

2 Background and Literature Review 16

2.1 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17

2.2 Australian Government schemes for emissions reduction and promotion of

renewable energy sources uptake . . . . . . . . . . . . . . . . . . . . . . . . 18

2.3 Microgrid resources operation management problem . . . . . . . . . . . . . 20

2.4 Demand forecasting in energy management systems . . . . . . . . . . . . . 21

2.4.1 Fuzzy modelling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22

2.4.2 Fuzzy subtractive clustering method . . . . . . . . . . . . . . . . . . 24

2.4.3 Adaptive Neuro Fuzzy Inference System (ANFIS) . . . . . . . . . . 26

2.5 Microgrids operation performance optimisation . . . . . . . . . . . . . . . 29

2.6 Multi-agent based microgrid management systems . . . . . . . . . . . . . . 32

2.6.1 Multi-Agent System (MAS) technology . . . . . . . . . . . . . . . . 33

2.6.2 JAVA Agent DEvelopment Framework (JADE) . . . . . . . . . . . . 34

vi
 2.7 Microgrids market participation and pricing mechanisms . . . . . . . . . . 37

2.8 Microgrids and Smartgrid . . . . . . . . . . . . . . . . . . . . . . . . . . . 39

2.9 Closing remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41

3 Load Forecasting Model 43

3.1 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44

3.2 Related works . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46

3.3 Research objectives and challenges . . . . . . . . . . . . . . . . . . . . . . . 48

3.4 Research assumptions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49

3.5 Modelling methodology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49

3.5.1 ANFIS model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51

3.5.2 FeedForward Tuning Fuzzy System (FFTFS) . . . . . . . . . . . . . 56

3.5.3 FeedBack Tuning Fuzzy System (FBTFS) . . . . . . . . . . . . . . . 60

3.6 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65

3.7 Closing remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72

4 Power Generation Management and Storage Devices Control System 74

4.1 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75

4.2 Related works . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76

4.3 Research objectives and challenges . . . . . . . . . . . . . . . . . . . . . . . 78

4.4 Research assumptions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78

4.5 Case study . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79

4.6 Planning for installing solar panels and wind turbines . . . . . . . . . . . . 80

4.6.1 Solar panel installation . . . . . . . . . . . . . . . . . . . . . . . . . . 83

4.6.2 Wind turbine installation . . . . . . . . . . . . . . . . . . . . . . . . 84

4.6.3 Other resources installation . . . . . . . . . . . . . . . . . . . . . . . 87

4.7 Power generation management and storage control . . . . . . . . . . . . . 88

4.7.1 Monitoring the microgrid's internal operational conditions . . . . . 88

4.7.2 Adaptation fuzzy system . . . . . . . . . . . . . . . . . . . . . . . . 91

4.7.3 Optimum generation cost performance . . . . . . . . . . . . . . . . . 93

4.8 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96

4.9 Closing remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100

5 Fuzzy Logic-Based Ancillary Service Adaptive Pricing System 101

5.1 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102

5.2 Related works . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103

vii
 5.3 Research objectives and challenges . . . . . . . . . . . . . . . . . . . . . . . 105

5.4 Research assumptions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106

5.5 Possible scenarios and rule base system . . . . . . . . . . . . . . . . . . . . 106

5.6 Decisions mechanism and Ancillary Services pricing equations formulation . 108

5.6.1 Pricing equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109

5.6.2 Pricing factors forecasting . . . . . . . . . . . . . . . . . . . . . . . 113

5.6.3 Adaptive pricing rules fuzzy system . . . . . . . . . . . . . . . . . . 114

5.7 Market simulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117

5.8 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119

5.9 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 124

6 Agent-Based Implementation of Energy Management Systems for Mi-

crogrids 126
6.1 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127

6.2 Related works . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 128

6.3 Research objectives and challenges . . . . . . . . . . . . . . . . . . . . . . . 130

6.4 Research assumptions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 130

6.5 Agent-based technology for implementing the proposed management strate-

gies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131

6.5.1 Demand Forecasting Model agent . . . . . . . . . . . . . . . . . . . 138

6.5.2 Generation and Storage Management agent . . . . . . . . . . . . . 143

6.5.3 Bids Analysis and Generation agent . . . . . . . . . . . . . . . . . 144

6.5.4 Master Controller agent . . . . . . . . . . . . . . . . . . . . . . . . 145

6.6 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 145

6.6.1 The role of forecasting in the proposed management strategy . . . . 145

6.6.2 The role of Generation and Storage Management agent in the pro-

posed strategy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 148

6.6.3 The role of Bids Analysis and Generation agent in the proposed

strategy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 153

6.7 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 155

7 CONCLUSION 157
7.1 The contribution of the thesis . . . . . . . . . . . . . . . . . . . . . . . . . 158

7.2 Future works . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 162

7.2.1 Developing adaptive membership function demand forecasting models 162

viii
7.2.2 Considering power quality and market participation in optimising

the microgrid performance . . . . . . . . . . . . . . . . . . . . . . . 163

7.2.3 Implementing adaptive pricing rules that aim to sell the power re-

serve in a limited time frame . . . . . . . . . . . . . . . . . . . . . . 164

7.2.4 Investigating the role of agent technology in implementing a mobile

distributed energy management system using the Android operating

system . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 164

ix
List of Tables

2.1 Hybrid learning passes directions [1] . . . . . . . . . . . . . . . . . . . . . . 27

3.1 ROI values and complexity of the 12-month electricity energy demand mod-

els . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52

3.2 Three, four, and ve-input systems forecasting accuracy . . . . . . . . . . . 56

3.3 Self tuning fuzzy rule-based system . . . . . . . . . . . . . . . . . . . . . . . 57

3.4 Membership function ranges design for the 12-month FFTFS . . . . . . . . 59

3.5 Statistical results for each month with the improvement rate made by FFTFS-

ANFIS model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59

3.6 Rule base table for the two-input-one-output FBTFS . . . . . . . . . . . . . 61

3.7 Year B Model feedback fuzzy tuning system gains selection . . . . . . . . . 63

3.8 The selected gains in the training data set of everyday data from January

to June . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66

3.9 The selected gains in the training data set of everyday data from July to

December . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67

3.10 Neural networks design parameters for the gain adaptation of the 12-month

forecasting FBTFS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68

3.11 The statistical analysis results in each month with improvement rate made

by the auto tuned FBTFS over the manually tuned and PI system . . . . . 69

3.12 Number of generators relating the predicted demand . . . . . . . . . . . . . 72

4.1 The daily average of missing generation hours . . . . . . . . . . . . . . . . . 87

4.2 Estimated generation capacity for the simulated resources . . . . . . . . . . 87

4.3 Local generation price based on the demand, the availability of the sources

and the dynamic generation price within the managed microgrid . . . . . . 90

4.4 Charging price fuzzy rule-base system . . . . . . . . . . . . . . . . . . . . . 93

4.5 Microgrid generation performance evaluation with FPSO . . . . . . . . . . 98

4.6 Microgrid generation performance evaluation with PSO . . . . . . . . . . . . 98

x
4.7 Storage devices performance comparison . . . . . . . . . . . . . . . . . . . . 99

5.1 Possible expected operation cost scenarios; with regards to the level of ex-

pected demand in the managed microgrid . . . . . . . . . . . . . . . . . . . 108

5.2 Formulated equations that evaluate the possible amount of Ancillary Ser-

vices in every operation scenario . . . . . . . . . . . . . . . . . . . . . . . . 112

5.3 Neural networks design parameters for the pricing factors forecasting models 114

5.4 Pricing adaptation fuzzy rule base system . . . . . . . . . . . . . . . . . . . 116

5.5 Pricing performance analysis . . . . . . . . . . . . . . . . . . . . . . . . . . 122

5.6 Performance analysis for discriminatory pricing rules . . . . . . . . . . . . . 122

5.7 Uniform pricing rules performance analysis . . . . . . . . . . . . . . . . . . 124

6.1 The cost of running the advanced management system for the resources

utilisation throughout the year . . . . . . . . . . . . . . . . . . . . . . . . . 151

6.2 The cost of running the basic management system for the resources utilisa-

tion under Connecting mode . . . . . . . . . . . . . . . . . . . . . . . . . . . 152

6.3 The cost of running the basic management system for the resources utilisa-

tion under Islanding mode . . . . . . . . . . . . . . . . . . . . . . . . . . . . 152

6.4 Comparing the amount of reserve sold with the amount of the reserve oered

in the market . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 153

xi
List of Figures

1.1 Demand, lost, and desired generation for a one week demand sample for a

specic microgrid . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7

2.1 Selecting suitable FSCM parameters in ANFIS modelling . . . . . . . . . . 27

2.2 ANFIS structure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28

2.3 PSO candidate's velocity updates mechanism . . . . . . . . . . . . . . . . . 32

2.4 Agent basic inter-connection structure . . . . . . . . . . . . . . . . . . . . . 33

2.5 Distributed agents and platforms installation using JADE [2] . . . . . . . . 34

3.1 The tuning fuzzy system . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46

3.2 A proposed model combination for January model . . . . . . . . . . . . . . 50

3.3 The electric power annual demand forecasting structure for ECU's Joon-

dalup campus . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51

3.4 Modelling data utilisation for the power consumption at ECU's Joondalup

campus . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53

3.5 The membership functions for the four inputs zero-order Sugeno fuzzy sys-

tem of January's power demand forecasting . . . . . . . . . . . . . . . . . . 54

3.6 Forecasting results comparison for demand in ECU from the 17th to the

21st of January, year C . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55

3.7 Fuzzy surface for the developed fuzzy models of the three, four, and ve-

input system . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57

3.8 Membership functions design for FFTFS of January demand forecasting model 58

3.9 Self tuning and ANFIS Demand forecasting in the ECU microgrid for the

17th to the 21st of January, year C . . . . . . . . . . . . . . . . . . . . . . . 60

3.10 Manually tuned FBTFS membership functions design . . . . . . . . . . . . 62

3.11 Auto tuned FBTFS membership functions design . . . . . . . . . . . . . . . 64

3.12 KP and Ki behaviour in feedback-tuning fuzzy system ANN . . . . . . . . 70

xii
3.13 The proposed model forecasting output for the period from the 17th till the

23'rd of January, year C . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71

4.1 Multi-Agent based smart microgrid layout . . . . . . . . . . . . . . . . . . . 80

4.2 Project development planning owchart . . . . . . . . . . . . . . . . . . . . 82

4.3 Summer, winter and optimal angles for solar panel installation . . . . . . . 83

4.4 Average monthly wind speed in the city of Perth [3] . . . . . . . . . . . . . . 85

4.5 Aerodynamic array eciency as a function of rotor distance in the wind

direction [4], where D is the wind turbine rotor diameter, dl is the space

length between the two installed wind turbines in project land, and dc is

the space width between the two installed wind turbines in project land. . . 85

4.6 A satellite image for the case study shows the available allocated installation

space for solar panels and wind turbines . . . . . . . . . . . . . . . . . . . . 86

4.7 The propsoed SDs charging management strategy block diagram . . . . . . 89

4.8 A sample of seven days generation price change for GP . . . . . . . . . . . . 90

4.9 The design of charging price generator fuzzy system membership functions . 92

4.10 The proposed Particle Swarm Optimisation update process . . . . . . . . . 96

4.11 Storage Devices charging/discharging status for a typical one week of oper-

ation in the managed microgrid . . . . . . . . . . . . . . . . . . . . . . . . . 99

5.1 The decision tree for the classied demand levels as compared with renew-

able energy sources and storage devices generation capacity scenarios . . . 107

5.2 Proposed adaptive pricing mechanism . . . . . . . . . . . . . . . . . . . . . 109

5.3 Proposed pricing adaptation process . . . . . . . . . . . . . . . . . . . . . . 110

5.4 Price decisions prole for a typical one week operation scenarios . . . . . . . 113

5.5 WS , P hG and UP forecasting models performance in a typical one week of

operation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115

5.6 Pricing adaptation fuzzy system membership functions . . . . . . . . . . . . 117

5.7 Proposed electronic negotiation software sequence (two types of sellers) . . . 118

5.8 Smart electricity trade process using Multi-Agent System technology under

the discriminatory pricing rules . . . . . . . . . . . . . . . . . . . . . . . . . 119

5.9 Fuzzy-based adaptation level performance . . . . . . . . . . . . . . . . . . . 121

6.1 The developed Multi-Agent System structure . . . . . . . . . . . . . . . . . 133

6.2 The messaging direction between agents in all operation modes . . . . . . . 136

6.3 Strategies and priority levels based on the microgrid's operational conditions 137

xiii
6.4 Information exchange between the proposed agents in Connecting mode . . 139

6.5 Framework initialisation sequence . . . . . . . . . . . . . . . . . . . . . . . . 140

6.6 Agents communication in Connecting mode . . . . . . . . . . . . . . . . . . 141

6.7 Agents communication in Islanding mode . . . . . . . . . . . . . . . . . . . 142

6.8 Agents communication in Emergency mode . . . . . . . . . . . . . . . . . . 143

6.9 Demand forecasting impact on reducing the economic cost ($) of the elec-

tricity supply in the case study . . . . . . . . . . . . . . . . . . . . . . . . . 147

6.10 Demand forecasting impact on reducing the CO2 emission (Kg) of the elec-

tricity supply in the case study . . . . . . . . . . . . . . . . . . . . . . . . . 147

6.11 The ratio of the renewable energy sources generation to the demand values

throughout the year in the case study . . . . . . . . . . . . . . . . . . . . . 149

6.12 Economic generation costs in ($) . . . . . . . . . . . . . . . . . . . . . . . . 149

6.13 Environmental generation costs in (Kg) . . . . . . . . . . . . . . . . . . . . 150

6.14 Comparing the amount of money achieved with the total amount of trans-

actions throughout the year . . . . . . . . . . . . . . . . . . . . . . . . . . . 155

xiv
Nomenclature

AP Achieved Prot

AS Ancillary Services

AI Articial Intelligence

T Charging and discharging time

EconCost Economic cost

Dmnd Electricity demand

SLP Electricity sale price

ET P Electricity Trade Participation

EM S Energy Management System

EnviCost Environmental cost

F LRdyF rSLElec Fuzzy Logic-based ready for sale electricity

F LAP R Fuzzy Logic-based Adaptive Pricing Rules

F LSLP Fuzzy Logic-based sale price

GG Gas gen-sets dispatch

M oAGc Gas gen-sets maintenance and operating annual xed cost (economic

or environmental)

M oV Gc Gas gen-sets maintenance and operating variable cost (economic or

environmental)

GC Gas generation cost (economic or environmental)

1
GlostG Gas generation lost reserve

GP Gas generation price

LGP Local generation price

LB Lower Bond

GGmax Maximum Gas gen-sets dispatch capacity

SGmax Maximum Storage Devices dispatch capacity

GGmin Minimum Gas gen-sets dispatch threshold

SGmin Minimum Storage Devices dispatch threshold

UGmin Minimum Utility dispatch threshold

M AS Multi-Agent System

N ewSaleprice New sale price

NPR Normal Pricing Rules

GN o. Number of Gas gen-sets

NOP Number Of Participation

P hG Photovoltaics (solar panels) dispatch

P hG Photovoltaics dispatch

PE Prediction Error

PH Prediction Horizon

ROI Range Of Inuence

RdyF rSLElec Ready for sale electricity

RF S Ready for sale energy

ST F S Self Tuning Fuzzy System

SI Solar Irradiation

SLDElec Sold Electricity

2
SC Storage Devices dispatch cost (economic or environmental)

SP Storage Devices dispatch price

SG Storage Devices dispatch

M oV Sc Storage Devices maintenance and operating variable cost (economic or

environmental)

Charging Storage Devices operation status (Charging) cost

Discharging Storage Devices operation status (Discharging) cost

SDs Storage Devices

SE Supplied Energy

TV Trade Value

UB Upper Bond

M oAUc Utility annual feeder utilisation percentage cost

UP Utility dispatch price

UG Utility dispatch

UC Utility generation cost (economic or environmental)

G Variable Gas price

W AM Weight adjusting method

WS Wind speed

WG Wind turbines dispatch

WS Wind turbines speed

3
Chapter 1

INTRODUCTION

Building a Smartgrid is a challenge which involves upgrading the current grid with the

necessary infrastructure in order to facilitate the new trends concerning electricity gener-

ation, transmission and distribution. These new trends are leading towards more reliable,

secure, economic, and environmentally friendly electricity generation. The Smartgrid is

expected to be comprised of a network of grids (microgrids), all of which are decentralised,

and which work independently to enhance the overall operation of the main grid. Looking

at the distribution side of the grid (Low Voltage), it is expected to have more penetration

from Distributed Generators (DGs). These estimates are based on the price increment

forecasts for oil and gas, and the price decrement forecasts for renewable energy sources.

Distributed generation involves the interconnection of small-scale, on-site generators, e.g.

diesel/gas, storage devices, and renewable energy resources. In this study, DGs will be

limited to solar panels, wind turbines, storage devices, and local gas generators. It is an-

ticipated that the smart utilisation of these resources within the utility grid, would lead

to a big economic and environmental benets for both the electricity provider and the

consumers. Accordingly, new market schemes will emerge along with the new trends of

demand and supply in the grid. So, providing the ideas and the strategies, for dealing with

the complexity of managing resources in these networks, is the targeted research topic of

this work. In this chapter, Section 1.1 points out the research signicance of this work,

and justies the reasons behind approaching this research problem. Section 1.2 briey

details the statement of the research problem targeted in this work, and also illustrates

the solutions required by this work to achieve optimum results. Section 1.3 introduces the

research questions raised by this thesis. Sections 1.4 and 1.5 highlight the main objectives

and the contributions aimed by this thesis respectively. Section 1.6 lists the publication

resulted of conducting this research, and nally Section 1.7 outlines the thesis structure.

4
1.1 Motivation

The motivation for this research was the obvious need to upgrade the current grids for more

economic and environmentally friendly electricity generation. Initially we considered the

levels of upgrade that were required, along with their relative importance from both eco-

nomic and environmental perspectives. We found that the rst challenge in the upgrade

process relates to the ability of the current grid to accommodate the higher number of

DGs. Financial, political, and geographical considerations must also be taken into account

when upgrading, which make this process even harder. When considering the technical

challenges, variable eciency is expected depending on the installation procedure for these

resources. Ultimately, all of the eciency and cost losses are expected to be compensated

for, and considered in the project planning. Having multiple sources of dispatch to com-

pensate for these losses would result in a power quality issue. Depending on the type of the

project and the complexity of managing the installed resources, variable generation cost

and uncertainty of demand prole are expected, and hence a complex management prob-

lem arises whilst operating the system. As a result, a modern infrastructure is required to

accommodate this complexity, and a greater level of project expenditure is anticipated in

order to control the operation and to maintain its eciency.

Choosing the appropriate management strategy necessitates a deep understanding of

the system's operational conditions, and requires the ability to apply the decisions to the

system. When building a management strategy which can cope with the uncertainty in

the demand and the generation cost in a power system, it is essential to categorise the

operational conditions based on their cost and safety factors; thus identifying the reasons

for, and solutions to, the problem. The problem can be solved using either classical or

intelligent management methods, depending on its level of complexity. The operational

scenarios for microgrids dier based on their load prole and the location where they

are installed. In the case of medium sized enterprise microgrids, the problem becomes

bigger depending on their operational scenario, specically when they are connected or

disconnected, as there is a chance to have two-way power ow. This leads to a challenging

process called the  Generation Scheduling Problem . The generation scheduling problem

can be dened as the challenge of accommodating the uncertainty and complexity of having

multiple sources of dispatch to provide a secure and reliable supply for a specic power

system. A brief introduction to this problem is provided in this thesis, along with some

supporting references.

Identifying the demand range in power systems is essential in making strategic decisions

5
about the reliability and sustainability of these systems. By studying a demand prole of

a typical microgrid that is usually supplied from DGs, a certain amount of lost energy

is identied when it is compared with the amount of supplied energy. This is needed

to secure the supply at all demand levels, because it is impractical to generate energy

to match the exact amount of variable demand only. Concurrently, this information is

needed when building energy management systems. Furthermore, demand forecasting is

essential in performing energy management system applications [5]. This can be achieved

by applying modern (computerised) infrastructure and implementing intelligent strategies

on the network. Figure 1.1 shows the full capacity and the desired generated supply, which

can be obtained by developing an eective Energy Management System (EMS). Problems

arise when a microgrid is supplied from a range of dierent types of generators in line with

the utility supply, each with variable generation availability and cost. Problems arise when

a microgrid is supplied from a range of dierent types of generators in line with the utility

supply, each having variable generation availability and variable cost. In the literature,

this problem with the uncertainties of parameters or constraints, have been discussed, and

a range of planning techniques has been presented [6, 7, 5]. This problem has also been

targeted by dierent optimisation techniques in [8, 9, 10, 7].

6
Figure 1.1: Demand, lost, and desired generation for a one week demand sample for a
specic microgrid

Ultimately, intelligence is essential in most critical issues in the microgrid's operation,

such as forecasting, demand response, self-healing, generation scheduling and markets par-

ticipation. Dealing with these issues has been widely addressed by researchers in this eld.

Fuzzy Logic, Articial Neural Network, Linear Programming, Particle Swarm Optimisa-

tion, and Genetic Algorithms have been applied extensively to solve these problems due to

their reputation for solving problems in many research areas. In the literature, Multi-Agent

System (MAS) technology has also been widely applied to solve microgrid management

problems.

Although Articial Intelligence (AI) based methods are found to be successful in solving

microgrid management problems, they still lack accuracy due to their very high demand

for computation resources. Consequently, using the right management structure with the

right AI-based method to solve microgrid management problems can reduce the required

computation resources and increase the eciency. By considering the uncertainty of gener-

ation of renewable energy sources and storage devices, a logical sequence of operations for

reducing the required computation resources is needed in tackling this uncertainty. Thus,

7
the right sequence of operation and the smart utilisation of resources can also increase

the operational eciency of the managed system. Storage devices and electric vehicles are

very important factors in improving power systems' operational eciency. So, operating

these resources, in a smart manner among other resources, would impose a substantial

environmental and economic cost reduction.

While cooperation between the centralised (Smartgrid) and the decentralised (micro-

grids) is expected, it is also expected that the challenge of participation under the compet-

itive market environment be encountered. Since development of standards and regulations

for a Smartgrid connection is an active research topic, contributing to the research with

novel ideas about connecting microgrids would have a positive impact on the literature.

In fact this work was inspired by the need for reliable management strategies that con-

nect microgrids with the Smartgrid; and as a result the proposed strategy carries a novel

strategic plan for performing reliable generation decision making on a simulated microgrid.

1.2 Problem statement

A recent visit by an ECU Smartgrid research group to Western Power, the public power

supplier for the state of Western Australia was very productive. The visit resulted in a suc-

cessful discussion of the roadmap and challenges for obtaining power from multiple sources

of dispatch system, as well as the intention to connect them to the grid in the near future.

Two major issues relating to running multiple sources of dispatch for a microgrid in parallel

with utility supply were identied, which are power quality and resources scheduling. It is

assumed that power quality is maintained at a satisfactory level, while managing variable

quality supply is a problem which can be solved in future extensions of this work. On the

scheduling side, the proposed management strategy is expected to deal with three kinds

of microgrid connection scenarios: 1) Islanding mode; 2) Connection mode when the mi-

crogrid is buying/selling energy from/to the main grid; and 3) Emergency mode when the

microgrid generation encounters an operational fault. These scenarios are always depen-

dent on the dynamic generation cost and the intermittent generation from the participant

resources. Dynamic utility price, solar irradiation, wind speed, and consumer demand are

the targeted variables to be managed to minimise the microgrid generation cost in this re-

search. As forecasting demand was found to be essential to provide the knowledge required

about the needed generation in managing resources, nding the most accurate modelling

method that suits the type of the demand prole (residential, commercial or industrial) is

another considerable challenge. Accordingly, a comprehensive study concerning modelling

8
the demand, and choosing the most eective parameters that aect the demand prole,

is needed in building the proposed management strategies. A more challenging state of

operation arises when local DGs propose to dispatch their reserve to the grid, by bidding

in a competitive market environment. This state of operation needs to be considered in

conjunction with a bidding strategy that works, depending on the microgrid's reserve sta-

tus. Hence these engineering problems can be solved by implementing a multi-processing

unit management system that comprises distributed sub-systems, each with its indepen-

dent computational resources, to maintain low-cost generation for the microgrid in the

shorter and the longer terms. A communication and synchronisation problems will then

arise out of incorporating decentralised processing units that are controlled by a central

unit. The main reason behind building distributed processing system is to enable mobility

and independence in making decisions, thus forming a multi-level of priority for the deci-

sion making. This structure enables the system to run under dierent operating modes,

and also allows the system to accommodate additional generators/loads when required.

Finally, the concept of building the system required in this work will be based on the

concept of implementing a Distributed Control System (DCS).

1.3 Research questions

The following research questions will be answered in this work:

ˆ How to introduce a demand forecasting model for a microgrid?

ˆ How to nominate the most eective parameters in modelling the demand for a mi-

crogrid?

ˆ How to enhance the robustness and the stability of the demand forecasting model?

ˆ How to make the forecasting adaptive to sudden and unexpected changes in the

microgrid operation?

ˆ What are the constraints and the problems involved in installing renewable energy

sources on the current traditional microgrids?

ˆ How to sequence the operation of sources based on the operational conditions, which

makes the generation cost lower in the longer terms of microgrid operation?

ˆ How to perform a benecial charging/discharging process for the storage devices

that run in parallel with other sources that deliver variable cost and intermittent

9
 generation?

ˆ How to oer the microgrid's generation reserve to the markets at competitive prices?

ˆ How to model the strategic thinking in placing reserve prices under a competitve

market environment?

ˆ How to integrate all of the above intelligent modules into one framework that works

towards achieving low-cost generation?

ˆ How AI-based MAS technology is utilised in implementing power system generation

management systems, and what are the limitations of the proposed power manage-

ment system for the microgrid?

1.4 Aims of this thesis

The thesis aims to provide AI-based solutions for energy management systems that target

low-cost generation for microgrids. More specically, the thesis introduces new techniques

to solve the identied problems for managing multiple sources of dispatch within medium-

sized enterprise microgrids. The main aims of this thesis can be explained as follows:

ˆ Identifying the impact of integrating demand forecasting into management strategies

that aim to reduce the generation cost for microgrids, and also applying an adaptive

forecasting model to adapt the forecasts to sudden unexpected demand changes in

microgrids.

ˆ Studying the possibilities of installing renewable energy sources on a medium-sized

enterprise microgrid, and also highlighting the challenges of implementing the pro-

posed solutions. The study and analysis address the role of geographical location,

weather conditions, utility price rises and expected demand growth in a targeted case

study. In addition, estimates about the economic and environmental benets based

on a set of conservative engineering assumptions will be reported. Accordingly, the

cost of upgrading the infrastructure in microgrids will be justied.

ˆ Implementing an optimisation technique that utilises the information provided every

30 minutes about the instantaneous cost of generation of utility, storage devices

and gas generation, coupled with the availability of installed solar panels and wind

turbines generation. This technique targets the smart utilisation of storage devices

10
 among other resources, to reduce the generation cost in the longer term for the

managed microgrid.

ˆ Reecting the human attitude in achieving competitive power trading with the util-

ity. Setting sale prices for the energy in a competitive market environment is imple-

mented by utilising AI in four stages: 1) generating initial sale prices; 2) forecasting

the weather conditions and price changes around the microgrid; 3) estimating the op-

erational conditions of the microgrid based on the delivered forecasts; and 4) adapting

the sale prices to these conditions, which make the prices adaptive and attractive to

sellers in the market.

ˆ Applying management strategies that include all of the above goals to achieve low

economic and environmental generation costs for the simulated microgrid. The pro-

posed strategies will be evaluated in a simulated environment based on a set of

engineering hypotheses. The thesis will also cover the steps of the development that

encompass the logical sequence of running the intelligent modules to help make op-

timum decisions.

ˆ Investigating the success of utilising Multi-Agent System (MAS) technology in the

microgrid resources management system. At this level, the thesis will cover the

stages of building a MAS for the proposed system. This will include the details

of embedding the proposed intelligent modules in software agents, to perform the

required optimum operation for the microgrid resources.

Ultimately, the thesis will provide to the research community, novel ideas about: 1) demand

forecasting; 2) optimising the cost of generation; 3) embedding the human sense of trading

within a competitive market environment; and 4) modelling strategic plans for running the

microgrid resources under dierent scenarios of operation.

1.5 Thesis contributions

The contributions of this thesis come from developing new methods for solving the most

critical issues in the eld of smart microgrid management systems. The novelty of the

work is found in:

ˆ Detailing a novel demand forecasting strategy, that is comprised of an adaptive struc-

ture to accommodate the unexpected demand prole on the managed microgrid,

based on a 30-minute sampling interval. The modelling stages are covered in detail,

11
 including nominating the most eective modelling parameters, dividing the demand

patterns into 12-month-based patterns, working on increasing the robustness of the

model, and considering the real-time demand changes in the forecasting.

ˆ Introducing a novel optimisation technique, that tackles the uncertainties of inter-

mittent generation of the installed renewable energy sources, in line with the variable

generation cost of the participant gas gen-sets, utility and the storage devices. The

proposed optimisation technique aims at achieving the lowest economic and environ-

mental generation cost from the participant DGs in line with utility supply. The

novelty of this work is also introduced by implementing a novel storage device charg-

ing/discharging control method, which reacts based on the daily generation cost

pattern, making the storage devices important players in reducing the microgrid

generation cost.

ˆ Proposing novel Ancillary Services (AS) pricing rules, that help in selling microgrid

electrical power reserve to the utility in a competitive market environment. The thesis

will detail the pricing rules, along with their adaptation mechanism that makes them

adaptive based on the operational conditions in microgrids, such as the availability of

generation and the attitude of the markets. Estimation these parameters will result

from forecasting models that are embedded into the pricing rules.

ˆ Implementing a strategic management system that operates all of the above-listed

modules in a reliable manner, in order to optimise the operation of the microgrid un-

der dierent types of operational conditions. The implementation includes modelling

the behaviours of the proposed management system using MAS technology. There-

after, a set of decision priority levels is embedded into the system to enable running

recongurable optimisation structure for all the operating modes. The optimisation

structure can be recongured by identifying the microgrid operational conditions,

and/or by reading the operational status of the intelligent modules.

1.6 Publications from this research

1. T. Mahmoud, D. Habibi, and O. Bass. Upgrading a medium size enterprise power

system with wind and solar sources: Design, nancial and environmental perspec-

tives. In ENERGY 2011, The First International Conference on Smart Grids, Green

Communications and IT Energy-aware Technologies, Venice, ITALY, pages 3339,

12
 2011.

2. T. Mahmoud, D. Habibi, O. Bass and S. Lachowics. Fuzzy Inference System in En-

ergy Demand Prediction, Fuzzy Inference System - Theory and Applications, InTech,

(2012).

3. T. Mahmoud, D. Habibi, O. Bass, and S. Lachowicz. Load demand forecasting:

Model inputs selection. In Innovative Smart Grid Technologies Asia (ISGT), 2011

IET, Perth, AUSTRALIA, pages 16, 2011.

4. T. Mahmoud, D. Habibi, O. Bass, and S. Lachowicz. Tuning fuzzy systems to

achieve economic dispatch for microgrids. In Innovative Smart Grid Technologies

Asia (ISGT), 2011 IET, Perth, AUSTRALIA, pages 16, 2011.

5. T. Mahmoud, D. Habibi, O. Bass. Fuzzy-Based Adaptive Pricing Rules for a Typical

Microgrid Energy Management System. In The 9th IET International Conference on

Advances in Power System Control, Operation and Management, IET, Hong Kong,

2012.

6. T. Mahmoud, D. Habibi, O. Bass. Fuzzy Logic for A Smart Utilisation of The

Storage Devices in A Typical Microgrid. In International Conference on Renewable

Energy Research and Applications, ICRERA-2012, Nagasaki, IEEE, JAPAN, pages

16, 2012.

7. T. Mahmoud, D. Habibi, O. Bass. An Adaptive Real-Time Forecasting Model for

A Medium-Sized Enterprise Electricity Demand. International Journal of Electrical

Power and Energy Systems, Elsevier, 2012. (Sumbitted)

8. T. Mahmoud, D. Habibi, O. Bass. Adaptive Ancillary Services Pricing Rules for A

Typical Medium Size Enterprise Microgrid. IEEE Transactions on Power Delivery,

IEEE, 2013. (To be submitted)

9. T. Mahmoud, D. Habibi, O. Bass. Application of Multi-Agent System for Microgrid

Autonomous Distributed Management System. IEEE Transactions on Smart Grid,

IEEE, 2013. (To be submitted)

1.7 Thesis outline

The organisation of this thesis is as follows:

13
ˆ Chapter 1 has introduced power generation trends, including microgrid roles and

operation mechanisms based on their structures, types and connection modes, along

with the challenges of operation and performance control. This chapter has also intro-

duced the economic and environmental impact of the decentralised power generation

units on the main utility grid. Several management strategies to utilise multi-type

decentralised generation units have been proposed in this context. Concurrently, a

brief explanation of the strategies needed to perform microgrid operation has been

provided in this chapter.

ˆ Chapter 2 presents the evolution, the technical background of power systems and

the generation problems, along with their prospective solutions. A comprehensive

review of the methods of management strategies applied to control power systems,

including microgrid control and optimisation techniques, is provided in this chap-

ter. In addition, a review of the articial intelligence methods used in controlling

and managing the distributed generators in microgrids is presented. The review also

includes modelling the demand in power systems, managing multiple sources of dis-

patch for microgrids, and performing energy trade with the utility based on dierent

utility and trading rules.

ˆ Chapter 3 introduces a novel modelling method that targets the demand forecasting
in the managed microgrid, and also investigates the utilisation of forecasting in other

roles within the proposed management system. The chapter also reviews the current

trends for modelling the demand, and provides a summary of the other works in

the eld. The details of the design of the proposed model are presented, along with

the details of the AI techniques used in the modelling. The modelling process is

proposed to cover 12 months of operational demand types throughout the year. The

results and verication of the modelling technique, which are used in other intelligent

modules of this work, are also presented in this chapter.

ˆ Chapter 4 illustrates the second intelligent module of the management system,

which is responsible for making optimised and low-cost generation decisions. The

chapter starts by detailing the problem and reviewing other people's works in this

context. In addition, it details the proposed optimisation technique that targets

achieving low economic and environmental generation costs by exploring the range

of opportunities for changing the variable generation factors for the participant gen-

erators. The chapter also introduces a new method for utilising the storage devices

14
 among the participant generators, which imposes a lower generation cost in the longer

term for the targeted case study. As part of that, all the simulation data, which have

been used, have been generated based on several engineering assumptions that make

the data semi-realistic for testing the proposed optimisation technique and the overall

management system. Highlights of the proposed optimisation structure, the design

outlines and the AI utilised are also covered in this chapter.

ˆ Chapter 5 proposes a new bidding strategy and pricing rules for the case study's

AS. In this chapter the steps for extracting the pricing rules for selling AS to the

utility are detailed. Furthermore, the modes of adapting these rules to the microgrid

operational conditions, the market supply and the demand levels are detailed.

ˆ Chapter 6 covers the proposed management structure that connects all the proposed
intelligent modules to build a management system. The chapter addresses the logi-

cal sequence of running the intelligent modules and exchanging information among

them. It also shows the impact of the proposed management strategies in reducing

the economic and environmental costs through utilising the forecasting models, the

optimisation techniques, and the bidding strategies. It also addresses the impact of

MAS technology in supporting the recongurable management strategies.

ˆ Finally, Chapter 7 summarises the contribution of this thesis in providing a mi-

crogrid, multiple sources of dispatch management solutions that aim to minimise

the generation cost in autonomous type microgrids. Furthermore, several research

directions are recommended in this chapter as future extensions of this work.

15
Chapter 2

Background and Literature Review

This chapter introduces the problem of managing multiple sources of dispatch in a typ-

ical microgrid, and also lists a historical background about the trends of power system

technologies involved to solve this problem. It was essential to investigate the current

technology trends that have been utilised by researchers and engineers to solve microgrids

management and control problems. Building the knowledge about these technologies has

been aligned in this work with our experience in articial intelligence implementation, sys-

tem modelling and software development to conduct a valuable investigation that answers

some of the highly regarded research questions in this area. In this chapter, the literature

has been critically revised and the missing parts of knowledge have been identied in order

to introduce our research proposal. Ultimately, this chapter has utilised the facts and ev-

idences introduced by other researchers to develop new advanced management strategies

for microgrids that can cope with the current research problems. In this chapter, we rstly

list a historical background about power systems and their problems in Section 2.1. Sec-

tion 2.2 details the Australian Government initiatives for reducing emission and installing

renewable energy sources that have been considered in the planning part of this work.

Section 2.3 introduces the generation scheduling and the resources utilisation problems.

Section 2.4 explains the role of demand forecasting in managing the resources for power

systems. Sections 2.5 details the methods of performing optimal operation for microgrid

resources, and also highlights the role of integrating the microgrid with electricity markets

under various market policies. Section 2.8 covers the role of Multi-Agent System (MAS)

technologies in managing microgrids resources while detailing the modelling and simula-

tion tools used by research communities. Finally Section 2.9 summarises the literature and

identies the limitations of the current research methodologies in answering the research

questions in this eld.

16
2.1 Background

The denition of an electric power system basically represents a system that comprises gen-

erating, transmitting and distributing the electricity power via various types of generators,

transmission lines and distribution networks through various types of connection layouts.

The rst design in the electricity industry was proposed by Edison in 1881, and it has been

changed with several modications. Most of the implementations were made for the rst

time using the limited technology available in the 19th century until the second half of the

20th century, when the design of generation and transmission has faced rapid challenges

due the increased plants of power systems [11]. In the last 50 years, power networks have

witnessed an enormous amount of development to cope with the modern challenges such as

security threats, high demand prole, reliability and power quality issues. The increased

network complexity that resulted from the diversity of load and generator types has forced

the power industry to plan for grid management advancement. Since then, industry has

come up with strategies to use the communication and information technology to increase

the network reliability. In the 1960s, the industry started to use the computer to imple-

ment monitoring and control for the power systems [12]. In the last 25 years, the developed

countries have modernised their electrical power systems with the use of communication,

control and automation to become a complex multi-level voltage grid. Concerns over catas-

trophic climate change, increased industrialisation, CO2 emissions and fuel price rises were

the reasons behind creating incentives for reliable, safe and environment-friendly energy

sources. To help achieve a clean and secure electric power industry, the current power

grid has been modernised to accept a range of renewable energy sources [13]. These are

expected to become more competitive with conventional generators, as they were involved

to help integrating the renewable energy resources within the power grid. The key point in

automating the network is dependent upon incorporating advanced management strategies

and reliable sensors and actuators. Since the 1980s, the automatic meter reading has been

used to monitor the loads, and in 1990s Advance Metering Infrastructure (AMI) has been

evolved. Smart meters have been used later to achieve real time monitoring with con-

tinuous communications. Synchronisation and wide area networks monitoring have been

revolutionised in the early 1990s. Demand response and demand side management are the

developed control strategies for the power grid by the use of smart metering.

Recent technological advances in the power generation and information technologies

areas are helping to change the modern electricity supply system in order to comply with

higher energy eciency and sustainability standards. The term  Smartgrid started to

17
be in use in 2005 [14], and is currently known as the use of smart meters, communications,

computational abilities and control in the form of enhancing the overall functionality of

two-way electric power delivery system. Smartgrid is an emerging trend of intelligence in

the power grid to optimise the network resources operation. In order for this intelligence

to be eective, it is necessary to retrieve enough information about the grid operation

together with other context data such as environmental variables. It is also essential to

intelligently modify the behaviour of the network elements accordingly. Currently, many

design challenges of Smartgirds are resulting from the advanced integration of digital

technology with power grid and its information ows. Economic generation and electric

market participation are the other concerns that are currently under considerations in

Smartgrid development.

