Download as pdf or txt
Download as pdf or txt
You are on page 1of 256

Advanced methods for offshore windfarm planning

Author:
Sun, Yilun
Publication Date:
2021
DOI:
https://doi.org/10.26190/unsworks/22606
License:
https://creativecommons.org/licenses/by-nc-nd/3.0/au/
Link to license to see what you are allowed to do with this resource.

Downloaded from http://hdl.handle.net/1959.4/70956 in https://


unsworks.unsw.edu.au on 2023-10-22
ADVANCED METHODS FOR OFFSHORE WINDFARM
PLANNING

Yilun Sun
B.E. Honors I

A thesis in fulfilment of the requirements for the degree


of Doctor of Philosophy

School of Electrical Engineering and Telecommunications

Faculty of Engineering

February 2021
1. THESIS TITLE & ABSTRACT

2. ORIGINALITY, COPYRIGHT & AUTHENTICITY STATEMENT


3. INCLUSION OF PUBILICATIONS STATEMENT
CONTENTS

CONTENTS ........................................................................................................................... I

ACKNOWLEDGEMENT .................................................................................................. VI

ABSTRACT ...................................................................................................................... VIII

ABBREVIATIONS .............................................................................................................. X

NOMENCLATURE .......................................................................................................... XII

LIST OF FIGURES ...................................................................................................... XXIII

LIST OF TABLES......................................................................................................... XXVI

CHAPTER 1 INTRODUCTION......................................................................................... 1

1.1 PROBLEM STATEMENT AND RESEARCH OBJECTIVES .......................... 1

1.1.1 Conventional offshore wind farm planning ................................................. 2

1.1.2 The impact of unstable wind power on the main grid ................................. 4

1.1.3 Research objective ....................................................................................... 5

1.2 RESEARCH CONTRIBUTIONS ................................................................................ 6

1.3 THESIS OUTLINE .................................................................................................. 8

CHAPTER 2 STATE OF THE ART OF OFFSHORE WIND POWER PLANNING . 10

2.1 OVERVIEW OF CURRENTLY OPERATING OFFSHORE WIND FARMS ...................... 10

2.1.1 Background of offshore wind farms .......................................................... 10

2.1.2 Electricity system configuration ................................................................ 17

2.2 COLLECTOR SYSTEM PLANNING OF OFFSHORE WIND FARMS ............................ 20

2.2.1 Collector system topology ......................................................................... 20

I
2.2.2 Submarine inter-array cable ....................................................................... 24

2.2.3 Modelling choices ...................................................................................... 27

2.2.4 Solution method ......................................................................................... 32

2.3 OFFSHORE SUBSTATION OPTIONS FOR OFFSHORE WIND FARMS ......................... 36

2.3.1 Configuration ............................................................................................. 36

2.3.2 Power transformer ...................................................................................... 38

2.3.3 Power converter ......................................................................................... 39

2.3.4. Single substation V.S. multiple substation ................................................ 40

2.4 TRANSMISSION AND INTEGRATION PLANNING OF OFFSHORE WIND FARMS ....... 41

2.4.1 Overview of transmission technique .......................................................... 41

2.4.2 HVAC submarine cable .............................................................................. 42

2.4.3 HVDC submarine cable ............................................................................. 42

2.4.4 Multi-terminal SVC network ..................................................................... 44

2.4.5 Deterministic V.S. Probabilistic planning .................................................. 44

2.4.6 The impacts of integrating OWFs on the regional transmission network .. 45

2.5 CONCLUSION ..................................................................................................... 47

CHAPTER 3 AN ELECTRICAL SYSTEM LAYOUT OPTIMIZATION

FRAMEWORK FOR AN OFFSHORE WIND FARM ............................................ 49

3.1 PROBLEM DESCRIPTION ..................................................................................... 49

3.2 MATHEMATICAL MODELLING ............................................................................ 53

3.2.1 Hypothesis.................................................................................................. 54

3.2.2 Inputs.......................................................................................................... 56

3.2.3 Problem formulation .................................................................................. 57

3.3 ELECTRIC SYSTEM LAYOUT OPTIMIZATION FRAMEWORK ................................. 64

3.3.1 Data input and pre-processing.................................................................... 64

3.3.2 Discretization of Candidate Centre Location ............................................. 66


II
3.3.3 Clustering wind turbines into feeders ........................................................ 70

3.3.4 Avoiding cable crossing ............................................................................. 73

3.3.5 Natural Aggregation Algorithm ................................................................. 75

3.3.6 Flowchart for the layout optimization framework ..................................... 77

3.4 CASE STUDIES ................................................................................................... 80

3.4.1 Case description ......................................................................................... 80

3.4.2 Input data and pre-processing for the optimization ................................... 82

3.4.3 Optimization result and analysis for layout 1 ............................................ 86

3.4.4 Optimization result and analysis for layout 2 ............................................ 93

3.5 CONCLUSION ................................................................................................... 105

CHAPTER 4 HIERARCHICAL OPTIMIZATION FOR THE COLLECTOR

SYSTEM PLANNING OF LARGE OFFSHORE WIND FARMS WITH

MULTIPLE SUBSTATIONS.................................................................................... 107

4.1 PROBLEM DESCRIPTION .................................................................................. 107

4.2 PROBLEM FORMULATION .................................................................................112

4.2.1 Hypothesis ................................................................................................113

4.2.2 Mathematical model .................................................................................115

4.3 HIERARCHICAL LAYOUT OPTIMIZATION FRAMEWORK .................................... 121

4.3.1 Offshore substation optimization layer .................................................... 121

4.3.2 Wind turbine allocation............................................................................ 126

4.3.3 Transmission system optimization ........................................................... 127

4.3.4 Inner-cluster layout optimization ............................................................. 130

4.3.5 Cost definitions between wind turbines and cluster centres .................... 134

4.3.6 Flow chart for the hierarchical optimization framework ......................... 136

4.4 CASE STUDIES ................................................................................................. 137

4.4.1 Case 1: moderate-scale OWF .................................................................. 138


III
4.4.2 Case 2: large-scale OWF .......................................................................... 151