As part of the international trend towards restructuring electricity grids, many de-

veloped countries are establishing new investment support policies, and also developing

legislation for commercialising Green Energy. As new clean-energy markets have been

emerging around the world, opportunities exist for economic development, jobs creation

and energy diversication benets [15]. Thus, there is a need to develop ecient clean-

energy regulations, so that socio-economic benets are maximised [16]. Regulators are

initiating new programs for energy eciency, renewable energy capacity generation and

reliability improvements [17]. Liberalising power markets will also become more important

because of the advantages of pooling large generation stations, sharing spinning reserves

and using the most ecient energy resources [18]. Therefore, managing the grid resources

with this much complexity has become a strategic matter, which has also been subject

to the governmental regulations and policies. Identifying the regulations and the possible

governmental support are essential in implementing new technologies on the grid. Thus,

we rstly explore the Australian Government regulations before conducting this research

to justify the impact of our research outcomes.

2.2 Australian Government schemes for emissions reduction

and promotion of renewable energy sources uptake

The Australian Government has taken on a responsibility to reduce the carbon pollution

in all its states and territories. Its main target is reducing Australian's carbon footprint

by nearly one third to one half. They have also started to develop new policies and rules

which are needed to help Australian businesses and households reduce their carbon pollu-

18
tion [19]. Given that Australia has a very large ecological footprint, which is dominated

by carbon dioxide emissions from fossil funds, it is appropriate that Australia be very ac-

tive in responding to climate change. This indicates major research and initiative project

funding for energy balance and emissions reduction, for example the Low Emission Tech-

nology Demonstration Fund (LETDF) to support the Australian rms to commercialise

low emissions technologies [20]. Carbon Pollution Reduction Scheme (CPRS) would intro-

duce a price on carbon pollution and ensure that all businesses take this into account when

developing their business plan. Concurrently, the Australian Government is developing,

commercialising, and investing in clean energy technologies [19]. The Australian Govern-

ment has initiated major funding support plans to promote commercialisation of renewable

energy [21]. Australia is considered one of the most favourable nations for investment in

energy resource development because of low political and regulatory risks [22]. Foreign

investment is welcomed by the Australian Government and no mandatory local equity or

local content requirements are imposed on energy resources development [23]. The Oce of

the Renewable Energy Regulator (ORER) in Australia administers the Renewable Energy

(Electricity) Act 2000 (the Act), the Renewable Energy (Electricity) Charge Act 2000, the

Renewable Energy (Electricity) (Small-scale Technology Shortfall Charge) Act 2010 and

the Renewable Energy (Electricity) Regulations 2001 (the regulations) to increase renew-

able electricity generation from Australia's renewable energy sources by encouraging the

generation of an additional 45,000 GWh of renewable energy per year by 2020 [24]. The

Australian Government has established the Renewable Energy Target (RET) scheme to

ensure that 20 per cent of Australia's electricity supply will come from renewable sources

by 2020. The plan is to make the amount of energy coming from sources like solar, wind,

wave and geothermal sources be around the same as all of Australia's current household

electricity use [25]. These will be crucial for Australia's eorts to reduce its carbon pollu-

tion emissions. These technologies will also be important to the rest of the world as they

also need to reduce their carbon pollution. Australia's RET goal will be met from renew-

able sources such as wind, solar and geothermal power by 2020. The aim of the initiative

is to create more jobs in clean industries by giving investors the condence to back low

emissions technologies. In line with these initiatives, there is a need to promote research

and development in this area. Hence the mentioned schemes and initiatives are considered

in the planning for integrating renewable energy sources in the case study. Consequently,

this will also be considered in the simulation were the data are elaborated from the case

study based on the operational conditions and the governmental initiatives for installing

19
renewable energy sources.

2.3 Microgrid resources operation management problem

A microgrid is a low voltage, decentralised power system that comprises various types

of distributed generators (DGs), storage devices, and controllable loads. It can also be

dened as a cluster of loads and micro sources working as a single controllable system

that provides power to its local area [26]. The microgrid might incorporate distributed

generators of dierent sizes and types such as gas and/or diesel generators, wind turbines,

solar panels, fuel cells and storage devices. Depending on the location and the connection

requirement of the microgrid, it may also have a switchable connection that connects or

disconnects the microgrid from the utility grid depending on the operational conditions.

In islanding mode, a microgrid is fully supplied by its local resources, whereas in case of

connecting mode, it is supplied by both the local resources and the utility grid. Ultimately,

this would result in a complex optimisation problem called generation schedulling.

Basically, this problem is dened as the challenge of matching the amount of generation

with the demand in a power system. This might also include matching the demand with

the most cost eective generation from both the economic and environmental perspectives,

which makes it a signicant research topic. Generally, generation scheduling might be

encountered with utility grids, in addition to the microgrids. In this case, adequate man-

agement strategies are the key solutions for the performing ecient operation and control

of the microgrid [27]. In addition, robust control and management strategies are needed

to schedule these resources safely and sustainably [28].

By looking at the utilities, planning their generation has been introduced by [9], where

the generation planning and scheduling of the Iranian power plants has been investigated

using a new fuzzy hierarchical approach. A short-term scheduling of Fujian Electric hy-

dropower, to maximise economic dispatch benets was introduced in [10]. Adaptive Neural

Fuzzy Inference System (ANFIS) has been considered to estimate the next day generation

based on the last 10 days generation information in [29]. In contrast with utilities, gen-

eration problem in microgrids is relitevly more challenging. The major challenge initially

arises when the microgrid is supplied from various types of generators, each with dierent

generation cost, while having unknown amount demand. For this problem, an optimal

generation scheduling of a microgrid in the islanded mode has been presented in [7]. This

problem, with uncertainties in parameters or constraints, has been solved by various types

of optimisation techniques such as fuzzy optimisation-based method, which has been pre-

20
sented in [8]. Performing optimum microgrid power dispatch by controlling the amount of

reserve to reduce the amount of fuel consumption in the microgrid connection modes has

been addressed in [30], with a detailed formulation of the constraint for stable islanded op-

eration in accordance with two power-sharing principles: i) xed droop and ii) adjustable

droop. The formulated generation scheduling problem as a mixed integer non-linear prob-

lem has been computed by the generalised Benders decomposition technique to perform

economic dispatch in [6]. As the multiple sources of dispatch in the microgrid may operate

under three dierent reactive power control strategies, including PV, PQ and voltage droop

schemes, reactive power scheduling has been proposed by a new stochastic programming

approach that considers the uncertainty of wind farms [31]. A multi-objective function has

been proposed in [32] to maximise the reactive power reserve and voltage security margin

by mixed linear programming. The main objective of this approach was to manage the

resources operation under grid connecting mode by considering the purchase/sale power

from/to the grid and the limitations of the distributed generators capacity. The outcome

of the above mentioned works states that scheduling the generation in a microgrid that

comprises multiple sources of dispatch is a challenging process, yet it is also of signicance

to reducing the microgrid's economic and environmental costs. Furthermore, to extract

the benet of obtaining an electricity generation from multiple resources, analysing the

demand prole is found essential to help identify the gap between the generation capacity

and the expected demand. The intelligence of a microgrid has been enhanced by applying

distributed control agents to tackle the challenges of operation that includes forecasting

demand in managing the microgrid resources in [33]. Eventually, this will help identify

the amount of spinning and non-spinning reserve, which can be a solution for cutting the

generation cost in microgrids [34].

2.4 Demand forecasting in energy management systems

The substantial role of considering the demand forecasting in solving electrical systems

operational planning problems has been introduced in [35, 5, 36]. The demand forecasting

has been utilised to prepare the generation planning and storage devices utilisation for the

next day of operation in a university campus load type in [37]. Estimating demand values

has been utilised to control wind power in order to take action according to a foreseen

amount of power decit or excess in the system in [38]. Forecasting the demand has

also been found essential in the management applications of microgrids such as demand

side management and load and generation scheduling. Dynamic Demand Response and

21
generation management has been proposed for a microgrid of a residential community in

[39] to consider the stochastic elements such as stochastic load and wind power, to cut

the generation cost for the community. Demand response management in line with price,

load and renewable energy generation forecasting have been utilised in [40] to involve

community with the grid to reduce the generation cost on both the loads and the grid

sides on the concept of a Smartgrid. Oine demand side management that considers the

uncertainty of user's behaviour and weather forecasts in scheduling the energy consumption

has been proposed in [41], which has also considered the intermittence of renewable sources

generation in line with the storage devices capacity.

Among the most eective adaptive modelling methods, neuro fuzzy has been widely

applied due its high modelling eciency and low modelling complexity. Neuro fuzzy mod-

elling comprises two main stages: 1) developing the fuzzy membership functions that

represent the model by clustering the historical operating data, 2) training the developed

fuzzy membership functions with the Articial Neural Network (ANN) to normalise the

model output with the data attitude. Therefore in this chapter, we rstly cover the concept

of fuzzy modelling, and then follow it with the neuro fuzzy adaptation as follows:

2.4.1 Fuzzy modelling

Fuzzy logic is a widely utilised modelling method, as it has attracted attention throughout

the academic and industrial research sectors because of its applicability and exibility in

mimicking and embedding human decision making in a logical way. Despite its complexity

and high number of rules that can inuence performance, fuzzy modelling is still one of

the most ecient modelling techniques. It shares some modelling concepts with other

techniques, which is nesting mathematical expressions derived from a set of supplied data.

Moreover it is considered to be an eective technique for establishing a Fuzzy Inference

System (FIS) from a given nonlinear input-output data set. In fuzzy modelling, the data is

partitioned in the input space, and an optimum fuzzy rule table and membership functions

are developed. The data partitioning is performed using a data clustering method, which

partitions the input-output set of data into a set of clusters. Depending on the type of

clustering method, dierent types and numbers of clusters can be identied.

Various data clustering methods have been characterised in the literature, including;

the nearest neighborhood clustering [42], Gustafson-Kessel clustering [43], Gath-Geva clus-

tering [44], Fuzzy C-means Clustering (FCC) [45], the mountain clustering [46][47], and

Fuzzy Subtractive Clustering (FSC) [48]. Among the above mentioned clustering meth-

22
ods, FSC was nominated to cluster the historical load prole data for this work. FSC

was selected based on its reasonable computation resource requirements in comparison to

other methods that achieve the same clustering performance. The main problem with the

clustering comes from the diculty of choosing the right range of parameters, which in

turn inuences the rules that are produced. Inaccurate parameter settings undermine the

nal forecasting accuracy of the system. Appropriate fuzzy modelling parameter settings

arise from having a good understanding of the physical system and the challenges likely to

be encountered when modelling that physical system. As an example, when the number

of clusters is increased, the forecasting output will have strong alignment with the mod-

elled data. However, when the number of clusters equals the number of data elements,

the performance and operation of the clusters so developed will closely resemble the char-

acteristics of the training data , having lost the resemblance to the systems operational

characteristics. Consequently, the clusters will tend to resemble only a part of the overall

operational data. In this circumstance the modelling results in a high forecasting error. In

contrast, when the number of clusters is reasonable, the forecasting will cover the training

data regions, as well as any other types of operational data insofar as they are located

within the range of the training data. Accordingly, the forecasting will result in a much

narrower and more acceptable range of error. Thus, suitable parameter choices are the key

to successful fuzzy modelling, based on an optimised number of rules and a suitable level

of accuracy achieved in the forecasting. Suitable modelling parameters can be identied

by an analysis of the modelled system's operational history. While having experience with

fuzzy modelling can be helpful, satisfactory output tuning can be achieved by trial and

error in most of the modelling cases.

The major limitations of the typical fuzzy modelling approaches are: 1) the limited

forecasting scope within the supplied operational history data range, and 2) the system

ignores any external or unexpected changes. From the literature, it is clear that essential

knowledge points for enhancing the forecasting output could be added by fuzzy logic-based

systems, especially with highly non-linear systems. Support vector regression has proved

to be useful when dealing with non-linear forecasting problems in recent years, and it

has been utilised to nest supporting knowledge for forecasting [49]. It is anticipated that

the introduction of self tuning systems will improve the forecasting accuracy and add the

missing knowledge to the proposed models. In the development of the demand forecast-

ing model, the adaptation of the self-tuning fuzzy-ANFIS as a feature of the (predened

knowledge) forecasting model is proposed to enhance the overall forecasting accuracy. A

23
failure forecasting approach, which integrates a fuzzy logic-based adaptive inference system

with the learning ability of a neural network to generate knowledge in the form of a fuzzy

rule base, has been developed in [50], where it has been proved that utilising this structure

yields reducing the forecasting error.

When comparing fuzzy modelling with Articial Neural Networks (ANN), we conclude

that in order to select the right modelling method, it is crucial to consider the type and the

size of the system, the amount of historical operational data that is available, and also the

computation resources that will be required. An examination of the character and the size

of the modelled case study revealed that fuzzy modelling is a suitable mechanism for the

modelling process. More details about the case study and the data analysis are provided

in the case study section in Chapter 4 . Full details about the fuzzy modelling process,

including the clustering method that was utilised, are also provided in subsection 2.4.2.

2.4.2 Fuzzy subtractive clustering method

The Fuzzy Subtractive Clustering Method (FSCM) [48] has been utilised to cluster the

historical operational data in the modelling process for this work. The data supplied was

tested under the condition that it has the highest density among the tested individuals.

Every individual datum was considered to be a candidate for the cluster centering. The

individual density was evaluated as follows:

(2.1)
Pn 2
Pi = j=1 e−αkxi −xj k

where

4
α= (2.2)
ra2

and ra is a positive constant that represents the radius of the data neighbourhood. In

2.1, Pi is the density of each examined point, xi is the point that is examined at the time

of measuring the density pi of the same point, and xj refers to the other data within the

neighbourhood. After nding the density for each data point, the highest density point

will be selected to be the rst cluster centre. The data density for a specic cluster centre

candidate is evaluated from the number of nearer individuals that contribute to the cluster

24
centre. The highest density identied becomes the rst cluster centre. The cluster size is

decided when the FSCM parameters have been set to cover a range of data individuals

in the cluster's neighborhood. The radius ra , which is also referred to as the Range of

Inuence (ROI), denes the range of the neighborhood for the cluster's extraction. Each

of the clusters developed in this way becomes the base of a fuzzy rule that describes the

system's attitude; the number of these clusters being equal to the number of fuzzy rules

in the modelled network. When the rst cluster centre is found, the next highest density

is evaluated. Let the new investigated cluster centre be xi , and Pi be its density measure.

When every individual datum is xc1 , the next cluster centre is identied as follows:

2
Pi = Pi − Pc1 e−βkxi −xc1 k (2.3)

where Pc1 , is the next density point to be examined, c1 is the next data point to be

examined, and

4
β= (2.4)
rb2

rb = 1.5ra (2.5)

where rb is a constant, which inuences the density measure and the value of which

is established from the previous experience of data clustering, and from trial and error.

Usually, rb is larger than ra to avoid closely placed clusters. In this research work, the

value of rb was set to 1.5ra reecting the values employed in the literature [48], and the

value of ra was determined from our previous experience with data clustering. In the cases

we investigated, dierent values were applied, depending on the nature of the problem. It

was observed that the number of membership functions is a direct outcome of the value

of ROI, thus the ROI value controls the complexity of the developed network. Table 2.1

illustrates the full details concerning ra settings for the cases that we investigated.

The next stage is to repeat the above estimation process to identify other cluster centres.

The process of identifying clusters is repeated until the value of the newly identied density

is equal to or less than 0.15 of the highest identied density. More information about FSCM

parameter details is found in [48].

The data clusters that have been identied can then be easily utilised as fuzzy rule

25
centres in the zero-order Sugeno fuzzy models. When an individual datum is located

within the cluster range, a membership function between that particular datum and its

cluster centre is derived. Data aliation to the cluster centres is derived as follows:

kxi −xc k2
−( )
µi = e (ra /2)2 (2.6)

where xi is the cluster centre and xc is the input set of data.

At the end of the clustering process, a fuzzy reasoning system will be developed and

trained with the input/output data set. The Adaptive Neuro Fuzzy Inference System

(ANFIS) is used as a structure for the training phase of the development of the targeted

fuzzy models in this work. In the nal phase, the model is veried using the Mean Absolute

Error (MAE) to evaluate the nal modelling performance for the specic data set. The

complete modelling process is illustrated in Figure 2.1.

2.4.3 Adaptive Neuro Fuzzy Inference System (ANFIS)

ANFIS was selected to be our modelling method in this work; largely because of its ad-

vanced learning mechanism and its minimal training time. In contrast with other learning

systems such as the Articial Neural Network (ANN), ANFIS has a hybrid learning mech-

anism which provides signicantly faster convergence. The ANFIS structure was rstly

proposed by [51], where other models of ANFIS were proposed by [52] and [53]. Figure 2.2

illustrates the ANFIS structure with its learning mechanism,

where f is the output of the net, x and y are the inputs to this net. The weights of

layer 3 are represented by (w

¯1 ,w̄2 ), and the weights of layer 4 are represented by (w¯1 f1 ,w¯2
f2 ), where the used rules of Sugeno fuzzy in this model are expressed in the following form:
If x is A1 andy is B1 THEN f1 = p1 x + q1 y + r1
If x is A2 andy is B2 THEN f2 = p2 x + q2 y + r2
Where (p1 ,q1 ,r1 ) are the parameters that have been determined, and are referred to as

the consequent parameters. More details about ANFIS parameters can be found in [51]. In

conventional neuro-fuzzy networks, the Back-Propagation algorithm is used to adjust the

network parameters, while in ANFIS the adjusting mechanism is performed by the Hybrid

Learning Algorithm (HLA). The HLA is a composite of two identication methods: the

least-squares method, to identify consequent parameters for the forward pass in layer 4, and

the Back-Propagation method for the backward pass, to identify the premise parameters

26
 Figure 2.1: Selecting suitable FSCM parameters in ANFIS modelling

by the gradient descent in layer 2. This combination achieves faster convergence than that

of the original Back-Propagation method. Table 2.1 illustrates the hybrid learning passes

with their identied parameters.

This combination has been utilised for modelling in various types of applications under

dierent modelling mechanisms. ANFIS has been utilised as part of a hybrid modelling

mechanism to cope with the uncertainty of process operational conditions in [54, 55, 56],

and also in a modelling method that incorporates one step ahead concept into ANFIS,

which has been built to a fusion ANFIS model to enhance the forecasting for electric-

Table 2.1: Hybrid learning passes directions [1]

Parameters Forward pass Back pass

Premise Parameters Fixed Gradient Descent

Consequent Parameters Least Square Estimator Fixed
Signals Node Outputs Error Signals

27
 Figure 2.2: ANFIS structure

ity loads by adapting the forecasting equation in [57]. It has been proved that ANFIS

was successful in being part of hybrid modelling mechanisms, which allows having time-

varying characteristics-based adaptive forecasting. ANFIS modelling structure has also

been utilised as a base for developing adaptive models. Adaptive models have been intro-

duced to cover long and short term demand forecasting in power systems. In [58], adaptive

fuzzy rule-based systems, in a top-down modelling framework, for a long term forecasting

of daily time series in the Neural forecasting competition has been presented. [59] has pro-

posed self-adaptive technique that addresses the shortcomings of the adaptive evolutionary

techniques for time series forecasting in non-static environments.

In the next stage of the modelling procedure, we will discuss how to stabilise the newly

developed models, and how to improve their forecasting accuracy through the addition

of the Self Tuning Fuzzy System. To increase the accuracy of the forecasting and cope

with the unexpected behaviours, adaptive forecasting mechanism was necessary in build-

ing the microgrids energy management systems. An automatic parameter tuning based

on the monitored attitude of the model has been proposed in [60] to perform an adaptive

forecasting for next day electricity price and load in a microgrid. A new self-adaptive

technique that addresses shortcomings identied during recent studies investigating adap-

tive evolutionary techniques for time series forecasting has been presented in [59]. An

adaptive model based on wavelet transform and adaptive learning time series models has

been proposed in [61] to forecast the daily energy consumption and generation price. By

analysing the above mentioned works, it has been proved that there is a need to include

28
the demand in building energy management systems, and an intelligent method for fore-

casting is needed in this work. Investigating ANFIS role in performing demand forecasting

has resulted in attractive outcomes that nominate ANFIS for modelling the demand and

integrating the model in the proposed management system. Due to the complexity of the

modelled demand proles, an adaptive mechanism has also been proved to be essential for

more accurate forecasting as part of the proposed interactive management system.

2.5 Microgrids operation performance optimisation

Managing the microgrid operation is the main part of this work and the core of analysis

required to achieve low cost generation in the microgrid. Hence this stage of analysis and

processing is subject to forecasting the demand and obtaining the generation instanta-

neous cost and availability of the distributed generators. In literature, this problem has

been targeted by researchers and engineers introducing new technologies and methods for

managing the microgrid resources under various types of operation scenarios. The power

dispatch in an islanded microgrid with solar panels and wind turbines has been managed in

[62] by modelling the generation cost and correlating the cost function to the investment

cost, lifetime and the uctuant energy of wind and solar resources. An optimal energy

management system that aims at providing sustainable low environmental cost generation

through controlling the generators based on utilising the demand forecasts and the amount

of renewable energy sources generation and the storage devices data has been proposed in

[63], which has also covered the charging/discharging process for the distributed storage

devices in the microgrid. Where charging/discharging the storage devices is a challenging

process in providing low cost generation for microgrids, a set of control strategies have

been tested to dene the charging or discharging rates of the batteries in [64]. Microgrids

operation management design and testing using Fuzzy Logic Energy Management System

(FLEMS) has been proposed in [65, 66] to manage the electricity, transport and water

needs in an autonomous polygeneration microgrid. Ant colony optimisation method has

been proposed to manage the generation in a typical microgrid by considering the complex

constraints such as generation availability and cost by nding Pareto optimal dispatch

solutions in [34].

Reducing the generation cost in microgrids, in most of the cases, is the art of re-

ducing the generators fuel consumption by matching the amount of generation with the

demand. Therefore, controlling the fuel consumption in a typical microgrid based on an

optimised operation of the reciprocating gas engines, a combined heat and power plant, a

29
photovoltaic array and a wind generator in a typical microgrid has been proposed in [67].

Fuel consumption reduction has also been introduced via planning strategies based energy

management system that incorporates the optimum placing and utilisation for distributed

generators in [68]. Apart from that, Genetic Algorithms based optimisation model with

cost and reliability cost functions that works based on a nested strategy and a limited

information objective functions has been proposed in [69], to provide low cost genera-

tion decisions based on Pareto optimal trade-o curves between cost and reliability. A

niching evolutionary algorithm (NEA)-based optimisation procedure has been proposed

to minimise the generation cost in a medium level voltage microgrid that encompasses

distributed renewable energy sources and storage devices in [70]. Basically this problem

is dened as an optimisation problem that requires a quick solution in order to match the

generation with the dynamic cost factors variation. On this purpose, Linear Programming

(LP) was identied as an ideal method for representing the problem and generating initial

results for further computation stages within our management system. To provide a linear

programming-based solution for a complex problem, four main components have to be

considered:

1. The decision variables, which represent the amount of resources to be determined.

2. The developed rules, which represent the backbone of the decision strategies.

3. The type of objective functions (single or multi), which dene the size and direction

of the solution (achieving maximum or minimum).

4. The constraints and the bonds, which control the solution limits and thus reduce the

time and computation resources needed to nd a solution.

Generally, linear programming is dened as follows:

Let X = {x1 , x2 , . . . , xn } be the set decision variables and C = {c1 , c2 , . . . cn } denote

the cost of one unit of decision variables, respectively. The value of ci xi denotes the cost

of using resource i. The general form of an LP model is described as:

ˆ The objective function:

n
X
Minimise or Maximise: F = ci .xi (2.7)
i=1

ˆ Subject to constraints:

 Equalities:

30
 a11 x1 + a12 x2 + · · · + a1n xn = b1
a21 x1 + a22 x2 + · · · + a2n xn = b2
(2.8)
. . .
. . .
. . .

am1 x1 + am2 x2 + · · · + amn xn = bm

 Inequalities:

a011 x1 + a012 x2 + · · · + a01n xn ≤ (≥) b01

a021 x1 + a022 x2 + · · · + a02n xn ≤ (≥) b02
(2.9)
. . .
. . .
. . .

a0k1 x1 + a0k2 x2 + · · · + a0kn xn ≤ (≥) b0k

ˆ Bounds:

xi ≥ LB (Lower Bound), j = 1, . . . , n (2.10)

xi ≤ U B (Upper Bound), j = 1, . . . , n (2.11)

A linear programming method, Particle Swarm Optimisation (PSO), has been pro-

posed for producing the nal optimum results due to its convergence speed, ro-

bustness and simplicity of implementation [71]. PSO is a well known optimisation

method that has been increasingly utilised in solving optimisation problems. In power

systems, PSO has been applied to provide solutions for the operation optimisation

problems [72, 73, 74]. It has also been proposed to achieve optimal dispatch for loads

and generators in a typical microgrid in [75]. PSO, which has been inspired by the

ocking social behaviour of birds, explores the best location in the swarm (solutions

matrix) by updating its place and velocity. By utilising the dened LP problem rep-

resentation by PSO, each solution candidate of the rules called  particle , represents

a member of the solutions set i.e., a swarm of candidates. Finding optimum solution

is performed by substituting the candidate's best positions in the cost function [76].

A guided update mechanism is applied to the best achieved solution in the swarm

to investigate the possibilities of nding a better solution around the best solution.

The update mechanism is illustrated in Figure 2.3.

The velocity equations that represent the update for the selected values are illustrated as

31
 Figure 2.3: PSO candidate's velocity updates mechanism

follow:

k+1 k k k
Vi,j = w.Vi,j + (Pbest.i,j − Xi,j )C1 .rand1 + (Gkbest.i,j − Xi,j
k
)C2 .rand2 (2.12)

k+1 k k+1
Xi,j = Xi,j + Vi,j (2.13)

k+1
where Vi,j is the updated velocity for the input variable j at iteration i. w is the

inertia weight, rand1 and rand2 are random generated numbers bewteeen 0 and 1, and C1
and C2 are the learning factors. The update process is repeated based on the predened

number of iterations i, or based on a predened targeted cost value.

2.6 Multi-agent based microgrid management systems

The need for using the computational methods in performing high level control and man-

agement for the microgrids with the Smartgrid has been detailed in [77] as the need

for multiple levels of distributed intelligence was claried, along with the potentials and

promises of the computational intelligence to realise an intelligent Smartgrid. This clar-

ies the role of distributed intelligent systems in achieving high level performance in the

next generation of power systems that includes the smart distributed power system that

is called the Smartgrid. Ultimately, Multi-Agent System (MAS) technology has proved

its eciency in achieving satisfactory performance in complex power systems, especially in

the control and management side.

32
 Figure 2.4: Agent basic inter-connection structure

2.6.1 Multi-Agent System (MAS) technology

The term Agent in computer science refers to a software component that behaves in a

human manner to perform tasks autonomously. What make the agent unique from other

software components are mainly: 1) its ability to perform tasks independently, or being

a part of environment that performs tasks based on a sequenced operation mechanism

and data exchange with other local or external agents. 2) its ability to sense and react

to changes in the environment based on a predened timing setting and a developed rule

base. 3) its ability to learn and establish a new rule base to cope with its environmental

changes.

Depending on the type and the reason why the agent is developed, dierent operation

mechanisms can be implemented to perform the tasks. Therefore, predened settings

for the type and the size of the inter-connectivity messages based on a local predened

vocabulary (Ontologies) can be implemented to serve and report coding and decoding

among the developed agents. Predened settings for the behaviours to model a specic

attitude can also be implemented within the agent. Eventually, agent's architecture varies

depending on the task that it is built for. However, the agent's most common architecture

details are the sensing and reacting parts, in addition to the processing part. Figure 2.4

illustrates the basic agent inter-connection structure with external information source.

Building a bigger software that incorporates common or additional objectives with

other agents yields building a software network called: Multi-Agent System (MAS), which

can perform more complicated tasks. Performing a centralised and decentralised decision

making among distributed systems, in line with distributed sensors and actuators needs a

standardised communication protocol to allow compatibility and ease of application. Even-

tually, MAS has been widely integrated with various types of applications such as robotics

33
 Figure 2.5: Distributed agents and platforms installation using JADE [2]

and manufacturing, electronic trade and negotiation, health care systems, information

management, trac management and telecommunication systems.

Depending on the MAS implementation tool, various design and communication stan-

dards applied in agents development were found in literature. Since this work was im-

plemented using JAVA Agent DEvelopment Framework (JADE), the next subsection will

detail this tool along with its standards compliance details.

2.6.2 JAVA Agent DEvelopment Framework (JADE)

Java Agent DEvelopment Framework (JADE) is a software-based framework, which is

developed using Java programming language [2]. JADE is an easy way to develop MAS

using a middle ware that complies with the Foundation for Intelligent Physical Agents

(FIPA) [78] design specications with a set of graphical user interface tools. FIPA is an

IEEE standards organisation community that promotes developing agent-based technology

standards. JADE allows the agents to work on dierent computers at the same time, which

will allow decentralised monitoring and easy control implementation. Figure 2.5 illustrates

the structure of distributed agents and platforms installation on dierent separate machines

using JADE.

34
 As it is shown in Figure 2.5, each main agents platform container comes with Agent

Management System agent (AMS), to monitor the platform, and a Directory Facilita-

tor (DF) agent to create the Yellow Pages services to allow services registration for the

participant agents. Including these two agents in the agents platform is a FIPA-agent

management standard, which is needed in this work to implement the distributed energy

management agents.

Apart from that, the following functionalities are enabled in JADE framework to allow

developers embed their proposed agent-based strategies: 1) It allows running same plat-

form's agents on dierent remote machine by creating Agent IDentier (AID) to address

the agents, transparently monitor them and communicate with them. 2) Its agents frame-

work operation and design complies FIPA standards. 3) It has the ability to create White

Pages and Yellow Pages services within the platform to help addressing agents' services

and status. 4) It has the ability to create and manage the life-cycle for its agents based

on the tasks that agents are developed for. 5) It supports agents mobility to allow agents

migration between the process and the machines when needed. 6) It supports subscrip-

tion mechanism, which allows external applications to subscribe in the platform and be

addressed and classied by local agents. 7) It has the ability to develop content languages

and Ontologies, as this will increase the security applied to the data transferred between

the agents and the environment. 8) It permits the web applications integration with the

agents to allow remote monitoring and control. 9) It supports Graphical User Interfacing

(GUI) to ease agents monitoring and debugging. 10) Finally, it allows for Kernel extension

to facilitate extending framework functionalities. These functionalities were found viable

tools for implementing management systems using MAS technology and JADE.

The architecture of context aware FIPA [78] compliant MAS for the lightweight devices

called SAGE-Lite has been presented in [79] to illustrate the lightweight devices and their

communications methods in implementing a framework for applications like business or

e-commerce using MAS. The proposed architecture has been enriched with many ideas,

which will help in future research and development of the framework. Consequently in the

management side of the microgrids, an agent-based intelligent energy management system

has been proposed to facilitate power trading among microgrids and allow customers to

participate in demand response. The proposed intelligence comprises demand response,

diversity in electricity consumption patterns of the customers and availability of power

from distributed generators as the vital means in managing power in the system. A smart

power restoration mechanism was developed using multi agent technology in [80] that oers

35
the prototype of a monitoring platform that exploits the agent approach for computation

capability improvement. An interesting concept where microgrids and other production

or consumption units form a virtual power plant control has been presented in [81]. MAS

technology role has been evaluated through implementing electronics markets on a real test

sites using JADE in [82, 83]. JADE has been utilised in developing intelligent management

frameworks for managing microgrid resources over dierent types of management strate-

gies. An intelligent management framework has been implemented using JADE in [84]

to allow active participation for the microgrid distributed storage devices in the demand

side management, and eventually in the market. The power trading among microgrids has

been implemented through an agent-based intelligent energy management system, devel-

oped using JADE in [85], to allow customers to participate in demand response that can

cope with all type of demand patterns in line with the intermittence of renewable energy

source generation.

To enable microgrids to operate in parallel to the grid, as an autonomous power system,

a robust control for the local voltage and frequency, and network protection are needed

to facilitate demand side management and resynchronization. Various types of microgrids

around the world (North America, Europe and Asia) with their test systems and con-

trol options have been summarised in a table comparing and contrasting the technologies

utilised in performing smart microgrids in [86]. MAS have been developed and imple-

mented to control power on microgrid by [87], which has been proposed to be compatible

with IEEE standard on FIPA IP-based network. This work has been implemented using

simulation, where the results showed that MAS was successful in achieving seamless transi-

tion from the grid connection to islanding mode when an upstream outage is detected, and

it also showed the possibilities of the MAS in managing microgrids. JADE framework has

been implemented on control systems to achieve decentralized control objectives. It has

been implemented on microgrids control to achieve multi-tasking on substation automation

and controls. The developed agents control strategies were implemented using JADE and

DIgSILENT. A decentralized control-based scheme has been introduced to handle the large

number of renewable components by implementing distributed control system using JADE

as the platform for agent communications, as well as developing customisable agents for

specic microgrid requirements such as ancillary services, power trading and negotiation

and network security [88]. A reinforcement learning method to minimise the generation

cost and also maintaining power balance in microgrids under connecting mode has been

presented by [89] to build an objective function that targets achieving optimal cost and

36
control for microgrids. Hence this method has been proposed to include dynamic hierarchi-

cal reinforcement learning to establish optimal policy based on the microgrids operational

scenarios, where the simulation results have been presented using JADE.

JADE-based agent development characteristics also allow for markets modelling, which

has also been an active research topic in literature. Virtual markets have been simulated

using MAS based architecture to enable customers to participate in demand response

and trading power with the grid using intelligent agents and JADE in [90]. MAS have

been proposed as distributed intelligent systems in [91] to schedule the operation of the

distributed resources to the local demand in the microgrid in line with the demand that

incurred from selling power to the grid. A priority index has been proposed for customers

participating in the market based on frequency and size of load in the demand response.

In order to validate the proposed method, a case study with two interconnected microgrids

is simulated. Based on extensive simulation results of the system developed using JADE,

it has been found that multi-agent based demand response is successful in reducing the

system peak in addition to cost benet for the customers with high priority index [85].

By identifying the role of JADE in modelling the markets attitude, we found it attractive

to investigate integrating the microgrid market participation after identifying the critical

research points for this issue.

2.7 Microgrids market participation and pricing mechanisms

Apart from optimising the operation of the microgrid through controlling the distributed

generators around the microgrid, there is another possibility to reduce the microgrid gen-

eration cost, which is the optimal utilisation for the reserve through market participation

under dierent policies. Integrating the microgrid with electricity market is a timely re-

search topic in the power engineering. The role of integrating microgrids with the electricity

market based on the sustainability index has been investigated in a large scale studies to

include the North West European market in [92]. In [93], the inclusion of exible CHP

production in balancing the electricity in smart energy system as part of the Smartgrid in
the Danish electricity market has been illustrated. In addition, the eect of the microgrid

market participation has been tested with a smaller scale, as it has been described in [94]

by implementing a central controllers that runs based on an optimal power ow under

dierent market pricing policies. The structure of the microgrid and the characteristics

of MAS have been outlined in [95]. A novel approach called multi agent reinforcement

learning has been introduced in [96] to increase the intelligence and the eciency of the

37
microgrid.

Furthermore, microgrids market participation has been addressed in literature with

various types of methods for active and reactive power. Microgrids reactive power market

participation has been introduced by [97] by bidding to the VAR market that operates

based on an optimal price sensitive power ow that controls the day ahead operation of

the distributed generators. In active power, the generation reserve in a typical microgrid

has been utilised through an optimal price sensitive dispatch and oered to the market by

[75]. Microgrid market participation may involve optimising the resources operation be-

fore generating a benecial bids to the market, though involving optimisation techniques

has been essential. The microgrid market participation with a hybrid market model has

been compared with the pool electricity market by optimizing the production of local DGs

and power exchanges with the main distribution grid in [98]. A multi-period optimisation

model for connecting a microgrid comprising a multiple sources of dispatch with market

under dierent operational scenarios has been presented in [99]. PSO has been proposed

to optimise the operation of the microgrid based on the capital and operating costs, and

interconnect the microgrid with the market as an independent generation company which

sells electricity to the main grid under dynamic selling price and dierent market policies

in [100]. Markets participation has also been subject to load control and demand response

in advance management strategies. Fully connected Neuron networks combined with opti-

mal operation for the microgrid have been implemented as an Energy Management System

that comprises adaptive tuning using Genetic Algorithms and Fuzzy Clustering to perform

Demand Side Management and Active Management Schemes on High Voltage (HV) net-

work in [101]. A special case for the development of Dynamic Stochastic Optimal Power

Flow technology has been shown in this work to be needed in Smartgrid design. MAS

was developed in context of intelligent distributed autonomous power systems using Mat-

lab and Zeus agent development toolkit [102]. Infotility's GridAgents— software [103] has

been used for large-scale integration of distributed energy and renewable energy resources.

The developed agents software was proposed to achieve microgrid management, intelligent

load control and smart charging applications. Finally MAS-based two level management

architecture has been proposed to solve the microgrid optimal power ow problems in

[104] to perform active market participation for the distributed generators under dierent

operational scenarios. It has been indicated in this Section how importance of each of

the listed technologies are imposing signicance to managing the resources in microgrid,

and eventually helping microgrids to be part of the Smartgrid under common operational

38
policies and communication protocols.

2.8 Microgrids and Smartgrid

Integrating the microgrids with the Smartgrid is the challenge of incorporating advanced

management and control strategies in the microgrid to allow safe and reliable connection

that can cope with the requirement and operation standards of the Smartgrid. More

specically, connecting the microgrids to the Smartgrid targets three major goals: market

participation, improving the power quality and increasing customer role in the grid power

generation and stability. The capabilities of MAS technology for connecting microgrids

to Smartgrid have been presented in [105], while this topic has been largely discussed by

researchers to develop new implementation methods and simulation tools.

By investigating the role of managing microgrids in connecting microgrids with Smartgrid,

it has been found that asset management techniques have been used to develop an au-

tomation system for medium voltage network in [106] to show the application of a new

automation concept that results in network reliability and enables the temporisation of

asset replacement programs for Smartgrid. Microgrid management systems development

using agent based technologies and application to the eective management of generation

and storage devices has been described in [14]. Active Distribution networks with full inte-

gration of Demand and Distributed energy RESourceS (ADDRESS), the European project

that aims to deliver a comprehensive commercial framework for the active demand devel-

opment for the Smartgrid has been presented in [107] to investigate the development of

the participation of domestic and small commercial consumers in the Smartgrid energy

markets. ADDRESS represented a unique and timely opportunity to coordinate and com-

bine the expertise and resources of partners from all Europe. A new type of smart demand

side technology to allow the customer to have a more dynamic role in improving the grid

voltage control, as this technology can be provided by thermostatically controlled loads as

well as other types of loads, has been introduced by [108]. This technology has been applied

on distribution systems with large composition of induction motors. Ultimately its mod-

elling technique, discussion and simulation results proved the eectiveness of the demand

as voltage controlled reserve technology for short-term voltage control under Smartgrid
connection.

Furthermore, a framework for Smartgird operation that comprises control objectives

to incorporate the participation of the distributed generators and exible AC transmissions

technologies to the Smartgrid has been applied to achieve voltage and stability control

39
for microgrids in [109]. In this framework, the control strategies have been developed in

such a way, that the voltage controller is operated by a global coordinated strategy, and

stability control is applied by a decentralized controller. Intelligent control strategies have

been applied to enhance stability in all modes of operation of the microgrid by [110], to

show that the stability in dierent modes (Parallel, Transient, and Autonomous modes)

has been improved. Agent technology has also been utilised in this role, where a control

structure has been proposed using agent-based technology to be implemented and tested

on a pilot microgrid of Kythnos Island, Greece. The developed autonomous agents have

been used to control the power grid while being connected to Smartgrid based on several

conversations and scenarios among agents [111]. Agent-based control framework has been

proposed to control the distributed energy resources microgrids [112, 113, 114] by consid-

ering Smartgrid connection, while a control scheme has been implemented with MAS for

the operation of the microgrids in [115].

As stated earlier in this section, the Smartgrid research community has investigated

the challenges of achieving active participation of the distributed generation with energy

markets. In this eld particularly, increasing energy eciency, under Smartgrid connec-

tion, by introducing a market mechanism that facilitates the ecient matching of energy

demand and supply through a double auction mechanism for the allocation and pricing of

energy resources has been presented in [116]. Another research has investigated the use of

automatic bids generation for allocating resources for demand in their various applications

in microgrids for ecient Smartgrid connection in [117], where an automatic bidding gen-
erating strategies were presented to use reinforcement learning techniques in generating

bids. Bidding language for communicating consumer has been provided using MAS tech-

nology as a simulation tool for performing specic control and management applications

for microgrids Smartgrid connection. PowerMatcher, a Smartgrid market-based control

concept for supply and demand matching in electricity networks has been applied on the

microgrid to achieve supply and demand matching with high share of distributed genera-

tions in [118]. A number of results from two eld tests performed with the PowerMatcher

concept has been discussed in [119].

The identied technologies in the this section clarify the possibilities and the limitations

of simulating a proposed management strategy that operates microgrids resources under

Smartgird connection. This section has also stated the level of research conducted on

performing smart management systems for microgrids under Smartgrid connection, which
that by its turn helps identifying the critical research questions in this eld.

40
2.9 Closing remarks

This chapter outlined the most up to date methods and technologies utilised to perform

smart management strategies to optimise microgrids operation. By studying the opera-

tional scenarios and the characteristics of the microgrids connection with the utility grid,

further details about the operation challenges were identied, and therefore, the literature

exploration converged towards investigating the technical challenges relating the smart

utilisation of microgrids' resources to reduce their economic and environmental costs. To

operate the microgrid in parallel with the Smartgrid, other implementation challenges

have been highlighted, thus new automation technologies have also been identied to cope

with these challenges. Identifying the role of demand forecasting in performing resources

management has been stated with the needs of involving advance methods for more accu-

rate forecasting in conjunction with the high uncertainties of users' behaviour and weather

conditions. Utilising the demand value in line with other microgrid operational details

will help analysing the operation, and then nding a suitable solution based on a prede-

ned rule base. The literature has highlighted this problem and also identied the role of

various types of optimisation techniques in solving this problem. However, this part has

included dierent implementation methods depending on the complexity of the operational

scenarios, like having markets participation, which would encompass more cost eective

parameters that need to be considered in the management decisions. To combine these

management decision stages in a comprehensive management framework, MAS technology

has been found a successful candidate for implementing energy management systems. MAS

technology has also been targeted in this research to test the possibility of implementing

a specic management strategy that relies on the local demand prole and the dynamic

generation costs. From the literature it has also been noticed, that the researchers have

investigated the impact of using MAS technology on the microgrids management and con-

trol as well. The highlighted challenges in performing reliable management on microgrids

resources were as follow: 1) the operation reliability, since microgrid can be connected or

disconnected to or from the utility grid depending on the operational scenarios, its power

supply should always be reliable, 2) the need for obtaining instantaneous information about

the demand prole in the microgrid, as this will help following the demand patterns instan-

taneously, which will eventually match the generation capacity with the demand prole,

3) the need for scheduling resources to reduce their operating maintenance and generation

costs, especially when their cost is dynamic and dependent on other local and/or external

cost factors, 4) the possibilities of involving the microgrid with the market depending on

41
the oer and the demand in the market itself, 5) nally, it has been found that there is a

need for developing comprehensive commercial framework which has the ability to achieve

economic and market participant energy production based on running multi-stage analysis

and decision making system. Ultimately, articial intelligence algorithms are needed to be

applied for better framework operation, which can be implemented as distributed intelli-

gent systems for optimising the microgrid operation. In conclusion, this thesis is targeting

all of the listed challenges by introducing a system developed to make optimum generation

decisions for the microgrid's resources.

42
Chapter 3

Load Forecasting Model

Estimating the amount of electrical demand is the key to minimising generation cost in

power systems. In this work, utilising the demand forecasts is essential in our planning to

maximise the utilisation of the generated power, and also to maximise the benet of hav-

ing variable generation cost in line with the intermittent generation from renewable energy

sources. It is therefore important to introduce a novel demand forecasting technique, that

can cope with the challenges encountered with the dynamic operational conditions of mi-

crogrids, prior to introducing other intelligent modules in the proposed EMS. Ultimately,

this chapter details the development of an adaptive demand forecasting model. This model

aims to generate adaptive demand forecasting, that reects the real-time demand values

in the forecasts. The proposed demand forecasting model was implemented to provide

a 30-minutes in-advance forecast of the electricity demand on the Joondalup campus of

Edith Cowan University (ECU). The proposed modelling mechanism aims to utilise the

historical operational data to build a base model for forecasting, and then adjusts the base

model forecast in response to measured changes in the load prole. These measurements

are taken on the real-time load. The model requires three stages of computation in order

to perform the forecasting, which are as follows: 1) forecasting demand, 2) controlling the

forecasting accuracy, and 3) adapting the results to the real-time demand values. In the

forecasting stage, the Adaptive Neural Fuzzy Inference System (ANFIS) was used to build

the model by utilising the historical operational data. Concurrently, a fuzzy system for

tuning was used to control the model's output. For the forecasting output adaptation, an

additional fuzzy system was used to consider the instantaneous load prole in the fore-

casting. Successfully implementing the third fuzzy system for the adaptation stage was

challenging because of the unexpected user behaviour on the campus. As a consequence,

an Articial Neural Network (ANN) based optimisation technique was proposed to tune

the adaptation parameters; based on the identied real-time load prole at time of fore-

43
casting. In this chapter the modelling strategy is discussed, and we organise it as follows.

Sections 6.1 and 5.2 give a historical background about modelling trends with respect to

the work of other researchers in modelling the load proles of power systems. Section 3.3

introduces the problems encountered with forecasting the demand in microgrids, and also

highlights the objectives aimed by the proposed modelling method. Section 3.4 lists the

research assumptions considered in the modelling procedure. Section 3.5 covers the pro-

posed modelling procedure, including some verication tests for the methods which were

utilised. Section 3.6 provides a discussion of the results, and in Section 3.7 we evaluate the

ndings and the outcomes of this work, and nominate extended works within this research

eld.

3.1 Background

Modern power generation management systems may include the Smartgrid applications

to perform more complicated tasks such as power trading, demand response, and resources

scheduling. In all of the above mentioned applications, forecasting the load prole, in both

short and long terms of operation, is essential. Providing accurate forecasts of the load

prole is a critical research target for mathematicians and engineers in this area.

In modern automation, adaptability has become crucial factor when implementing

smart applications. The addition of adaptability enables the smart applications to more

closely resemble the human sense of adaptive thinking. Including an adaptive demand

forecasting within a power generation management system would result in an interactive

application that resembles the human behaviour in making critical decisions. Investigat-

ing the implication of including an adaptive demand forecasting model within the power

generation manageemnt system is of signicance to the research community due to the ne-

cessity to obtain accurate forecasts prior to making decisions concerning managing power

systems.

Articial Neural Networks (ANN) are typically utilised when implementing adaptive

systems; however, self tuning and adaptive algorithms are not restricted to ANN, as they

can also be implemented through a range of optimisation techniques, as well as Fuzzy Logic

(FL). The latter has been proved to be one of the most ecient modelling techniques due

to its ability to reect the human sense in making decisions on a particular problem. The

specic implementation of the tuning mechanism is governed by the nature of the problem

to be solved, and/or the system which is to undertake the processing . The tuner and

main systems may share the same input parameters, or they may receive two dierent

44
types of inputs from external sources, depending on the nature of the operation. Self-

tuning systems have practically unlimited applications, and to date they have been widely

utilised in various academic and industrial applications. Eventually, Self Tuning Fuzzy

System (STFS) is hypothesised to achieve a satisfactory modelling performance when it is

attached to a forecasting model.

In essence a STFS is an on-line adaptive output fuzzy system, where the output is

changed depending on the type of input and predened knowledge rules. In general terms,

a fuzzy system is called tunable when any of its parameters (input/output scaling factors,

membership functions shape and type, or rules) are changeable on an instananeous basis.

A tunable fuzzy system is a combination of a general and a tuner Fuzzy Inference System

(FIS), where the tuner FIS tunes the general (main) FIS parameters. Although at times

both systems have the same input parameters, they still perform independent roles. The

reason STFS is used in modelling is to enable short term forecasting to be conducted, which

improves the overall system exibility by adding the safe forecasting estimation knowledge

to the forecasting; either by adding extra knowledge to the forecasting, or by adapting the

forecasting to the external eects.

The case study that we have modelled has highly non-linear characteristics, so develop-

ing a model for high precision forecasting is a major challenge. Hence it becomes necessary

to focus on the model's forecasting accuracy in order to consider its weak points. By con-

sidering the electricity demand in the targeted case study, modelling knowledge could be

added regardless of whether or not that knowledge is initially available within the supplied

operation data.

Through the use of the STFS, so of the gaps in the knowledge can be added back to the

operational history data. For such systems, a design with external input parameters for

tuning the main fuzzy model output, based on a knowledge base, could be implemented.

In this chapter, the real-time demand change measure was utilised in order to investigate

the FIS' ability to adapt the forecasting output to the actual demand change. The main

fuzzy system's input parameters were also used to tune the forecasting, based on a knowl-

edge base system. Similarly, the tuning part may use alternative mechanisms, e.g. rules,

membership functions, or output scale tuning. The Weights Adjusting Method (WAM), is

the mechanism for adjusting the output of the main system, enabling the system to adapt

to changing circumstances. WAM adjusts the weights of the output of the main system

with its tuner, according to the level of forecasting needed. By giving consideration to

the tuner's fuzzy rule base, a suitable WAM can be derived. For simplicity, an output

45
 Figure 3.1: The tuning fuzzy system

scale adaptation was applied. Figure 3.1 illustrates the general structure of the STFS.

The full design details of the particular STFS being employed are explained in subsection

3.5.2, where the results of the implementation will be discussed from the perspective of

improving both forecasting and adaptation performance.

More specically, STFS will be implemented in two computation stages within these

models: 1) in improving ANFIS forecasting accuracy by applying the FeedForward Tuning

System (FFTFS). 2) in adapting the demand forecast values to the external operational

condition changes through the FeedBack Tuning Fuzzy System (FBTFS).

The nal model structure will be built and veried with simulation results in this

chapter. In addition, it will also be implemented as part of the proposed management

system in Chapter 6, to work with other intelligent modules to solve the more complicated

problems concerning minimising the power generation cost for the microgrid.

3.2 Related works

At the time of application, ANFIS was successfully employed to perform short term demand

forecasting for wide range of applications in a variety of elds, such as: 1) nance- where it

was used to create a forecasting model for the domestic debt based on 10 years of monthly

values of currency, total money supply, consumer price index and interest rate in [120];

2) tourism- where the ANFIS model was used to forecast the tourist arrivals coming to

Taiwan in [121]; 3) control engineering- where ANFIS modelling was introduced in order

46
to reduce a large number of FL controller rules in [122], in [123] where ANFIS model

was utilised as classier for control decision making, and in [124], an ANFIS based space

vector modulation technique was utilised to control the voltage source inverter; 4) Natural

Gas consumption- an ANFIS stochastic frontier analysis approach for long-term natural

gas (NG) consumption forecasting and analysis of the behavior of NG consumption was

proposed in [125].

ANFIS has already achieved a successful track record in modelling the forecast for

electricity demand. Under various modelling structures, ANFIS was included when mod-

elling the electricity demand for a number of power utilities. A two-dimensional wavelet

based state dependent parameter modelling approach, was proposed to produce a compact

mathematical model for this complex nonlinear dynamic system; for short-term forecast of

daily demand in the state of Victoria, Australia in [126]. A new hybrid ANFIS model was

proposed for short-term load forecasting for two Brazilian power companies in [127], while

in Taiwan, ANFIS was applied to forecast regional loads in Taiwan and to demonstrate the

forecasting performance of the same model in [128]. For smaller load sizes, a number of

studies evaluating the forecasting of energy consumption within buildings were conducted,

in order to select the best control strategies to manage the excessive energy consumptions.

A model of short term forecasting based on variables such as; maximum and minimum

temperature, climate change, and the previous days consumed load; for forecasting the

power load consumption was developed using ANFIS in [129]. ANFIS was also applied to

forecast the demand change in the cold regions in [130]. A short term forecasting model

was developed using the neuro fuzzy system Locally Liner Model Tree (LoLiMoT) learning
algorithm in [131]. Next-day load demand forecasting in electrical power generation was

developed using ANFIS; improving the power system as an application of articial neu-

ral networks and FL-based hourly load demand forecasting, with linear polynomial and

exponential equations in [132]. The demand forecasting using time series modelling and

ANFIS estimator was developed in [133], where an ANFIS model based on the data eld

was proposed to solve the drawbacks of the general fuzzy neural network, and to optimise

the FL rules in [134].

The ANFIS-based models were found to present opportunities for managing power de-

mand, as a response to electricity market participation. ANFIS was proved to be suitable

for undertaking demand forecasting based on its sensitivity to real-time external param-

eters, such as uctuations in the market price of electricity [135], and real-time pricing

which was integrated in the short-term energy demand forecasting modelling in [136].

47
Real-time electricity pricing and bidding were used in modelling demand in [137] and [138]

respectively.

Although ANFIS has already been established as a successful modelling technique,

it was previously used with only a simple modication to deal with highly non-linear

load proles. ANFIS modelling accuracy, however, depends on many factors including

the ANFIS structure parameters selection. An approach based on the combination of

Particle Swarm Optimisation (PSO) and adaptive-network based FIS was proposed to

forecast the electricity prices in a competitive market in [139]. Importantly, the results

obtained show signicant improvement in both price and load forecasting. The other

important factor that aects ANFIS modelling accuracy is the data clustering parameters.

When evaluating existing clustering techniques and clustering methods, it was found that

a technique utilising multivariate inputs for electrical load forecasting on hybrid neuro-

fuzzy and fuzzy C-means forecaster has previously been proposed. It used a neuro-fuzzy

approach, with additional fuzzy c-Means clustering before the inputs enter the networks

[140]. A clustering-based, genetic fuzzy expert system for power demand forecasting has

also been presented, which has a novel load forecasting approach that integrates genetic

fuzzy systems and data clustering for extracting a load forecaster expert system [141]. A

clustering algorithm based neuro-fuzzy method for estimating wind speed prole, based on

knowledge of wind speed at dierent heights, was applied in [142]. Among the proposed

modelling methods, the ANFIS based Fuzzy Subtractive Clustering Method (FSCM) was

deemed relevant to this project due to its applicability and ease of utilisation for the

purpose of research. Ultimately, the experience in clustering the system's operational data

would result in a more eecient ANFIS model.

3.3 Research objectives and challenges

Obtaining accurate demand prole forecasts for a typical power system is the challenge

of understanding the consumer's behaviour throughout the year. Hence identifying the

type of the demand prole (residential, commercial or industrial) facilitates evaluating the

demand diversity factor, which is needed in making reliable estimates about the demand

forecasts. However, this cannot be used in autonomous management systems unless it is

added to a mathematical model that has been developed in accordance with the historical

trends of that particular demand. The other major challenges, encountered with modelling

a demand prole for a power system, are:

1. Selecting the modelling technique and embedding the knowledge required in order

48
to cope with unpredictable operational scenarios.

2. Reecting the system operator experience about the system's operational scenarios.

3. Dealing with the unpredictable load prole that may result in a higher forecasting

error, which might aect the management system performance.

4. Monitoring the forecasting error rate and controlling it by adapting the forecasts to

the actual demand values.

3.4 Research assumptions

This model was initially intended to be part of our proposed generation management

system, in line with other intelligent modules. Thus the model gives considerations to

the amount of uncertainty presents at each part of the system, including the forecasting

component. Based on the expected uncertainty of the load prole in the case study, the

following assumptions underpinned the research methodology:

1. The proposed demand forecasting model is targeting a university type load prole.

2. The model is proposed to deal with 12 dierent demand pattern changes throughout

the year. Ultimately, 12 models are required to forecast the demand throughout the

year for the targeted case study.

3. All university function- or event-type load changes resulting from on campus activities

are ignored in the modelling process. Therefore, it is assumed that the system may

operate with full load capacity when such kinds of activities take place.

4. During faults and emergencies, the forecasting model is immediately suspended and

the system operates in emergency mode. The load prole forecast is ignored when

supplying power for the case study where the system has faulted.

3.5 Modelling methodology

This section covers the methodology for modelling the electricity demand prole, and

is based on a series of 30-minute intervals in the case study. The proposed model was

developed in two distinct modelling stages, encompassing three computation modules,

each of which performs an essential role in contributing to the overall modelling. In the

rst modelling stage, the demand is forecast and then corrections are made based on the

added knowledge gained through FFTFS. Two FIS are integrated together in the rst

49
 Figure 3.2: A proposed model combination for January model

modelling stage: 1) the main FIS which is developed from modelling the input-output

data using FSCM and ANFIS, and 2) the second FIS that represents FeedForward Tuning

Fuzzy System (FFTFS). The FFTFS is either developed by using the correlation between

the power demand and the temperature throughout the day into future decision making,

or developed by using knowledge about the real-time demand.

In the second modelling stage, the demand forecast is adapted to the real-time changes

relating to the case study. At this stage of computation, two systems are utilised; a two-

input-one-output FIS FeedBack Tuning Fuzzy System (FBTFS), in conjunction with an

ANN-based optimisation technique, to tune the FIS parameters in the second stage of

computation.

Figure 3.2 describes the full details for the proposed modelling mechanism of a one-

month, randomly selected January model. The whole year model is illustrated in Figure

3.3.

To improve the forecasting accuracy and reduce the model complexity, the annual power

demand of ECU's Joondalup campus was split into twelve monthly models, represented

by twelve discrete demand pattern models. Each model represents a one month demand

model, which can be activated depending upon the type of the input set. Figure 3.3

illustrates the annual power demand forecasting structure for ECU's Joondalup campus.

Splitting the annual demand model into twelve sub-models expands the system's fore-

casting ability by being able to draw from twelve dierent load change patterns. In ad-

dition, the twelve sub-models reduces the computation resource requirement, since only

50
Figure 3.3: The electric power annual demand forecasting structure for ECU's Joondalup
campus

one month's model is active at a time. Thus the modelling process uses twelve separate

modelling methodologies depending on the load change analysis for the individual months.

A switching control is applied to automate the running of the one month model, based

on the type of input parameters, and extended for the forecasting throughout the year in

this composite model. Each one month model consists of the main modelling structure,

including the forecasting and the FeedForward and FeedBack adaptations. The following

subsections detail the model's structure for each month:

3.5.1 ANFIS model

In this subsection, we discuss the use of FIS in the modelling process, and also provide

specic details about the stages of ANFIS modelling of the power demand in the targeted

case study. In this investigation, data clustering methods are nominated to perform fuzzy

modelling. More specically, FSCM and ANFIS are used to model the demand in the

51
Table 3.1: ROI values and complexity of the 12-month electricity energy demand models

Months\Membership Functions ranges ROI Rules Membership Fctn.

January 0.35 28 112
February 0.4 23 92
March 0.5 14 56
April 0.33 40 160
May 0.44 17 68
June 0.4 25 100
July 0.45 20 80
August 0.48 19 76
September 0.43 18 72
October 0.5 11 44
November 0.5 16 64
December 0.41 20 80

case study, based on the detailed methods of modelling from Section 6.1. Generally, the

optimum selection of the modelling parameters is the key factor in achieving the highest

performance from the developed models. Based on experience gained in modelling and pa-

rameters selection, dierent clustering parameters for each month were selected, depending

on the amount of non-linearity of the data-set. By clustering; ambient temperature, hour,

day, and load change data; random FSCM parameters values (e.g. Range OF Inuence

(ROI), Squash, Accept Ratio, and Reject Ratio) are applied. These selected values may

increase the complexity of the models that are developed. Table 3.1 illustrates the selected

modelling parameters, along with the complexity achieved for each of the 12-month power

demand models.

After clustering is complete, the membership functions previously developed are then

trained. Following this step, a simple test is carried out to verify the forecasting accuracy

of the models that have been developed. In this work, each month's model is developed

by using the microgrid's historical operation data for that particular month in three sub-

sequent years (A, B, and C). The three years of data are then divided into three dierent

sets. The rst set is used to extract the clusters, which is taken as 90% of the A and B

years historical data. The second set of data, which is used to train the fuzzy systems

which have been developed, used the whole set of A and B data. Finally, the third set of

data, which is used to verify the success of the developed model, was taken as the year C

operation data. Fig. 3.4 shows the data utilisation in developing the demand models in

this work.

After the rules which relate the input-output data have been developed, the clusters

are then utilised in neuro-fuzzy networks to develop a zero-order Sugeno fuzzy inference

52
Figure 3.4: Modelling data utilisation for the power consumption at ECU's Joondalup
campus

system, which can then perform a 30-minute ahead short-term forecast. In conventional

fuzzy systems, a trial and error approach is applied to tune the membership functions

of the input-output universe of discourse of the fuzzy system. When ANN is used to

tune the membership functions, an automated selection process is performed based on the

performance index. The membership functions are trained to resemble the training data

characteristics. In neuro-fuzzy networks, the structure is changed in accordance with the

operational scenarios. Neuro-fuzzy networks, however, utilise the neural network's ability

to learn to achieve the best tuning process, with better performance and less time [143].

Since fuzzy systems have the property of universal approximation [51, 144], it is expected

that the equivalent neuro-fuzzy networks representation will have the same property.

Finally, when a verication result is within an acceptable error bound, the modelling

procedure is complete. Figure 3.5 illustrates the input membership functions that were

developed for the four inputs, zero-order, Sugeno fuzzy system of January's operation for

ECU's Joondalup campus power system.

From Figure 3.5, and from the Sugeno-fuzzy system for January demand forecasting,

the rules that were developed are explained as follows:

If (Temperature is Temperature in Cluster n) and (Hour is Hour in Cluster n) and (Day

is Day in Cluster n) and (Day-type is Day-type in Cluster n), then (Demand is Demand

in Cluster n)
where 0 <n ≤ number of developed rules.

Finally, for the other eleven months of the year, it was found that each of the models

had a dierent range of inputs, based on the pattern of operation and weather change

across the four seasons throughout the year, in the city of Joondalup. Although other

eective modelling parameters have been nominated for the proposed models, experimental

investigations were applied to use three, four, and ve-input modelling parameters to

53
Figure 3.5: The membership functions for the four inputs zero-order Sugeno fuzzy system
of January's power demand forecasting

54
Figure 3.6: Forecasting results comparison for demand in ECU from the 17th to the 21st
of January, year C

achieve demand forecasting performance. Ultimately, the four-input modelling parameters

approach was chosen, which proved to be an optimal selection.

At this stage, modelling approach robustness was tested with the nominated input

parameters. The results obtained demonstrated the dependence of the forecasting accuracy

on both the number of inputs and on the the type of inputs. Figure 3.6 illustrates the

dierent forecasting accuracy for the three, four, and ve-input systems.

Table 3.2 presents modelling complexity for all investigated cases. Modelling complexity

is shown in terms of extracted number of membership functions (MMFcn) and the value of

(ROI). The complexity of the models was raised to double or triple in most cases, indicating

that the input does not have a big inuence on the load changes at ECU. As a result,

adding the fth input to the models will not have a signicant inuence on forecasting

improvement, or may even have a negative aect. Figure 3.6 shows the forecast results for

the three, four, and ve-input systems for the period from the 17th to the 21st of January,

year C.

55
 Table 3.2: Three, four, and ve-input systems forecasting accuracy

Three inputs Four inputs Five inputs

Month
MMFcns ROI MMFcns ROI MMFcns ROI
January 63 0.35 76 0.43 150 0.5
February 69 0.38 72 0.4 150 0.5
March 33 0.45 56 0.5 145 0.5
April 78 0.33 160 0.33 140 0.5
May 39 0.44 68 0.44 145 0.5
June 45 0.48 100 0.4 320 0.4
July 36 0.5 80 0.45 145 0.5
August 36 0.5 76 0.48 145 0.5
September 54 0.45 76 0.43 120 0.55
October 24 0.5 44 0.5 125 0.5
November 33 0.48 64 0.5 135 0.5
December 45 0.41 60 0.41 182 0.45

Studying system operational conditions e.g., the operation time, the number of people,

and the amount of consumed power during the day or night, may help develop supportive

variables to improve the forecasting accuracy, as long as the signicance of the input is

assessed prior to its use in the model development. Figure 3.7 presents the surfaces of the

developed fuzzy systems for the three, four, and ve input variables.

3.5.2 FeedForward Tuning Fuzzy System (FFTFS)

In this subsection, we investigate improving the forecast results, based on knowledge of

the power demand conditions, which could have been imperfectly represented in the given

set of historical operational data. The system is required to enhance the performance of

the forecasting model, by using the knowledge of the system performance, safe operational

estimations, and the actual decisions that are required. In this work, FFTFS is proposed

to improve the forecasting accuracy. Two of the model's inputs are selected to develop

the fuzzy rule-based system, so that there is a smooth transition between the specied

operational cases in the decision making. In this work, we typically investigate use of a

one rule based system for application to all of the 12-month models. Table 3.3 illustrates

the proposed rule based system in this investigation.

To cope with changes to the operational patterns throughout the year, dierent ranges

of membership functions universe of discourse are proposed for each month's model. Trian-

gular type membership functions have been proposed in developing FFTFS. The rule base

56
Figure 3.7: Fuzzy surface for the developed fuzzy models of the three, four, and ve-input
system

Table 3.3: Self tuning fuzzy rule-based system

Hour Temperature V. Cold Cold L. warm Room temp. Warm Hot V. hot
Midnight S. low Normal Normal S.High High V. High V. High
Dawn Normal S. High High V. High V.High Vv High Vv High
Morning Low S.Low Normal Normal S. High High V. High
Afternoon V. Low V.Low Low Low S.Low Normal Normal
Sunset V. Low Low S. Low Normal S. High High V. High
Evening Low S. Low S. Low Normal S.l High High V. High
Night S. Low Normal S. High High V. High V. High Vv High

57
system consists of seven membership functions; in both universe of discourse of the Hour,

and the Temperature: Vv (very very ) High, V (very) High, High, Normal, Low, V (very)

Low and Vv (very very) Low. The output of the rule base system consists of nine member-

ship functions, which are the same as the input's membership functions, except that there

are two extra membership functions added (S (small) High and S (small) Low), in order

to increase precision of the output. All twelve monthly models have the same membership

functions shape, however, they also have dierent input/output ranges. Figure 3.8 shows

the proposed membership functions design for FFTFS in the January forecasting model.

Figure 3.8: Membership functions design for FFTFS of January demand forecasting model

Table 3.4 illustrates the membership function design for all 12-month FFTFS.

The twelve monthly models have dierent FFTFS designs. Use of each of the twelve de-

signs results in dierent forecasting improvement outcomes. By considering the forecasting

reults from January operational data, variable amount of forecasting error is encountered

throughout the month, however conservatively, the weakest forecasting region throughout

January, which presents that highest amount of forecasting error, is shown in the results.

Figure 3.9 shows the demand forecasting for the 17th to the 21st of January, year C, using

ANFIS and FFTFS. The amount of forecasting improvement is calculated by evaluating

the Root Mean Square Error (RMSE), Integral Square Error (ISE), and Mean Absolute

Percentage Error (MAPE). From the results, it is found that FFTFS has an enhanced

58
 Table 3.4: Membership function ranges design for the 12-month FFTFS

Months Temperature Hour Output

January [-10 30] [0 24] [-75 75]
February [-10 35] [0 24 ] [-75 75]
March [-10 20] [0 24] [-50 50]
April [15 35] [0 24] [-30 30]
May [0 20] [0 24] [-40 40]
June [0 25] [0 24] [-50 50]
July [-20 20] [0 24] [-50 50]
August [5 20] [0 24] [-30 30]
September [-20 20] [-4 24] [-30 30]
October [30 70] [0 18] [-200 200]
November [10 50] [-4 18] [-100 100]
December [-10 20] [-4 18] [-100 100]
Table 3.5: Statistical results for each month with the improvement rate made by FFTFS-
ANFIS model

ANFIS FFSTFS-ANFIS
Month
Rules RMSE ISE MAPE Rules RMSE ISE MAPE

January 28 0.0058 4331 9.3% 77 0.006 4076 12.75%

February 23 0.0069 3310 7.2% 72 0.0067 3097 7%
March 14 0.0049 6157 11% 63 0.0048 6308 11%
April 40 0.005 5943 11% 89 0.005 5902 10.8%
May 17 0.0059 4141 11% 66 0.0059 4160 11.1%
June 25 0.007 3051 8.9% 74 0.0069 3023 8.9%
July 20 0.0072 2851 8% 69 0.0072 2800 8.2%
August 19 0.0063 3734 9.7% 68 0.0063 3717 9.9%
September 18 0.0064 3605 9.2% 67 0.0064 3543 9.4%
October 11 0.0058 4366 11.3% 60 0.0058 4314 11.2%
November 16 0.006 4039 10.3% 65 0.006 3742 9.9%
December 20 0.0055 4908 11% 69 0.0055 4764 11%

forecasting accuracy. Table 3.5 lists the statistical results evaluated for each month, and

the improvement achieved by FFTFS.

It is shown from the applied statistical analysis, that the proposed design can success-

fully maintain the forecasting accuracy for year C, where its design was based on year A

and year B data. The results obtained showed that the model has the ability to improve

the forecasting accuracy and maintain it within a desired range. However, table 3.5 also

shows that the range of improvement varies depending on the demand pattern in each

month, where for certain months, FFTFS causes a deterioration of the forecasting accu-

racy as a result of some statistical measures. The deterioration of the monitored accuracy

59
Figure 3.9: Self tuning and ANFIS Demand forecasting in the ECU microgrid for the 17th
to the 21st of January, year C

in these months is still acceptable, providing that the model is showing a positive attitude

throughout the 12 months of the year. It is expected that any changes to future demand

will cause a deterioration in the model. Even so, the overall performance after applying

FFTFS to increase the model's robustness is encouraging, and providing that this dete-

rioration is always found to be within an acceptable error range, the overall results are

signicant and worthwhile in terms of improvement. Ultimately, FFTFS was added to the

forecasting model to increase its robustness in parallel with the feedback adaptation tuning

system.

3.5.3 FeedBack Tuning Fuzzy System (FBTFS)

The utilisation of actual demand as a feedback to the model will introduce a real-time

adaptive demand forecasting capability. In order to introduce an external parameter so

that we can improve the forecasting, the forecasting model is treated as a process, while

the real time demand value is treated as the desired set-point. The dierence between the

forecast and the actual demand values is fed back to the model via a controller, in order to

improve the subsequent forecasting accuracy, so that the system's range of operation could

60
be established. In order to gain further insights into the design considerations, the system

was simulated with a classical Proportional Integral (PI) system. After introducing PI to

control the accuracy error, we monitored the performance of the model. In this work, the

proposed PI controllers are manually tuned based on the error performance. At this stage

a decision was made to replace the classical PI with an intelligent system (fuzzy logic),

largely because of its ability to achieve better transient response. The proposed intelligent

system uses the measured error in order to: 1) adapt the subsequent forecasting output

to the actual demand changes; and 2) control its accuracy. In this part of the model, the

proposed two-inputs-one-output PI-fuzzy system was applied to redo the PI system's job.

For the sake of simplicity, we proposed that a one rule based system for the 12-monthly

demand models be deployed. The proposed intelligent system in this modelling strategy

is the FeedBack Tuning Fuzzy System (FBTFS). In this component, the proposed rule

base system was developed utilising knowledge about the power demand changes in the

case study. Accordingly, the fuzzy systems have a set of other design parameters to cope

with the 12-monthly demand patterns. Table 3.6 illustrates the rule base system for the

proposed two-inputs-one-output FBTFS.

Table 3.6: Rule base table for the two-input-one-output FBTFS

∆Error
Vv High V High High Normal Low V Low Vv Low
Error
Vv High Vv Low V Low Normal S High High V High V High
V High V Low S High High V High V High Vv High Vv High
High Low S Low Normal Normal S High High V High
Normal Vv Low V Low Low Low S Low Normal Normal
Low V Low Low S Low Normal S High High V High
V Low Low S Low S Low Normal S High High V High
Vv Low S Low Normal S High High V High V High Vv High

The rule base system consists of seven membership functions; in both universe of dis-

course of the error, and change of error (∆Error) : Vv (very very ) High, V (very) High,

High, Normal, Low, V (very) Low and Vv (very very) Low. The range of the universe of

discourse is set by analysing the error, and the change of error behaviour. Therefore the

error and the change of error operation for the proposed model is set between -1650 and

1650 kW, in order to cover the maximum expected dierence between the forecast and

the actual demands, in the +ve and the -ve sides range. The output of the rule base sys-

tem consists of nine membership functions, which are the same as the input's membership

61
 Figure 3.10: Manually tuned FBTFS membership functions design

functions, except that there are two extra membership functions added (S (small) High

and S (small) Low), in order to increase precision of the output. These have a range of

action between -600 and 600 kW. The same design parameters are applied for year C, as

it is impractical to monitor and tune the gains manually based on the real time demand

changes. In the design of the membership functions, triangular type membership functions

were applied. Figure 3.10 shows the proposed membership function design for the FBTFS

that we have developed.

For the proposed FBTFS, a manual gain selection was derived from the experience with

the system's operation, while trial and error tuning was also applied at the nal stages of

the design. Table 3.7 illustrates the gain values for the PI-FBTFS.

In proposing that the FBTFS will adapt the forecasting to the actual demand changes,

the aim is to produce a model that can cope with the unmonitored demand changes, at

the stage that it becomes impractical to tune the gains manually. The proposed FBTFS

has the ability to cope with the monitored demand change, based on the manual dened

settings. However the FBTFS, requires an optimised parameter selection in order to cope

with the wide range of demand changes. The major obstacle that arises when performing

62
 Table 3.7: Year B Model feedback fuzzy tuning system gains selection

Month KP Ki
January 0.01 0.1
February 0.01 0.05
March 0.01 0.05
April 0.01 0.06
May 0.01 0.06
June 0.01 0.2
July 0.1 0.15
August 0.1 0.15
September 0.1 0.1
October 0.1 0.15
November 0.05 0.1
December 0.01 0.005

the optimisation, similarly to other optimisation techniques, is the parameter searching

time. The proposed FBTFS was designed to adjust itself to the demand changes automat-

ically in a one iteration optimised gain selection mechanism. This mechanism will make

the searching time the lowest as compared with other optimisation techniques. An ANN

optimisation technique was used to tune the fuzzy input parameters. The design for the

ANN to take into consideration the 12-month demand patterns, thus 12 ANNs have been

developed so that a high level of accuracy can be maintained throughout the year. In this

modelling strategy, each ANN was trained to tune the fuzzy parameters, depending on the

input pattern. Consequently an ANN model was implemented to perform the proposed

online optimised FBTFS gains selection. Although ANFIS was successful in performing the

modelling; when consideration was given to its training time and modelling performance;

in this instance the optimisation technique was developed with back-propagation ANN.

Back-propagation ANN was chosen because of the potential for ANN to achieve better

performance with Multi Input Multi Output (MIMO) systems. In the proposed model it

was found that fuzzy input parameters are the most eective parameters for controlling

the forecasting accuracy. To implement an adaptive (tuneable gains) fuzzy system, three

main issues must be considered simultaneously: 1) dynamic characteristics of a plant; 2_

self-selection of the performance index; and 3) self-tuning of the controller parameters. In

this work, FBTFS gain tuning was targeted. The proposed ANN was designed to optimise

the gains selection by learning from the experience of selecting the relevant gain values

for every expected demand pattern. For this auto-tuned FBTFS, a membership function

design was rstly proposed having the same membership functions' number of that in the

manually tuned FBFTS, accept that a normalised universe of discourse is proposed, for

63
 Figure 3.11: Auto tuned FBTFS membership functions design

both the inputs and the output of this system as shown in Figure 3.11.

In order to collect the data that trains the ANN to perform an optimised gain selection,

a range of historical operational data was generated. A software was developed to automate

the process of generating and collecting the relevant data for the training purposes. The

process of collecting the data was developed by setting a history of suitable gain selection for

each specic range of operation. The suitable gain setting is arrived at by searching a range

of gains with a specied precision, for that particular range of operation. The best selected

gain values are determined by achieving the lowest statistical results for the forecasting

error, within the investigated operation period. The search range converged around the

manually tuned parameters, following which, the gains can then be set for the manually

tuned FBTFS. This process reduces the searching time and resembles the behaviour of

human-operator tuning. The limits and the precision of the search are established by

setting the gains manually in the rst instance, then by monitoring the attitude of the

model. A data set of the most accurate results are then saved, along with their selected gain

values for use in the next stage for the ANN training. In this phase of the work, demand

is divided into a day-by-day demand change, where 29, 30 or 31 data sets are carried

64
out for each monthly model, depending on the month in question. The structure of the

proposed ANN for the 12-monthly demand patterns consists of ve-inputs (Temperature,

Hour, Date, Date-type and month number) and two-outputs (K P and K i ), which represent
the gains of the proposed PI-like fuzzy system. Tables 3.8 and 3.9 show the best nominated

gain values for the FBTFS in everyday tuning throughout the year. These tables are used

to build the optimisation ANN for tuning (K P and K i) of the proposed FBTFS. The

results in these tables are also used to identify the operational limits of KP and K i. In

addition, they are the key solutions for converging the optimisation decisions applied by

the ANN.

From Tables 3.8 and 3.9, it can be seen that some of the selected gains have been

chosen around the manually predened gains values for the proposed models, while the

rest have been selected as a result of having a large disparity in values. The variety present

in the selection proves that the automated search process was successful in performing the

scan over a wide range of operation by employing high resolution searching. Tables 3.8

and 3.9 also shows that Ki aects the accuracy of the forecasting more than K p, as it

shows a wider range of change within the same month as compared with KP . The set

of historical operational data is used to develop the ANN for the proposed model. The

selection of the ANN parameters is based on modelling experience and the type of system

being modelled. The proposed ANN also includes one hidden layer network, with a bi-polar

log-sigmoid activation function in the hidden layer, and a bi-polar linear type activation

function in the output layer. Tables 3.8 and 3.9 illustrate the set of the best selected

gains in the training data, out of the everyday data. For all of the proposed ANN, the

training parameters including the used learning rate, number of nodes, and the resulted

Mean Square Error (MSE) are illustrated in Table 5.3.

3.6 Results

The completed model was veried by employing realistic demand change data from ECU.