4.5 CONCLUSION ................................................................................................... 167

CHAPTER 5 COMPREHENSIVE INTEGRATION PLANNING MODEL FOR

MULTIPLE REMOTE WIND FARMS................................................................... 169

5.1 PROBLEM DESCRIPTION ................................................................................... 169

5.2 COST MODEL DESCRIPTION............................................................................. 174

5.2.1 Hypothesis................................................................................................ 174

5.2.2 Inputs........................................................................................................ 176

5.2.3 Problem Formulation ............................................................................... 176

5.3 INTEGRATED TRANSMISSION PLANNING FOR MULTIPLE LARGE-SCALE REMOTE

WIND FARMS ................................................................................................................... 179

5.3.1 Stochasticity ............................................................................................. 179

5.3.2 Voltage Stability ....................................................................................... 181

5.3.3 Multi-Objective and Normalization Process ............................................ 183

5.3.4 MTDC Power Flow Model ...................................................................... 185

5.3.5 Simplified mathematical model ............................................................... 187

5.3.6 Implementation of the Model ................................................................... 189

5.3.7 Flow Chart for the Integrated Transmission Planning Model .................. 190

5.4 CASE STUDIES ................................................................................................. 191

5.4.1 Case description ....................................................................................... 191

5.4.2 Case A: deterministic optimization without uncertainty .......................... 196

5.4.3 Case B: stochastic optimization with multiple objectives ........................ 199

5.4.4 Case C: stochastic optimization with single objective ............................. 201

5.4.5 Analysis .................................................................................................... 202

5.5 CONCLUSION ................................................................................................... 207

IV
CHAPTER 6 CONCLUSIONS AND FUTURE WORK .............................................. 208

BIBLIOGRAPHY ............................................................................................................. 212

V
ACKNOWLEDGEMENT

Thanks to the guidance, and supervision from my supervisors, I would not be able to

keep on track and finalize my research study without their constant source of academic

support and encouragement.

My initial thanks and appreciation go to my primary supervisor, Professor Zhao Yang

Dong. I am grateful indeed for all the support and patience from him. Whenever I feel puzzled

about my research direction, he can always offer me several constructive suggestions and

point out to me how to get onto the right path. It is helpful indeed for me to speed up and get

back to my original direction. He seems always dynamic and energetic to promptly response

whenever I have a request. His rigorous scientific attitude, open-mindedness, and humble

character also inspires me in my future work.

I would also express my sincere gratitude to my secondary supervisors, Dr. Ke Meng

and Dr. Ziyuan Tong, both from University of New South Wales. Their solid research

knowledge always provides me with more inspiration in my own research work. Dr. Ke Meng

has ever paved the way for me to decide my research topic. Thanks Dr. Ke Meng for his

patient and elaborate revision to my paper. Dr. Ziyuan Tong has also cared about both my

study and life, and she has offered me her invaluable patience and helped me to manage the

past stressful research years.

My thanks and gratitude is extended to some of my friends, especially Ms. Danlin Li,

Ms. Jinzhu Yang, Ms. Jingjie Huang, Ms. Kexin Wang and many other friends for their

dependable companion, generous patience over the past few years. I would be grateful indeed

to have all of them in my life.

Particularly, I am grateful to Australian government, The University of Sydney and The

University of New South Wales. The Commonwealth of Australia has provided financial

support to my Ph.D. study with Australian Government Research Training Program

VI
Scholarship and Australian Postgraduate Award.

Most importantly, I have to express my heartfelt to my family, my parents Mr. Xiuzong

Sun and Ms. Xiuhua Song, who have always had faith in me during the ups and downs in my

life and provided me with a solid backing for me to pursuit my dream. Love you all forever.

VII
ABSTRACT

There have been increasing interests and projects of Offshore wind farm (OWF)

development across the world given the rich wind resources in order to achieve carbon neutral

objectives. Appropriate electrical system design of OWF is of key importance in terms of cost

saving and improving system efficiency. Two novel electric system layout optimization

models for OWF planning are proposed to optimize the topology of collector system and

connected transmission system simultaneously in OWFs with single and multiple substations.

For OWF with single-substation, a novel mathematical model to represent the system

topology is proposed to reduce the number of variables so as to effectively decrease the search

space of the optimisation problem, where the continuous substation sitting problem is

discretized by a 2-step rasterization method. For large-scale OWFs, the overall electric system

optimization problem has been classified into 3 levels: substation optimization, feeder

selection, and cable determination. Fuzzy clustering technique and wind turbine allocation

method has been proposed to effectively divide the large offshore windfarms into partitions.

Both HVDC and HVAC cables are considered as alternatives used in the associated

transmission system, which can be optimized at the substation level. The concept of clustering

is further applied in feeder level to cluster wind turbines into appropriate feeders. The

proposed model and the optimization algorithms are tested and validated in two large-scale

offshore winds.

A comprehensive decision support model is proposed which covers three key factors

that characterize OWF integration: investment cost, system stability and the interactions

between MTDC and local AC system, all of which are concerned to characterize the optimal

integration location of wind turbines into AC bus location and appropriate converter size

installed at the corresponding MTDC terminals. To better fit into the real-world situation,

various wind speed and load scenarios have been considered. Validity and effectiveness of the

VIII
proposed model has been tested to integrate two wind farms to a benchmark AC system via a

MTDC grid.

The research methodologies presented in the thesis form a rather comprehensive

approach for OWF design and planning. With the rapid development in OWF technologies,

future research needs are also identified and presented in the thesis.

IX
ABBREVIATIONS

AC Alternating Current

BIP Binary integer problem

DC Direct Current

DFIG Doubly-fed induction generator

EEZ Exclusive Economic Zone

GA Genetic Algorithm

GIS Gas Insulated Switchgear

HV High voltage

HVAC High voltage alternating current

HVDC High voltage direct current

LCC Line-commutated converter

LM Loadability margin

LP Linear problem

MILP Mixed-integer linear problem

MINLP Mixed-integer nonlinear problem

MIP Mixed integer programming

MIQP Mixed integer quadratic programming

MV Medium voltage

MST Minimum spanning tree

MTDC Multi-terminal VSC-HVDC network

MVAC Medium-voltage alternating current

NAA Nature Aggregation Algorithm

OWF Offshore wind farm

PCC Point of common coupling

X
PMSG Permanent magnet synchronous generator

PSO Particle swarm optimization

RIBBP Reverse Interlocked Busbar Protection

RTN Regional transmission network

SF6 Sulphur Hexafluoride

STATCOM Static synchronous compensator

MIP Mixed integer programming

VSC Voltage source converter

SO System operator

TSO Transmission system operator

WT Wind turbine

XLPE Crossed-linked polyethylene

XI
NOMENCLATURE

Sets

N wt Set of wind turbines

N cp Set of collector point

N pcc Set of PCC

I Set of nodes/buses in the system

PWT Set of cartesian coordinate of all the wind turbines

F Set of feeders

S Set of substations

Set of available MV and HV submarine cables (used


K ,H
in Ch 3)

C Set of available cable types (used in Ch 4)

C ac / C dc set of available AC/DC cable types

WF Set of Windfarm-side terminals in MTDC grid

M Set of grid-side terminals in MTDC grid

B Set of MTDC grid buses

N Set of AC buses

 Set of all the buses in the system

Indices

ij Wind turbines (used in Ch 3)

ia ja Collector point

i* j * PCC

i, j Bus/node

XII
p Wind turbines (used in Ch 4)

f Feeder

k, h MVAC/HVAC cable type

c Cablet type (used in Ch 4)

ac / dc AC/DC system

s Cable section on feeder

ws Wind speed scenario

ls Power demand scenario

ss System state

Parameters

N Number of WTs

y Lifetime span of OWF (years)

r Interest rate

Closs Cost of unserved energy ($/MWh)

POS Permission region for offshore substation

2-dimension cartesian coordinate/x coordinate/y


PCC / PCC1 / PCC2
coordinate of PCC

pij / pi' j' Cartesian coordinate of wind turbine ij ( i' j' )

pi* j* Cartesian coordinate of PCC

PC Penalty multiplier for crossed cable

nW Number of offshore wind farms

nm Number of grid-side terminals in MTDC grid

nb Number of ac buses

ng Number of generators in ac system

XIII
T Hours per year

Dij ,i' j' Distance between two wind turbines ij and i' j'

Distance between WT ij and collector platform


Dij ,ia ja
ia ja

Distance between collector platform ia ja and PCC


Di ,i* j*
a ja

i* j *

Length of cable section s (or between node i and


d s / dij
j)

D Distance matrix between busses

Rated power, cost, resistance, reactance, susceptance,


Sk , Rk , X k , Bk ,Vk
rated voltage, respectively of a specific cable type k

Rated power, cost, resistance, reactance, susceptance,

Sh , Rh , X h , Bh ,Vh rated voltage, respectively of a specific HVAC cable

type h

Unit cost, rated resistance, rated capacity, rated


UMCc , MRc , MCc , MVc
voltage of MV cable type c

Unit cost, rated resistance, rated capacity, rated


UHCc , HRc , HCc , HVc
voltage of HV cable type c

CC Investment cost of HVDC cable (M$/km)

Investment cost of unit capacity of converter


CW
(M$/MW)

Sc Rated power of cable type c

max
I MTDC Maximum current allowed for HVDC cable

XIV
SWT Rated capacity of wind turbine

Maximum number of feeders can be installed on each


N max
f
platform

Maximum number of wind turbines can be connected


max
NWT ,f
to one feeder

N smax Maximum number of wind turbines in a substation

CSI Shipping and installation cost for submarine cables

 ,  , Cost parameters for MV submarine cable

kc , k s Maintenance cost factor for cables/substation

 Estimated efficiency of transformer (or converter)

UCMVSG / UCHVSG Unit cost of MV switchgears/ HV switchgears

Regulation index for collector cost, transmission cost,


c , t , s
substation cost

m Exponential index for fuzzy clustering

A ratio between transmission cable unit cost and its


coef
unit resistance

M A large auxiliary number

ai ,bi , ci Cost coefficients of generator on bus i

Punit Unit capacity of converter (MW)

Minimum or maximum allowed voltage in MTDC


U ,U
grid (p.u.)

Minimum or maximum allowed voltage in AC grid


v ,v
(p.u.)

Minimum or maximum real power generation of


PGi ,PGi
generator i (MW)

XV
Minimum or maximum reactive power generation of
QGi ,QGi
generator i (MW)

Minimum or maximum apparent power flow between


Fij ,Fij
bus i and bus j in ac system (MVA)

Droop gain of converter installed at terminal i in


Ki
MTDC grid

Decision variables

Installation or not of a collector platform at ia ja . If

bia ja a collector platform is set up at ia ja , bia ja = 1 ;

otherwise, bia ja = 0 .

Determines if a WT ij (WT i ) is located on a

feeder f . If the WT ij (WT i ) is clustered to


xij , f / xi , f
feeder f , xij , f (or xi , f ) = 1 ; otherwise,

xij , f (or xi , f ) = 0 .

Determines if a pair of WTs ij and i' j' are

uij ,i' j ' , f ,k interconnected by means of cable type k on feeder

f .