The year C demand change data is utilised to perform the model testing and verify the

robustness of the modelling strategy. The simulation started with the classical PI system

to establish an understanding of the forecasting error behaviour, which resulted in various

ranges of demand accuracy. The accuracy of the forecasts was evaluated by the statistical

methods: RMSE, ISE, and MAPE. By looking at the results in Table 3.11, various demand

improvement rates can be identied, depending on the non-linearity of the demand in each

month, and depending upon the precision of the model. Starting with the classical PI

65
Table 3.8: The selected gains in the training data set of everyday data from January to
June

Day Jan Feb Mar Apr May Jun

KP Ki KP Ki KP Ki KP Ki KP Ki KP Ki

1 0.007 0.075 0.01 0.02 0.05 0.011 0.005 0.11 0.02 0.003 0.11 0.06
2 0.007 0.125 0.01 1 0.05 0.011 0.005 0.11 0.02 0.003 0.01 1
3 0.007 0.125 0.01 1 0.05 0.0095 0.005 0.0125 0.02 0.015 0.01 1
4 0.011 0.075 0.01 0.02 0.05 0.011 0.005 0.0125 0.02 0.015 0.01 1
5 0.011 0.075 0.01 0.02 0.05 0.011 0.005 0.0125 0.02 0.003 0.01 0.5975
6 0.011 0.125 0.01 0.02 0.05 0.011 0.005 0.11 0.02 0.003 0.01 1
7 0.007 0.125 0.01 0.03 0.05 0.011 0.005 0.11 0.02 0.003 0.01 0.1125
8 0.007 0.012 0.01 0.02 0.05 0.011 0.005 0.0125 0.02 0.003 0.01 0.06
9 0.007 0.075 0.01 0.02 0.05 0.001 0.005 0.01 0.02 0.003 0.01 1
10 0.007 0.075 0.01 0.02 0.05 0.0105 0.005 0.11 0.02 0.003 0.01 0.5975
11 0.007 0.075 0.01 0.02 0.05 0.011 0.005 0.11 0.02 0.015 0.01 1
12 0.007 0.0763 0.01 0.02 0.05 0.011 0.005 0.11 0.02 0.003 0.01 0.6475
13 0.011 0.125 0.01 0.02 0.05 0.011 0.005 0.11 0.02 0.003 0.01 1
14 0.007 0.075 0.01 0.02 0.05 0.011 0.005 0.11 0.02 0.003 0.01 0.06
15 0.007 0.075 0.01 0.02 0.05 0.001 0.005 0.11 0.02 0.003 0.01 0.1
16 0.004 0.075 0.01 0.015 0.05 0.011 0.005 0.11 0.02 0.003 0.01 1
17 0.007 0.017 0.01 1 0.05 0.001 0.005 0.11 0.02 0.003 0.01 0.5975
18 0.007 0.012 0.01 0.0175 0.05 0.011 0.005 0.11 0.02 0.015 0.01 1
19 0.011 0.075 0.01 0.8075 0.05 0.011 0.005 0.11 0.02 0.003 0.01 1
20 0.007 0.075 0.01 0.8275 0.05 0.011 0.005 0.11 0.02 0.003 0.01 1
21 0.007 0.012 0.01 0.0175 0.05 0.011 0.005 0.11 0.02 0.003 0.01 0.06
22 0.007 0.125 0.01 0.27 0.05 0.001 0.005 0.11 0.02 0.003 0.01 0.06
23 0.007 0.012 0.01 0.0175 0.05 0.001 0.005 0.11 0.02 0.003 0.01 0.9475
24 0.007 0.075 0.01 0.0175 0.05 0.011 0.005 0.11 0.02 0.015 0.01 1
25 0.007 0.075 0.01 0.02 0.05 0.001 0.005 0.0125 0.02 0.015 0.01 1
26 0.007 0.125 0.01 0.03 0.05 0.0105 0.005 0.015 0.02 0.003 0.01 1
27 0.007 0.125 0.01 0.02 0.05 0.011 0.005 0.015 0.02 0.003 0.01 1
28 0.007 0.012 0.01 0.03 0.05 0.011 0.005 0.015 0.02 0.003 0.01 0.06
29 0.007 0.017 0.01 1 0.05 0.001 0.005 0.11 0.02 0.003 0.01 0.06
30 0.007 0.0763 - - 0.05 0.001 0.005 0.0125 0.02 0.003 0.01 1
31 0.007 0.075 - - 0.05 0.001 - - 0.02 0.015 - -

66
Table 3.9: The selected gains in the training data set of everyday data from July to
December

Day Jul Aug Sep Oct Nov Dec

KP Ki KP Ki KP Ki KP Ki KP Ki KP Ki

1 0.2 0.07 0.1 0.15 0.01 0.51 0.01 0.15 0.01 0.03 0.001 0.0028
2 0.2 0.07 0.1 0.15 0.01 0.51 0.01 0.15 0.01 0.02 0.001 0.011
3 0.2 0.07 0.1 0.15 0.01 0.51 0.01 0.15 0.01 0.3 0.001 0.011
4 0.2 0.07 0.1 0.15 0.01 0.51 0.01 0.02 0.01 0.3 0.001 0.0018
5 0.2 0.08 0.1 0.15 0.01 0.51 0.01 0.02 0.01 0.3 0.001 0.0048
6 0.2 0.08 0.1 0.15 0.01 0.4975 0.01 0.15 0.01 0.3 0.001 0.0058
7 0.2 0.07 0.1 0.15 0.01 0.51 0.01 0.15 0.01 0.3 0.001 0.0038
8 0.2 0.07 0.1 0.15 0.01 0.51 0.01 0.15 0.01 0.1 0.001 0.011
9 0.2 0.07 0.1 0.15 0.01 0.51 0.01 0.15 0.01 0.3 0.001 0.011
10 0.2 0.07 0.1 0.15 0.01 0.51 0.01 0.15 0.01 0.3 0.001 0.011
11 0.2 0.07 0.1 0.15 0.01 0.51 0.01 0.15 0.01 0.3 0.001 0.0038
12 0.2 0.08 0.10.1 0.15 0.01 0.51 0.01 0.02 0.01 0.3 0.001 0.011
13 0.2 0.08 0.1 0.15 0.01 0.51 0.01 0.15 0.01 0.3 0.001 0.011
14 0.2 0.07 0.1 0.15 0.01 0.06 0.01 0.15 0.01 0.3 0.001 0.011
15 0.2 0.07 0.1 0.15 0.01 0.51 0.01 0.15 0.01 0.02 0.001 0.011
16 0.2 0.07 0.1 0.15 0.01 0.51 0.01 0.0175 0.01 0.02 0.001 0.0048
17 0.2 0.07 0.1 0.15 0.01 0.51 0.01 0.15 0.01 0.3 0.001 0.0028
18 0.2 0.07 0.1 0.15 0.01 0.51 0.01 0.15 0.01 0.3 0.001 0.0028
19 0.2 0.08 0.1 0.15 0.01 0.51 0.01 0.15 0.01 0.3 0.001 0.011
20 0.2 0.08 0.1 0.15 0.01 0.51 0.01 0.15 0.01 0.3 0.001 0.0018
21 0.2 0.07 0.1 0.15 0.01 0.51 0.01 0.0175 0.01 0.3 0.001 0.0018
22 0.2 0.07 0.1 0.15 0.01 0.51 0.01 0.15 0.01 0.3 0.001 0.011
23 0.2 0.07 0.1 0.11 0.01 0.51 0.01 0.15 0.01 0.3 0.001 0.011
24 0.2 0.07 0.1 0.15 0.01 0.51 0.01 0.15 0.01 0.3 0.001 0.011
25 0.2 0.07 0.1 0.15 0.01 0.51 0.01 0.0175 0.01 0.3 0.001 0.011
26 0.2 0.08 0.1 0.15 0.01 0.51 0.01 0.15 0.01 0.1 0.011 0.0015
27 0.2 0.08 0.1 0.15 0.01 0.51 0.01 0.15 0.01 0.3 0.011 0.002
28 0.2 0.07 0.1 0.15 0.01 0.06 0.01 0.15 0.01 0.1 0.011 0.002
29 0.2 0.07 0.1 0.15 0.01 0.06 0.01 0.15 0.15 0.02 0.001 0.0038
30 0.2 0.07 0.1 0.15 0.01 0.51 0.01 0.15 0.15 0.02 0.001 0.0088
31 0.2 0.07 0.1 0.11 - - 0.01 0.15 - - 0.001 0.011

67
Table 3.10: Neural networks design parameters for the gain adaptation of the 12-month
forecasting FBTFS

Month Nodes Learning Rate epochs MSE

January 25 0.05 7 0.0022287

February 10 0.05 135 0.00935
March 10 0.05 2 0.0000481
April 15 0.05 2 0.00602
May 10 0.05 2 0.0011696
June 20 0.05 3 0.0083366
July 20 0.05 2 0.0027826
August 7 0.05 2 0.0002594
September 20 0.05 1 0.008631
October 11 0.05 4 0.007355
November 17 0.08 4 0.005618
December 15 0.05 1 0.0002097

system, a wide range of statistical results were identied, as dierences arise from the

number of days (29, 30 or 31) in the month, in addition to the dierent demand patterns

that are encountered throughout the year. From the control theory point of view, a slow

response was observed when the tested interval time was less than the required transient

time, in order to completely eliminate the error of the forecast demand. That was why

the system recorded this value of the steady state error for the simulated periods. This

problem was partially resolved by adding the intelligent FBTFS, as it was proposed that

the applied rule based system, with its control action, was able to achieve a faster response

than it could achieve with the classical PI system.

In the case of the intelligent system, manual and automatically tuned two-inputs-one-

output FBTFS were tested. When the automatic tuning is considered, the target becomes

achieving an optimum forecasting accuracy with the lowest evaluated statistical error. A

safety margin is added to the forecast output to ensure that the forecast is always above

the actual demand. Thus safety can be guaranteed after nominating the required number

of generators for every 30 minutes interval. The result is presented in Table 3.11 in terms

of the type of the used feedback system in each month models.

From Table 3.11, it was shown that the proposed FBTFS, in its manual and automatic

tuning, was superior to the classical PI system in improving the forecasting accuracy. It

is also shown that the FBTFS had a faster tracking response than that of the classical PI

system. This proves that the knowledge embedded in the FBTFS rule base system could

achieve better transient response for the forecasting error control system. Therefore, for

68
Table 3.11: The statistical analysis results in each month with improvement rate made
by the auto tuned FBTFS over the manually tuned and PI system

Month Classical PI Manual Tuning Automatic Tuning

RMSE ISE MAPE RMSE ISE MAPE RMSE ISE MAPE
January 0.0024 24160 29% 0.0026 21830 27% 0.0033 20180 18.5%
February 0.0025 23590 25.1% 0.0024 24580 24.6% 0.0038 24120 13.4%
March 0.0024 25130 27% 0.0022 29190 29% 0.003 28010 18.1%
April 0.0024 24150 29% 0.0031 14670 18% 0.0039 13990 14.7%
May 0.0024 24640 31.5% 0.0029 17320 23.6% 0.0037 16300 17%
June 0.0024 24590 32.9% 0.003 15390 24% 0.005 4354 12.8%
July 0.0024 24720 33% 0.003 16140 21.7% 0.0038 15760 18.5%
August 0.0024 24820 32.8% 0.0028 18840 25.7% 0.0039 16440 17.2%
September 0.0024 23940 31% 0.004 12280 14% 0.005 5459 13.2%
October 0.0024 24480 31% 0.0025 23210 29% 0.0047 21910 14.8%
November 0.0025 23910 31% 0.0027 19790 26% 0.004 18960 15.4%
December 0.0024 24690 32.7% 0.0022 29780 24% 0.0047 29480 14.7%

the same simulated periods, better demand forecasting accuracy had been achieved. A

dierent range of improvement was recorded when simulating the proposed model with

the FBTFS throughout the year; mainly because of the dierence in the demand change

patterns throughout the year. It is observed that the results are generally better, except

in December and February. Realistically, an imperfect result is still expected, largely

because of the unmonitored behaviour of the controller, along with the unexpected type of

demand pattern. The results can be enhanced by tunning the model's parameters based

on the forecasting error rate. This task can be formulated as an optimisation problem

that aims to nd optimum modelling values based on the amount of error measured at

time of forecasting. Yet the proposed model is also expected to encounter this limitation

with other months when a completely dierent demand change is encountered. From

simulating the model with year C data and comparing Figure 3.12 with Table 3.8, a clear

understanding about the robustness of the proposed ANN in this role was established. This

reveals that Ki is reacting more than KP , as the demand changes, which shows that the

developed ANN was successfully reecting the smart search mechanism for the optimised

gain selection. Ultimately, the proposed optimisation technique was well modelled, thus

the forecasting model was achieving the required level of accuracy. Figure 3.12 shows KP
and Ki behaviour for the same simulation period that was presented in Figure 5.5. As

it is clearly shown in this Figure, KP was not active in changing the demand forecasting

accuracy in the working days from Monday till Friday; whereas in the weekend, it had

a large role in changing the demand forecasting output. Importantly these results reect

the attitude of the ANN, which has been taught how to select the best gain for each

69
demand pattern; consequently KP was found with higher values in the weekends. As it

is observed that the forecast error rate was generally higher on weekends than on working

days, changing K P was more benecial in achieving lower error rate. However, the results

show that Ki had more eect on the demand changes throughout the week. The results of

the analysis will give pertinent information for building a more accurate demand forecasting

model in future extensions to this work. Table 3.11 shows that the accuracy was improved

over the classical PI system by adding intelligence, especially by implementing the ANN

based, auto tuned, two-input-one-output fuzzy systems. The results also show that the use

of optimisation techniques to perform automatic gain selection, achieved better forecasting

with an average improvement rate of 5.6% compared with the manually tuned systems. To

show the results of the proposed forecasting model, Figure 5.5 illustrates the forecasting

output for the period from the 17th until the 23rd of January, year C.

Figure 3.12: KP and Ki behaviour in feedback-tuning fuzzy system ANN

70
Figure 3.13: The proposed model forecasting output for the period from the 17th till the
23'rd of January, year C

It also shows the added safety margin value over the actual load, in order to prepare

the forecasting for the generation scheduling application. In Figure 5.5, the results were

also shown in terms of the load prole for the normal working days and the weekend.

In the results shown in Figure 5.5, it is observed that there is an array that indicates

the number of the assigned generators, evaluated as a result of bundling the evaluated

demand into 500 kW lots. The nominated number of generation units was either increased

or decreased in each 500 kW demand increment or decrement as shown in Figure 5.5. It

is shown that the number of nominated generation units was evaluated by adding the

safety generation margin to the demand. Table 3.12 reects more details about the actual,

forecast and supplied electric energy to the case study, along with the generation schedule

based on the amount of demand. In this Table, (1) indicates active generation and (0)

indicates inactive generation.

Based on the demand (of multiples of 500 kW), a lost generation (reserve) amount equal

to the dierence between the actual demand and the generated energy from the nominated

71
 Table 3.12: Number of generators relating the predicted demand

Hour Generators Demand kW

G1 G2 G3 G4 G5 G6 G7 Actual Forecast Supplied

10 1 1 1 0 0 0 0 634 1162 1500

23 1 1 1 1 0 0 0 1212 1687 2000
55 0 1 1 1 0 0 0 795 1150 1500
92 0 1 1 1 1 0 0 1105 1596 2000
115 0 1 1 1 1 1 0 2151 2385 2500
124 0 0 1 1 1 1 0 1545 1715 2000
134 0 0 0 1 1 1 0 1028 1395 1500
142 0 0 0 1 1 1 1 972 1528 2000
157 1 0 0 1 1 1 1 1960 2285 2500
176 1 0 0 0 1 1 1 1276 1520 2000
205 1 1 0 0 1 1 1 1990 2450 2500
224 1 1 0 0 0 1 1 975 1491 2000
247 1 1 1 0 0 1 1 1970 2231 2500
256 1 1 1 1 0 1 1 2224 2545 3000
264 1 1 1 1 0 0 1 1996 2242 2500
280 1 1 1 1 0 0 0 1207 1605 2000
287 1 1 1 1 1 0 0 2067 2305 2500
294 1 1 1 1 1 1 0 2057 2595 3000

generation units was expected. A future extension to this work may be in researching

the reduction in the amount of lost generation through further optimisation. Figure 5.5

also shows that the assigned generator duty is rotated over the installed generators to

achieve longer waiting and operation times, which leads to a longer life time cycles for

the installed generation units. In the case study where the safety margin is added to the

demand value, it was anticipated that the lowest possible switching time from OFF to ON

or from ON to OFF for any generation units be not less than 3 hours; which is indicated

in the demand change between hour 100 and 150 in Figure 5.5. A clearer view about the

relationship between the demand forecasts and the generation scheduling with multiple

sources of dispatch is provided in Chapter 4.

3.7 Closing remarks

In this chapter, we presented an adaptive demand forecasting model where the actual real-

time load changes were included in the short term forecasting. Firstly, an investigation

was conducted into the design of a model of the load prole in a university type power

system. In addition, model stability was tested and improved by adding a self tuning

fuzzy system. However, the main aim was to develop a model which included the real-time

load changes in the forecast output. These changes were incorporated within the forecast

72
through the actions of an intelligent system performing adaptive demand forecasting. The

model employs an optimised parameter selection to suit the demand characteristics. The

results demonstrated that the optimised parameters selection for tuning the adaptation

parameters has been implemented successfully. Moreover, the results also proved that this

auto-tuned adaptation system was superior to the manually tuned systems in providing

an accurate forecast. Ultimately, in this chapter, we have successfully introduced a novel

demand forecasting model that adapts itself to the real-time load variation and reduces

the subsequent forecasting error accordingly. After scaling the forecasting values with the

necessary safety factors, the potential for utilising the adaptive forecasting in the generation

scheduling problem was investigated. Finally, the model output is then sent to the other

intelligent modules, responsible for scheduling and managing the installed resources in the

case study. For future extensions to this work, scope exists for an investigation into tuning

the adaptation fuzzy membership functions in order to further improve the accuracy of the

system.

73
Chapter 4

Power Generation Management and Storage

Devices Control System

This chapter details the second intelligent module in our management system. Since Stor-

age Devices (SDs) can contribute most to improving power generation performance when

they include other sources of dispatch such as renewable energy sources and micro-turbines,

the potential for utilising SDs more eciently is critically important. In this chapter, we

will look at the possibilities of achieving the lowest economic and environmental costs for

the targeted case study. The major challenges that we expected to be confronted with

in order to achieve these goals were: 1) tackling the uncertainty of demand, 2) genera-

tion cost, 3) availability of renewable energy sources, and 4) (charging/discharging) time,

and price for the installed SDs. Therefore, the goal of this research was an optimisation

technique, that encompassed a strategy to achieve the lowest generation cost from multiple

sources, and was also able to perform smart utilisation for the SDs with these sources. The

rst task was to estimate the best charging price for the SDs, in order to achieve a prof-

itable charging and to maximise the opportunity to participate during the SDs' lifetime;

as this would result in a greater eciency for the installed SDs. A Fuzzy Logic (FL) based

adaptive charging price was set for charging the SDs, which was based on monitoring the

microgrid's local generation price at the time of charging, and the amount of the daily SDs

participation in the microgrid dispatch. Thereafter, when the SDs status was evaluated,

another stage of calculation was conducted in 30-minute intervals, in order to evaluate

the optimum amount of dispatch from each participant source. A multi-objective Particle

Swarm Optimisation (PSO) method was applied to achieve the targeted low generation

costs. This chapter will detail the design and testing of the proposed technique as follows:

Sections 6.1 and 5.2 consider the historical background and relate other researchers' work

to the proposed method. Section 5.3 details the targeted problems and the main solution

74
objectives of the proposed method. Section 4.4 discusses the research assumptions that

underpin the design considerations in order for the research results to have signicant and

reasonable justications. For the planning side of this study, Sections 4.6 and 4.5 cover the

installation of renewable energy sources for the case study. Section 4.7 details the proposed

optimisation technique in this work Sections 6.6 and 6.7 discuss the results and provide

recommendations for future research.

4.1 Background

Just as the deregulation of electricity power markets has empowered the business of power

generation in the recent years, the prospective Smartgrid will move residential and small

business electricity power consumers/producers beyond the role of passive price accepters

in the near future. Therefore, dealing with the complex management of a network of

networks will be one of the most active research elds in power engineering. When con-

sidering our case study, the provision of management solutions for the power ow of the

decentralised resources will facilitate future connections with the Smartgrid. Although

sucient intelligence in the current traditional grid exists, performance can still be im-

proved by introducing more sophisticated infrastructure. Therefore, starting to upgrade

the decentralised power systems is a major participation in building the Smartgrid. The

current level of intelligence can support a high level of performance, including demand

response and smart utilisation for the installed resources. However, running a benecial

microgrid requires initial management strategies to control the resources within the grid.

Due to the high non-linearity in system operation models, there are diverse options for

improving the performance of the electricity supply for such power systems. The problem

arises due to the uncertainty of demand, generation cost, availability of renewable energy

sources and (charging/discharging) time and price for the participant SDs, if there is any.

This problem is considered to be a power engineering optimisation problem, especially

when the power ow encompasses intermittent generation in line with highly non-linear

demand. Eventually, ecient utilisation of SDs within microgrids among multiple sources

of dispatch has a substantial impact on reducing the economic and the environmental

generation costs. The dynamic charging price for the SDs was identied to be the key to

achieve a protable charging, and also to maximise the opportunity of participation during

the SDs lifetime. By considering both the economic and environmental generation costs

in the decisions made to optimise the performance of SDs, a multi-objective optimisation

method is required for this role. Therefore, studying the operational conditions of the case

75
study is crucial before selecting the right method and nominating the operational factors

that can reduce the generation cost. In this work, a general review about the targeted

case study will be covered in Section 4.5, to highlight the eective generation cost factors,

and also to identify the constrains that need to be considered in the calculation required

to make the optimum generation decisions.

4.2 Related works

The operational management of microgrids and large power grids was compared, and it

was found that from both the economic and environmental perspective, microgrids could

successfully reduce generation costs and CO2 emissions [145]. Computational intelligence

methods such as Mesh Adaptive Direct Search (MADS), FL, PSO, and Ant Colony Opti-

miser (ACO) have been introduced to deal with this problem by minimising the generation

cost by considering the environmental issues in [146, 147, 148, 149]. A mixed integer

nonlinear cost function was proposed to minimise the operation cost in [150], while the

potential nancial return for utilising electrical vehicles as a grid resource was addressed

in [151].

Successful microgrid management strategies can include other factors that result in a

better operational performance, such as smart utilisation of the resources and loads within

the managed microgrid. Optimal operation strategies have been formulated for smart

microgrids in [152], to accommodate the integration of renewable energy sources and max-

imise the prot for microgrid systems in Taiwan. On this basis, SDs have been nomiated

as a solution to maximise the prot and increase the microgird operation eeciency. SDs

have been proved to play a key role in improving microgrids' DER performance, and also

in providing more reliable ancillary services to the power grid in [153, 154]. Moreover SDs

have been utilised to stabilise the operation of microgrid by tackling the uctuation and

intermittence resulted from unstable micro-sources and nonlinear loads in [155].

In smart buildings, SDs have been applied to reduce energy cost in [156]. This proves

that SDs can tackle the variable generation price problem and can benet microgrids that

include autonomous generators, renewable sources and controllable loads.

The concept of demand response or smart load shedding can enhance the overall man-

agement performance in microgrids. Dealing with the behaviours of numerous consumers

in an intelligent fashion can result in a low cost generation for such microgrids. This

was proved by using a mixed integer linear program to include on-site electricity power

generation facilities for an organisation with numerous employees in [157].

76
 Since SDs can work as either loads or sources, depending on the situation, managing

them yields a complex optimisation problem. Microgrid operation uncertainty problem,

as a result of incorporating multiple type of DER including SDs, has been presented in

[158]. As discussed earlier in this chapter, while the complexity is elevated by the un-

certainty associated with their assigned dispatch, charging time and price, and life time;

their smart utilisation yields rapid performance improvement. Power quality issues are

still the major problems associated with incorporating SDs in the grid, so a decentralised

electricity SDs may solve power quality problems in microgrids [159, 160]. Microgrid power

quality problem has been illustrated in [161], where a novel approach has been proposed

to compensate for the voltage harmonics in a grid-connected microgrid. This problem has

been resolved using a proper control for the distributed generators interface converters to

provide a high voltage quality at the point of coupling. Electric vehicles have also been

utilised in balancing the microgrids' voltage and frequency. The control and management

strategies for unbalanced microgrid voltage that incorporates electrical vehicles have been

identied in [64]. The role of integrating SDs in building a reliable microgrid power dis-

patch has also been discussed in the literature, as they can dictate the economic benets

for the current grid, if they are upgraded with the needed infrastructure for Smartgrid
integration [162]. SDs (electric vehicles) have had major impact on reducing the capital

costs of microgrids [163]. SDs have also been utilised to manage the power ow in mi-

crogrids in [164]. Ultimately, the key issue for performing smart utilisation for the SDs is

their charging/discharging time and cost. A smart charging method with a load control

management module has been presented in [165].

A literature analysis has revealed that eorts to enhance the performance of the mi-

crogrid's operation through the use of numerous employees, is a key topic in the power

engineering research eld. Therefore, the study proposes a power generation management

system that seeks an optimum generation cost for the microgrid. Since the major chal-

lenge will be the SDs charging/discharging time and price uncertainties, their utilisation

(charging/discharging) control was targeted at this stage. The proposed control method

will consider demand and price changes, in order to achieve the optimum utilisation for the

SDs and the lowest possible generation cost for the same microgrid. This method will be

implemented and tested in this chapter as an independent system, works autonomously to

provide optimum decisions concerning the power generation in a microgrid. This method

will also be implemented as a decentralised module, controlled by a central processing

unit, to deal with more complicated operational scenarios in making optimum decisions

77
concerning the power generation of a microgrid. This implementation is not covered in

this chapter, and it will be detailed in Chapter 6.

4.3 Research objectives and challenges

Performing low cost generation for a microgrid that encompasses multiple sources of gen-

eration, including intermittent renewable sources (wind and solar), SDs (batteries and

electrical vehicles), distributed gen-sets (gas or diesel), and utility grid; is a complex op-

timisation problem. Therefore, to cope with the uncertainty of the demand and supply

from dierent power sources, a comprehensive study was needed prior to nominating this

optimisation method. The study was mainly conducted due to the necessity to inden-

tify the challenges encountered with operating the distributed generators in the microgrid.

Ultimately, the challenges identied are:

1. The intermittent nature of renewable generation sources.

2. The variable demand.

3. The dynamic utility price change.

4. The uncertain charging price and time SDs.

5. The diculty of nding simultaneous solutions for achieving low economic and envi-

ronmental generation costs.

The following sections will broadly identify the solutions for these challenges, along with

the design details for the proposed optimisation technique.

4.4 Research assumptions

The following assumptions were made prior to testing our power generation management

method:

1. Data are read from source les which have been pre-prepared to behave like real-time

systems.

2. The eective price change sampling time that is needed to set the SDs charging price,

is seven operational days.

3. SDs' lifetime is not considered in the calculations of generation cost.

78
 4. The electricity power generated from the microgrid resources is with reasonable power

quality.

5. Both microgrids' local generation cost and external utility cost are proposed to be

competitive with each other; varied in response to the operational conditions.

4.5 Case study

In this work we have targeted the power grid of Edith Cowan University (ECU), Joondalup

Campus in Western Australia as a case study for simulating the proposed optimisation

method. The case study is a High Voltage (HV) customer for the major power utility

supplier in Western Australia. In Western Australia, HV network connection is obliged

for customers consuming more than 2 MVA. This kind of connections forces customers

to install their transformers on the network, but it also allows customers to freely cong-

ure their Low Voltage (LV) network. The nature of connection of our case study allows

the system operator to freely select the design of the power network, including installing

distributed generators. On this basis, we study the possiblities of including solar panels,

wind turbines, storage devices, distributed gas gen-sets, and utility connection for supply-

ing the case study with the required power, as illustrated in Figure 4.1. From monitoring

the operational conditions around the Joondalup's campus power grid, a range of possible

generation scenarios were estimated. Accordingly, in order to obtain more precise results

about the generation scenarios so that simulation data for testing could be accurately

compiled, some planning studies were made for installing renewable energy sources on the

existing grid. Consequently, semi-realistic operational data for the generation scenarios

will be available to test the proposed optimisation technique for the case study. Although

some of the required simulation data were collected directly from the system's archive, the

rest were generated based on several engineering assumptions underpinning the planning

results. The simulation data were compiled using realistic historic operational conditions

pertaining to the case study, such as the weather conditions, generation cost, and the

internal demand.

The rst challenge in compiling the needed simulation data was in forecasting the

internal electricity demand for the case study. In the simulation, the demand data will be

received from the demand forecasting model agent every 30-minutes, based on predened

communication settings amongst the proposed agents. The other challenge in compiling the

simulation data was in evaluating the relative amounts of distributed generator penetration

79
 Figure 4.1: Multi-Agent based smart microgrid layout

for the case study. Based on the design considerations for installing distributed generation

units, a comprehensive study was made of solar panel and wind turbine installations for

the case study, as detailed in the next section.

4.6 Planning for installing solar panels and wind turbines

For this project, realistic operational scenarios were not available; therefore the research

was aligned with the planning for installing distributed generators on the existing ECU's

microgrid. The major planning limitation was the availability of installation space. A

sucient share of renewable energy supply, requires a high number of generation units

located within new or existing facilities; or at more remote installation sites. There is still

an alternative solution for solving the installation space problem, which is to increase the

utility grid share in such a way that is compatible with utilising all the available space

in the project land, and then supplying the rest from the utility grid. The Smartgrid
connection is another planning consideration. Since Smartgrid is a foundation for com-

mercialising energy suppliers, it will be a key solution for solving the installation space

problem. Smartgrid will allow the private sector to feed energy into the public grid. The

power consumers will be able to get their powerback from the Smartgrid from dierent

80
substations in dierent locations covered by the Smartgrid connection. Selling and buy-

ing power to and from the grid from dierent locations will impose other expenses, which

will be analysed from an economic perspective, and added to the project cost. Figure 4.2

shows the proposed planning strategy for adding renewable energy sources to an existing

microgrid.

ECU's microgrid at its Joondalup campus was selected as a case study for project

development planning. The average climate from minimum to maximum values and safety

limits are considered in the studies. The City of Joondalup is located in one of the world's

ve Mediterranean climate zones.

The geographical information for this city is

Country: Australia

State: Western Australia

City: Joondalup

Latitude: 31.44 °S

Longitude: 115.45 °E

Elevation / Altitude: 25 m

Average solar irradiation: 5.35 kWh/m²/day

Average wind speed: 12.24 km/h

The above details are helpful for identifying the installation challenges and likely solu-

tions when employing solar panels and wind turbines. For maximum output, solar-electric

(photovoltaic; PV) modules or solar thermal collectors need to be located where they re-

ceive the most sunshine. Ideally, the installation site should not be shaded by hills, trees,

buildings, or other obstructions at any time during the year; so identifying the best location

can be a complex undertaking.

Based on the university's power demand readings, the highest amount of demand per

hour occurs on summer's afternoons, reaching 3 MWh; while the lowest energy occurs on

winter nights or holidays, falling to a low of 0.5 MWh. To meet the existing demand at

ECU's Joondalup campus, a specic number of solar panels and wind turbines are needed.

As the highest demands are always during the day time, more emphasis should be placed

on solar panels, rather than on wind turbines.

81
Figure 4.2: Project development planning owchart

82
4.6.1 Solar panel installation

Based on the installation space available in the case study, we found that it is possibile

to utilise 5193.2 m² for installing solar panels. After an investigation on the cost and the

amount of generation to identify the most feasible solar panels for this project, the  Sharp
NUS5E3E 185 W mono  was selected. Each solar panel unit requires (1318 x 994 x 46)

mm2 of the available installation space. This results in an installation of 3964 solar panel

units, which is estimated to deliver up to 735 kWh at it maximum generation capacity.

Generation outcomes were estimated based on the City of Joondalup climate data,

and the generation eciency of solar panels from a variety of manufacturers. For best

generation outcomes, the installation angles of the solar panels must be adjusted according

to the location of the project, and the angle of the sun during the winter and summer

seasons. To optimise the installation of solar panels for the city of Joondalup, the angle of

o
inclination should be adjusted to 23 , facing north, as depicted in Figure 4.3.

Figure 4.3: Summer, winter and optimal angles for solar panel installation

83
 The average lost generation for solar panel was estimated to be about 10% of the overall

annual generation capacity; arising from such installation challenges as sub-optimal angles

of inclination and shading by trees and built-structures

4.6.2 Wind turbine installation

The useful energy contribution (work) gained by capturing the available wind energy, is

calculated by summing the amount of generated power over time, taking into account

variation in wind and operational conditions. The relationship between the size of the

wind turbine and the generation outcomes may vary, depending on the threshold point

for wind speed and the amount of wind speed available annually. From the climate study

results, it has been shown that wind speed drops to zero from midnight till sunrise on a

regular basis. The average wind speed is 12.24 km/h, while the minimum and maximum

(monthly average) wind speeds are 9.36 km/h and 15.84 km/h respectively. The minimum

wind speed required to activate generation in the nominated unit is 9 km/h. For a secure

power supply design, the utility supply will cover the missing hours of generation during

situations when both the solar panels and wind turbines are ineective. The wind speed

readings were collected from the City of Perth station, which is located about 20 km south

from the project location.

As a conservative estimate, the average monthly wind speed at 9 am for the City of

Perth, since 1993 is illustrated in Figure 4.4.

From the climate study, the estimated average of missing wind generation hours varied

between 20 hours in the winter season, and zero hours per day in the summer season. For

wind turbines, the technical layout and design of the wind park installations are important.

When installing several wind turbines concentrated in an array, the major challenge which

arises is the spacing between turbines. Hence in this study, wind turbines noise has been

ignored. The relationship between rotor distance and aerodynamic array eciency is illus-

trated in Figure 4.5. A minimum clearance between the wind turbines must be guaranteed,

otherwise power losses will be so high that the wind park will not operate economically.

84
 Figure 4.4: Average monthly wind speed in the city of Perth [3]

Figure 4.5: Aerodynamic array eciency as a function of rotor distance in the wind
direction [4], where D is the wind turbine rotor diameter, dl is the space length between
the two installed wind turbines in project land, and dc is the space width between the two
installed wind turbines in project land.

85
 In this case, the  WinPower 48V DC/240V 2000Watt  product was selected after

studying the suitability of many other products for this project. The selected unit carries

a 4 m rotor diameter. Accordingly, 300 m2 of installation space is required for each unit,

which provides for the 83 wind turbines for this project. Therefore, it is expected to receive

up to 250 kWh from the proposed wind turbines.

This inuences the required space needed for this project at Joondalup, where space

can be limiting. However, for the Smartgrid connection, an o-campus location is used

and land cost becomes a factor. For this project, a satellite image for the project's land is

added to identify available allocated installation space for solar panels and wind turbines

as shown in Figure 4.6.

Figure 4.6: A satellite image for the case study shows the available allocated installation
space for solar panels and wind turbines

Other calculation assumptions for the missing generation hours due to clouds and low

speed wind have to be considered under generation capacity for the installed renewable

energy sources. The average calculated missing hours of generation are summarised in

Table 4.1, based on the 10 years of weather data obtained from the Australian Bureau of

86
 Table 4.1: The daily average of missing generation hours

Energy Source Day time (8:00-17:00) Night time (17:00-800) Total missing hours

Wind turbine 3 hours 7 hours 10 hours

Solar panel 2 hours 13 hours 15 hours

Table 4.2: Estimated generation capacity for the simulated resources

Source Capacity (kWh)

GP 0-3500
P hG 0-735
WG 0-250
SG 0-350
UG 0-Dmnd

Meteorology [3].

4.6.3 Other resources installation

In case of gas gen-sets; other studies were applied in order to identify the design require-

ment for installing distributed gas generators on the campus property. The major concern

regarding the installation of the distributed generators, comes from the demand limits.

Depending on structure and the load distribution in the case study, it was more suitable

to bundle the demand into 0.5 MW lots; thus a set of 0.5 MWh gas generators have been

identied for this design. Therefore in order to cover this demand, the design requires

the installation of local generation units that provide 3.5 MWh, based on the inclusion of

an extra generation unit for safety and emergency requirements. Finally, in the proposed

management software, seven gas gen-sets, each of 0.5 MWh of capacity, were incorporated

to meet the maximum demand, including the added safety dispatch. For storage devices

needs, it was estimated that 0.35 MWh would be needed to provide the total required

generation capacity.

The nal simulation data ranges are illustrated in Table 4.2.

Subsequently, gas generation costs were estimated based on gas price information for

Western Australia in [166]. It was necessary to convert the units of gas supply into N m3 /hr
to follow the gen-sets fuel consumption units standard:

1N m3 /h = 0.031736GJ (4.1)

There is also an additional variable cost resulting from the operating and starting-up

87
and shutting-down costs for the installed units; in addition to the xed annual maintenance

and insurance costs, where the nal gas generation cost is evaluated as follows:

Gp = G + M oV + M oAn (4.2)

Based on a management decision, the gas generation was prioritised as the second

major supplier after the utility, due to the sustainability of the utility supply. The future

electricity utility prices in Western Australia can be predicted by multiplying the price by

the dynamic energy price change pattern observed in Australia's eastern states. The price

change pattern data was provided by AEMO [167].

4.7 Power generation management and storage control

The proposed management method encompasses three stages of calculation: 1) microgrid

internal operational conditions monitoring, 2) SDs operation control using FL system, and

3) overall optimum generation cost management decisions using PSO. Figure 4.7 illustrates

the proposed SDs charging management strategy block diagram.

The three stages of calculation are illustrated as follows:

4.7.1 Monitoring the microgrid's internal operational conditions

In this work, a method for monitoring the microgrid's internal operational conditions, is

applied to build a rule base for charging the SDs. Based on evaluating the economic and

environmental cost factors in the most recent seven days of operation, an adaptive rule

base can be obtained to tackle the non-linearity of the operational conditions. The SDs

charging price is set as the dierence between the mean of the maximum daily prices, and

the mean of the minimum daily prices for Up or GP . These values are updated every 24

hours, to follow the most recent daily price change trends based on monitoring the most

recent seven days of operation. Figure 4.8 shows a sample of seven days of price change,

which helps to identify the minimum and maximum price values, thus nding GChP and

UChP , which will in turn set the discharging price for SDs.

For discharging; decisions are made by the PSO in the later stages of calculation which

will be discussed in subsection 4.7.3. However, a decision will then be made, based on

comparing the SDs dispatch cost with other dispatch parameter cost; e.g., gas, and utility

dynamic generation price. To implement smart involvement for the SDs in dispatching

the electricity power for the managed microgrid, a fuzzy system described in subsection

88
Figure 4.7: The propsoed SDs charging management strategy block diagram

89
 Figure 4.8: A sample of seven days generation price change for GP

Table 4.3: Local generation price based on the demand, the availability of the sources and
the dynamic generation price within the managed microgrid

Price RE>Dmnd RE <Dmnd<RE +SG Dmnd > RE +SG

GP <UP <SP 0
GSG
GP ∗ ( Dmnd ) GP ∗ ( Dmnd−(P
Dmnd
hG +W G )
)
GSG GSG
GP <SP <UP 0 GP ∗ GG +P hG +WG GP ∗ Dmnd
P −SP )
SP <GP <UP 0
SP ∗SG
Dmnd GP − ( SG (G
Dmnd )
UP ∗UG UG
UP <GP <SP 0
Dmnd UP ∗ ( Dmnd )
UP ∗UG UG
UP <SP <GP 0
Dmnd UP ∗ ( Dmnd )
SP <UP <GP 0 SP ∗ ( Dmnd−(P
Dmnd
hG +W G )
) UP ∗ ( Dmnd−(P hG +W G )
Dmnd )

4.7.2 was developed with two inputs. The rst input is the local generation price, which

was initially evaluated by studying each and every operational scenario in the microgrid.