Determines if ij is directly connected to offshore


uij ,ia ja ,k
substation ia ja via a cable type k

A binary variable to determine whether wind turbine

u psfc p is located in section s of feeder f via cable

type c

XVI
Determines if ia ja is connected to onshore PCC
ui * *
a ja ,i j ,h

i* j * via a cable type h

Determines if a feeder f has been chosen for the


ia ja i
z f /z f
substation at ia ja (or substation i )

Number of MV feeders installed on collector


Nf
platform at ia ja

Number of MV (HV) cables installed on collector


N MV / N HV
platform at ia ja

Number of transmission cables (HVAC or HVDC)


AC DC
nHV ,i / nHV ,i / nHV ,i
required for substation i

Number of wind turbines can be connected to feeder


nWT , f
f

Number of clusters (collector systems) for the

ns / nsac / nsdc general system/ for AC alternative/ for DC

alternative

nWT ,s Number of wind turbines in cluster (substation) s

An integer variable to represent number of

nc ,S transmission cable of type c installed at substation

CS Rated capacity of cluster (substation) s

n f ,s Number of feeders in cluster (substation) s

A binary variable to determine whether HVAC or


iAC ,iDC
HVDC is chosen for substation i

XVII
A binary variable to determine whether wind turbine
z ij
j is clustered to substation i

A decision variable to determine whether a cable


 ij ,c
type c is to connect bus i and bus j

AC bus no. integrated with grid-side terminal i in


xi
MTDC grid

No. of unit capacity of converter installed at terminal


yi
i in MTDC grid

Variables

Cinvestment Investment cost

Cop Operational cost

CCollect Investment cost of collector system

Cop _ MV Operational cost for losses in MV submarine cables

Cop _ HV Operational cost for losses in HV submarine cables

CCollector Total cost of multiple collector systems

Investment/ maintenance / operational cost for


Investment Ma int ain Operation
CCollector / CCollector / CCollector
collector system i
i i i

COS Investment cost of substation

CSUB Total cost of multiple substations

CSUBi Total cost of substation i

Investment/ maintenance / operational cost for


Investment Ma int ain Operation
CSUB / CSUB / CSUB
substation i
i i i

CTrans Cost of transmission system

XVIII
CTransi Total cost of transmission system of substation i

Investment/ maintenance / operational cost for


Investment Ma int ain Operation
CTrans / CTrans / CTrans
transmission system of substation i
i i i

CSGMV / CSGHV Cost of MV switchgears/ HV switchgears

Total switchgear cost/ MV switchgear cost/ HV


CSG ,i / CMVSG ,i / CHVSG ,i
switchgear cost for substation i

Operational cost of the whole electric system/


Cop / Closs _ MV / Closs _ HV
collector system/ transmission system

C1 Investment cost (M$)

C2 Operation cost of MTDC grid (M$)

C3 Operation cost of AC grid (M$)

F1 Total annualized cost (M$)

F2 Overall stability of the system

di Distance between substation i and onshore PCC

Current flow of a MV inter-array feeder f / cable


I sf / I fc
section c on feeder f

Current flow of a HV submarine transmission cable


Is
for substation s

Apparent power flow on section between WT ij and


Sij ,i' j' / Sij ,ia ja
WT i' j' / substation ia ja

Apparent power flow in cable section s of feeder


Ssf
f

Sia ja ,i* j* Apparent power flow between substation ia ja and

XIX
PCC i* j *
Si Apparent power flow from substation i

Cartesian coordinate of substation ia ja (or


pia ja / PCi
substation i )

 ij Membership degree of WT j to substation i

CCollectorij Collector system cost for WT j to substation i

CTransij Transmission system cost for WT j to substation i

CSUBij Substation cost for WT j to substation i

p ss Probability of system scenario ss

p ws Probability of wind speed scenario ws

pls Probability of power demand scenario ls

MTDC
Ploss Losses in MTDC grid (MW)

Real power generation of wind farm i under wind


ws ss
PWFi / PWFi speed scenario ws or system state scenario ss

(MW)

Real power from MTDC grid-side terminal i into


ws ss
PMTDCi / PMTDCi ac grid under wind speed scenario ws or system

state scenario ss (MW)

Reference injected power at grid-side terminal i of


0
P MTDCi
MTDC grid

Real power output of generator installed at bus i


ws ,ls ss
P G ,i /P G ,i
under system scenario of ws and ls or system

XX
state scenario ss (MW)

Reactive power output of generator installed at bus

QGws,i,ls / QGss,i i under system scenario of ws and ls or system

state scenario ss (MW)

Real/ reactive power under power demand state ls


ls ls
P /Q
D ,i D ,i
at bus i

Real terms of element (i , j ) in bus admittance


Gij
matrix

Imaginary terms of element (i , j ) in bus admittance


Bij
matrix

Ui Bus voltage of terminal i in MTDC grid (p.u.)

Reference terminal voltage at terminal i in MTDC


0
U i
grid (p.u.)

Current flow between terminal i and j in MTDC


I MTDC ,ij
grid (p.u.)

Vi Voltage magnitude of bus i in AC system (p.u.)

i Voltage angle of bus i in AC system (p.u.)

zl Line impedance (p.u.)

 ij Line impedance angle between bus i and j (p.u.)

Apparent power flow in branch between bus i and


Fij
j (MVA)

Pl / Ql Real power and reactive power flow on branch l

Pl ws ,ls ws ,ls
max / Ql max Maximum real power and reactive power transfer

XXI
capability of branch l under wind speed ws and

power demand ls

VCPI index under wind speed ws and power


VCPI lws ,ls
demand ls

XXII
LIST OF FIGURES

Figure 1.1 Annual installation for onshore and offshore wind from 2001 to 2018................. 2

Figure 2.1 Cumulative installed capacity of wind farms ...................................................... 11

Figure 2.2 Annual installed capacity of offshore wind farms ............................................... 11

Figure 2.3 Schematic diagram of a collector system ............................................................ 21

Figure 2.4 Standard configuration for offshore wind farm ................................................... 24

Figure 2.5 Equivalent model of AC cables ........................................................................... 27

Figure 2.6 Equivalent circuit of a power transformer ........................................................... 39

Figure 3.1 Schematic diagram of an OWF electric system................................................... 50