Eventually, it was represented by the equations illustrated in Table 4.3. These equations

were derived to reect the amount of generators' participation in supplying the power

required to match the microgrid's demand. The operational scenarios are identied by

three factors: 1) the demand value, 2) the resources generation availability, 3) and the

dynamic generation cost. By relating these factors to each other, a decision tree was

established to represent the LGP at each scenario.

where

90
 RE = WG + P hG (4.3)

After intialisation, LGP is evaluated by equation 4.6. The details about the proposed

cost functions are utilised in this thesis are explained in sub-section 4.7.3.

The second input is the actual daily accumulated SDs participation in the microgrid

dispatch. The local generation price should aect the operation status for the SDs in the

microgrid electricity power dispatch. This favours creating an adaptive charging price;

to follow the operational conditions, and to maximise the chance of participation in the

microgrid's electricity power dispatch. For the best amount of SDs participation within

the microgrid, half of the total operational time should be the ideal target for the SDs,

with the other half being ideal for charging. As per optimum operation, storage devices

are anticipated to work continuously between charging and discharging, thus dividing the

operational time 50%:50% between charging and discharging.

4.7.2 Adaptation fuzzy system

This section discusses the switching mechanism, which is integrated with the PSO in

subsection 4.7.3 to control the behaviour of the storage devices. A signal indicator was

developed that indicates the SDs status (charging, discharging or standby). The indicator

compares the SDs instantaneous capacity with the full practical SDs capacity, and generates

a signal whenever the capacity goes below 100%. This signal will be treated under other

pricing conditions:

1) when the local generation price is below the adaptive pricing threshold, a charging

signal will be activated.

2) when SDs charging price is less than the price of other generation sources, a dis-

charging signal is activated to dispatch the power to the microgrid's loads.

3) when SDs charging price is greater than the local generation price, or greater than

other generation units generation price at the time of comparison, a standby signal is

activated.

To normalise these values for the fuzzy system input, the actual accumulated amount of

SDs participation value is multiplied by two, to establish an input membership functions'

universe of discourse with a range of one. This resulted from combining the charging and

discharging times in the microgrid's operational participation. The optimum accumulated

amount of the SDs for the most recent seven days of operation is evaluated as follows:

91
 336
1X SG (h)
T he Amount of P articipation = 2∗ (4.4)
h SG max
h=0

where h is the half hour operation interval in the seven days of operation.

It is proposed that the output of the fuzzy system carries the amount of the charging

price change ∆GChP , which updates the charging price value as follows:

GChP (t + 1) = GChP (t) + ∆GChP (4.5)

where t is a 30-minutes operational interval.

The proposed fuzzy system included triangular and trapezoidal membership functions.

Figure 4.9 illustrates the proposed membership functions design, for the proposed charging

price adaptation fuzzy system.

Figure 4.9: The design of charging price generator fuzzy system membership functions

The relationship between the two fuzzy inputs and their output is detailed in Table 4.4,

which illustrates the rules that adapt the SDs charging price in the managed microgrid.

92
 Table 4.4: Charging price fuzzy rule-base system

Amount of Participation
LGP
Low Medium High
Low Low Low Medium
Medium Medium Medium Medium
High Medium High High

4.7.3 Optimum generation cost performance

The Particle Swarm Optimisation (PSO) method is applied to solve the generation cost

optimisation problem in the managed microgrid. It also covers the changes to the dynamic

variables and the solution constraints in the economic and environmental generation costs

within the PSO's considerations. Accordingly, the proposed PSO will consider SDs as

loads, sources or not counted, based on the generated signals from the charging price FL

system at the time of processing. Hence the proposed PSO aims to nd an optimum

solution for minimising the economic and environmental costs by introducing a general

multi-objective cost function, which is described as follows:

n SwarmSize
P P
(GG + GlostG )i ∗ (Discharging + Gc + M oAGc + M oV Gc )M oAi
Fmin (Cost) = c=1 swarm=1
+SGi ∗ (Sc + M oV Sc ) ∗ Charging + UGi ∗ (Uc + M oAUc + Discharging)
(4.6)

where Fmin is a multi-objective cost function, c is the cost function to be tested,

n is the number of the tested cost functions, and swarm is the PSO swarm size. In

all applied cost functions, there are variable and xed cost factors for each source of

dispatch. For utility, M oAU represents the cost of the annual utility feeder utilisation

percentage reserved to supply the power for the microgrid ($/kWh). In the environmental

cost considerations, it was assumed that M oAU =0. In Gas gen-set generation, M oAG
represents the xed annual insurance, inspection and operating cost ($/kWh). In economic

cost considerations (M oV G) represents the variable maintenance and starting and shutting
down costs ($/kWh), while in environmental cost considerations, it represents the gas

emission cost (Kg/MWh).

Furthermore, the environmental and economic cost functions are subject to the follow-

ing constraints:

a) SDs charging/discharging values are evaluated as follows:

93
 
SDsn,t + An ∗ Pn,t ∗ ∆T if SDs are charging






SDsn,t+1 = SDsn,t − An ∗ Pn,t ∗ ∆T if SDs are discharging (4.7)




Sn,t if SDs are idle




1/ηC






An = 0 (4.8)




 ηD


where Sgn,t is the power stored in the nth storage unit at time t. An is the nth storage
units charging/discharging eciency. ηC and ηD are the charging and discharging factors

respectively. Storage device charging/discharging time is written as follow:

SD
T = (4.9)
IP n

where SD is the theoretical capacity, I is current, and Pn is the Peukert number.

b) All updated velocities of the selected initial values of each generation unit must not

exceed the capacity of the source itself:

UGmin 6 UG
SGmin 6 SG 6 SGmax (4.10)

GGmin 6 GG 6 GGmax

c) When each gen-set is covering 500 kWh, the nominated dispatch values maximise the

spinning reserve utilisation from the running gen-sets. Therefore, the gas gen-sets dispatch

cost is formulated as follows:

(int( G
500 ) + 1) ∗ 500 − GG
G
GlostG =( ) (4.11)
500

d) The sum of all of the tested dispatch should always be greater or equal to the value

of demand at the time of decision (t) in the managed microgrid:

GG (t) + SG (t) + UG (t) + WG (t) + P hG (t) ≥ Demand(t) (4.12)

e) The cost function must consider the SDs operational status, which is represented by

environmental and economic cost function indicators. This indicator is applied to switch

the searching mechanism, and thus prioritise the utilisation of the storage devices when

94
they are ready for dispatch. Another cost indicator was integrated within the economic

and the environmental cost functions, to indicate the charging-discharging status for SDs.

f ) All velocity equation updates result in positive values to guide the searching towards

selecting positive dispatch values:

k+1 k + (P k k
Vi,j = w.Vi,j best.i,j − Xi,j )C1 .rand1 ()+
(4.13)
(Gkbest.i,j − Xi,j
k )C .rand ()
2 2

k+1 k k+1
Xi,j = abs(Xi,j + Vi,j ) (4.14)

k+1
where Vi,j is the updated amount of the dispatched power (velocity) for each dispatch

source j at iteration i, C1 and C2 are the learning factors and w is the inertia weight. By

referring to other people works [168, 169], in line with simple empirical studies applied to

nominate PSO parameters, w has been set to 0.4, C1 and C2 have been set to 1.4 and

rand1 () and rand2 () are random number generators, set to provide numbers between 0

and 1 .

g) The update process is repeated based on a predened number of iterations i as shown

in the ow chart illustrated in Figure 4.10.

95
 Figure 4.10: The proposed Particle Swarm Optimisation update process

4.8 Results

We show in detail the microgrid's overall management performance within the proposed

method. Accordingly, the integrated optimisation roles were tested by simulating the sys-

96
tem with the operational scenarios data. The results show that the initial proposed PSO

successfully maximised the utilisation of the spinning reserve of the installed gas gen-sets.

It also showed that the method could control the dynamic generation cost dispatch, with

the intermittent generation from the renewable energy sources. Tables 4.5 and 4.6 illus-

trate the performance of the proposed PSO and Fuzzy-based Particle Swarm Optimisation

(FPSO) method for a typical 12-hours daily operation. From both Tables 4.5 and 4.6,

it is shown that this work was rstly successful in modelling the SDs attitude by reect-

ing the charging process in line with evaluating the variable SP as the SDs are charged.

Secondly, the studies conducted on the microgrid's operational conditions could identify

the charging/discharging cost parameters. Eventually, controlling these parameters using

FPSO could make the charging price cheaper as compared with the traditional PSO. These

results highlight the robustness of the proposed adaptive optimisation method (FPSO) in

switching its calculation mechanism based on the operational scenarios that were encoun-

tered. By following the charging process from hour 1 to 7, it is shown that while SDs is

charged, their price is inuenced by the local generation cost that results from evaluat-

ing the share of generation of each participant generator at time of charging. It is also

shown, by comparison, that SP is relatively cheaper when the microgrid is ran by FPSO

than it is with the traditional PSO. Although, the dierence is tiny, but there will be a

larger dierence in the longer term of operation, and also a higher amount of participation,

which in turn results in a lower cost for the power generation in the microgrid. The cost

dierence is further detailed in Table 4.7. The results also show that the interpreted cost

function was successfuly utilised to reduce the amount of lost reserve, by maximising the

share of gas gen-sets and matching the number of active units with the power required by

the microgrid. Hence the additionl safety unit cost was divided over the amount of power

supplied by the microgrid, and eventualy considered in the local generation price.

Figure 4.11 shows the control for the SDs generation capacity during their charging

and discharging times.

The results also proved that the SDs can play a substantial role in balancing the opera-

tion within a remote self-supplied microgrid. Moreover, the role can be utilised to enhance

SDs eciency. FPSO method was tested for its ability to control the charging/discharging

of the SDs, based on monitoring their operational performance. The method was evaluated,

based on its ability to increase the amount of SDs participation in the microgrid electricity

power dispatch with the lowest possible generation cost. It was found to bring a range of

benets to the system in terms of Number of Participation (NOP), Supplied Power (SE),

97
 Table 4.5: Microgrid generation performance evaluation with FPSO

Unit Generation capacity (kWh) Price ($/kWh)

H Dmnd P hG WG UG GG SG GP UP SP GN o.
1 1883 0 92 0 1883 10.5- 0.0991 0.1971 0.0782 5
2 1862 0 208 0 1863 6- 0.1180 0.2043 0.0940 5
3 1846 0 235 0 1846 3- 0.1200 0.1966 0.1026 5
4 1879 0 235 0 1787 2- 0.0120 0.1956 0.1026 5
5 2188 65 235 1882 0 1- 0.2376 0.2208 0.1015 0
6 2425 187 227 0 2253 0.5- 0.1376 0.2520 0.1040 6
7 2550 300 92 2157 0 0.5- 0.2146 0.2556 0.1041 0
8 2671 429 201 1949 0 177+ 0.2140 0.2772 0.1044 0
9 2709 544 0 0 2459 10+ 0.1140 0.3212 0.1044 6
10 2727 646 0 2012 0 67+ 0.2146 0.3367 0.1044 0
11 2708 642 0 0 2458 7+ 0.1363 0.3319 0.1044 6
12 2589 520 0 1961 0 45+ 0.2146 0.3123 0.1044 0

Table 4.6: Microgrid generation performance evaluation with PSO

Unit Generation capacity (kWh) Price ($/kWh)

H Dmnd P hG WG UG GG SG GP UP SP GN o.
1 1883 0 92 0 1807 15- 0.0991 0.1971 0.0782 5
2 1862 0 208 0 1851 13- 0.1180 0.2043 0.0940 5
3 1846 0 235 0 1846 8- 0.1200 0.1966 0.1026 5
4 1879 0 235 0 1809 4- 0.0120 0.1956 0.1026 5
5 2188 65 235 1887 0 3- 0.2376 0.2208 0.1015 0
6 2425 187 227 0 2329 1- 0.1376 0.2520 0.1040 6
7 2550 300 92 2157 0 1- 0.2146 0.2556 0.1041 0
8 2671 429 201 1864 0 91+ 0.2140 0.2772 0.1044 0
9 2709 544 0 0 2449 3+ 0.1140 0.3212 0.1044 6
10 2727 646 0 2015 0 131+ 0.2146 0.3367 0.1044 0
11 2708 642 0 0 2375 4+ 0.1363 0.3319 0.1044 6
12 2589 520 0 2024 0 58+ 0.2146 0.3123 0.1044 0

98
Figure 4.11: Storage Devices charging/discharging status for a typical one week of oper-
ation in the managed microgrid

Table 4.7: Storage devices performance comparison

Performance PSO FPSO

N OP 89 124
SDs SE kW 19801 20727
Money $ 4450 4659
AP 237 481
Microgrid's January total demand (kW) 1088225

Achieved Prot (AP), and generation cost. The simulation results are discussed below.

Table 4.7 illustrates the dierence in the values resulting from adding the proposed fuzzy

pricing method to the typical PSO.

From Table 4.7, it is shown that generally running SDs would result in a higher cost

as compared with UP and GP . However, due to the disparity of the UP and GP , SDs can

be smartly utilised to make a cheaper power dispatch as compared with UP and GP . It

is also shown that the traditional PSO could manage this uncertainty and could make the

SDs protable by selecting the suitable dispatch time to maintain optimum dispatch cost.

However, the proposed fuzzy charging mechanism was successful to guide the traditional

PSO towards making optimum decisions concerning the charging and discharging process.

Eventually our method has increased the N OP by 28.2%, SE by 4.5% and AP by

50%. These increments resulted from following the pattern of weekly prices around the

99
microgrid's operation. Hence, investigating the best time domain for the prices change

could be an optimisation problem to be addressed in future research. The performance

results should prove to be attractive to potential investors in this sector.

4.9 Closing remarks

In this chapter the target was to test a novel optimisation method aimed to perform smart

utilisation of the resources in generating electricity power for a typical autonomous micro-

grid. The proposed method is applied to control the charging process of the SDs through

a fuzzy charging price threshold system. The proposed system included monitoring the

microgrid's local generation price, and the total amount of SDs dispatch in the most recent

seven days of operation, to establish an optimum charging/discharging prices for SDs. The

generated charging prices aim to maximise the opportunity to charge the SDs with low

prices, thus maximising the likelihood of achieving cheaper dispatch throughout the SDs

lifetime. After evaluating the charging price, a switchable PSO mechanism was applied to

control the power dispatch, based on the dynamic generation cost of producing electricity

within the microgrid. The results showed a higher amount of SDs participation, thus a

higher amount of potential prot can be achieved with the proposed system. The higher

prot in turn reduces the overall generation cost of the managed microgrid. Future research

should investigate the demand response, and the ancillary services trade considerations,

under a competitive prices environment; for the given optimisation method.

100
Chapter 5

Fuzzy Logic-Based Ancillary Service Adaptive

Pricing System

This chapter details the third intelligent module in our management system. At this stage

we aim at performing two way power ow for our microgrid, and therefore, we have assumed

that the micgrogrid has two-ow meters for importing and exporting electricity, from and to

the utility grid. It is clearly understood that importing electricity was less problematic than

exporting it, due to the complexity of markets, the capacity of dispatch, administrative fees

and quality issues. Therefore, even if there is a chance of exporting electricity to the grid, it

will always be sold at a lower rate. In this chapter, we are going to illustrate and analyse the

possibilities of increasing the electricity export rate from microgrids with more competitive

Ancillary Services (AS) pricing. The chapter will detail the pricing principles used for AS.

In addition, it will look at the impact of adding appropriate Articial Intelligence (AI)

methods to increase the chance of market participation in the longer run. The proposed

pricing mechanism aims at adapting xed initial pricing rules to the microgrid's past,

present and future operational conditions. Accordingly, several levels of intelligence have

been integrated with the proposed pricing module to enhance the eciency of the applied

rules. In this chapter, Section 6.1 introduces the types of AS, and detailing their role in

maintaining safety and reliability in power systems. Section 5.2 studies the technologies

and the methods utilised by research community to perform smart energy trading. Section

5.3 will detail the targeted research problems and the objectives aimed at this work. Section

6.4 will simply show the research assumptions that have been considered before performing

our simulations. Section 5.5 will detail the exploration of the operational conditions in the

microgrid. Section 5.6 will discuss the methods of extracting the adaptive pricing rules

which includes: 1) subsection 5.6.1, where the main pricing equations are formulated, 2)

subsection 5.6.2, where the relevant forecasting models are detailed, 3) subsection 5.6.3,

101
where a fuzzy rule base system based on correlating eective pricing variables to achieve

satisfactory prot is detailed. Section 5.7 has covered the simulation tools utilised to

create a virtual competitive market for testing the proposed pricing method. Section 6.6

will discuss the recorded results, and nally Section 6.7 carries the conclusions and the

recommended future extensions of this work.

5.1 Background

The electricity rules force the market operators to ensure achieving safe and reliable op-

eration for the power systems. In this context, the frequency and voltage proles are the

major considerable issues. Ancillary Services (AS) can play a major role in balancing the

voltage and frequency proles in the power system. AS are generally classied as: 1) Load

Following - the power required to balance the amount of demand and generation, and en-

sures supporting the system when any of its scheduled generation units fails; 2) Dispatch

Support - the ancillary power required to maintain the voltage level; 3) Load Rejection

Reserve - the generators ability to reduce their dispatch when a load operational fault is

encountered; 4) System Restart, the power that can be supplied from the running gen-

erators to black start additional generation unit without importing extra power from the

transmission source; and 5) Spinning Reserve - is the power held as part of the generators

capacity to respond to the network demand when any other generation unit experiences

an outage. By considering our case study (microgrid), we targeted a combination of Load

Following, Dispatch Support and Spinning Reserve types of AS that can be oered to the

market. Specically, two types of reserve was identied to be oered as AS to the market:

1) the fast reserve, which is the rapid and fast response active power supply that results

from increasing the torque of the operational generation units, or from a planned load

shedding within the managed microgrid; 2) standing reserve, which is the active power

supply that can be obtained from extra synchronous or asynchronous generation units in

the microgrid. However, AS can be integrated with the power grid after a set of negoti-

ation between the grid and sellers. Depending on the amount required from the market

(utility grid), and the amount oered to the market at any particular time, variable and

competitive price might be encountered for the AS trading. Hence oering the reserve to

the market becomes a challenging process. Therefore, our pricing rules aim at maximising

the amount of AS trading with the market.

A comprehensive study for the market operation mechanism was needed to guide the

development of the AS pricing. Generally in the electricity market, auctions are divided

102
vertically and horizontally. In the latter, the daily demand in a particular spot is parti-

tioned by its duration, which is considered as a distinct lot for each time frame. Whereas

in case of the vertical auction, the daily demand is divided into hourly demand lots, where

each hour demand lot contains the full demand of that particular demand zone. In both

types of auctions, our pricing equations were formulated based on two pricing rules: 1) the

uniform rules, which allow the auction winners to pay the highest accepted bidding price;

and 2) the discriminatory pricing rule, which allows the winners to pay their own proposed

bids. However, although the uniform price auction is expected to achieve a higher num-

ber of transactions, which makes it a more ecient auction format [170], it is restricted

to the amount of supply and demand for electricity in the market. The auction design

has been tested with the ability of learning in the context, where symmetrically informed

adaptive agents with common evaluations learn to bid for a good in [171]. The auction

types (First-price Sealed-bid, Vickrey, English or Dutch), based on the complexity of the

markets and the number of participant players, have been investigated by [172]. In Aus-

tralia, the English type auction is the auction that electricity market operators use, where

all the bidders use their best strategies to increase their bid until it reaches their minimum

protable capability. Essentially both sellers and buyers wish to maximise their prot,

where the buyers look to buy from the cheapest selling price. The current Australian en-

ergy market relies on the Australian Energy Market Commission's (AEMC) rules, which

are applied depending on the place and the size of energy trade. In the eastern and south-

ern Australian states, the Australian Energy Market Operator [167] is the pool that hosts

the electricity generators' bids to the retailers, who in turn sell the electricity to the end

consumers. In the Northern Territory, electricity reforms have been introduced, but at

present there is no competition in the generation or retail markets. By way of contrast, in

the state of Western Australia, the Independent Markets Operator [173] operates both the

Wholesale Electricity Market (WEM) and the Gas Information Services Project (GISP).

5.2 Related works

Price modelling has been extensively covered by researchers in both engineering and eco-

nomics research elds. A new price modelling approach has been presented in [174] to

construct noisy supply and demand processes, and then equate them to nd an equilib-

rium price. An accurate pricing model that combines social, technical, environmental,

and economic aspects using an analytic network process to show the impact of the renew-

able energy pricing has been discussed in [175]. Integer-programming [176] and non-linear

103
programming [177] have been introduced respectively to formulate prices, based on unit

commitment of electric power generators, and to maximise the retailer's prot under a

deregulated electricity market. Generic mixed integer linear programming mode has been

proposed to resolve residential microgrid generation cost in [178] by involving a predictive

control method to reduce the generation cost and cope with the uncertainty of installed

DERs under dierent market pricing rules.

Various marketing strategies for the microgrid power market, such as spot marketing

and pricing, which can be applied to forecast and ensure its economic viability, have been

addressed in [179, 180]. A buy-back program that incorporates the compensations and

the setup cost over nite planning horizons has also been modelled in [181]. The strategy

of the spot market price of microgrids and the new pricing formulae have been detailed

in [179] to show how the microgrid can foster an ecient market. Microgrids AS market

participation has been represented as an optimisation problem in [182], which has been

modelled using fuzzy sets concept to allocate three AS types among microgrid agents

in electricity markets. As a result, the commercialisation of the microgrid generation in

context of smart, economic and market participant generation was the major identied

challenge. . Various marketing strategies, such as bidding, spot marketing, and pricing,

for ensuring economic viability of a microgrid scenario have been investigated in [183].

Developing an energy management system that relies on the spot electricity market

in its optimum decisions has been proposed in [184]. [185] has investigated a case where

microgrids are considered as market competitors for generation companies in Fort Collins.

The operation of microgrid as part of Colombian energy market has been investigated

in [186], where a linear programming algorithm has been proposed to perform an opti-

mum operation for microgrids under the uncertainty of electricity market participation. A

market-based mechanism which allows a single microgrid operator to control the behaviour

of the internal loads, has also been indicated in [187, 188].

By considering the dynamic electricity demand changes and monitoring the factors that

inuence prices, dierent approaches can be utilised to build smart integration for the local

generation units within the utility grid. A pricing mechanism for supplying the microgrid

electricity in a competitive electricity market, participating in the bidding process, and

settling the market-clearing price has been proposed in [189]. A price-based open-loop

control signal to facilitate enhanced penetration by distributed energy resources in the

power system; by coordinating their participation in electricity markets while also main-

taining the local system energy balance, has been considered in [190]. In addition, having

104
the right method to generate competitive bids in a competitive market environment can

increase the range of electricity trades for the managed power system.[191] has proposed

an electricity market oriented tool that creates bidding strategies in a competitive market

environment by combining fuzzy logic and deterministic approaches, while [192] has pro-

posed a pricing analyser and generator to initiate competitive bids in electricity markets.

Microgrid market participation with distribution system has been investigated under dif-

ferent market policies and pricing rules in [193]. Accordingly, to increase the performance

of eeciency of the microgrid's DER and increase market participation, this problem has

been solved using PSO. A novel fuzzy modelling approach, to generate strategic bidding to

handle uncertainty in market parameters, such as load demand, generator bidding, power

dispatch, price, cost, revenue, and prot, has been detailed in [194]. In this section, we

have critically reviewed the methods and technologies utilised in trading the AS with the

markets. Hence in this work we will also refer to the pricing method utilised by [194] to

develop our pricing method supported by the operational conditions forecasts. Ultimately

this will build more ecient pricing method that suits our case study and also the market

rules. Finally, our pricing rules will be implemented to cope with the problems discussed

in the literature, and also to reect our strategic ideas relating to selling AS to the market.

The rules will be implemented and tested in this chapter as an independent seller, which

tries to sell the AS in a competitive market. The rules will also be implemented as a bigger

management system that consider other operational conditions in the decisions, which will

be covered in Chapter 6.

5.3 Research objectives and challenges

Selling the AS to the market requires advanced network infrastructure, which in turn

results in political restrictions due to the high cost of development and implementation.

Therefore, justifying the substantial economic and environmental benets will help attract

investment when building a more reliable grid like the Smartgrid [195, 196, 197]. In

the meantime, researchers and engineers seek solutions for solving this problem, either

by nding alternatives or by utilising intelligence in the power grid in order to cut these

expenses. By aligning our research with other researchers and engineers, we are targeting

the following objectives by proposing our AS pricing rules:

1. Identifying the operational scenarios that maximise the chance of market participa-

tion.

105
 2. Maximising the level of market participation through minimising the selling price in

the longer term.

3. Maximising the amount of power oered to the market through the achieved market

participation.

5.4 Research assumptions

The assumptions that we are relying on at this stage of our investigations are:

1. The microgrid negotiation is connected directly to the market and not to retailers or

any other third party.

2. We are targeting the UP as the indicator of the market balance point for buying and

selling electricity.

5.5 Possible scenarios and rule base system

Controlling the microgrid's resources under the complexity of manipulating multiple sources

of dispatch, is the art of selecting the appropriate management strategy. Therefore, a com-

prehensive analysis of the system's operational conditions is needed before nominating the

right strategy. Our approach aims at classifying the expected operational conditions based

on the cost of generation. However, due to the complexity and the variety of sources;

including the installed onsite renewable energy sources, with dynamic demand, utility and

gas/diesel dispatch price, we related the amount of demand within the microgrid to the

variable generation cost that results from the intermittent generation and dynamic utility

price. Based on the extracted operation scenarios, a suitable pricing equation is set for

each operational scenario. Hence the complexity of the rules increases when the SDs are

considered in operation in the case study; due to the uncertainty of charging/discharging

price and time. The multiple sources of generation within the case study are initially clas-

sied based on three dierent demand levels: 1) less than the available renewable energy

sources dispatch; 2) more than the renewable energy sources dispatch and less than the

renewable energy sources together with the SDs dispatch; 3) more than the renewable

energy sources together with the SDs dispatch. After classifying the operation scenarios

based on the demand levels, we then sub-classify the operation at each level based on the

generation cost for each generation unit: GP , SP and UP . Eventually these dynamic cost

factors would complicate the problem and increase the number of rules that are required.

106
Figure 5.1: The decision tree for the classied demand levels as compared with renewable
energy sources and storage devices generation capacity scenarios

Depending on the importance of each of the generation sources to the case study, the rules

are developed based on comparing gas generation cost with the utility generation cost,

and comparing the gas generation price with SDs generation price. As a result, a suitable

pricing that aims at maximising the benets and cutting the generation cost was initially

generated for each operation scenario.

The extracted decision tree for the expected operational scenarios from the case study

is illustrated in Figure 5.1.

Table 5.1 was utilised to represent the relationship between the demand levels and the

expected generation cost scenarios illustrated in Figure 5.1.

where RE is the total amount of P hG and WG generation. The relationship identied

in this table will be utilised in the next section to formulate the initial pricing rules for the

AS.

107
Table 5.1: Possible expected operation cost scenarios; with regards to the level of expected
demand in the managed microgrid

Price scenarios (1) RE>Dmnd (2) RE <Dmnd<RE +SG (3) Dmnd > RE +SG
GP <UP <SP A G M
GP <SP <UP B H N
SP <GP <UP C I O
UP <GP <SP D J P
UP <SP <GP E K Q
SP <UP <GP F L R

5.6 Decisions mechanism and Ancillary Services pricing equa-

tions formulation

Our proposed pricing mechanism initially reects our trading attitude, which will set an

initial sale price and monitor the further opportunities for generating more competitive

prices based on the market and the microgrid operation conditions. Since the ability to

change the sale price is typically needed to cope with these conditions, we have considered

dierent three implementation stages to build the proposed pricing system. In the following

subsections, we describe the pricing rules in detail, along with their ability to adapt to

external conditions.

As weather conditions and energy prices are highly correlated [198, 199, 200], the rst

eective parameters identied regarding the prices for the case study the amount of SG ,
P hG , WG and Dmnd. Evaluating the amount of these factors will result in counting the

amount of RdyF rSLElec (reserve). To increase the scope for monitoring the operational

conditions, forecasting models and optimisation techniques have been utilised to estimate

the limits of the pricing factors in the decision making. Forecasting the generation pa-

rameters has been considered as a substantial factor for increasing the net earning of the

generation in [201], and the future electricity price classication forecasts have been utilised

in the decision-making process in [202]. To estimate the long term risks in the electricity

markets, a swarm-intelligence, meta-heuristic, optimisation technique have been proposed

in [203]. Our optimisation technique is aimed at estimating operational conditions based

on the weather and prices forecasts. Figures 5.2 and 5.3 illustrate our proposed adaptive

pricing mechanism with its relative operation sequence. Basically, the proposed pricing

mechanism starts by reading the information about the renewable energy sources, the util-

ity price, the weather conditions, and the time and the date of the pricing. The forecasting

module then starts forecasting WS , SLRI , and UP for up to 4 iterations (two hours) in

advance. 2) Since SG , SP and LGP are strongly related to the information supplied to the

108
 Figure 5.2: Proposed adaptive pricing mechanism

forecasting module, the PSO is called from the Power Generation Management and Storage

Control System (PGMSCS), which was discussed in Chapter 4 to utilise these forecasts

in order to estimate SG , SP and LGP throughout the four-iterations forecasting horizon.

Figure 5.3 illustrates SG , SP and LGP estimation sequence in a specied forecasting hori-

zon. 3) The estimates and forecasts are then utilised by the third computation stage (the

fuzzy adaptation) to adapt the initial sale price, which is resulted from the decision tree

that has been developed in Sub-section 5.6.1, based on the operational scenarios explained

in Section 5.5. Subsequently, the nal price value will be fed back to the PSO in line with

forecasts to estimate SG , SP and LGP in the subsequent iterations.

5.6.1 Pricing equations

The developed pricing equations are the core of our pricing mechanism. Basically, the

pricing equations were developed to suit every possible operational scenario that holds a

RdyF rSLElec (reserve) in the case study. Just like any successful business, the pricing

rules were developed to make a prot, by evaluating the cost of the product and the possible

sale value derived from the perceived market aordability.

Firstly, we discuss the formulation of the LGP equations. These were developed af-

ter: 1) studying the decision making rules at each operational scenario; 2) evaluating the

amount of participation for each of the generation units; and 3) assessing the percentage

109
Figure 5.3: Proposed pricing adaptation process

110
contribution of each of the generation units in supplying the power required to match the

demand in the microgrid. By giving consideration to the use of renewable energy sources

in LGP would always result in it being cheaper than UP and GP .

In case of evaluating the amount of reserve that is proposed for sale (RdyF rSLElec),

the evaluation was always based on the condition that LGP is less than UP , by subtracting
the demand value from the amount of local generation at each operational scenario. Hence

the value of RdyF rSLElec depends mainly on the demand in the case study; along with the
type of utilised cost function e.g., economic, environmental, or economical-environmental.

In this work the proposed pricing method was tested with economic, environmental, and

economic-environmental operation in the case study in section 6.6.

Finally, SLP rules were simply developed from the logic of achieving prot by evaluating

the LGP and UP , and making a fair decision about setting a competitive price between

these values. Although SLP was proposed to be adaptive, as an initial value, we have set

it to start from the half dierence between the two prices: LGP and UP . Since this initial

pricing point is aimed to make SLP competitive and protable.

To relate the operational scenarios with their assigned pricing rules, we have utilised

the identied operational scenarios in Table 5.1, to list the pricing rules, as illustrated in

Table 5.2.

To evaluate the robustness of the proposed pricing rules, we have conducted a simple

simulation test with the operational scenarios in the January demand pattern of the case

study. For simplicity we have presented seven working days for the operation prole.

Figure 5.4 illustrates the range of generated sale prices over that week.

111
Table 5.2: Formulated equations that evaluate the possible amount of Ancillary Services
in every operation scenario

Sc. RdyF rSLElec SLP

A GG + P hG + WG − Dmnd (UP −GP )∗GG
( 2∗RdyF rSLElec )
((GP +(SP −GP )∗(SG )∗(GG +SG )
(UP − )
B GG + SG + P hG + WG − Dmnd UP − G ∗RdyF rSLElec
G
2
((S +(G −SP )∗(SG )∗(GG +SG )
(UP − P GP ∗RdyF rSLElec )
C GG + SG + P hG + WG − Dmnd
UP − G
2

D P hG + WG − Dmnd SP − ( GP S∗(SP −GP )

P +GP
) − 0.02
if (P r >= UP ) then UP ∗ 0.98

E P hG + WG − Dmnd SP + ( GP S∗(G P −SP )

P +GP
) − 0.02
if (P r >= UP ) thenUP ∗ 0.98

F SG + P hG + WG − Dmnd (UP −SP )∗SG

SG + ( 2∗RdyF rSLElec )
G GG − GSG UG − ( (UP −G
2
P)
)
((GP +(GP −SP )∗(SG )∗(Dmnd−(P hG +WG +SG ))
(UP − )
H GG − GSG + SG UP − GG ∗RdyF rSLElec∗(GG −GSG )
2
I SG + P hG + WG − Dmnd + GG P −GP GG −GG SP
UP − ( UP GG +SP S2∗G G
)
J 0 /
K 0 /

L SG + P hG + WG − Dmnd UP − UP −S2
P

M GG − GSG UP − UP −GP
2
N GG − GSG −SG UP − ( 2UP +S4P −GP )
O GG − GSG UP − ( UP −G
2
P
)
P 0 /
Q 0 /
R 0 /

112
 Figure 5.4: Price decisions prole for a typical one week operation scenarios

Figure 5.4 shows that as long as that there is a RdyF rSLElec, the generated SLP is

always found between LGP and UP , which proves the success of the initial pricing rules in

creating benecial SLP under all operational conditions in the case study. Subsequently,

decisions are subject to the forecasting conditions as shown in sub-section 5.6.2, which will

nally be evaluated by calling the (PGMSCS).

5.6.2 Pricing factors forecasting

The second part of the smart pricing mechanism was forecasting the operational conditions

that aect pricing in a competitive market environment. The rst eective parameter

in the case study was the power demand. Therefore, we called the Demand Forecasting

Model (DFM), that has been introduced in Chapter 3, to provide this system with demand

information. To estimate the prot, and to balance it with the amount of RdyF rSLElec
to generate competitive prices, there is a need to estimate LGP and to forecast UP . To

support our proposed pricing mechanism, which depends on the short-term forecasting

for the operational scenarios, we have introduced a forecasting model for each pricing

factor. We have used dierent modelling approaches for these parameters, depending on

the non-linearity of each model. The proposed forecasting models are to perform short term

forecasting, and to align the forecasting output with the instantaneous real-time parameters

113
Table 5.3: Neural networks design parameters for the pricing factors forecasting models

Variable Nodes Learning rate epochs MSE

WS 8 0.05 250 0.0022287

SLRI 10 0.05 135 0.00935
UP 8 0.05 145 0.00481

change. Therefore, we have developed three forecasting models for WS , SLRI , and UP .
For WS and SLRI , we obtained four years of historical records from the Australian Bureau
of Meteorology [3]. In order to deal with the high nonlinearity of the recorded WS and

SLRI , we implemented the models by dividing the generation times into dierent zones:

1) in case of forecasting WG , we utilised the WS model to forecast the expected amount

of generation from the nominated wind turbines on the project location as explained in

Chapter 4; 2) in case of P hG , we did the same, except that solar panels are expected to have
zero generation from sunset to sunrise, which will make the SLRI model less complicated

than that for the WS . In both models, we have also utilised the instantaneous real-time

measured values, in line with the date (hour/day/month) values, as a base for the short

term forecasting.

For many forecasting horizons, Articial Neural Network (ANN) has been proved to

be a robust multi-step forecasting [204]. We have used ANN with a Backpropagation

learning algorithm. In all developed ANN models, we proposed one hidden layer network,

with a bi-polar log-sigmoid activation function in the hidden layer, and a bi-polar linear

type activation function in the output layer. Table 5.3 details the design of the proposed

forecasting models. Figure 5.5 shows that the forecasting accuracy for the developed models

is relatively high, due to forecasting alignment with the real-time values.

The RdyF rSLElec estimate was then evaluated by calling the PGSMCS, based on

utilising the WS , SLRI and UP forecasts. Hence the most challenging part of estimating

the factors in the forecasting horizon for the case study were, SG and SP ; due to the

uncertainty of charging/discharging time and cost as explained in Chapter 4. As a re-

sult, the Power Generation and Storage Management Controller Agent would return the

RdyF rSLElec estimates for the near future as specied by the forecasting horizon, which

would be eventually considered in the nal SLp.

5.6.3 Adaptive pricing rules fuzzy system

To utilise the factors forecast in our pricing mechanism, we proposed a fuzzy system that

reects our experience in generating competitive prices within a competitive markets en-

114
 (a) UP forecasting model performance

(b) P hG generation model performance based on (c) WS forecasting model performance

the SLRI forecasts

Figure 5.5: WS , P hG and UP forecasting models performance in a typical one week of

operation

115
 Table 5.4: Pricing adaptation fuzzy rule base system

RdyF rSLElec S M L
SLp

IncandDec(+) S S S M
M S M L
L M L L

RdyF rSLElec S M L
SLp

IncandDec(−) S S S M
M S M L
L M L L

vironment. Our proposed fuzzy system was utilised to adapt the pricing rules to the

microgrid's operational conditions. In the design, we related the amount of RdyF rSLElec
to the simultaneously rule based generated SLp, to have an adaptive SLp, thus keeping

it advantageous and competitive. The other input to the proposed fuzzy system is the

direction of IncandDec (increment/decrement) of the eective pricing factors in the fore-

casting horizon, whether the estimation results in an increment/decrement in the pricing

factors. The direction of parameters change (increment/decrement) was evaluated from

the values dierence at each iteration in the forecasting horizon. However, to evaluate the

slope of the increment/decrement, the values were compared with their next and previous

forecasts. The proposed Fuzzy rule based system was extracted and illustrated in Table

5.4. In the same table, the increment/decrement were represented by (+/-).

To introduce the nominated fuzzy inputs, we normalised the rst input (SLP ) by

relating it to the UP :

(UP − SLp)
F LSLp = (5.1)
UP

and the second input (F LRdyF rSLElec) to the amount of full generation capacity

(3000 kWh) in the case study (without considering the standby 0.5MW gen-set):

(3000 − RdyF rSLElec)

F LRdyF rSLElec = (5.2)
3000

The proposed membership functions that represent the normalised input values and

the range of increment/decrement for the eective pricing parameters are illustrated in

Figure 5.6. In this work, we have utilised two types of fuzzy membership functions, the

triangular and the trapezoidal, to suit the objectives of the proposed fuzzy system. The

nal sale price N ewSaleprice was evaluated by multiplying the value of P riceAdjustment

116
 Figure 5.6: Pricing adaptation fuzzy system membership functions

by the dierence between LGP and UP as explained in the following equation:

N ewSaleprice = ((UP − LGP ) ∗ P riceAdjustment) + LGP (5.3)

5.7 Market simulation

The proposed negotiation mechanism is built to start the trading process by registering

the participant agents within the pool under a common predened service. In this work

we have dened the common service in the MAS framework as  ElectricityT rade, where

each pool participant must be registered under this service to receive the appropriate

information messages. The implementation details of MAS technology on this system will

be covered in Chapter 6 Thereafter, the buyer was designed to look for the participant

sellers in the framework's Yellow Pages. Messages with Call For Proposal (CFP) type are

sent to the identied agents asking them to submit a proposal for their registered service.

When CFP is received, the sellers will reply with messages of INFORM type that carries

117
 Figure 5.7: Proposed electronic negotiation software sequence (two types of sellers)

either acceptance or rejection, based on their predened pricing analysis. In the case

of acceptance, the agreed seller's proposal should include informative details about the

proposed oer. Figure 5.7 illustrates the proposed negotiation software sequence. Our

simulation also allowed for one of the sellers to receive a rejection of their proposal, where

the ability to regenerate more competitive prices within a strict time frame was attached

to the seller's pricing mechanism. At this stage, this mechanism was not discussed, but

can be extended as future work arising from this thesis.

Our pricing system has been tested using MAS technology, where a virtual market

was created and tested under both the uniform and discriminatory pricing rules. In the

created market, each player was represented by a smart agent that has the ability to

negotiate and perform smart trading decisions based on its pricing rules. For the buyer

agents, we proposed that they have randomly generated CFPs based on their predened

settings that reect the market's attitude at that point of time. In addition, each buyer

was programmed to select the winning seller based on the uniform and discriminatory

pricing rules. In this work, we used Java Agent DEvelopement Framework (JADE) [2] to

simulate the proposed agent system, since JADE-based MAS technology role has already

been evaluated through implementing electronic markets on a real test sites [82, 83]. Hence

the data for the simulated markets have been collected from [167]. Figure 5.8 illustrates

118
Figure 5.8: Smart electricity trade process using Multi-Agent System technology under
the discriminatory pricing rules

a JADE simulation based sellers selection process, by the utility (buyer) based on the

competitive oered proposals.

5.8 Results

To evaluate the robustness of our pricing method, we have conducted several simulation

studies for a month of operation time on our case study. Specically, we have selected

January operational data for the simulation, as it contains dierent demand patterns that

reect public holidays, normal working days, and weekends. Therefore for simplicity and

clarity of results, our gures show a one week of operational time results, while our tables

show the whole month of operational time results. In this section, we will show the impact

of adding each of the intelligent modules to the agents that carry the main pricing rules.

So we will test the system based on its level of intelligence. In the simulation we have

used three sellers, where the rst seller enters the market and generates proposals with the

initial pricing rules. The second seller enters the market and generates proposals based

on the fuzzy system adaptive pricing rules. The third seller enters the market with initial

pricing rules in line with a fuzzy adaptation mechanism based on forecasting the very near

future operational conditions; making the proposals adventurous, though more competitive

in the longer term.

The simulation studies start with the initial pricing rules under the economic, envi-

ronmental, and economic-environmental cost considerations in the case study. These cost

considerations based decisions are generated by the PGSMCS. The simulation studies have

119
been conducted using JADE to act in an electronic competitive market as explained in sec-

tion 5.7; where the case study has been represented as a competitive seller, who tries to

achieve a maximum prot with a reasonable amount of transactions. From the simula-

tion, we found that the programmed buyer reecting the predened trends for the actual

market attitude, was generating dierent CFP signals to the sellers. Eventually, based on

the nature of the released CFP, we found our pricing agents oering dierent amounts at

dierent times and prices. Therefore, the robustness of each of the tested pricing rules

depends on their response to the generated CFP in a specic simulation time frame. In

fact, the developed initial pricing rules were found to achieve a set of transactions, and

thus a range of achieved prot, and exchanged electricity with the utility through these

successful transactions. Tables 5.6 and 5.7 identify dierent numbers of transactions and

dierent amounts of achieved prot in a month of simulation time. At each operational

cost consideration there were dierent SLp and RdyF rSLElec, depending on the assigned
number and the type of generation units for the local generation at that operational cost

mode. Comparing these results with the adaptive fuzzy based rule results, we can identify

a higher amount of participation under the uniform pricing rules for the fuzzy adaptation

based pricing system. Thereafter, when we added the forecasting considerations to the

pricing rules, we also found a higher amount of participation in the economic cost consid-

eration under the uniform pricing rules. The proposed fuzzy system has been tested to

evaluate its range of adaptation, which was found to actively increase and decrease the

sale price based on the evaluated amount of SLp and RdyF rSLElec as shown in Figure

5.9.

Moreover, the nal results proved that this logic was successful in making more prot

even with less amount of participation than with the initial pricing rules. This is because

the fuzzy system was acting when a higher amount of RdyF rSLElec was available, to

generate more competitive prices. This was based on the logic that when the amount

was increased, a sucient price reduction can make more prot. That will eventually

compensate the missed transactions that occurred when a high SLp was oered with a low
amount of RdyF rSLElec. The main reason for increasing the prices with a low amount

of RdyF rSLElec, was to try selling the small amount at a higher price; based on the low

probability of there are no other sellers oering cheaper services at that time. When

comparing the amount with the possibility of participation, we found that it was worth

trying to sell at these high prices.

The number of transactions under the uniform pricing rules, was evaluated from the

120
 Figure 5.9: Fuzzy-based adaptation level performance

number of successful electricity exchanges based on the condition that the sellers provide

equal or less than the expected price (market balance point). Eventually that was also

depending on the CFP signal generation trends from the buyers. In case of the discrim-

inatory rules, the number of transactions was evaluated from the number of successful

electricity exchanges based on the conditions: 1) that the seller records lowest price among

the participants; and 2) the seller records an equal or lower price than that expected by

the buyer. Based on our studies of performing a smart sale decision making process, we

would like to test our extended strategy which adds the forecasting module, to achieve a

higher rate of successful transactions. Adding the forecasting module will act to generate

more competitive prices by considering the near future operational conditions; and possibly

make more prot in the longer term. To verify the robustness of adding the forecasting

module to the pricing mechanism, we investigated the future forecasting horizon, based on

the number of iterations with 30-minutes based samples. However the major problem that

resulted from adding forecasting to the pricing decisions, was the forecasting error; which

resulted from the lack of accuracy in the developed forecasting models. Increasing the per-

centage of error would eventually reduce the accuracy of the pricing decisions. Table 5.5

illustrates the importance of adding the forecasting to the pricing system under dierent

operating cost modes; by showing the number of transactions, money value, and amount

of exchanged electricity with the buyer. From the same table we can relate the percentage

of forecasting error to the evaluated pricing performance.

121
 Table 5.5: Pricing performance analysis

Forecasting horizon 0.5 1 1.5 2

Cost PE % 2.2 4.84 10.648 23.42

ET P 89 91 93 93
Economic
TV $ 15724 15123 15778 16070
ET P 55 63 61 65
Environmental
TV $ 11011 12032 11249 12302
ET P 90 85 71 71
Eco-Environ.
TV $ 18918 16921 13363 13534

Table 5.6: Performance analysis for discriminatory pricing rules

Cost consideration Results NPR F LAP R AN N − F LAP R

ET P 9 15 91
Economic SLDElec kW 13770 24960 160700
TV $ 5777 10545 15123
ET P 50 2 63
Environmental SLDElec kW 42550 1823 58100
TV $ 273 473 12032
ET P 43 1 85
Eco-environmental SLDElec kW 35920 542.5 80870
TV $ 109 135 16921

We subsequently tested the pricing system under dierent operation cost considerations

for both pricing rules. To evaluate the importance of adding the stages of intelligence to

the pricing system, we created three sellers, each carrying the proposed pricing system

with dierent level of intelligence. We also created a buyer that searches for the cheapest

prices in this simulated environment. As discussed before, we have utilised JADE to

represent MAS in performing this simulation, and have created an agent for each of the

three sellers and the one buyer. The recorded amount of successful ET P ,SLDElec and

TV are illustrated in Table 5.6.

In the simulation studies, we used January operational data to simulate this spot virtual

market based on the relevant operational conditions for the case study. In the simulation

it is assumed that all sellers have the same generation capacity and operational conditions.

However they are also assumed to have dierent pricing mechanisms. The results showed

that in a month of simulation time the economic operation mode recorded 115 transactions,

199.43 MW of sold electricity, and eventually $AUD 31445 of the total achieved money

value. It is notable that adding the forecasting to the system led the seller to take 79%

of the achieved transactions, 80% of the exchanged amount of electricity, and 48% of the

total transactions money value. In addition, the initial pricing rules that did not include

the forecasting in their decisions could achieve 33% of the total money value for all of

122
the 115 transactions. This means that the highest eciency was achieved with the initial

rules and the fuzzy adaptation system. However the forecasting was still needed to cover

those critical pricing regions in the operational conditions, where the initial pricing rules

and the adaptation fuzzy system lack the ability to do so. Thus the initial basic pricing

rules could achieve only 18% of the total transactions money value. The 33% and 18% of

the achieved money values mean that although the forecasting and the fuzzy system could

generate more competitive prices in the markets; they were still missing those transactions

due to the careful considerations about maximising the prot. The number of the missed

transactions can be reduced by applying an online optimisation technique to the generated

prices, by monitoring the success rate of the trade with the buyer in the competitive

environment, and trying to adjust the prices accordingly. This part is recommended as

a future research undertaking arising out of this work. From analysing the results along

with other cost considerations, we have set the forecasting zone to one hour, when it

carried a precision error of 4.8%, and took about 48% of total achieved money value in the

simulated time frame. However, the results showed the same rates of improvement with

environmental and economic-environmental cost considerations, except that in these cases

there was a higher initial local generation cost due to the environmental concerns. Thus a

fewer number of trade participations was expected, as shown in Table 5.6. Hence there was

a noticeable state in the results where the fuzzy based pricing system recorded a very low

number of transactions, compared with the initial pricing rules and the forecasting based

pricing rules. However, it was also clear from the table that although the initial pricing

rules recorded a higher number of transactions, the recorded total money value was also

higher. Therefore, although there might be no transactions from the fuzzy based pricing

rules for a certain trading time frame, these missed transactions can be compensated by

few higher price ones when there is a higher demand in the market. The results showed

that the proposed pricing system was robust in maximising the prot, by exploiting some

of the possible opportunities, and logically reducing the sale price to increase the chance

of participation.

In the uniform pricing rules we had a higher amount of transactions from the participant

sellers, as the rules were applied to buy all the available electricity in the market; as long

as the oered prices were below what the buyer expected. Therefore, we expected to

have closely related values from all sellers. The simulation studies using JADE presented

a buyer agent who generates CFP signals based on predened random function settings;

and three sellers, who analyse their operational conditions and generate their sale prices

123
 Table 5.7: Uniform pricing rules performance analysis

Cost consideration Results NPR F LAP R AN N − F LAP R

ET P 98 107 110
Economic SLDElec kW 74576 82472 84640
TV $ 20620 20472 18286
ET P 117 113 117
Environmental SLDElec kW 104676 102889 104676
TV $ 29443 15176 23942
ET P 117 81 117
Econ-environmental SLDElec kW 104766 75061 104766
TV $ 29561 19261 24592

based on the tested pricing system, with dierent level of intelligence. It was also assumed

that all sellers share the same generation capacity and operational conditions. Table 5.7

records the resulting values achieved from testing the pricing system with dierent levels

of intelligence, under three operation cost considerations with the uniform pricing rules.

The results showed that the economic cost considerations recorded 315 transactions

from the three participant sellers, where each seller had a closely related amount of partic-

ipation. Eventually these results led to the same ratio of total amount of sold electricity for

the total value of money. The dierence in the number of participations results from the

amount of price reduction achieved by the adaptation system. Since it was expected that

the buyers would always buy the electricity as long as the sale price is below the buying

price; reducing the price can lead to less prot, although a higher number of participations

can be achieved. However the proposed logical price adaptation has proved its success with

a large number of transactions, and achieved prot in doing so.

5.9 Conclusion

In this chapter we developed adaptive pricing rules that aim at generating competitive AS

sale prices for a typical autonomous microgrid. The pricing rules were developed in a way

that reects the business strategies in achieving a maximum possible prot from successful

transactions with the grid. The pricing rules have been developed with three levels of

intelligence. First, the initial pricing rules were developed using linear programming, and

a derived decision tree that details the operational scenarios in the case study. Second, a

fuzzy system was added to adapt the initial rules based on monitoring the microgrid SLp
and the availability of RdyF rSLElec. Third, a forecasting module was added to enlarge

the monitoring horizon and include the near future scenarios in the careful prices adapta-

tion. The developed pricing rules have been simulated with a virtual competitive market

124
environment, where dierent amount of transactions have been recorded, and estimated

money values have also been recorded. The results highlighted the importance of adding

each level of intelligence to the pricing rules. By using MAS, we have simulated the intel-

ligent behaviour of the electronic negotiation between the three microgrids (sellers), and

the utility (buyer) in a competitive market pool. The results provided dierent numbers of

transactions and dierent amounts of exchanged electricity, and thus dierent total money

values. The results also proved that the proposed pricing strategy was largely successful,

yet it has encompassed few limitations such as the lack of forecasting modules' accuracy

and the diculty to cope with other seller' changable strategies. Overcoming these chal-

lenges can be achieved through the inclusion of a monitoring strategy that can adapt the

pricing strategy to the changable pricing parameters accordingly. This piece of intelligence

can be included as a future extension to this work.

125
Chapter 6

Agent-Based Implementation of Energy

Management Systems for Microgrids

In this chapter, we investigate the economic and environmental impacts of the intelligent

utilisation for the microgrid resources, through a set of strategies which are invoked, de-

pending upon the microgrid's operational conditions. The purpose of these strategies is to

maintain a secure, and low economic and environmental cost generation for our case study.

In this chapter we describe how we applied agent-based technology to model the proposed

management strategies. The proposed management strategies aim to provide feasible de-

cisions concerning power generation, in order to cope with the problems which arise from:

1) the uncertainty of demand; 2) weather conditions; and 3) electricity markets. Hence

testing the proposed management strategies includes testing of the software components

which perform the agent-based modelling. These software components are installed on

distributed machines and are connected via a standardised communication protocol to al-

low for future extension. The proposed Multi-Agent System (MAS) implementation aims

to reect the human factors involved, when making team-decisions based on the observed

operational conditions around the managed microgrid. In addition, the proposed MAS

implementation provides several emergency operational modes, which are activated when

any of the distributed software components fails. Dealing with the uncertainties of the op-

erational modes has been successfully achieved by having a dynamic management system

structure, which is autonomously congured to cope with the variable operational condi-

tions. Although all the software components are needed in order to provide full system

functionality, testing of each of the components was conducted in turn, in order to evaluate

the individual importance of each of them. In this chapter, Sections 6.1 and 6.2 address

the role the energy management systems in improving the power systems performance,

and introduce the current challenges encountered with this process. Section 6.3 details the

126
targeted research problems and the objectives aimed at in this work. Section 6.4 briey

shows the research assumptions considered before performing the simulation. Section 6.5

utilises a case study to address the proposed management strategies for the smart utilisa-

tion of the multiple sources of dispatch; and in addition, it details the modelling procedure

for implementing the proposed management system. Finally sections 6.6 and 6.7 present

the results which have been captured, and conclude the research outcomes of this chapter

respectively.

6.1 Background

Power systems are required to apply ecient solutions to reduce the generation cost for

both generators and consumers. So far, Energy Management Systems (EMS) have shown

to be key in the power systems development, and has had a signicant role in improving the

power systems eciency. This is mainly through providing planning and security solutions

at the High Voltage (HV) level networks; and optimising the power ow and the generators

scheduling at the Low Voltage (LV) level networks of the power systems. Therefore, EMS

development has been investigated in this research, based on the fact that EMS includes

two major research topics in the electrical engineering eld: 1) Supervisory Control And

Data Acquisition (SCADA) system to monitor and control the network; 2) and the specic

strategies embedded in the management systems advised to optimise the performance of

the power systems. This chapter has targeted the second component of the EMS, which is

the strategies required to perform low cost generation; more specically, the chapter has

investigated the role of this technology in managing the power generation of our case study.

From both the economic and the environmental perspective, microgrids have proved

their reliable performance by reducing the generation cost [145]. In this context, building a

reliable EMS for a microgrid encounters the challenge of incorporating multiple sources of

dispatch; with the attendant problems of uncertainty of generation cost and availability of

the participant generators. In addition to this challenge, it is necessary that the generation

facility performs successfully in a market under competitive pricing conditions.

As a result, the process of identifying the cost factors for generating the electric power

highlights the challenges being faced, which go far beyond simply performing low cost gen-

eration. Thus, identifying the cost factors of generation helps to guide the strategies being

embedded within the EMS design, orienting it more towards providing a practical solution

for this problem. Hence dealing with more than one complex problem simultaneously re-

quires the deployment of a real-time distributed management system. Microgrids' EMS can

127
be implemented by having real-time distributed management system. Such a deployment

allows the network operators to optimise the use of distributed generation resources, and

enables real-time communication between customers and utility service providers in order

to allow optimisation and balancing of energy usage. It is only possible to improve the

exibility and control of the distribution network in this way if the control center operator

has very accurate knowledge of the real-time situation of the network [205]. EMS and

distributed decision making systems have been widely used for providing power ow man-

agement and economic dispatch solutions for microgrids. The details of the technologies

which have been deployed, and the applied strategies, are explained in the next section.

6.2 Related works

In recent times, investigating the impact of EMS on reducing the economic and environ-

mental costs for microgrid-based power generation has been active research topic. A survey

of various existing microgrid energy management strategies has been presented in [75] to

highlight researchers' opinion on microgrids' enery management systems. AC and DC mi-

crogrid systems have been compared in [206] in terms of the main features and the control

methods utilised to maintain their optimum operation. Furthermore, a comprehensive re-

view of the control and operation management strategies utilised in both microgrid systems

has been presented in [207, 208]. In terms of the economical, technical, and environmen-

tal benets, microgrid operation has been comprehensively reviewed in [208]. Ultimately,

microgrid operation management problem has been considered as a reliability problem

in [209], which has also been considered as a multi-objective self scheduling optimisation

problem. Thus, microgrids performance has been identied to be mainly dependent on the

robustness of the management strategy and the level of intelligence required to achieve the

optimum operation.

EMS development has been considered within an economic analysis of distributed sup-

ply in [210], while a determinist EMS for a microgrid, including advanced photovoltaic

generators with embedded storage units and gas gen-sets, has been proposed in [211].

A new method of mesh adaptive direct search has been employed to minimise the cost

function of managing the operation of a microgrid designed for residential application in

[149].

The literature has addressed the challenges of managing microgrid resources encoun-

tered by the system operators, and has also nominated the technologies and the design

standards required to tackle these challenges.

128
 Therefore, the integration of distributed decision making systems has been identied

as a successful method for optimising the performance of the microgrid generation. A

distributed control framework for the integration of distributed wind turbines and solar

panels has been previously addressed [212]. Multi-Agent technology was found to be a

viable solution for implementing and/or simulating the distributed control systems in or-

der to optimise the performance of the microgrids. The capabilities of Multi-Agent System

(MAS) technology in performing the microgrids generation management has been presented

in [105]. MAS has also been used as a solution for managing a distributed hybrid renew-

able energy generation system in [213], and performing a microgrid power management in

[214]. MAS-based control architecture, for achieving microgrids power management ob-

jectives, has been proposed in [215, 216]. A decentralised MAS, and a hybrid algorithm,

combined with an Evolutionary Algorithm (EA) and a Linear Programming (LP) have

been developed to manage a power distribution system in [217].

For market participant-based microgrid management systems, agent technology has also

been widely utilised due to their social and autonomous characteristics. In addition, due

to agent's learning characteristics such as creating behaviours and following specic rules

in making decisions, they have been proved to be a right modelling tool to model the high

uncertainties of various market factors [201]. Eventually, agent technology has been proved

as one of the most applicable modelling techniques, and has been successfully utilised in

modelling power systems, in order to foster their integration into the electricity market in

[218, 219, 220, 221], and also in [222], where it has been used to model the UK market

mechanism. For Smartgrid research, agents have been proposed for use when modelling

and automating Smartgrid markets in [223]; taking into consideration power dispatch

limits, generation and transmission constraints, and spinning reserve requirements.

By considering the environmental issue in the market participation, agent-based plan-

ning systems for distributed energy management have been proposed in [224] in order

to process energy trading based on energy decisions, with application to market oriented

programming for controlling CO2 emissions. Finally, it has been realised that implement-

ing smart management strategies to perform microgrid optimum performance is a timely

research topic, and it has been increasingly targeted by power engineering research com-

munities. In this context, MAS technology has been identied as one of the most successful

tools for modelling and simulating power systems. However, some critical issues, such as in-

tegrating interactive demand forecasting in the decisions, and dynamically conguring the

proposed management system based on the operational scenarios, have not been covered

129
in literature yet, and hence introducing a new method that overcomes these limitations

would have a high impact in this eld.

6.3 Research objectives and challenges

The aim of this research is to evaluate the potential of the intelligent management sys-

tems in reducing the economic and environmental generation costs. Concurrently, MAS

technology has been found to be a successful solution to model the power system man-

agement systems; and therefore, we will utilise this technology in the implementation of

the proposed system. Hence Articial Intelligence (AI) is also needed in the implementa-

tion. Ultimately, our research will contribute to the knowledge by presenting a novel and

adaptive MAS-based microgrid management system, which is implemented as distributed

AI-based modules in order to achieve low economic and environmental generation costs.

The major identied challenges in implementing this system are:

1. Reducing the complexity of the system and the high computation resources required

to implement it.

2. Managing the diculty of synchronising the operation of the distributed intelligent

software components.

3. Implementing a hierarchy of decision levels in order to prioritise the decisions based

on the operational scenarios.

4. Enabling a remote access and control for the proposed management system and the

distributed generation units.

6.4 Research assumptions

The assumptions that were considered in the investigations are:

1. Renewable energy sources are assumed to experience a dramatic drop in price in the

coming years, whereas electricity and gas prices are expected to increase rapidly.

2. Negotiations for microgrid market participation are conducted directly with the mar-

ket itself and not with the retailers or any other third party.

3. The market balance point is considered to be an indicator of UP when buying and

selling electricity, by assuming that the market operator is a non-prot body.

4. The nominated generation units for this project are assumed to have the highest

eciency, in comparison with the other products in the market.

130
 5. It was assumed that the weather conditions test data are valid for the simulations in

this work.

6. It was hypothesised that the simulation data will accurately reect the nature of the

near future interaction between the utilities and their customers.

6.5 Agent-based technology for implementing the proposed

management strategies

The management system for this work was developed in order to implement agent-based

applications, so that the performance of the microgrid power generation can be optimised,

and can enable the microgrid generation to be integrated with the power markets. The

proposed architecture was enriched with ideas for increasing the intelligence of the system,

thus increasing its operational prociency. By analysing the operational scenarios in the

targeted microgrid, we found that it was essential to consider the demand values in all of

the computation stages, when optimising the microgrid's performance. Consequently, a

demand forecasting model was also required as part of the proposed management strategy.

Furthermore, after studying the possibilities of running multiple sources of dispatch, other

issues were subsequently identied. These issues centre around the uncertainty of generat-

ing electricity from renewable sources; and the diculty of forecasting the generation cost

for SDs, unless a proper control technique is applied to control the charging and discharging

process. As a result, it became clear that an alternative optimisation model was necessary

in order to optimise the microgrid's performance. Finally, the microgrid power generation

includes lost reserve power generation capacity that can be utilised, if it can be delivered

to potential buyers in a timely and cost eective fashion. While exploring an opportunity

to sell this reserve to the grid under competitive market environment, it became clear that

there is a need to include a smart bidding strategy that can integrate with other models

in the overall management strategy, to facilitate selling the electricity to the grid. After

identifying the critical operational issues concerning the provision of optimum dispatch for

the microgrid, the challenge becomes building a strategy that successfully integrates these

models in order to synthesise a solution delivering the optimum performance.

The proposed strategy was designed around a 30-minute operational life-cycle, which

ensures that a new decision is made every 30 minutes. The implementation of the proposed

management architecture is based on a real case study data from Edith Cowan University

(ECU), where the on-campus power network with its everyday load prole was utilised

131
in the simulation of this work. The weather conditions throughout the year in the city of

Joondalup have been studied over the previous 5 years, and formed the basis of the weather

data employed within the case study. In the actual simulation, the MAS was utilised to

perform the required generation management objectives on the case study. As the demand

values must be included in all of the computation stages of this strategy, the operation

cycle starts by forecasting the demand. The process of creating the demand forecast begins

by reading the information about the demand of the microgrid, such as time, date and real

time demand. After the demand has been forecast, it is then sent to the subsequent

computation stages for further processing. The next computation stage is the creation of

the power generation schedule and the SDs control. During this stage, all of the information

concerning the operational conditions of the microgrids are supplied every 30 minutes, and

are then employed to determine the optimum decisions for scheduling the power generation

within the microgrid. Once a decision has been made during this stage, all the results are

sent to the third computation stage, where the bids are generated to participate within

markets under competitive prices. Hence the power generation is planned, and the resource

utilisation is based on economic and environmental cost considerations.

This sequence of operation highlights the need for the computational stages to be con-

trolled by superior decisions. The proposed MAS has multiple levels of advanced decision

making capability, which helps in prioritising the specic decision making process within

the microgrid. The implementation of the proposed agents was based on four major deci-

sion making objectives: 1) master controller decisions; 2) demand forecasting; 3) generation

and storage management; and 4) bids analysis and generation. In addition to this struc-

ture, various external units are also present; representing the grid with other generators,

and other microgrids that represent the market competitors. Consequently a network of

control agents, that achieve the detailed management objectives, was subsequently pro-

posed. The MAS that has been developed includes three main sections: 1) processing;

2) communication; and 3) input/output interface (monitoring and control). Figure 6.1

illustrates the overall MAS structure.

Generally, the implementation of all agents in this system encompasses the following:

1. Service Registration: At this stage, after the agents are created, named and ad-

dressed, they are also registered under the framework's Yellow Pages. This allows

agents to identify each other by the services they provide. In this work, the agents

were registered under various types and names, depending on the task they assigned

to do.

132
Figure 6.1: The developed Multi-Agent System structure

133
 2. Behaviour setup: During this stage, all of the behaviour parameters for each of the

agents are set. More specically, cyclic behaviours are applied to perform repetitive

operations, which unfortunately, also have the drawback of being extremely CPU

extensive. In order to reduce the CPU workload, the action() and block() methods

were added under the same behaviour class. The action() and block() methods cause

the the behaviour cycle to be marked as being blocked; and no longer available for

agent processing. The behaviour can be over-ridden if the agent receives a message

instructing it to undertake the processing.

A typical pattern for receiving messages code looks like this [2]:

public void action() {

ACLMessage Message= BidsGeneratorAgent.receive(mt1);

if (msg!=null){
// If message recieved, process it!
}
else block();

3. Message template: This is a predened messaging communication template that is

used to classify the received messages type. For instance, if the template carries a: 1)

performative type  REQUEST , then the message carries a request to the receivers; 2) an

 INFORM , then the message carries information to the receivers; 3) either a  PROPOSE

or  CALL FOR PROPOSAL (CFP) , then it is either proposing or is calling for a proposal.

More details about FIPA complaint message template types can be found in [2, 78]. A

message template creation sample looks like the following:

MessageTemplate
mt1 = MessageTemplate.MatchPerformative(ACLMessage.INFORM);

The specic content of messages sent between the agents and the environment, or within

the agents themselves, were discussed in section 6.5. More specically, agent structures

depend on the complexity of their tasks, and therefore, dierent structures have been

proposed for the agents.

4. Ontology: This is a predened messaging vocabulary; shared among all the par-

ticipant agents to code/decode the information exchanged within the agents framework.

Building ontologies in our proposed system allows both the information senders and re-

ceivers to categorise the content of the messages based on the role they assigned to do.

134
The ontologies also allow the agents to exchange a set of details simultaneously. In this

work, the information exchange is depicted in Figure 6.4. More information about JADE

ontologies and their implementation methods can be found in [2].

Finally, the proposed MAS has been congured to run in three connection modes, which

results in a dierent operation strategy for the distributed agents. For the Connecting

mode, the microgrid is connected to the utility grid; while for Islanding and Emergency

modes, the microgrid is disconnected from the utility grid and is self supplied from its

integrated distributed generators. Each of the operational modes has its own individual

processing and communication scenario. Figure 6.2 shows the direction that messages take

when being sent between the agents, for each of the three operational modes. The specic

strategy to be applied is chosen because of its connection mode.

In Connecting mode, the sequence of operations begins by forecasting the demand.

This step is followed by the creation of a report of the demand value, along with the status

and cost of the other generators, including the utility grid supply, which is sent to the

Generation and Storage Management agent (GSM). Once a decision has been made by

this agent, the results are then reported to each of the Master Controller (MC) agent,

Bidding Analysis and Generation (BAG) agent, and to the generators; for scheduling the

generation. The BAG agent may analyse and generate bids, and then send them to the grid

agent after receiving information from: 1) the GSM agent stating the amount of power that

is ready for sale; 2) weather conditions; 3) generators status; 4) utility and gas generation

cost; and 5) a CFP from the grid agent for bidding. When the grid agent responds to this

unit with acceptance or rejection, another analysis stage is commenced and other reports

are sent to the GSM agent, the MC agent, and the utility grid agent to prepare the system

export the power to the grid in the next 30 minutes of operation; providing that there is

an agreed deal between the grid and the BAG agents. At that time another processing

cycle commences.

In case of Islanding mode; the operation sequence starts by forecasting the demand

and reporting the demand value with other generator's status and costs, including the

grid, to the GSM agent for nal decision about the generation scheduling. Thereafter, the

results are reported to the generators for operation, and to the MC agent for updating

the microgrid operation status. Meanwhile, the MC agent will report the operational

status to all the processing agents to keep or change the management scenario on the

microgrid. Finally, in Emergency mode, the operation can be categorised under Islanding-

Emergency mode or Connection-Emergency mode. In both cases the operational decisions

135
Figure 6.2: The messaging direction between agents in all operation modes

136
Figure 6.3: Strategies and priority levels based on the microgrid's operational conditions

will be generated from the GSM agent, without considering the demand value to run full

generation capacity from the distributed gas gen-sets (in Islanding mode), or from the gird,

(in Connecting mode). At this time, the MC agent will be updated with the microgrid's

operational status from the GSM agent.

Based on the outcome of the decision concerning the microgrid's connection mode

and status, four layers of strategies, resulting in multi-level of decisions priority, have

been proposed to cope with operational conditions around the managed microgrid: 1)

basic management strategy; 2) utilising demand in the generation scheduling strategy;

3) advanced management strategy; and 4) market participation strategy. It is proposed

that the MC agent makes the high priority decisions for the microgrid's operation, based

on three operational modes: 1) Connecting mode; 2) Emergency mode; and 3) Islanding

mode. In addition, it was proposed to include the basic management strategy that runs

other strategies, based on the microgrid's operational conditions. The priority of the

decisions, and their related strategies based on the operational conditions, are depicted in

Figure 6.3.

In Figure 6.3, it is shown that the operational strategies are congured in such a

manner that the operation is sustained, even in the case of faults occurring in one of the

processing agents. To involve the microgrid with market participation requires that all

other agents must provide their solutions before generating bids. At the same time, the

advanced management strategy, which is operated as the GSM agent, is not permitted to

operate without identifying the demand in the microgrid. Finally, the basic management

strategy (bias) is essential for running all the subsequent decision agents, including running

basic utility and gas gen-sets generation scheduling.

137
 The roles of each agent were described previously in Chapters 3, 4 and 5, however a

simplied description concerning the connection and the exchanged messages among the

proposed agents, are illustrated in Figure 6.4. The gure also shows the exchange of data

between the environment and the agents in the proposed management framework. The

gure clearly details the framework for processing the information about the operational

conditions of the microgrid; including the attached learning and adaptation mechanisms

that were incorporated within the agents. The same gure also illustrates the MC agent

role, which is crucially important when reconguring the agents structure, based on the

specic operational conditions. It also controls the microgrid grid-connection, based on

the received results from other agents in the framework.

The management system relies on the utility in supplying the power to the microgrid,

as this will provide relatively more secure back-up for the microgrid. Figure 6.5 illustrates

the initialisation of the strategy program that runs all agents. To perform the sequence

control operation of the proposed management system, messages are to be sent between the

agents and external data sources. The purpose of the sequence of operations is to control

the conversations needed by the agents, both within and outside of the framework. Several

conversations and negotiations must occur before the nal decision is made. In Figure 6.5

it is shown that the system supplies all of the electricity demand from the utility, until

decisions about optimising the generation have been made. Finally, the decisions are made

by the GSM agent, which is also the interface between the management system and the

resources. The connection modes with their operation strategies have been rstly tested

using JADE as shown Figures 6.6, 6.7, and 6.8.

The implementation of the agents, including their interaction with external data sources

and software, is described next.

6.5.1 Demand Forecasting Model agent

The Demand Forecasting Model (DFM), which was explained in detail in Chapter 3, was

inter-connected to the proposed MAS through a safe and reliable middle ware, as shown

in Figure 6.1. This integration will enable this model to interact with other agents in the

proposed management system, as the agent will be considered as a buer for the forecast

values. In addition, the forecast values can also be encrypted and/or modied based on

the operational requirement, before they are sent to other agents. Ultimately, this agent

development encompassed the following issues:

1. Data streaming - in this part, the data are to be supplied to the model from the date

138
Figure 6.4: Information exchange between the proposed agents in Connecting mode

139
Figure 6.5: Framework initialisation sequence

140
 Figure 6.6: Agents communication in Connecting mode

141
 Figure 6.7: Agents communication in Islanding mode

142
 Figure 6.8: Agents communication in Emergency mode

and load metering agents, based on 30-minute intervals of operation, which can be a future

extension of this work. In this work the data was supplied from the database. However,

the superior decisions are proposed to be received from the MC agent. The streaming data

consists of date, time, type of the day (working day or holiday) and real-time demand.

2. Forecasting process - at this stage the model will process the month, day, hour,

temperature, and the real-time load data to estimate the next 30-minute demand value in

the case study.

3. Data beaming - nally, the demand forecast values are beamed to the targeted

agents for further processing stages.

This agent is registered as demand forecasting model agent under

 OptimumGenerationP erf ormance  type, and

 DemandF orecastinAgent name in the framework's Yellow Pages of the proposed

management system. There is also a synchronisation setting between the forecasting model

in Matlab and the forecasting agent in JADE. This setting allows the agent to read the

amount of demand from the model, in a predened and timely manner. A timing setting

for reading information from the model, is essential for all further stages of calculations in

the proposed management system.

6.5.2 Generation and Storage Management agent

This agent is the integration of the PGMSCS with other modules in the proposed system.

This agent's role depends mainly upon the information exchanged with other agents. It

143
will initially acquire the information about the demand and the generators' status from

the DFM agent, and also from the generators' agents. Thereafter, it will proceed with the

required optimisation calculation. Ultimately, the results will be sent to other agents as

explained in Section 6.5. The implementation of this agent includes the following:

1. Data streaming - in this part the data is sourced from other agents based on 30-

minute intervals of operation. The streaming data consists of the generators' status and

costs; in addition to the superior decisions that are generated from the MC agent.

2. Processing - all received data are processed with the proposed optimisation technique

explained in Chapter 4.

3. Data beaming - all results of the optimisation and the decision making should be

beamed to other agents as explained in Section 6.5.

This agent will be registered under  OptimumGenerationP erf ormance type,

and  GenerationStorageM anagement  name in the framework's Yellow Pages. The

type and the name will help the seeker agents identify their specic targets for information

exchange.

6.5.3 Bids Analysis and Generation agent

This agent will connect the Fuzzy Logic-Based Ancillary Service Adaptive Pricing System

(FLASAPS) with other intelligent modules in the proposed system. The implementation

of this agent considers the competitiveness when selling electricity in a spot market pool.

Therefore, building this agent included:

1. Data streaming - the data are sourced from other agents based on 30-minute intervals

of operation. The streaming data consists of weather conditions, demand, generators'

status and cost, markets Call For Proposal (CFP) requests, availability and readiness for

selling electricity; in addition to the superior decisions that are generated from the MC

Agent.

2. Processing - all received data are processed with the proposed pricing and bidding

methods explained in Chapter 5. Receiving orders from the market and GSM agent as

discussed in Chapter 5.

4. Data beaming - all results of the pricing decisions are beamed to the markets agents,

and to other agents in the framework as discussed in Section 6.5.

144
6.5.4 Master Controller agent

The implementation of this agent involves setting the type and the sequence for the signals

generated from this agent for the purpose of controlling other agents in the framework.

Hence these signals are generated based on the operational conditions (connection modes)

of all the active agents at the time of decision. The agents connection modes, and the

messaging directions created in these modes, are illustrated in Figure 6.2. The MC agent

development included:

1. Data streaming - the data are sourced from other agents to indicate their operation

status based on 30-minute intervals. The streaming data consists of the decisions generated

from the GSM agent and BAG agent; in addition to the grid agent that reports the

connection status.

2. Processing - all received data are processed with the proposed operational condition

rules. The processing includes running the basic management strategy in emergency modes,

and setting the orders to other agents to run the advance management strategy in normal

operation modes.

3. Data beaming - all decisions applied on the microgrid's agent operation strategies

and messaging directions are beamed to all other agents in the framework as discussed in

Section 6.5.

6.6 Results

In this section, the signicance of using the proposed strategy with MAS for managing

resources in an autonomous microgrid was presented. Ultimately, the results target two

main issues for the microgrid operation management: 1) the importance of the proposed

strategy in reducing the generation cost in the targeted case study; and 2) the role of

agents technology in modelling the proposed strategy. In both considerations, the results

cover the economic and the environmental costs throughout a year of simulation time. The

subsections below evaluate the impact of the agents on reducing the generation costs.

6.6.1 The role of forecasting in the proposed management strategy

The demand forecasting was integrated with the proposed management system, to max-

imise the utilisation of generation reserve and to schedule the operation of the generators

as explained in 3. In the basic management strategy, the demand forecasting is utilised in

order to manage the operation of utility and gas gen-sets. Accordingly, the demand was

145
fully supplied from gas gen-sets in Islanding modes. The Islanding modes are assumed to

be activated in emergency cases, and when the GP becomes less than UP . Ultimately, full

generation capacity is supplied to the load, thus 6 gas gen-sets run all the times in these

modes. Hence the environmental cost is evaluated based on the following function:

EnviCost = (GG + GlostG ) ∗ GE )

(6.1)

In the economic cost function, there are additional variable and xed costs for each

dispatch source, added to the main cost function: (M oV ) variable maintenance and start-

ing and shutting down cost, and (M oA) annual maintenance and insurance cost. The

economic cost is evaluated based on the following function:

EconCost = (GG + GlostG ) ∗ (GP + M oVG + M oAnG )

(6.2)

The Islanding mode is achieved when the following condition becomes true:

(((GG + GlostG ) ∗ GE ) + ((GG + GlostG ) ∗ (GP + M oVG +

M oAnG )))>(((UG ) ∗ UE ) + ((UG ) ∗ (UP ))

based on this condition, Figures 6.9 and 6.10 show the impact of utilising the demand

values in reducing the economic and environmental costs throughout the year. In the

simulation, the microgrid's mode of operation was switched automatically between Con-

necting and Islanding modes, based on the variation of economic and environmental costs.

The evaluation of economic cost was evaluated from the Gp and UP data. To evaluate the

environmental cost, modern gas and steam turbine power plants were found to emit 435

kg (960 lb) CO2/MWh, while coal-red power plants were found to emit 900 kg (1980 lb)

CO2/MWh [225]. Figures 6.9 and 6.10 show that by taking the forecasting into consider-

ation, the economic and environmental costs would be reduced by 25%.

In both Figures 6.9 and 6.10, it is shown that in Islanding mode, the full generation

capacity is identied; which results in a variable amount depending on the month. There-

fore, when considering the demand, the system would be more eective in winter, when a

low demand prole is identied.