Figure 3.2 Schematic diagram for layout 1 ........................................................................... 52

Figure 3.3 Schematic diagram for layout 2 ........................................................................... 53

Figure 3.4 Connection configuration for network nodes ...................................................... 54

Figure 3.5 Raster map of a permissible region ..................................................................... 68

Figure 3.6 Illustrative example for coarse-fine rasterization for a permissible region ......... 70

Figure 3.7 Illustrative display of cable crossing determinant ............................................... 74

Figure 3.8 Flow chart of the layout optimization framework ............................................... 79

Figure 3.9 Wind turbine sitting in the OWF ......................................................................... 82

Figure 3.10 Evolution of the best individual in each generation (without losses) ................ 88

Figure 3.11 Electric system topology of wind farm collector system (without considering

losses) ........................................................................................................................... 88

Figure 3.12 Evolution of the best individual in each generation (with losses) ..................... 91

Figure 3.13 Electric system topology of wind farm collector system (with losses) ............. 91

Figure 3.14 Evolution of the best individual in each generation for coarse tuning in layout

2(without losses) .......................................................................................................... 95

Figure 3.15 Electric system topology of an OWF under coarse tuning (without losses) ...... 96

XXIII
Figure 3.16 Electric system topology of an OWF under fine tuning (without losses) .......... 98

Figure 3.17 Electric system topology of an OWF for case 4 (coarse raster map) ............... 100

Figure 3.18 Electric system topology of an OWF for case 4 (fine raster map) ................... 101

Figure 3.19 Evolution of the best individual in layout 2(with losses) ................................. 101

Figure 4.1 General schematic diagram for a large-scale OWF ........................................... 109

Figure 4.2 Illustrative diagram for multi-layer in offshore substation ................................ 110

Figure 4.3 Inter-array feeder layer encoding ....................................................................... 133

Figure 4.4 Illustrative example for collector cost between Node j and its own Substation 135

Figure 4.5 Illustrative example for collector cost between Node j and other Substations .. 135

Figure 4.6 Flow chart of the Hierarchical optimization framework for OWF layout design

.................................................................................................................................... 137

Figure 4.7 Wind turbine sitting in the OWF “Banc de Guèrande”...................................... 139

Figure 4.8 Initial Wind farm partition result ....................................................................... 143

Figure 4.9 Optimal connection topology of the wind farm network. .................................. 146

Figure 4.10 Turbine layouts of the 150-WT wind farm ...................................................... 152

Figure 4.11 Initial Wind farm partition result for HVAC transmission .............................. 155

Figure 4.12 Initial Wind farm partition result (HVDC) ...................................................... 156

Figure 4.13 Optimal connection topology of the wind farm network (HVAC) .................. 159

Figure 4.14 Optimal connection topology of the wind farm network (HVDC) .................. 159

Figure 5.1 Illustrative transmission options for Offshore windfarms (a) Multiple HVAC

transmission (b) Multiple HVDC transmission (c) A simplified MTDC network ..... 170

Figure 5.2 Illustrative example of integrating OWFs into an onshore utility grid .............. 171

Figure 5.3 A simplified MTDC diagram ............................................................................. 178

Figure 5.4 Two-stage stochastic model of wind speed and power demand ........................ 181

Figure 5.5 PV curve of an arbitrary bus .............................................................................. 182

Figure 5.6 Encoding format of chromosome ....................................................................... 189

XXIV
Figure 5.7 Flowchart of methodology ................................................................................ 191

Figure 5.8 14-bus RTN ....................................................................................................... 193

Figure 5.9 5-terminal MTDC grid topology ....................................................................... 193

Figure 5.10 Power output characteristic curve of wind turbine .......................................... 194

Figure 5.11 Hourly wind speed of a year ............................................................................ 194

Figure 5.12 Discretization of the probability density function of wind speed .................... 195

Figure 5.13 Hourly load percentage to the peak load of one year ...................................... 195

Figure 5.14 Pareto front of Case A: Deterministic Optimization ....................................... 198

Figure 5.15 Optimal topology of Case A: Deterministic Optimization .............................. 198

Figure 5.16 Pareto front of Case B: Stochastic Optimization ............................................. 200

Figure 5.17 Optimal topology of Case B: Stochastic Optimization.................................... 200

Figure 5.18 Optimal topology of Case C: Stochastic optimization with single objective .. 202

Figure 5.19 Generation versus demand curve for a case with different wind speed level and

high demand (a) low wind speed (b) moderate wind speed (c) high wind speed

(d)extremely high wind speed .................................................................................... 204

XXV
LIST OF TABLES

Table 2-1 Available wind turbines and specifications .......................................................... 13

Table 2-2 Cost coefficients for MV submarine cables .......................................................... 26

Table 2-3 Installation and transportation cost of submarine cables [k € /km] ..................... 26

Table 2-4 Cost coefficient for power transformer ................................................................. 38

Table 2-5 Cost coefficients for HVAC submarine cables ..................................................... 42

Table 2-6 Cost coefficients for HVDC cables....................................................................... 43

Table 3-1 Possible voltage level for collector and transmission network ............................. 57

Table 3-2 Cost parameter for MV cables .............................................................................. 58

Table 3-3 Cost parameter for HV cables ............................................................................... 60

Table 3-4 Data input for the optimization framework........................................................... 65

Table 3-5 Data preprocessing for the optimization framework ............................................. 66

Table 3-6 MV cable parameters and unit cost ....................................................................... 83

Table 3-7 HVAC cable parameters and unit cost .................................................................. 85

Table 3-8 Optimization result for layout 1 (without the consideration of losses) ................. 89

Table 3-9 Optimization result for layout 1 (with losses) ....................................................... 92

Table 3-10 Optimization result for layout 2 (raster cell 1*1/without losses) ........................ 99

Table 3-11 Optimization result for layout 2 (raster cell 1*1/with losses) ........................... 103

Table 4-1 Comparison of proposed framework with other techniques ............................... 112

Table 4-2 Parameters of necessary input of the OWF ......................................................... 140

Table 4-3 Cable parameters of the utilised three-core copper conductor medium voltage

submarine cables......................................................................................................... 141

Table 4-4 Parameter values worked out from input ............................................................ 144

Table 4-5 Optimal cluster result .......................................................................................... 146

Table 4-6 MV submarine cable cost breakdown ................................................................. 148

XXVI
Table 4-7 Substation cost breakdown ................................................................................. 149

Table 4-8 Transmission cost breakdown ............................................................................ 150

Table 4-9 Parameters of necessary input of the OWF ........................................................ 153

Table 4-10 Cable parameters of submarine HVAC cables ................................................. 153

Table 4-11 Cable parameters of submarine HVDC cables ................................................. 154

Table 4-12 Parameter values worked out from input .......................................................... 157

Table 4-13 Cost and volume of overall collector system .................................................... 161

Table 4-14 MV submarine cable cost breakdown .............................................................. 162

Table 4-15 Substation cost breakdown ............................................................................... 163

Table 4-16 Transmission cost breakdown .......................................................................... 165

Table 5-1 Pareto optimal solution of deterministic optimization (Case A) ........................ 197

Table 5-2 Optimal decision variables for the best compromise solution (Case A)............. 198

Table 5-3 Pareto optimal solution of stochastic optimization (Case B).............................. 199

Table 5-4 Optimal decision variables for the best compromise solution (Case B) ............. 199

Table 5-5 Optimal solution for Case C ............................................................................... 201

Table 5-6 VCPI index for Case A, Case B and Case C ...................................................... 202

XXVII
Chapter 1 INTRODUCTION
1.1 PROBLEM STATEMENT AND RESEARCH OBJECTIVES

With growing populations and rapid economic growth, the continuously increasing

power demand has already exerted pressure on the existing power system to evolve to satisfy

the basic needs and to attain improved standards of human welfare. However, to tackle the

global warming, ambitious carbon reduction targets have recently been proposed, which has

started to revolutionize the power system to transit from conventional nonenvironmental

friendly energy source to renewables. Literally, considerable thermal power plants will retire

and be replaced by renewable power resources in the coming years. Despite environmental

advantages of such resources, incorporation of renewable energy also poses certain technical,

economic and reliability issues that can challenge the conventional operational and planning

procedures of power systems. How to utilize renewables to satisfy society’s energy needs is a

worldwide issue to be dealt with. Wind energy, one of the most attractive renewables, have

been widely adopted to replace traditional fuel generators and meet a remarkable percentage

of global power demand. In the last decade, the global cumulative installed wind power

capacity has dramatically increased and reached up to 650GW in 2019 [1]. Generally,

windfarms are classified into onshore and offshore windfarms (OWF) according to their

locations. It is presented the annual installation of onshore wind power and offshore wind

power from 2021 till now in Fig.1.1, which clearly implies that the installation of offshore

wind power is far less than that of onshore wind power. The less installed capacity cannot

imply that onshore wind is superior to offshore wind. In fact, a better wind source with a

higher and regular wind speed is available for offshore wind farm, let alone its advantages of

no visual and noise pollution. It is the higher capital cost of offshore wind infrastructure that

still impedes a burden for potential owners to exploit the abundant wind resources of high
1
quality off the coast.

Figure 1.1 Annual installation for onshore and offshore wind from 2001 to 2018
This chapter identifies the problems of existing offshore wind farm planning from its

collector system planning to transmission and integration planning, and also briefly states the

impact on existing utility grid of incorporating large amount of wind power from offshore

wind farm, raises the inadequacies and limitations in the traditional wind farm planning

strategy, followed by the final sections where the objectives and contributions of this research

are presented.

1.1.1 Conventional offshore wind farm planning

The majority of currently operating offshore wind farms adopt a symmetric layout

design and follow one standard configuration type such as simply radial or simply ring

configuration. However, given the fact that each standard configuration has its own merits

and demerits, the fully optimized topology of an offshore wind farm is usually a combination

of several standard configurations and violating the symmetric connection layout design. In

addition, the number of wind turbines in an offshore wind farm is relatively small. The

common practice is to distribute the limited wind turbines in several rows and artificially

connect wind turbines in the same row by one array cable of same type. Despite its ease of
2
use by following the rule of thumb, the wind farm planning result obviously cannot guarantee

its strength in terms of either economic or reliability. To further exploit the high-quality wind

resources off the coast, an increased number of wind turbines will be expected in offshore

wind farms. Besides, the nominal capacity of wind turbine has been increasing rapidly.

Correspondingly, the offshore wind farm could reach giant size, which poses challenges on

not only the offshore installation for turbines but also the electric system design with a larger

wind farm size. Accordingly, an optimum electric system matters more as the offshore wind

farm size increases. A dedicated electric system optimization is of greater necessity for an

offshore wind farm to achieve an efficient, economical, and reliable performance, considering

the high expense on offshore infrastructure. Investment cost of the components are widely

regarded as the only metric to evaluate a wind farm design, which is not sufficient considering

the long operation lifetime of an offshore wind farm. Inside the offshore wind farm, MV inter-

array cables with one or two cross-sectional areas are considered, which has the benefits of

simplicity to implement, however, leads to avoidable savings on cables.