146
Figure 6.9: Demand forecasting impact on reducing the economic cost ($) of the electricity
supply in the case study

Figure 6.10: Demand forecasting impact on reducing the CO2 emission (Kg) of the elec-
tricity supply in the case study

147
6.6.2 The role of Generation and Storage Management agent in the

proposed strategy

The role of this agent was evaluated according to its ability to maintain low economic

and environmental generation costs throughout the year. The most challenging part in

performing optimum power generation in this agent, is determining SP . This agent was

also evaluated according to its ability to control the SDs, in order to perform a benecial

charging/discharging process. The failure of this agent will result in activating the emer-

gency operation mode, that runs based on the basic management strategy. In the basic

management strategy, the load is supplied from the utility only at the Connecting mode,

and from the full capacity of the installed gas generators in Islanding mode. This fact will

result in a huge dierence in the generation cost when the advanced management strategy

is enabled. Furthermore, operating the microgrid without controlling the storage devices'

charging/discharging process will result in the SDs running continuously, which will re-

sult in a higher generation cost that includes the operating cost and the cost incurred by

the utility and gas generation prices. In Figures 6.12 and 6.13 and Table 6.1, the results

illustrate the role of this unit in controlling SG , reducing the overall cost and increasing

the utilisation of reserve over all types of connection modes. Hence the wind turbines and

the solar panels generation cost over the 25 years of operation, with the estimated mainte-

nance and insurance cost can be added to the overall achieved costs. The renewable energy

sources' maintenance and insurance costs were estimated by [4, 226]. These estimates have

been reected in our case study's generation cost by considering the size of the generation

units in line with their number of operation hours. The ratio of supplied energy from wind

turbines and solar panels is illustrated in Figure 6.11 as compared with demand values

throughout the year.

Finally, the economic and environmental costs are compared, based on the utilisation of

the advanced management strategy by the GSM agent over the basic management strategy

by the MC agent. Figures 6.12 and 6.13 compares the evaluated economic and environ-

mental costs under Islanding and Connecting modes, with basic management strategy and

both connection modes with the advanced management strategy.

The results show that the evaluated costs under all microgrid connection modes are

mainly dependent on the demand in each month. The maximum demand was identied in

February during the summer time, with full load utilisation in the university; diering from

December and January, as these two months usually encounter public holidays, which would

result in a low power consumption. The lowest demand values can be identied during

148
Figure 6.11: The ratio of the renewable energy sources generation to the demand values
throughout the year in the case study

Figure 6.12: Economic generation costs in ($)

149
 Figure 6.13: Environmental generation costs in (Kg)

the winter season from June till September, as shown in Figures 6.12 and 6.13. Tables

6.1, 6.2 and 6.3 compare the utilisation of resources in the basic management system and

the advanced management system under all types of connection modes. After evaluating

the average UP and GP , which are 0.28 $/kWh and 0.2 $/kWh respectively, we found that

although UP is greater than GP , UG has more penetration due to its cheaper operating

cost, and the fact that it does not waste reserve.

In Table 6.1, it is also noted that SDs utilisation was controlled, by monitoring cost

variation in UG and GG to have optimum amount and time for charging/discharging, to

maximise their eciency for both economic and environmental costs. Running the SDs

without controlling them would result in a big economic and environmental loss, due to UG
and GG cost variation; in addition to their operating cost. Ultimately, GSM agent could

save these losses, making the SDs of benet to the ecient running of the system.

The other identied cost parameter is the gas gen-sets lost generation reserve. The

proposed GSM agent aimed to utilise the generation reserve in accordance with the safety

requirements for the microgrid's power supply. In the Connecting mode under the basic

management strategy, the demand is fully supplied from the utility. In this scenario, the

lost reserve that was incurred by the microgrid as shown in Table 6.2 will not take place;

instead, the lost reserve is incurred by the utility generators, an outcome which is not

150
 Table 6.1: The cost of running the advanced management system for the resources utilisation throughout the year

SG kW UG kW GG kW Lost Reserve kW
Month
AP $ CO2 (Kg) $ CO2 (Kg) $ CO2 (Kg) $ CO2 (Kg)
Jan 466 4659 3.14 313860 629.06 258596 344.08 136636 167.58
Feb 825 3887 2.67 395380 736.72 294299 378.69 137267 167.07
Mar 694 4621 3.11 324230 661.12 255213 339.30 135447 165.80
Apr 323 4577 2.92 239644 488.6 258607 337.23 145129 176.45
May 519 2071 3.12 361093 691.1 259468 345.60 136293 167.01
Jun 561 4582 2.99 322910 642 280337 365.07 145248 176.27
Jul 664 2081 3.04 402363 770.45 265775 354.96 136124 166.82
Aug 824 4854 3.07 291522 550 234194 308.95 136291 167.10
Sep 751 4606 2.83 234080 452.55 236577 305.37 141811 172.42
Oct 614 4763 3.00 369004 719.74 258963 344.41 136257 167.06
Nov 1152 4455 2.92 214040 400.5 238191 308.52 139684 170.05
Dec 673 4602 3.16 25799 527.8 233840 308.30 137977 169.52

151
Table 6.2: The cost of running the basic management system for the resources utilisation
under Connecting mode

Ug
Month Gg Lost Reserve
$ CO2 (Kg)
Jan 572976.5 1144.6875 0 0
Feb 758705 1388.0545 0 0
Mar 579256.5 1176.9185 0 0
Apr 524867.5 1064.4255 0 0
May 632015.5 1213.1575 0 0
Jun 637223 1270.1615 0 0
Jul 683871 1318.1785 0 0
Aug 532442 1001.007 0 0
Sep 498206.5 963.5535 0 0
Oct 640246 1235.546 0 0
Nov 523229 970.2705 0 0
Dec 469999 960.446 0 0

Table 6.3: The cost of running the basic management system for the resources utilisation
under Islanding mode

Ug Gg Lost Reserve
Month
$ $ CO2 (Kg) $ CO2 (Kg)
Jan 0 826800 1003.2185 373295 449.58
Feb 0 746099 906.105 191856 235.211
Mar 0 826800 1003.2185 359594.5 434.374
Apr 0 800867 970.811 378248.5 456.338
May 0 826800 1003.2185 345609 416.859
Jun 0 800867 970.811 293225 356.899
Jul 0 826800 1003.2185 300975.5 366.099
Aug 0 826800 1003.2185 429250 519.398
Sep 0 800867 970.811 416939.5 505.093
Oct 0 826800 1003.2185 330604.5 501.846
Nov 0 800867 970.811 416360.5 501.846
Dec 0 826800 1003.2185 443804.5 539.003

considered in this study at this stage. In Islanding mode under the basic management

strategy, it is assumed that the demand is fully supplied from the distributed gas gen-sets.

In this sense, the generation will result in a spinning reserve, which is incurred by the

gas gen-sets. This lost reserve is found in the advanced management strategy, but with

relatively small amount. By comparing the results in Tables 6.1 and 6.3 it is shown that the

advanced management strategy could utilise up to 6% of the total lost reserve throughout

the year.

152
Table 6.4: Comparing the amount of reserve sold with the amount of the reserve oered
in the market

Month
RdyF rSLElec kW SLDElec kW
30 days 31 days 28 days

Feb
240196 117621
331950 138730 Jul
363319 134030 May
397058 170177 Oct
411850 63231 Mar
458499 104766 Jan
526950 217790 Aug
709219 315927 Dec
905687 425920 Sep
936272 428840 Apr
1259486 174973 Nov
2863598 288030 Jun

6.6.3 The role of Bids Analysis and Generation agent in the proposed

strategy

It is intended that this agent performs electronic negotiations to sell the generated elec-

tricity to the utility under various competitive pricing rules. Although the amount of

electricity actually sold will be based initially on the amount of generated, CFP signals

from the utility, and the oered proposals from the sellers; the system proved its ability to

attract the attention of the market, when tested with other sellers under same operational

conditions. Due to unexpected markets and the behaviour of other sellers, it was easier to

test this unit under the uniform pricing rules, so that the oered proposals can be accepted

when they become lower than the market balance point. Table 6.4 compares the amount of

the sold reserve with the amount of the total reserved oered to the market by considering

the type of the month.

In Table 6.4, it is shown that the amount of reserve mainly varies depending on the

season, in addition to the type of the month (31 days, 30 days or 28/29 days). In winter, a

relatively low amount of demand is encountered, when the distributed generators can oer

higher amount of reserve, except in July, as the demand will be higher when more activities

are expected to take a place in the university as the semester starts. In contrast to summer,

a lower amount of reserve is oered to the market as shown in the Table 6.4. Generally,

the demand is high in summer; however, the wind speed and the solar irradiation are also

high, which makes the amount of reserve ranked in the middle. Having the semester break

as in summer will also help increase the amount of reserve. The results show that February

153
has the lowest amount of reserve in this study. Since February has 29 days of operation,

in addition to the extensive utilisation for load in this month due the starting of semester

activities around the campus. However, in February, there is still some amount of reserve

that can be oered to the market. This is due to the high amount of solar generation

encountered at that particular period as shown in Figure 6.11 in line with the temperature

drop at the second half of the month. In case of both seasons: spring and autumn, there will

be higher dierence between the demand and the full generation capacity of the microgrid,

especially in semester breaks as in April, September and November. This is due to the

fairness of the ambient temperature in these seasons, which reduces the amount of power

required to run the air-conditioning. This dierence has been indicated by the amount of

reserve oered to the market in Table 6.4. The other considerable result is the relationship

between the amount of the transactions and the money value achieved out of selling the

power to the market. Figure 6.14 compares the amount of money achieved with the total

amount of transactions. By comparing the results illustrated in Figure 6.14 and Table 6.4,

it is shown that the higher amount of power sold to the market is encountered in accordance

with the high amount of reserve. Figure 6.14 also shows a non-linear relationship between

the amount of money and the transaction achieved out of selling the power to the market.

This is mainly due to the variation between the local generation price and the market price;

in addition to the adaptation in the sale price which reduces the amount of money achieved

in the favour of increasing the amount of sold power in a competitive market environment.

154
Figure 6.14: Comparing the amount of money achieved with the total amount of transac-
tions throughout the year

From the results in Table 6.4 and Figure 6.14, it was found that the system could sell

27.8% of the oered amount for sale in the market throughout the year with an average

price of 0.13 $/kWh. As comparing this price with the UP , which is of an average of 0.28

$/kWh throughout the year, we found that oering the electricity with 46% of the market

balancing price will only sell 27.8% under the uniform pricing rules. This value can be

changed, depending on the market conditions, accuracy of the pricing method that cope

with the markets requirement, and the variation between the market and the microgrid's

operational conditions. In this work we assumed that the market is ready to buy all the

oered electricity from the microgrid, if its generated bid was accepted.

6.7 Conclusion

In this chapter we presented the design and the development of a management strategy,

that encompasses distributed intelligent systems for optimising the generation cost in a

155
typical microgrid. The strategy was tested so that the importance of cutting the cost

of generation could be evaluated; where the simulation data was based on actual histor-

ical operational conditions. As was discussed in the literature, the MAS was eective in

modelling intelligent systems and participating systems with markets under dierent com-

petitive pricing rules. Eventually the MAS was widely applied to the management of power

systems. Therefore, it was utilised to model the proposed management strategy, and inte-

grate it with a simulated market environment to show the eectiveness of the intelligent

systems that we developed. In this chapter we presented the development of the proposed

management system, which included the integration of MAS technology with AI. The re-

sults showed that the proposed layers of strategies were eective in performing optimised

economic and environmental generation cost, and also providing remote monitoring, access

and control for the intelligent systems under various operational conditions. Furthermore,

it proved that the proposed strategies can run under dierent modes, and the proposed

MAS has the ability to recongure the strategies to add or remove agents to or from the

system depending on the operational conditions. Ultimately, adding MAS characteristics

to the proposed management strategy, provided exibility for system maintenance and up-

grade; which overcomes the limitations of the traditional management systems. As veried

by the realistic operational conditions data, the MAS was ecient, and could achieve the

humanisation management of the microgrid power generation. As a future work, AI can

be embedded in the generator's agents, to increase the quality for the generated electricity

and optimise the dispatch for the microgrid. Since this work required a set of software

licenses to implement the proposed strategy, using MAS technology to build a system with

less number of licenses by directing all the distributed system requests to a one licensed

server agent, can be implemented as a future extension of this work. In addition, it is

intended that a suite of smart agent-based Android applications be developed, that can

provide remote mobile and instantaneous access, monitoring and control for the manage-

ment system.

156
Chapter 7

CONCLUSION

This thesis has introduced a new management strategy for power generation within a mi-

crogrid, where there are multiple sources of dispatch. The proposed management strategy

is based on the deployment of multiple intelligent software modules, which have been de-

veloped to perform individual management tasks simultaneously. The intelligent modules

undertake decision making, in order to achieve the lowest possible economic and environ-

mental costs, for power generation within a microgrid. The requirement for integrating

this level of intelligence into the management system, in order to operate the microgrid's

resources; has been established by analysing the microgrid's operational conditions, and

by critically reviewing the recommendations of other researchers in this eld. Short term

demand forecasting has been identied as an essential part of the proposed system, as

it provides the other stages of analyses, with the necessary information concerning the

requirements for power generation in the microgrid. Due to the uncertainty of both user

behaviour and weather conditions, obtaining an accurate demand prole has become a

challenging process. Therefore, it is crucial that an advanced modelling technique that

incorporates an adaptation of the demand forecasts be deployed, which can provide the

instantaneous operational conditions to the model. The implementation of a novel adap-

tive demand forecasting model which meets this criteria, has been illustrated in this thesis.

The second major issue we were confronted with when attempting to optimise power gen-

eration within the case study, was in controlling the attitude of the SDs under complex

operational scenarios. This problem has been addressed in this thesis by introducing a

novel charging/discharging process, which aims to improve the operation of the storage

devices. After optimising the operation of multiple sources of dispatch in the microgrid,

another cost eective factor has been identied, the generation reserve. A further stage of

investigation and analysis became essential in order to successfully incorporate this factor

when optimising the generation cost. Oering the reserve for sale to the grid became the

157
next challenge, the process also needing an extensive pricing analysis. Before implementing

the required pricing algorithm to optimise the bidding, it was also essential that the mar-

kets policies and the bidding rules were studied in detail. Ultimately, adaptive pricing rules

have been added to the proposed management system to respond to the competitiveness of

the market, and also to ensure that the amount of power oered at a competitive price is

maximised. Finally, since the proposed management system is needed to achieve reliable,

safe, and low cost generation for the microgrid; MAS technology has been implemented

to operate the distributed intelligent modules in a reliable manner. To ensure reliability

in the management decisions, the management system has been set to run the intelligent

modules in a complementary fashion, where all the modules are dependent upon each other

when making the nal and optimum decisions. In addition, the management system struc-

ture included a number of active agents that work together to support the system when

it encounters an operational fault. In this case, a recongurable management strategy is

activated to run the remaining agents in a dierent operating mode. The simulation results

illustrate the generation cost under dierent operational conditions, proving the success of

MAS technology to model the proposed management strategy with distributed processing

stages. This chapter summarises the ndings and the contributions of this thesis in Sec-

tion 7.1, and suggests several research directions that can be undertaken as future works

in Section 7.2.

7.1 The contribution of the thesis

The main contributions of this thesis are concluded as follows:

ˆ Introducing a novel modelling method that performs adaptive demand forecasting

for the case study. The modelling method that has been developed, has been im-

plemented to perform short term demand forecasting for the case study, by reading

the information relating to the forecasts such as: the air ambient temperature, date,

type of day, and time of forecasting. This combination has been nominated to suit

a university type demand prole after studying the demand patterns in the case

study. It has been identied that the demand prole is aected by individual and

correlated factors. After these factors have been identied, they have been included

in the model, though not through the traditional modelling training methods. The

factors have been included through an adaptation method, that tunes the forecasts

based on the knowledge about the real eect of the correlated parameters. This

158
 adaptation mechanism has been implemented via Fuzzy Logic (FL) with a rule base

system, developed to impose the eect of these parameters on giving more accurate

forecasts. We have also made note of the inuence that our experience in modelling

has, when attempting to set the correct membership functions and universe of dis-

course settings. (These functions and settings are in turn applied to the fuzzy tuning

system in order to enhance forecasting accuracy). Apart from the main forecasting

model, another intelligent component has also been added to the model; to return

the real-time demand values to the model's forecasts, through an optimum tuning

process that aims to achieve the highest forecasting accuracy. This part of the mod-

elling includes another adaptation method; to ensure that the system can react to

unexpected changes in the real-time demand, based on optimum adaptation values

achieved with low optimisation searching times. The optimisation method utilised

in this model was developed by training the Articial Neural Network (ANN) to

react to changes by providing suitable tuning values. The results obtained from test-

ing have enabled the conclusion to be drawn that the proposed demand forecasting

model achieves more accurate forecasting in comparison to the traditional model. In

addition, the results have also proved that the adaptation mechanism we added, was

a viable solution for utilising the forecasts with the generation scheduling problem

in the case study; when further analysis and studies involved.

ˆ Implementing a new optimisation method, to optimise the performance of the mi-

crgrid's power generation, while being supplied from various types of distributed

generators. The implementation of the proposed method was undertaken to cope

with the variations in the cost of generation that result from: variable utility tari

($/kWh), variable gas generation cost ($/kWh), dynamic demand prole (kWh), in-

termittent solar and wind generation (kWh), unexpected SDs charging/discharging

price ($/kWh), and eventually the unpredicted SDs generation capacity (kWh). Op-

erating wind turbines, solar panels, SDs, gas gen-sets, and utility supply simulta-

neously would result in high generation costs; unless a robust management system

is involved to track the dramatic changes to the cost of generation, and match the

supplied power to the level of demand being experienced in the microgrid. To cope

with these challenges, our proposed optimisation technique is comprised of a combi-

nation of intelligent methods, applied to optimise the performance of the microgrid

power dispatch. Traditional Particle Swarm Optimisation (PSO) has been applied, to

ensure that optimum decision- making is occurring when scheduling the power gener-

159
 ation in the microgrid. However, to deal with the challenging SDs operation, a price

monitoring mechanism has been integrated with the PSO to enable making protable

charging/discharging decisions, while maintaining optimum performance for other re-

sources. Since utility supply and gas gen-set power generation are the main supply

points for the microgrid, their prices have been nominated as being the index for the

charging/discharging process. The method updates the charging/discharging price

threshold, by employing a daily price change trend; following this, a weekly average

of price change trends is calculated, which in turn becomes the charging price for the

utility supply or gas generation. This value has been considered in parallel with the

amount of storage device SDs participation in the microgrid dispatch, to ensure that

their protable operation time is maximised. In order to include these factors for

consideration when optimising the decision-making, a hybrid PSO-FL structure has

been implemented. In this instance, the results are accumulated in a daily pattern so

that the cost variations can be analysed, and then a rule base system is invoked, to

correlate the price variation with the amount of SDs utilisation. Finally this method

has been compared with the traditional PSO, giving consideration to the overall gen-

eration cost achieved, and also to the SDs' level of operational eciency. As a result,

this method has demonstrated that it can achieve both lower costs, and a better

operational eciency.

ˆ Developing novel pricing rules for oering the microgrid's spinning, and non-spinning

generation reserve, as AS to the electricity market. This part of the thesis includes

a basic economic analysis to place benecial, yet also competitive reserve prices, so

that both its prospects of market participation can be increased, and that the amount

of reserve sold to the market can be maximised. The pricing rules have been built to

reect the strategic thinking of a seller, who might adjust his/her prices according

to his/her operational circumstances, in parallel with the market conditions. The

rules have also been designed to include the near future operational conditions in the

pricing. It has been suggested that this strategy may have a better outcome in the

longer term, than that which is possible from the xed pricing rules. The proposed

pricing mechanism is comprised of four stages of processing: 1) the decision tree

stage, 2) the forecasting stage, 3) the estimation stage, and 4) the adaptation stage.

At each stage of processing, a proper AI method is invoked to achieve the objectives

required by the stage itself. In the main part of the pricing, i.e. the decision tree,

the initial pricing rules have been devised in such a way, that they can evaluate the

160
 dierence between the microgrid's local generation cost point, and the market bal-

ancing point. Depending on the operational scenarios, variable local generation price

is often encountered, thus variable prices (SLp) are also expected from the micro-

grid. In order to maximise the prot and increase the chance of market participation,

the initial prices have been set so that they are midway between the two points. It

was proposed that the middle point move between the market balancing and the

local generation cost points, as the operational conditions are explored; looking for

possibilities to maximise the prot and the chance of market participation. To this

end, a simple forecasting module has been implemented; to forecast the near future

weather conditions (wind speed and solar irradiation), the utility supply tari and

gas generation cost, and to include these estimates in the nal prices. The proposed

forecasting models have been built to consider the real-time operational conditions

in the forecasting, which would benet from having accurate short term forecasts.

To utilise these forecasts in the pricing, the third processing stage has been designed

to estimate the operation of the microgrid resources, based on the forecast values.

PSO has been incorporated in the design, to utilise the forecasts when estimating

the optimum operation prole for the microgrid. This estimation was necessary for

building predictions about the reserve conditions, to make the pricing more strate-

gic. These estimates are utilised by the fourth processing stage, to adapt the initial

prices and make them market competitive. Basically, the adaptation is subject to

three elements that adjust the prices in the appropriate direction: the amount of

reserve oered for sale RdyF rSLElec, the evaluated microgrid local generation price
SLp, and the future direction of price change IncorDec (up/down). These factors

are correlated through a fuzzy system that operates according to a predened rule

base, to reect the desired decisions, corresponding to each operational scenario.

MAS technology has been utilised to simulate a competitive market environment,

that represents three competitive sellers, including our case study, and a one buyer

(utility grid). The proposed pricing method has been tested under various market

policies, where the proposed method has consistently sold a higher amount of reserve.

ˆ Implementing a Multi-Agent based management System (MAS) that holds strate-

gic behaviours, to maintain reliable, secure, and cost eective operation for the case

study. Initially, the system was implemented as a collection of distributed intel-

ligent agents, controlled via a central agent, and working together to achieve the

management system objectives. The distributed agents were centrally controlled to

161
 keep them running in a timely fashion under predened strategies. Furthermore, the

operational structure of the agents was set to be recongurable, depending on the

scenarios of the microgrid and the management system. The system that we have im-

plemented, is comprised of four levels of increasing sophistication of decision making;

so that the microgrid can be operated under dierent economic and environmental

cost levels. The decision levels are enabled, according to the information supplied

about the case study, and the distributed agents status. The proposed agents were

implemented within three major categories: Input/output interface, communication,

and processing. Within each category, agents were assigned to deal with the op-

erational scenarios, reacting to changes in the operating conditions, so that they

provide the optimum decision. The system implementation has proved its success in

running low cost generation for the microgrid, under dierent operational scenarios.

The simulation studies have also proved that the proposed strategy was successful,

in making the microgrid participate appropriately under dierent market policies.

From the reliability perspective; implementing MAS based technology for manag-

ing resources, could overcome the traditional management systems, by including the

exibility in maintenance and upgrade when required. The system has been veried

with detailed operational conditions data, which have been derived from a set of engi-

neering assumptions. In conclusion, the MAS was able to reect the human strategic

behaviour, in managing the power of generation for a microgrid with multiple sources

of dispatch.

7.2 Future works

The possible future extensions of this work can be undertaken in the following areas:

7.2.1 Developing adaptive membership function demand forecasting mod-

els

Our proposed demand forecasting has been implemented via an intelligent adaptive model

that adapts the forecasts to the operational conditions. By analysing the operation of

the models that we developed, several opportunities for investigating how to improve the

adaptation performance have been identied. The rst possible research direction of this

work, can be the fuzzy model membership functions update based on the demand accu-

racy. In this case, the demand accuracy can be represented as the cost of a function, that

162
encompasses the shape and the range of discourse of the fuzzy model's membership func-

tions. The rationale being that this approach will build a dynamic modelling structure,

capable of coping with the uncertainties of the variable modelling parameters. The other

research direction extended from this work, can be implementing an optimisation technique

to tune the feedback fuzzy tuning parameters, based on the operational conditions in the

microgrid. In this case, the modelling accuracy will be represented by a cost function, for

the feedback adaptation membership functions of the fuzzy system. In both cases, the

demand forecasting accuracy will be the index for updating the model's parameters. Yet,

the limitation of implementing the optimisation technique to tune the model's parameters,

is the searching time that should match with demand forecasting intervals.

7.2.2 Considering power quality and market participation in optimising

the microgrid performance

In the second intelligent module, we have presented a novel optimisation method, for op-

timising the microgrid performance with various types of distributed generators. In this

context, four distinct research investigations can be launched from this chapter. 1) Mod-

elling the generation price change pattern for the microgrid resources. 2) Considering the

microgrid market participation in prioritising the operation of the resources in the mi-

crogrid. For instance, making the SDs operation attractive to the market; or including

forecasting modules, to guide the optimisation technique towards estimating the opera-

tional conditions, required to prioritise the operation of the resources. 3) The traditional

PSO utilised in this chapter has been implemented with x searching parameters, how-

ever there is an opportunity for including online parameters adaptation, that adjusts the

searching mechanism, according to the operational scenarios. This part of the work will

help improve the traditional PSO performance, which would result in a lower generation

cost. Adapting the PSO parameters has been addressed using FL in [148] on managing

microgrid resources. The future work arising from this chapter is in teaching the PSO how

to select its best searching parameters, based on the historical values of the achieved cost,

using Articial Neural Networks (ANN). 4) Considering the voltage and frequency droop

control, in line with the economic and environmental dispatch in the power management

strategies, to improve the reliability and eciency of the power generation in microgrids.

163
7.2.3 Implementing adaptive pricing rules that aim to sell the power

reserve in a limited time frame

In this work, we have implemented an adaptive pricing mechanism that adjusts both the

initial prices oered to the market, and the microgrid's operational conditions. The suc-

cessfulness of the method was tested under a variety of dierent market rules. However,

in all the market rules, the power trade is made after a set of electronic negotiations are

conducted; resulting in either the acceptance or rejection of the bids which were previously

placed. This was made in a xed transaction time frame. The rst instance of a possible

future work, is to investigate the application of adaptive pricing rules to nd the optimum

deal within the market, after receiving a market rejection for the microgrid's initial bid;

while utilising the lowest possible negotiation time. This can be implemented by converg-

ing the price change towards the lowest expected market price, in order to achieve the

least possible negotiation time. The second possible extension to our research work is in

adjusting the current implemented pricing factors, by monitoring the rate of successful

transactions achieved by the pricing parameters. Accordingly, adjusting these parameters

by making the maximum number of transactions, is the cost for the pricing parameters

function. This would help maximise the number of successful power trade transactions.

The investigation would also include an analysis of the specic optimisation methodology

which would be most appropriate for this research.

7.2.4 Investigating the role of agent technology in implementing a mo-

bile distributed energy management system using the Android

operating system

It has been demonstrated that the MAS with distributed AI, is successful in reecting the

human attitude in making strategic decisions for optimising the microgrid's power genera-

tion performance. As a future work, the MAS implementation can be extended to include

sensors, relays, and generators; enabling remote access, a greater degree of control over the

devices, and providing for a more sensitive management system. Furthermore, AI can be

deployed on the generator's agents, in order to provide a multi-objective optimisation func-

tionality that includes power quality control. In this case, we are expecting high amount

of computation resources required by the intelligent distributed systems. Ultimately, in-

vestigating the role of MAS technology in reducing the amount of these resources can be

another extension to this work. MAS technology can be utilised in building a centralised

system that directs all the distributed agents' tasks to a centralised server agent. This

164
agent by its turn can be developed to process the requests sent from the distributed sys-

tems in a timely fashion, thus ensuring systematic operation for the distributed agents.

Further extensions could also include the design of control systems capable of coping with

voltage and frequency instabilities in the dispatched power. It is also possible to extend

the mobile-based agent technology to Android devices, allowing for more exible remote

access capabilities for controlling the system.

165
Bibliography

[1] J. R. Jang, C. Sun, E. Mizutani, Neuro-Fuzzy and Soft Computing: A Computational

Approach to Learning and Machine Intelligence. MATLAB CURRICULUM SERIES,
Upper Saddle River, New Jersey: Prentice Hall, 1997.

[2] Fabio Bellifemine, Giovanni Caire, Tiziana Trucco, Giovanni Rimassa, Jade (java

agent development framework). http://jade.tilab.com/. [Online; accessed 15-

April-2013].

[3] Bureau of Meteorology, National weather forecasts. http://www.bom.gov.au/.

[Online; accessed 15-April-2013].

[4] E. Hau, Wind Turbines: Fundamentals, Technologies, Application, Economics.

Berlin: Springer, 2006.

[5] F. Javed, N. Arshad, F. Wallin, I. Vassileva, and E. Dahlquist, Forecasting for

demand response in smart grids: An analysis on use of anthropologic and structural

data and short term multiple loads forecasting, Applied Energy, no. 0.

[6] K. H. Chung, B. H. Kim, and D. Hur, Distributed implementation of generation

scheduling algorithm on interconnected power systems, Energy Conversion and

Management, vol. 52, no. 12, pp. 34573464, 2011.

[7] T. Logenthiran and D. Srinivasan, Short term generation scheduling of a microgrid,

in TENCON 2009 - 2009 IEEE Region 10 Conference, pp. 16.

[8] H. Siahkali and M. Vakilian, Fuzzy generation scheduling for a generation company

(genco) with large scale wind farms, Energy Conversion and Management, vol. 51,

no. 10, pp. 19471957, 2010.

[9] R. Tanha Aminloei and S. F. Ghaderi, Generation planning in iranian power plants

with fuzzy hierarchical production planning, Energy Conversion and Management,

vol. 51, no. 6, pp. 12301241, 2010.

166
[10] J. Wang, Short-term generation scheduling model of fujian hydro system, Energy
Conversion and Management, vol. 50, no. 4, pp. 10851094, 2009.

[11] W. H. Kersting, Distribution System Modeling and Analysis. The Electric Power

Engineering, Boka Raton: CRC press, 2006.

[12] P. Clark W. Gellings, The Smart Grid ENABLING ENERGY EFFICIENCY AND
DEMAND RESPONSE. Lilburn: The Fairmont Press Inc., 2009.

[13] B. Roberts, Capturing grid power, Power and Energy Magazine, IEEE, vol. 7, no. 4,
pp. 3241, 2009.

[14] S. M. A. Wollenberg and B. F., Tward a smart grid, IEEE Power and Energy
Magazine, vol. 3, no. 3, pp. 3441, 2005.

[15] C. W. JOEL MAKOWER, RON PERNICK, Clean energy trneds 2009, tech. rep.,

2009.

[16] S. Honkapuro, Business impacts and incentives of regulation for dsos and smart

grids, in Electricity Distribution - Part 2, 2009. CIRED 2009. The 20th Interna-
tional Conference and Exhibition on, pp. 110, 2009.

[17] D. Novosel, Emerging technologies in support of smart grids, in Power and Energy
Society General Meeting - Conversion and Delivery of Electrical Energy in the 21st
Century, 2008 IEEE, pp. 12, 2008.

[18] D. R. C. U. W. Breuer, D. Povh and M. Weinhold, Prospects of smart grid technolo-

gies for a sustainable and secure power supply, tech. rep., World Energy Council,

2007.

[19] "Australian Government", Department-of-Climate-Change-and-Energy-Eciency,

Reducing australia's emission. http://www.climatechange.gov.au/government/

reduce.aspx/. [Online; accessed 15-April-2013].

[20] Department of Resources Energy and Tourism, Low emissions technol-

ogy demonstration fund. http://www.ret.gov.au/energy/clean/low_emissions_

technology_demonstration_fund/. [Online; accessed 15-April-2013].

[21] Australian Government, Smart grid, smart city. http://www.

smartgridsmartcity.com.au/. [Online; accessed 15-April-2013].

167
[22] Oce of the Renewable Energy Regulator, Legislation. http://ret.
cleanenergyregulator.gov.au/For-Industry/Legislation/. [Online; accessed

15-April-2013].

[23] Australian Government, Securing australia's energy future. http://www.efa.com.

au/Library/CthEnergyWhitePaper.pdf/. [Online; accessed 15-April-2013].

[24] Department of Climate Change and Energy Eciency, Renewable energy tar-

get. http://www.climatechange.gov.au/reducing-carbon/renewable-energy/
renewable-energy-target/. [Online; accessed 15-April-2013].

[25] WAtoday, renewable energy taris announced. http://www.watoday.com.au/

wa-news/renewable-energy-tariffs-announced-20100527-wgnn.html/. [Online;

accessed 15-April-2013].

[26] M. Pipattanasomporn, H. Feroze, S. Rahman, Multi-agent systems in a distributed

smart grid: Design and implementation, in IEEE PES 2009 Power Systems Con-
ference and Exposition, (Seattle, Washington, USA), pp. 18, IEEE, 2009.

[27] M. Zamarripa, J. C. Vasquez, J. M. Guerrero, and M. Graells, Detailed Operation

Scheduling and Control for Renewable Energy Powered Microgrids, vol. Volume 29,

pp. 18191823. Elsevier, 2011.

[28] G. Platt, A. Berry, and D. Cornforth, Chapter 8 - What Role for Microgrids?,
pp. 185207. Boston: Academic Press, 2012.

[29] M. B. Ammar, M. Chaabene, and A. Elhajjaji, Daily energy planning of a household

photovoltaic panel, Applied Energy, vol. 87, no. 7, pp. 23402351, 2010.

[30] S. J. Ahn, S. R. Nam, J. H. Choi, and S. I. Moon, Power scheduling of distributed

generators for economic and stable operation of a microgrid, Smart Grid, IEEE
Transactions on, vol. PP, no. 99, pp. 18, 2013.

[31] B. Khorramdel and M. Raoofat, Optimal stochastic reactive power scheduling in a

microgrid considering voltage droop scheme of dgs and uncertainty of wind farms,

Energy, vol. 45, no. 1, pp. 9941006, 2012.

[32] P. K. Naraharisetti, I. A. Karimi, A. Anand, and D.-Y. Lee, A linear diversity

constraint application to scheduling in microgrids, Energy, vol. 36, no. 7, pp. 4235
4243, 2011.

168
[33] L. Hernandez, C. Baladron, J. M. Aguiar, B. Carro, A. Sanchez-Esguevillas, J. Lloret,

D. Chinarro, J. J. Gomez-Sanz, and D. Cook, A multi-agent system architecture for

smart grid management and forecasting of energy demand in virtual power plants,

Communications Magazine, IEEE, vol. 51, no. 1, pp. 106113, 2013.

[34] C. M. Colson, M. H. Nehrir, and C. Wang, Ant colony optimization for microgrid

multi-objective power management, in Power Systems Conference and Exposition,

2009. PSCE '09. IEEE/PES, pp. 17.

[35] A. M. Melamed, Y. I. Morzhin, V. F. Timchenko, E. V. Tsvetkov, and V. F. Zharkin,

Improvement of electric power system load forecasting and operational planning

for control automation, International Journal of Electrical Power and amp; Energy
Systems, vol. 11, no. 3, pp. 218222, 1989.

[36] Javed, Fahad and Arshad, Naveed and Wallin, Fredrik and Vassileva, Iana and

Dahlquist, Erik

[37] Y. Hida, R. Yokoyama, K. Iba, K. Tanaka, and K. Yabe, Load forecasting on demand

side by multi-regression model for operation of battery energy storage system, in

Universities Power Engineering Conference (UPEC), 2009 Proceedings of the 44th

International, pp. 15.

[38] A. Tascikaraoglu, M. Uzunoglu, and B. Vural, The assessment of the contribution of

short-term wind power predictions to the eciency of stand-alone hybrid systems,

Applied Energy, vol. 94, no. 0, pp. 156165, 2012.

[39] B. Jiang and Y. Fei, Dynamic residential demand response and distributed gener-

ation management in smart microgrid with hierarchical agents, Energy Procedia,

vol. 12, no. 0, pp. 7690, 2011.

[40] S. C. Chan, K. M. Tsui, H. C. Wu, H. Yunhe, W. Yik-Chung, and F. F. Wu, Load-

price forecasting and managing demand response for smart grids: Methodologies and

challenges, Signal Processing Magazine, IEEE, vol. 29, no. 5, pp. 6885, 2012.

[41] A. Barbato and G. Carpentieri, Model and algorithms for the real time management

of residential electricity demand, in Energy Conference and Exhibition (ENERGY-

CON), 2012 IEEE International, pp. 701706.

[42] L. Wang, Adaptive fuzzy systems and control: design and stability analysis. NJ:

Prentice-Hall, 1994.

169
[43] E. G. Donald and C. K. William, Fuzzy clustering with a fuzzy covariance matrix,

in Decision and Control including the 17th Symposium on Adaptive Processes, 1978
IEEE Conference on, vol. 17, pp. 761766.

[44] I. Gath and A. B. Geva, Unsupervised optimal fuzzy clustering, Pattern Analysis
and Machine Intelligence, IEEE Transactions on, vol. 11, no. 7, pp. 773780, 1989.

[45] F. Hoppner, F. Klawonn, R. Kruse and T. Runkler , Fuzzy Cluster Analysis: Methods
for Classication, Data Analysis and Image Recognition. 1999.

[46] R. R. Yager and D. P. Filev, Approximate clustering via the mountain method,

Systems, Man and Cybernetics, IEEE Transactions on, vol. 24, no. 8, pp. 12791284,
1994.

[47] Y. Ronald and F. Dimitar, Essentials of Fuzzy Modeling and Control. New York,

NY, USA: John Wiley & Sons, Inc., 1st ed., 1994.

[48] S. L. Chiu, Fuzzy model indentication based on cluster estimation, Journal of

Intelligent and Fuzzy Systems, vol. 2, pp. 267278, 1994.

[49] J. Wang, L. Li, D. Niu, and Z. Tan, An annual load forecasting model based on sup-

port vector regression with dierential evolution algorithm, Applied Energy, vol. 94,
no. 0, pp. 6570, 2012.

[50] V. Behbood, L. Jie, and Z. Guangquan, Adaptive inference-based learning and

rule generation algorithms in fuzzy neural network for failure prediction, in Intel-
ligent Systems and Knowledge Engineering (ISKE), 2010 International Conference
on, pp. 3338.

[51] J. S. R. Jang, Ans: adaptive-network-based fuzzy inference system, Systems, Man

and Cybernetics, IEEE Transactions on, vol. 23, no. 3, pp. 665685, 1993.

[52] Chin-Teng Lin, C. S. Goerge Lee, Neural Fuzzy Systems: A Neuro-Fuzzy Synergism
to Intelligent Systems. New Jersey: Prantice-Hall, 1996.

[53] L. X. Wang and J. M. Mendel, Back-propagation fuzzy system as nonlinear dy-

namic system identiers, in Fuzzy Systems, 1992., IEEE International Conference

on, pp. 14091418.

[54] S. Zhu, J. Wang, W. Zhao, and J. Wang, A seasonal hybrid procedure for electricity

demand forecasting in china, Applied Energy, vol. 88, no. 11, pp. 38073815, 2011.

170
[55] Y. Yulai, W. Yongwei, Y. Guangwen, and Z. Weimin, Adaptive hybrid model for

long term load prediction in computational grid, in Cluster Computing and the Grid,
2008. CCGRID '08. 8th IEEE International Symposium on, pp. 340347.

[56] Purwanto, C. Eswaran, and R. Logeswaran, An adaptive hybrid algorithm for time

series prediction in healthcare, in Computational Intelligence, Modelling and Simu-

lation (CIMSiM), 2010 Second International Conference on, pp. 2126.

[57] C.-H. Cheng and L.-Y. Wei, One step-ahead ans time series model for forecasting

electricity loads, Optimization and Engineering, vol. 11, no. 2, pp. 303317, 2010.

[58] I. Luna and R. Ballini, Top down strategies based on adaptive fuzzy rule based sys-

tems for daily time series forecasting, International Journal of Forecasting, vol. 27,

no. 3, pp. 708724, 2011.

[59] N. Wagner and Z. Michalewicz, Adaptive and Self adaptive Techniques for Evolution-
ary Forecasting Applications Set in Dynamic and Uncertain Environments Founda-
tions of Computational Intelligence Volume 4, vol. 204 of Studies in Computational
Intelligence, pp. 321. Springer Berlin, Heidelberg, 2009.