As for transmission system of offshore wind farm, some near-coast offshore wind farms

are connected to onshore point of common coupling via medium voltage AC cables. Most

offshore wind farms are connected to shore via high voltage AC cables. Recently, the evolving

floating wind turbines allow developers to access previously unreachable waters (also further

from coast) to capture better-quality wind resources, which, accordingly, should revolutionize

the transmission of bulky wind power to shore. With remarkable development of wind energy

technology, an increase in both wind turbine capacity and number of wind turbines in wind

farm will revolutionize the most widely-used transmission option, HVAC for offshore wind

farms, given its inadequate capacity due to its capacity charging current. HVDC transmission

technology can be considered as an alternative of HVAC to be responsible for the bulk-power

long-distance transmission in offshore wind farm planning. Regardless of used transmission

options, offshore wind farms are currently connected radially to onshore PCC via independent

3
transmission cables, either HVAC or HVDC transmission technique. Literally, the

transmission topology is a “one-to-one” manner between offshore wind farms and PCC. Its

drawback cannot be neglected in consideration of the overall reliability of the whole system,

that this transmission topology will contribute to loss of generation from the whole offshore

wind farm in case of any failure in the submarine transmission cable. In addition, the PCC is

usually regarded as a specified point at a fixed location, where the information of the

connected regional transmission network is not provided. It is functioning as a slack bus to

accept however much wind power generated from offshore wind farm is, without considering

the impact of variable wind power on the connected grid.

1.1.2 The impact of unstable wind power on the main grid

High wind power penetration has impacts on connected power system in terms of

stability, operational security, reliability, and efficiency.

Initially, with regard to transmission planning of offshore wind farms, the impacts of

wind power on the connected utility grid is dependent on its integrated location, to be more

specific, the correlation between wind power production and electricity power demand. The

injection of wind power probably changes the power flow in the transmission network,

thereby alleviating or worsening operational losses and bottleneck in the network.

Accordingly, a balance between electrical generation and power demand must be

achieved at any time to maintain grid stability, which, however, is threatened by integrating

large amounts of variable wind power into the grid. The penetration of renewable resources

would have a significant impact on the original generation dispatch of the regional

transmission network. Planning reserve is supposed to be considered seriously to be capable

of instantaneously making up the mismatch between generation and demand after integrating

a large amount of unstable wind power. Even though more advanced forecasting tools for

wind are developed and applied, power imbalances are still occurring more frequently with

increasing penetration of wind power.


4
The reactive power consumption of some wind generating systems are related to its real

power output. Literally, the variations in real power output of wind power system contributes

to the variation in its reactive power consumption, thus causing voltage variations in the

connected electric system.

In the current transmission planning from the point view of offshore wind farm, it is

assumed that however much the generated wind power can be able to be absorbed by the

connected regional transmission network. In fact, the existing electrical transmission systems

have already been forced to approach their stability margin by increasing power demand,

retirement of conventional thermal power plants. The increasing penetration of unstable

renewable energy, such as wind power, would push the system stability further to the limit

and probably contributes to a cascading blackout in the worst case.

Besides, the corporation of wind power has the issues such as flickers and aggregate

the power quality and transient voltage stability. It also introduces current harmonics into the

electric system by employing large amount of power electronic devices in wind generating

systems.

1.1.3 Research objective

Based on above limitations of conventional offshore wind farm planning and impact of

unstable wind power on the main grid, the main objectives to be tackled in this research are

highlight as follows:

(1) Propose a comprehensive and explicit cost model for offshore wind farm
planning.

(2) Fully optimize the electric system layout of an offshore wind farm rather than
relying on one simple standard configuration.

(3) Simplify the complexity of collector system optimization model by proposing


reasonable reduction of variables thereby reducing the search space and
improving the computational efficiency.

(4) Provide a guideline optimization tool adaptable for large-scale offshore wind
5
farms, which receives the locations and capacity of wind turbines and PCCs and
returns an optimal connection topology with proper cable types and substation
types placed at the optimal location

(5) Propose a decision tool to determine the choice between HVAC and HVDC for
OWF transmission planning.

(6) Research on the impact of OWF integrations on the connected utility grid and
involve stability performance to determine the best integration location for
multiple wind farms

1.2 Research contributions

This research focus on the problems of offshore wind farm collector system planning,

transmission planning and integration planning. The stochastic features of wind speed and

power demand is involved. The detailed contributions that distinguish from existing research

can be described in following aspects:

1) The electric system topology of the collector system is modelled by the connection

topology of the MV system, where each possible connection will introduce a binary variable.

Given N wind turbines and the two directions of power flow, N*(N-1) binary variables are

required and the maximum number of possible connection topologies among these wind

turbines will be 2 N , many of which can be defined as infeasible topologies, since it is either

not a fully connected topology able to form a reasonable result or filled with unacceptable

cable crossings. The main objective is to reduce the number of variables representing the

topology and reduce the search space by introducing the concept of allocating wind turbines

to feeders, and further optimizing the connections within feeders by minimum spanning tree

algorithm, in order to make sure each generated feeder is reasonable and not internally crossed.

A cost adjacency matrix is introduced to realize the penalty mechanism for crossing between

feeders.

6
2) The optimal connections of collector system and transmission system are

simultaneously considered, the number and locations of substations, the number and types of

transmission cables, and the connection scheme and MV cable type between any 2 WTs in

collector system. The continuous sitting problem of offshore substation is discretized by a 2-

step rasterization method.

3) A multi-level optimization framework has been proposed to optimize the electric

system layout of a large-scale offshore wind farm with multiple substations by proposing a

more detailed cost model including investment and maintenance cost for all necessary

components and operational cost, where the problem is divided into substation optimization

layer, inter-array feeder optimization layer and submarine cable section optimization layer to

effectively solve the problems caused by the large number of wind turbines.

4) The concept of clustering technique is applied to offshore wind farm planning where

wind turbines can be clustered into proper substations and further clustered into appropriate

feeders, in order to simplify the mutually affected and complicated offshore wind farm

planning problem. Fuzzy clustering algorithm, Nature aggregation algorithm and Prim’s

algorithm are applied in substation layer, inter-array feeder layer and submarine cable section

layer, respectively. In the substation layer, distance is not the only objective to carry out the

partition between wind turbines and substations. Corresponding costs of allocating wind

turbine to substation are considered as objective function, where a novel cost definition

mechanism between wind turbines and substations is provided, to provide a more proper

partition among wind turbines.