[60] A. Karsaz, H. R. Mashhadi, and M. M. Mirsalehi, Market clearing price and load

forecasting using cooperative co-evolutionary approach, International Journal of

Electrical Power and amp; Energy Systems, vol. 32, no. 5, pp. 408415, 2010.

[61] H. T. Nguyen and I. T. Nabney, Short-term electricity demand and gas price fore-

casts using wavelet transforms and adaptive models, Energy, vol. 35, no. 9, pp. 3674
3685, 2010.

[62] N. Augustine, S. Suresh, P. Moghe, and K. Sheikh, Economic dispatch for a micro-

grid considering renewable energy cost functions, in Innovative Smart Grid Tech-
nologies (ISGT), 2012 IEEE PES, pp. 17.

[63] E. R. Sanseverino, M. L. Di Silvestre, M. G. Ippolito, A. De Paola, and G. Lo Re,

An execution, monitoring and replanning approach for optimal energy management

in microgrids, Energy, vol. 36, no. 5, pp. 34293436, 2011.

[64] J. A. Peças Lopes, S. A. Polenz, C. L. Moreira, and R. Cherkaoui, Identication

of control and management strategies for lv unbalanced microgrids with plugged-in

electric vehicles, Electric Power Systems Research, vol. 80, no. 8, pp. 898906, 2010.

171
[65] G. Kyriakarakos, A. I. Dounis, K. G. Arvanitis, and G. Papadakis, A fuzzy logic en-

ergy management system for polygeneration microgrids, Renewable Energy, vol. 41,
no. 0, pp. 315327, 2012.

[66] G. Kyriakarakos, D. D. Piromalis, A. I. Dounis, K. G. Arvanitis, and G. Papadakis,

Intelligent demand side energy management system for autonomous polygeneration

microgrids, Applied Energy, vol. 103, no. 0, pp. 3951, 2013.

[67] C. A. Hernandez-Aramburo, T. C. Green, and N. Mugniot, Fuel consumption mini-

mization of a microgrid, Industry Applications, IEEE Transactions on, vol. 41, no. 3,
pp. 673681, 2005.

[68] A. K. Basu, Microgrids: Planning of fuel energy management by strategic deploy-

ment of chp based ders an evolutionary algorithm approach, International Journal

of Electrical Power and Energy Systems, vol. 44, no. 1, pp. 326336, 2013.

[69] C. Bustos, D. Watts, and H. Ren, Microgrid operation and design optimization with

synthetic winds and solar resources, Latin America Transactions, IEEE (Revista
IEEE America Latina), vol. 10, no. 2, pp. 15501562, 2012.

[70] S. Conti, R. Nicolosi, S. A. Rizzo, and H. H. Zeineldin, Optimal dispatching of dis-

tributed generators and storage systems for mv islanded microgrids, Power Delivery,
IEEE Transactions on, vol. 27, no. 3, pp. 12431251, 2012.

[71] Y. del Valle, G. K. Venayagamoorthy, S. Mohagheghi, J. C. Hernandez, and R. G.

Harley, Particle swarm optimization: Basic concepts, variants and applications in

power systems, Evolutionary Computation, IEEE Transactions on, vol. 12, no. 2,

pp. 171195, 2008.

[72] S. A. Pourmousavi, M. H. Nehrir, C. M. Colson, and W. Caisheng, Real-time energy

management of a stand-alone hybrid wind-microturbine energy system using parti-

cle swarm optimization, Sustainable Energy, IEEE Transactions on, vol. 1, no. 3,

pp. 193201, 2010.

[73] N. Phuangpornpitak, W. Prommee, S. Tia, and W. Phuangpornpitak, A study of

particle swarm technique for renewable energy power systems, in Energy and Sus-
tainable Development: Issues and Strategies (ESD), 2010 Proceedings of the Inter-
national Conference on, pp. 16.

172
[74] N. Gudi, W. Lingfeng, V. Devabhaktuni, and S. S. S. R. Depuru, Demand response

simulation implementing heuristic optimization for home energy management, in

North American Power Symposium (NAPS), 2010, pp. 16.

[75] A. K. Basu, S. P. Chowdhury, S. Chowdhury, and S. Paul, Microgrids: Energy

management by strategic deployment of ders a comprehensive survey, Renewable

and Sustainable Energy Reviews, vol. 15, no. 9, pp. 43484356, 2011.

[76] J. Kennedy and R. Eberhart, Particle swarm optimization, in IEEE International

Conference on Neural Networks: proceedings, vol. 4, pp. 19421948 vol.4, 1995.

[77] J. A. Momoh, Smart grid design for ecient and exible power networks operation

and control, 2009.

[78] Fipa, the foundation of intelligent physical agents. http://www.fipa.org/. [On-

line; accessed 15-April-2013].

[79] Bajwa A., Farooq S., Malik O., Khalique S., Suguri H., Farooq A. H. and Ali A.,

Persistent architecture for context aware lightweight multi-agent system, in Pro-

gramming Multi-Agent Systems (R. Bordini, M. Dastani, J. Dix, and A. Seghrouchni,
eds.), vol. 4411 of Lecture Notes in Computer Science, pp. 5769, Springer Berlin Hei-
delberg, 2007.

[80] L. Cristaldi, A. Monti, R. Ottoboni, and F. Ponci, Multiagent based power systems

monitoring platform: a prototype, in Power Tech Conference Proceedings, 2003

IEEE Bologna, vol. 2, pp. 5 pp. Vol.2, 2003.

[81] T. Nagata, H. Watanabe, M. Ohno, and H. Sasaki, A multi-agent approach to

power system restoration, in Power System Technology, 2000. Proceedings. Power-

Con 2000. International Conference on, vol. 3, pp. 15511556 vol.3, 2000.

[82] A. Dimeas and N. Hatziargyriou, Control agents for real microgrids, in Intelligent
System Applications to Power Systems, 2009. ISAP '09. 15th International Confer-
ence on, pp. 15, 2009.

[83] A. Dimeas and N. Hatziargyriou, Agent based control of virtual power plants, in

Intelligent Systems Applications to Power Systems, 2007. ISAP 2007. International

Conference on, pp. 16, 2007.

173
[84] H. Nunna and S. Doolla, An intelligent energy management framework to incorpo-

rate demand response and storage in microgrids, in India Conference (INDICON),

2012 Annual IEEE, pp. 10631068, 2012.

[85] H. S. V. S. K. Nunna and S. Doolla, Demand response in smart distribution system

with multiple microgrids, Smart Grid, IEEE Transactions on, vol. 3, no. 4, pp. 1641
1649, 2012.

[86] N. Lidula and A. Rajapakse, Microgrids research: A review of experimental micro-

grids and test systems, Renewable and Sustainable Energy Reviews, vol. 15, no. 1,

pp. 186  202, 2011.

[87] E. Peeters, R. Belhomme, C. Batlle, F. Bouard, S. Karkkainen, D. Six, and M. Hom-

melberg, Address: Scenarios and architecture for active demand development in the

smart grids of the future, in Electricity Distribution - Part 1, 2009. CIRED 2009.
20th International Conference and Exhibition on, pp. 14, 2009.

[88] F. Eddy and H. B. Gooi, Multi-agent system for optimization of microgrids, in

Power Electronics and ECCE Asia (ICPE ECCE), 2011 IEEE 8th International
Conference on, pp. 23742381, 2011.

[89] L. Fu-Dong, W. Min, H. Yong and C. Xin, Optimal control in microgrid using

multi-agent reinforcement learning, {ISA} Transactions, vol. 51, no. 6, pp. 743 

751, 2012.

[90] H. S. V. S. K. Nunna and S. Doolla, Demand response in smart microgrids, in

Innovative Smart Grid Technologies - India (ISGT India), 2011 IEEE PES, pp. 131
136.

[91] T. Logenthiran, D. Srinivasan, and A. M. Khambadkone, Multi-agent system for

energy resource scheduling of integrated microgrids in a distributed system, Electric

Power Systems Research, vol. 81, no. 1, pp. 138  148, 2011.

[92] C. Lo Prete, B. F. Hobbs, C. S. Norman, S. Cano-Andrade, A. Fuentes, M. R. von

Spakovsky, and L. Mili, Sustainability and reliability assessment of microgrids in a

regional electricity market, Energy, vol. 41, no. 1, pp. 192202, 2012.

[93] H. Lund, A. N. Andersen, P. A. Ostergaard, B. V. Mathiesen, and D. Connolly, From

electricity smart grids to smart energy systems a market operation based approach

and understanding, Energy, vol. 42, no. 1, pp. 96102, 2012.

174
 [94] A. G. Tsikalakis and N. D. Hatziargyriou, Centralized control for optimizing micro-

grids operation, Energy Conversion, IEEE Transactions on, vol. 23, no. 1, pp. 241
248, 2008.

[95] A. Saleem, K. Heussen, and M. Lind, Agent services for situation aware control

of power systems with distributed generation, in Power Energy Society General

Meeting, 2009. PES '09. IEEE, pp. 18, 2009.

[96] A. Dimeas, S. I. Hatzivasiliadis, and N. Hatziargyriou, Control agents for enabling

customer-driven microgrids, in Power Energy Society General Meeting, 2009. PES

'09. IEEE, pp. 17, 2009.

[97] A. G. Madureira and J. A. Peças Lopes, Ancillary services market framework for

voltage control in distribution networks with microgrids, Electric Power Systems

Research, vol. 86, no. 0, pp. 17, 2012.

[98] M. Mohammadi, S. H. Hosseinian, and G. B. Gharehpetian, Ga-based optimal sizing

of microgrid and dg units under pool and hybrid electricity markets, International
Journal of Electrical Power and Energy Systems, vol. 35, no. 1, pp. 8392, 2012.

[99] E. Mashhour and S. M. Moghaddas Tafreshi, Integration of distributed energy re-

sources into low voltage grid: A market based multiperiod optimization model,

Electric Power Systems Research, vol. 80, no. 4, pp. 473480, 2010.

[100] M. Mohammadi, S. H. Hosseinian, and G. B. Gharehpetian, Optimization of hybrid

solar energy sources/wind turbine systems integrated to utility grids as microgrid

(mg) under pool/bilateral/hybrid electricity market using pso, Solar Energy, vol. 86,
no. 1, pp. 112125, 2012.

[101] P. Siano, C. Cecati, Y. Hao, and J. Kolbusz, Real time operation of smart grids via

fcn networks and optimal power ow, Industrial Informatics, IEEE Transactions
on, vol. 8, no. 4, pp. 944952, 2012.

[102] D. A. Cohen, Gridagentstm: Intelligent agent applications for integration of dis-

tributed energy resources within distribution systems, IEEE Explore, pp. 15, 2008.

[103] J.K. Kok, C.J. Warmer and I.G. Kamphuis, Powermatcher: Multiagent control in

the electricity infrastructure, in International Conference on Autonomous Agents,

(Utrecht, Netherlands), ACM Portal, 2005.

175
[104] H. Kumar Nunna and S. Doolla, Multiagent-based distributed-energy-resource man-

agement for intelligent microgrids, Industrial Electronics, IEEE Transactions on,

vol. 60, no. 4, pp. 16781687, 2013.

[105] N. H. A. Dimeas, A multi-agent system for microgrids, SpringerLink,

vol. 3025/2004, 2004.

[106] X. P. Zhang, A framework for operation and control of smart grids with distributed

generation, in Power and Energy Society General Meeting - Conversion and Delivery
of Electrical Energy in the 21st Century, 2008 IEEE, pp. 15, 2008.

[107] M. J. C. Berende, J. G. Slootweg, J. Kuiper, and J. Peters, Asset management argu-

ments for smart grids, in SmartGrids for Distribution, 2008. IET-CIRED. CIRED
Seminar, pp. 14, 2008.

[108] C. Block, D. Neumann, and C. Weinhardt, A market mechanism for energy allo-

cation in micro-chp grids, in Hawaii International Conference on System Sciences,

Proceedings of the 41st Annual, pp. 172172, 2008.

[109] S. Bathaee and M. H. Abdollahi, Fuzzy-neural controller design for stability enhance-

ment of microgrids, in Universities Power Engineering Conference, 2007. UPEC

2007. 42nd International, pp. 562569, 2007.

[110] W. Friedl, L. Fickert, E. Schmautzer, and C. Obkircher, Safety and reliability

for smart-, micro- and islanded grids, in SmartGrids for Distribution, 2008. IET-
CIRED. CIRED Seminar, pp. 14, 2008.

[111] Z. Jiang, Agent-based control framework for distributed energy resources micro-

grids, in Intelligent Agent Technology, 2006. IAT '06. IEEE/WIC/ACM Interna-

tional Conference on, pp. 646652, 2006.

[112] C. xia Dou, S.-J. Jin, G.-T. Jiang, and Z.-Q. Bo, Multi-agent based control frame-

work for microgrids, in Power and Energy Engineering Conference, 2009. APPEEC
2009. Asia-Pacic, pp. 14, 2009.

[113] A. Dimeas and N. Hatziargyriou, Operation of a multiagent system for microgrid

control, Power Systems, IEEE Transactions on, vol. 20, no. 3, pp. 14471455, 2005.

[114] A. Dimeas and N. Hatziargyriou, A mas architecture for microgrids control, in

Intelligent Systems Application to Power Systems, 2005. Proceedings of the 13th In-
ternational Conference on, pp. 5 pp., 2005.

176
[115] A. Dimeas and N. Hatziargyriou, Agent based control for microgrids, in Power
Engineering Society General Meeting, 2007. IEEE, pp. 15, 2007.

[116] N. Borissov, D. Neumann, and C. Weinhardt, Automated bidding in computa-

tional markets: an application in market-based allocation of computing services,

Autonomous Agents and Multi-Agent Systems, vol. 21, no. 2, pp. 115142, 2010.

[117] G. Venayagamoorthy, Potentials and promises of computational intelligence for

smart grids, in Power Energy Society General Meeting, 2009. PES '09. IEEE, pp. 1
6, 2009.

[118] M. P. F. Hommelberg, C. Warmer, I. G. Kamphuis, J. K. Kok, and G. Schaeer,

Distributed control concepts using multi-agent technology and automatic markets:

An indispensable feature of smart power grids, in Power Engineering Society General

Meeting, 2007. IEEE, pp. 17, 2007.

[119] S. J. Chatzivasiliadis, N. D. Hatziargyriou and A. L. Dimeas, Development of an

agent based intelligent control system for microgrids, 20-24 July 2008 2008.

[120] A. Keles, M. Kolcak, and A. Keles, The adaptive neuro-fuzzy model for forecasting

the domestic debt, Knowledge-Based Systems, vol. 21, no. 8, pp. 951957, 2008.

[121] M.-S. Chen, L.-C. Ying, and M.-C. Pan, Forecasting tourist arrivals by using the

adaptive network-based fuzzy inference system, Expert Systems with Applications,

vol. 37, no. 2, pp. 11851191, 2010.

[122] T. S. Mahmoud, M. H. Marhaban, T. S. Hong, and N. Sokchoo, Ans controller with

fuzzy subtractive clustering method to reduce coupling eects in twin rotor mimo

system (trms) with less memory and time usage, in Advanced Computer Control,
2009. ICACC '09. International Conference on, pp. 1923.

[123] M. Han and Y. Zhao, Dynamic control model of bof steelmaking process based on

ans and robust relevance vector machine, Expert Systems with Applications, vol. 38,
no. 12, pp. 1478614798, 2011.

[124] G. Durgasukumar and M. K. Pathak, Comparison of adaptive neuro-fuzzy-based

space-vector modulation for two-level inverter, International Journal of Electrical

Power and amp; Energy Systems, vol. 38, no. 1, pp. 919, 2012.

177
[125] A. Azadeh, S. M. Asadzadeh, M. Saberi, V. Nadimi, A. Tajvidi, and M. Sheikalishahi,

A neuro-fuzzy-stochastic frontier analysis approach for long-term natural gas con-

sumption forecasting and behavior analysis: The cases of bahrain, saudi arabia, syria,

and uae, Applied Energy, vol. 88, no. 11, pp. 38503859, 2011.

[126] N.-V. Truong, L. Wang, and P. K. C. Wong, Modelling and short-term forecasting

of daily peak power demand in victoria using two-dimensional wavelet based sdp

models, International Journal of Electrical Power and amp; Energy Systems, vol. 30,
no. 9, pp. 511518, 2008.

[127] Vellasco, B. R. Marley Maria, Pacheco, C. Marco Aurélio, Neto, S. R. Luiz, de Souza,

J. Flavio, Electric load forecasting: evaluating the novel hierarchical neuro−fuzzy

bsp model, International Journal of Electrical Power and amp; Energy Systems,
vol. 26, no. 2, pp. 131142, 2004.

[128] L.-C. Ying and M.-C. Pan, Using adaptive network based fuzzy inference system

to forecast regional electricity loads, Energy Conversion and Management, vol. 49,

no. 2, pp. 205  211, 2008.

[129] Z. Souzanchi-k, H. Fanaee-T, M. Yaghoubi, and M.-R. Akbarzadeh-T, A multi adap-

tive neuro fuzzy inference system for short term load forecasting by using previous

day features, in Electronics and Information Engineering (ICEIE), 2010 Interna-

tional Conference On, vol. 2, pp. V254V257, 2010.

[130] B. Bektas Ekici and U. T. Aksoy, Prediction of building energy needs in early stage

of design by using ans, Expert Systems with Applications, vol. 38, no. 5, pp. 5352
5358, 2011.

[131] A. R. Koushki, M. Maralloo, C. Lucas, and A. Kalhor, Application of neuro-fuzzy

models in short term electricity load forecast, in Computer Conference, 2009. CSICC
2009. 14th International CSI, pp. 4146, 2009.

[132] C. Jaipradidtham, Next day load demand forecasting of future in electrical power

generation on distribution networks using adaptive neuro-fuzzy inference, in Power

and Energy Conference, 2006. PECon '06. IEEE International, pp. 6467, 2006.

[133] S. Mohammadi, H. Keivani, M. Bakhshi, A. Mohammadi, M. Askari, and

F. Kavehnia, Demand forecasting using time series modelling and ans estimator,

178
 in Universities Power Engineering Conference, 2006. UPEC '06. Proceedings of the
41st International, vol. 1, pp. 333337, 2006.

[134] K. Y. Tan and Lun-nong, Application of adaptive neuro-fuzzy inference system

based on data eld clustering in load forecasting, 2010.

[135] J. P. S. Catalao, H. M. I. Pousinho, and V. M. F. Mendes, Short-term electricity

prices forecasting in a competitive market by a hybrid intelligent approach, Energy

Conversion and Management, vol. 52, no. 2, pp. 10611065, 2011.

[136] H. E. Alireza Khotanzad, Enwang Zhou, A neuro-fuzzy approach to short-term load

forecasting in a price-sensitive environment, IEEE TRANSACTIONS ON POWER

SYSTEMS, vol. 17, no. 4, p. 10, 2002.

[137] A. H. Mohsenian-Rad and A. Leon-Garcia, Optimal residential load control with

price prediction in real-time electricity pricing environments, Smart Grid, IEEE

Transactions on, vol. 1, no. 2, pp. 120133, 2010.

[138] C. M. Ruibal and M. Mazumdar, Forecasting the mean and the variance of electricity

prices in deregulated markets, Power Systems, IEEE Transactions on, vol. 23, no. 1,
pp. 2532, 2008.

[139] H. M. I. Pousinho, V. M. F. Mendes, and J. P. S. Catalao, Short-term electricity

prices forecasting in a competitive market by a hybrid pso ans approach, Interna-

tional Journal of Electrical Power and amp; Energy Systems, vol. 39, no. 1, pp. 2935,
2012.

[140] F. Pasila, Multivariate inputs for electrical load forecasting on hybrid neuro-fuzzy

and fuzzy c-means forecaster, in Fuzzy Systems, 2008. FUZZ-IEEE 2008. (IEEE
World Congress on Computational Intelligence). IEEE International Conference on,
pp. 23072312, 2008.

[141] M. A. A. Arash Ghanbari, S. Farid Ghaderi, A clustering based genetic fuzzy expert

system for electrical energy demand prediction, 2010.

[142] M. Mohandes, S. Rehman, and S. M. Rahman, Estimation of wind speed prole

using adaptive neuro-fuzzy inference system (ans), Applied Energy, vol. 88, no. 11,
pp. 40244032, 2011.

[143] I. Kandel, & G. Langholz, Fuzzy Control Systems. CRC Press, 1993.

179
[144] C. T. S. J. S. R. Jang and E. Mizutani, Neuro-Fuzzy and Soft Computing: A Compu-
tational Approach to Learning and Machine Intelligence. Matlab Curriculum Series,

Upper Saddle River, NJ: Prentice Hall, 1997.

[145] Y. Wang, X. Ai, and Y. Gao, Microgrid's operation-management containing dis-

tributed generation system, in Electric Utility Deregulation and Restructuring and

Power Technologies (DRPT), 2011 4th International Conference on, pp. 703707.

[146] C. M. Colson, M. H. Nehrir, and S. A. Pourmousavi, Towards real-time microgrid

power management using computational intelligence methods, in Power and Energy

Society General Meeting, 2010 IEEE, pp. 18.

[147] G. Celli, F. Pilo, G. Pisano, and G. G. Soma, Optimal participation of a microgrid

to the energy market with an intelligent ems, in Power Engineering Conference,

2005. IPEC 2005. The 7th International, pp. 663668 Vol. 2.

[148] A. A. Moghaddam, A. Sei, and T. Niknam, Multi-operation management of a typi-

cal micro-grids using particle swarm optimization: A comparative study, Renewable

and Sustainable Energy Reviews, vol. 16, no. 2, pp. 12681281, 2012.

[149] F. A. Mohamed and H. N. Koivo, Power management strategy for solving power

dispatch problems in microgrid for residential applications, in Energy Conference

and Exhibition (EnergyCon), 2010 IEEE International, pp. 746751.

[150] H. Yun and W. Li, Optimization and analysis of distributed energy system with

energy storage device, Energy Procedia, vol. 12, pp. 958965, 2011.

[151] L. Zhuowei, H. Zechun, S. Yonghua, X. Zhiwei, L. Hui, J. Long, and L. Haiyan,

Economic analyses of plug-in electric vehicle battery providing ancillary services,

in Electric Vehicle Conference (IEVC), 2012 IEEE International, pp. 15, 2012.

[152] Y.-H. Chen, S.-Y. Lu, Y.-R. Chang, T.-T. Lee, and M.-C. Hu, Economic analysis

and optimal energy management models for microgrid systems: A case study in

taiwan, Applied Energy, vol. 103, no. 0, pp. 145  154, 2013.

[153] A. Rabiee, H. Khorramdel, and J. Aghaei, A review of energy storage systems in

microgrids with wind turbines, Renewable and Sustainable Energy Reviews, vol. 18,
no. 0, pp. 316  326, 2013.

180
[154] S. Koohi-Kamali, V. Tyagi, N. Rahim, N. Panwar, and H. Mokhlis, Emergence of

energy storage technologies as the solution for reliable operation of smart power sys-

tems: A review, Renewable and Sustainable Energy Reviews, vol. 25, no. 0, pp. 135
 165, 2013.

[155] X. Tan, Q. Li, and H. Wang, Advances and trends of energy storage technology

in microgrid, International Journal of Electrical Power & Energy Systems, vol. 44,

no. 1, pp. 179  191, 2013.

[156] Z. Xu, X. Guan, Q.-S. Jia, J. Wu, D. Wang, and S. Chen, Performance analysis

and comparison on energy storage devices for smart building energy management,

Smart Grid, IEEE Transactions on, vol. 3, no. 4, pp. 21362147, 2012.

[157] Y. Dashora, J. W. Barnes, R. Pillai, T. Combs, and M. Hilliard, Optimized energy

management for large organizations utilizing an on-site phev eet, storage devices

and renewable electricity generation, Energy Systems, pp. 119, 2012.

[158] A. Baziar and A. Kavousi-Fard, Considering uncertainty in the optimal energy

management of renewable micro-grids including storage devices, Renewable Energy,

vol. 59, no. 0, pp. 158  166, 2013.

[159] B. Romer, P. Reichhart, J. Kranz, and A. Picot, The role of smart metering and

decentralized electricity storage for smart grids: The importance of positive exter-

nalities, Energy Policy.

[160] L. Kyebyung, K. M. Son, and J. Gilsoo, Smart storage system for seamless tran-

sition of customers with intermittent renewable energy sources into microgrid, in

Telecommunications Energy Conference, 2009. INTELEC 2009. 31st International,

pp. 15.

[161] M. Savaghebi, J. C. Vasquez, A. Jalilian, J. M. Guerrero, and T.-L. Lee, Selective

compensation of voltage harmonics in grid-connected microgrids, Mathematics and

Computers in Simulation, vol. 91, no. 0, pp. 211  228, 2013. ELECTRIMACS 2011

- {PART} II.

[162] Y. Agarwal, T. Weng, and R. K. Gupta, Understanding the role of buildings in a

smart microgrid, in Design, Automation & Test in Europe Conference & Exhibition
(DATE), 2011, pp. 16.

181
[163] M. Mao, S. Sun, and C. Liuchen, Economic analysis of the microgrid with multi-

energy and electric vehicles, in Power Electronics and ECCE Asia (ICPE and
ECCE), 2011 IEEE 8th International Conference on, pp. 20672072, 2011.

[164] J. K. Kaldellis and D. Zarakis, Optimum energy storage techniques for the improve-

ment of renewable energy sources-based electricity generation economic eciency,

Energy, vol. 32, no. 12, pp. 22952305, 2007.

[165] A. Schroeder, Modeling storage and demand management in power distribution

grids, Applied Energy, vol. 88, no. 12, pp. 47004712, 2011.

[166] Synergy, Gas and electricity suppliers: Energy suppliers. http://www.synergy.

net.au/. [Online; accessed 15-April-2013].

[167] AEMO, Australian energy market operator. http://www.aemo.com.au/. [Online;

accessed 15-April-2013].

[168] B. S. Ahmed, K. Z. Zamli, and C. P. Lim, Application of particle swarm opti-

mization to uniform and variable strength covering array construction, Applied Soft
Computing, vol. 12, no. 4, pp. 13301347, 2012.

[169] K. E. Parsopoulos and M. N. Vrahatis, Parameter selection and adaptation in unied

particle swarm optimization, Mathematical and Computer Modelling, vol. 46, no. 1-
2, pp. 198213, 2007.

[170] D. S. Damianov, J. Oechssler, and J. G. Becker, Uniform vs discriminatory auctions

with variable supply experimental evidence, Games and Economic Behavior, vol. 68,
no. 1, pp. 6076, 2010.

[171] T. H. Noe, M. Rebello, and J. Wang, Learning to bid: The design of auctions

under uncertainty and adaptation, Games and Economic Behavior, vol. 74, no. 2,

pp. 620636, 2012.

[172] Z. Alibhai, W. A. Gruver, D. B. Kotak, and D. Sabaz, Distributed coordination of

micro-grids using bilateral contracts, in Systems, Man and Cybernetics, 2004 IEEE
International Conference on, vol. 2, pp. 19901995 vol.2, 2004.

[173] Independant market operator. http://www.imowa.com.au/. [Online; accessed 15-

April-2013].

182
[174] R. J. Elliott, M. R. Lyle, and H. Miao, A model for energy pricing with stochastic

emission costs, Energy Economics, vol. 32, no. 4, pp. 838847, 2010.

[175] I. Iskin, T. Daim, G. Kayakutlu, and M. Altuntas, Exploring renewable energy

pricing with analytic network process − comparing a developed and a developing

economy, Energy Economics, vol. 34, no. 4, pp. 882891, 2012.

[176] H. Daneshi, A. K. Srivastava, and A. Daneshi, Application of fuzzy logic to priced-

based unit commitment under price uncertainty, in Transmission and Distribution

Conference and Exposition, 2010 IEEE PES, pp. 15.

[177] P. Faria and Z. Vale, Demand response in electrical energy supply: An optimal real

time pricing approach, Energy, vol. 36, no. 8, pp. 53745384, 2011.

[178] P. O. Kriett and M. Salani, Optimal control of a residential microgrid, Energy,

vol. 42, no. 1, pp. 321  330, 2012. 8th World Energy System Conference, {WESC}

2010.

[179] A. Sinha, R. N. Lahiri, S. Neogi, S. Chowdhury, S. P. Chowdhury, and C. T. Gaunt,

Analysis of market price for distributed generators (dgs) in microgrid, in Energy

Market, 2009. EEM 2009. 6th International Conference on the European, pp. 15.

[180] D. R. Thiam, An energy pricing scheme for the diusion of decentralized renew-

able technology investment in developing countries, Energy Policy, vol. 39, no. 7,

pp. 42844297, 2011.

[181] X.-s. Ding, J.-h. Zhang, and X. Chen, A joint pricing and inventory control problem

under an energy buy-back program, Operations Research Letters, pp. 516520, 2012.

[182] M. H. Gomes and J. T. Saraiva, Allocation of reactive power support, active loss

balancing and demand interruption ancillary services in microgrids, Electric Power

Systems Research, vol. 80, no. 10, pp. 1267  1276, 2010.

[183] T. K. Panigrahi, A. K. Basu, S. Chowdhury, S. P. Chowdhury, N. Chakraborty,

A. Sinha, and Y. H. Song, Emerging power market in a microgrid scenario, in

Universities Power Engineering Conference, 2007. UPEC 2007. 42nd International,

pp. 10791085, 2007.

[184] N. M. M. Razali and A. H. Hashim, Microgrid operational decisions based on cfar

with wind power and pool prices uncertainties, in Universities Power Engineering
Conference (UPEC), 2009 Proceedings of the 44th International, pp. 15.

183
[185] M. Panwar, G. P. Duggan, R. T. Grin, S. Suryanarayanan, D. Zimmerle, M. Pool,

and S. Brunner, Dispatch in microgrids: Lessons from the fort collins renewable

and distributed systems integration demonstration project, The Electricity Journal,

vol. 25, no. 8, pp. 71  83, 2012.

[186] C. Correa, R. Bolanos, and A. Garces, Optimal operation of microgrids in the colom-

bian energy market, in Transmission and Distribution: Latin America Conference

and Exposition (T D-LA), 2012 Sixth IEEE/PES, pp. 15, 2012.

[187] I. C. Paschalidis, L. Binbin, and M. C. Caramanis, A market-based mechanism

for providing demand-side regulation service reserves, in Decision and Control and
European Control Conference (CDC-ECC), 2011 50th IEEE Conference on, pp. 21
26.

[188] I. C. Paschalidis, B. Li, and M. C. Caramanis, Demand-side management for reg-

ulation service provisioning through internal pricing, Power Systems, IEEE Trans-
actions on, vol. PP, no. 99, pp. 11, 2012.

[189] A. Sinha, A. K. Basu, R. N. Lahiri, S. Chowdhury, S. P. Chowdhury, and P. A.

Crossley, Setting of market clearing price (mcp) in microgrid power scenario, in

Power and Energy Society General Meeting - Conversion and Delivery of Electrical
Energy in the 21st Century, 2008 IEEE, pp. 18, 2008.

[190] J. B. Cardell and T. Chin Yen, Distributed energy resources in electricity markets:

The price droop mechanism, in Communication, Control, and Computing (Aller-

ton), 2010 48th Annual Allerton Conference on, pp. 5865.

[191] S. Massucco, L. Meini, F. Silvestro, G. Burt, S. Galloway, J. McDonald, and

T. Siewierski, Electricity market strategy characterisation with fuzzy logic: appli-

cation to dierent types of power plants, in Probabilistic Methods Applied to Power

Systems, 2004 International Conference on, pp. 356361.

[192] A. K. Basu, T. K. Panigrahi, S. Chowdhury, S. P. Chowdhury, N. Chakraborty,

A. Sinha, and Y. H. Song, Key energy management issues of setting market clearing

price (mcp) in micro-grid scenario, in Universities Power Engineering Conference,

2007. UPEC 2007. 42nd International, pp. 854860.

[193] M. Mohammadi, S. Hosseinian, and G. Gharehpetian, Optimization of hybrid solar

energy sources/wind turbine systems integrated to utility grids as microgrid (mg)

184
 under pool/bilateral/hybrid electricity market using {PSO}, Solar Energy, vol. 86,

no. 1, pp. 112  125, 2012.

[194] M. Widjaja, L. F. Sugianto, and R. E. Morrison, Fuzzy model of generator bidding

system in competitive electricity markets, in Fuzzy Systems, 2001. The 10th IEEE
International Conference on, vol. 3, pp. 13961399.

[195] H. Allcott, Rethinking real-time electricity pricing, Resource and Energy Eco-
nomics, vol. 33, no. 4, pp. 820842, 2011.

[196] H.-p. Chao, Ecient pricing and investment in electricity markets with intermittent

resources, Energy Policy, vol. 39, no. 7, pp. 39453953, 2011.

[197] C. Brandstatt, G. Brunekreeft, and N. Friedrichsen, Chapter 13 Smart Pricing to

Reduce Network Investment in Smart Distribution Grids Experience in Germany,
pp. 317342. Boston: Academic Press, 2012.

[198] M. Caporin, J. Pres, and H. Torro, Model based monte carlo pricing of energy and

temperature quanto options, Energy Economics, vol. 34, no. 5, pp. 17001712, 2012.

[199] A. Cruz, A. Munoz, J. L. Zamora, and R. Espinola, The eect of wind generation

and weekday on spanish electricity spot price forecasting, Electric Power Systems
Research, vol. 81, no. 10, pp. 19241935, 2011.

[200] S. Chakraborty and M. G. Simoes, Pv-microgrid operational cost minimization by

neural forecasting and heuristic optimization, in Industry Applications Society An-

nual Meeting, 2008. IAS '08. IEEE, pp. 18, 2008.

[201] G. Li and J. Shi, Agent-based modeling for trading wind power with uncertainty in

the day-ahead wholesale electricity markets of single-sided auctions, Applied Energy,

vol. 99, no. 0, pp. 1322, 2012.

[202] S. Anbazhagan and N. Kumarappan, A neural network approach to day-ahead

deregulated electricity market prices classication, Electric Power Systems Research,

vol. 86, no. 0, pp. 140150, 2012.

[203] F. Azevedo, Z. A. Vale, P. B. M. Oliveira, and H. M. Khodr, A long-term risk

management tool for electricity markets using swarm intelligence, Electric Power
Systems Research, vol. 80, no. 4, pp. 380389, 2010.

185
[204] H.-T. Pao, Forecasting electricity market pricing using articial neural networks,

Energy Conversion and Management, vol. 48, no. 3, pp. 907912, 2007.

[205] I. K. Song, S. Y. Yun, S. C. Kwon, and N. H. Kwak, Design of smart distribution

management system for obtaining real-time security analysis and predictive operation

in korea, Smart Grid, IEEE Transactions on, vol. PP, no. 99, pp. 18, 2012.

[206] E. Planas, A. G. de Muro, J. Andreu, I. Kortabarria, and I. M. de Alegria, General

aspects, hierarchical controls and droop methods in microgrids: A review, Renewable

and Sustainable Energy Reviews, vol. 17, no. 0, pp. 147  159, 2013.

[207] R. Zamora and A. K. Srivastava, Controls for microgrids with storage: Review,

challenges, and research needs, Renewable and Sustainable Energy Reviews, vol. 14,
no. 7, pp. 2009  2018, 2010.

[208] J. J. Justo, F. Mwasilu, J. Lee, and J.-W. Jung, Ac-microgrids versus dc-microgrids

with distributed energy resources: A review, Renewable and Sustainable Energy

Reviews, vol. 24, no. 0, pp. 387  405, 2013.

[209] J. Aghaei and M.-I. Alizadeh, Multi-objective self-scheduling of {CHP} (com-

bined heat and power)-based microgrids considering demand response programs and

{ESSs} (energy storage systems), Energy, vol. 55, no. 0, pp. 1044  1054, 2013.

[210] H. Muller, A. Rudolf, and G. Aumayr, Studies of distributed energy supply sys-

tems using an innovative energy management system, in Power Industry Computer

Applications, 2001. PICA 2001. Innovative Computing for Power - Electric Energy
Meets the Market. 22nd IEEE Power Engineering Society International Conference
on, pp. 8790.

[211] H. Kanchev, L. Di, F. Colas, V. Lazarov, and B. Francois, Energy management

and operational planning of a microgrid with a pv-based active generator for smart

grid applications, Industrial Electronics, IEEE Transactions on, vol. 58, no. 10,

pp. 45834592, 2011.

[212] W. Qi, J. Liu, and P. D. Christodes, A distributed control framework for smart grid

development: Energy/water system optimal operation and electric grid integration,

Journal of Process Control, vol. 21, no. 10, pp. 15041516, 2011.

186
[213] Z. Jun, L. Junfeng, W. Jie, and H. W. Ngan, A multi-agent solution to energy man-

agement in hybrid renewable energy generation system, Renewable Energy, vol. 36,
no. 5, pp. 13521363, 2011.

[214] T. Logenthiran, D. Srinivasan, A. M. Khambadkone, and H. N. Aung, Multi-agent

system (mas) for short-term generation scheduling of a microgrid, in Sustainable

Energy Technologies (ICSET), 2010 IEEE International Conference on, pp. 16.

[215] C. M. Colson and M. H. Nehrir, Algorithms for distributed decision-making for

multi-agent microgrid power management, in Power and Energy Society General

Meeting, 2011 IEEE, pp. 18, 2011.

[216] T. Logenthiran, D. Srinivasan, and D. Wong, Multi-agent coordination for der in mi-

crogrid, in Sustainable Energy Technologies, 2008. ICSET 2008. IEEE International

Conference on, pp. 7782, 2008.

[217] T. Logenthiran and D. Srinivasan, Multi-agent system for managing a power dis-

tribution system with plug-in hybrid electrical vehicles in smart grid, in Innovative
Smart Grid Technologies - India (ISGT India), 2011 IEEE PES, pp. 346351.

[218] M. Reeg, W. Hauser, S. Wassermann, T. Kast, U. Klann, K. Nienhaus, U. Pfenning,

and W. Weimer-Jehle, Amiris: An agent-based simulation model for the analysis of

dierent support schemes and their eects on actors involved in the integration of re-

newable energies into energy markets, in Database and Expert Systems Applications
(DEXA), 2012 23rd International Workshop on, pp. 339344.

[219] P. Trigo, P. Marques, and H. Coelho, (virtual) agents for running electricity mar-

kets, Simulation Modelling Practice and Theory, vol. 18, no. 10, pp. 14421452,

2010.

[220] E. Guerci, S. Ivaldi, S. Pastore, and S. Cincotti, Modeling and implementation of

an articial electricity market using agent-based technology, Physica A: Statistical

Mechanics and its Applications, vol. 355, no. 1, pp. 6976, 2005.

[221] G. Li, J. Shi, and X. Qu, Modeling methods for genco bidding strategy optimization

in the liberalized electricity spot market a state of the art review, Energy, vol. 36,

no. 8, pp. 46864700, 2011.

187
[222] V. H. Pakka, B. M. Ardestani, and R. M. Rylatt, Agent-based modelling of the

uk short term electricity market: Eects of intermittent wind power, in European

Energy Market (EEM), 2012 9th International Conference on the, pp. 18.

[223] S. Al-Agtash, Electricity agents in smart grid markets, Computers in Industry,

pp. 235241, 2013.

[224] Y. Sugimoto, T. Miyamoto, S. Kumagai, K. Mori, S. Kitamura, and T. Yamamoto,

Co2 emission basic unit control mechanism in a distributed energy management

system using the market oriented programming, in Sustainable Energy Technologies,

2008. ICSET 2008. IEEE International Conference on, pp. 520525.

[225] A. S. A. W. Martin Kaltschmitt, Wolfgang Streicher, Renewable energy: Technology,

economics and environment, 2007.

[226] M. Kaltschmitt, W. Streicher, and A. Wiese, Renewable energy : technology, eco-

nomics, and environment. Berlin; New York: Springer, 2007.

188