5) The transmission option based on the available HV cable data can be determined at

the substation level, which is adaptive to consider both HVAC and HVDC transmission.

6) An adaptive wind turbine allocation method is proposed to allocate the wind turbines

into the appropriate substation, maximizing the overall membership degree without

overloading the capacity limit of substation, which can be treated as a BIP problem. Fuzzy

7
clustering technique combined with the adaptive wind turbine allocation method provides the

final partition result of wind turbine under secure operation.

7) The offshore wind farm planning is mostly assumed to be integrated to one fixed and

specified point of common coupling. Besides, transmission system of multiple offshore wind

farms are assumed to be independent from each other through a HVAC or HVDC cable. Given

the advantages of MTDC network, a decision-making model for integrating offshore wind

farms via MTDC network is proposed to evaluate both its capital cost and its impact on

regional transmission network. The location of PCC in a provided regional transmission

network are optimized instead of specifying one fixed bus beforehand.

8) The stochastic wind speed and power demand are considered, both of which are

incorporated to formulate a two-stage scenario tree to represent the system scenarios, thereby

providing a more realistic and accurate evaluation of operational cost in both MTDC grid and

regional transmission network.

9) Voltage stability is taken as another objective beside costs to characterize the

integration of multiple offshore wind farms via MTDC grid so as to minimize the negative

impact of integrating offshore wind farms on the connected regional transmission network

when extra attention is needed for voltage stability enhancement.

1.3 Thesis outline

In this chapter, the problem statement and research objectives are briefly introduced,

and the research contributions are correspondingly summarized. The rest of the thesis is

organized as follows.

In Chapter 2, a literature review is provided on existing collector system planning

methods, transmission planning and integration planning methods, where the solution

methods are reviewed, and the merits and demerits of existing technique are provided.

In Chapter 3, a comprehensive optimization framework is proposed to simultaneously

8
optimize the collector and transmission system of an offshore wind farm with single

substation, with the aim to minimize its investment and operational cost. The substation

location is also optimized with the topology rather than being fixed and specified beforehand.

In addition, a recently proposed metaheuristic optimization algorithm (Nature Aggregation

Algorithm (NAA)) is used to solve the proposed model.

Chapter 4 proposes a multi-level optimization model for large-scale offshore wind

farms with multiple substations, where the optimization has been decomposed into substation

optimization, inter-array feeder selection and submarine cable type selection. The concept of

clustering is applied to simplify the problem step by step. Fuzzy clustering combined with a

proposed wind turbine allocation method is applied to partition wind turbines into proper

cluster in substation level, where the choice of optimal transmission method is determined as

well.

In Chapter 5, a transmission optimization model is proposed to integrate multiple

offshore wind farms into a utility grid via MTDC network, with the aim to minimize the

investment cost, operational cost in both DC grid and onshore AC transmission network.

Stability of the connected grid is also taken as an objective to determine the best integration

location for offshore wind farms. Stochasticity of wind speed and power demand are taken

into consideration to provide a more realistic model.

Finally, the thesis is concluded in Chapter 6, which also provides the potential future

work.

9
Chapter 2 STATE OF THE ART OF OFFSHORE

WIND POWER PLANNING

The state-of-art of research and engineering practices in offshore wind farm

technologies is reviewed in this chapter. A brief overview of offshore wind farms currently

operating is initially presented, with onshore wind farms on comparison. The following of the

chapter is divided into 3 sections in accordance with function division in offshore wind farms:

collector system, offshore substation and transmission, all of which includes the associated

technologies in design, operation and planning. Literally, this chapter forms the literature

review of the latest research in offshore wind farm planning.

2.1 Overview of currently operating offshore wind farms

2.1.1 Background of offshore wind farms

Ambitious emission reduction targets have been proposed by governments all over the

world to advance towards an environmental-friendly and low-carbon future. For instance, the

carbon emission of European is expected to cut down by 80% in 2050 [1]. The generation mix

is expected to transform from conventional climate-unfriendly thermal power plants to

renewable power, of which wind power is one of the fastest going to replace a significant

share of the retired thermal power plants. According to the locations, there are two types of

wind farms: onshore and offshore wind farms. The global cumulative installed wind power,

as shown in Fig 2.1, has increased to over 650GW in 2019, 27 times as that in 2001 [1]. The

total installed onshore wind power plants have a cumulative capacity of over 621GW, while

it is 29.1GW for installed offshore wind power plants.

10
Figure 2.1 Cumulative installed capacity of wind farms

Figure 2.2 Annual installed capacity of offshore wind farms


Although there is a huge gap in installed capacity between onshore and offshore wind

farms, the annual installed offshore capacity has reached its maximum of over 6GW in 2019,

increased by 16% from 2015 [1]. As reported in [2], the size of offshore wind farms have

almost doubled over the last decade, from 313MW in 2010 to 621MW in 2019 – see Fig 2.2.
11
The average distance to shore has been up to 59km in 2019, 24 km further from that in 2018.

Hornsea One in the UK is the offshore wind farm furthest from shore, with a distance of over

100km. The average water depth of offshore wind farms was 33km in 2019, while a pilot

project called Hywind demo has the deepest water depth of over 220m.

Offshore wind farms continue to get bigger. Size almost doubled over a decade from

313 MW in 2010 to 621 MW in 2019. The UK has the largest wind farms as a result of the

extensive Exclusive Economic Zone (EEZ) surrounding the country’s coastline. The East

Anglia (714 MW) and Hornsea One (1,218 MW) are both the largest offshore wind farms

already supplying electricity to the grid.

The offshore wind farm is expected to keep increasing up to gigawatt level in the future.

The increase in offshore wind farms comes from both the increasing wind turbine size and

increasing number of wind turbines in the offshore wind farms. Wind turbine is a device to

convert the kinetic energy of wind to electric energy. The blades are applied to capture that

kinetic energy, and the amount of captured energy is proportional to the blade size of wind

turbines. However, potential visual impact of onshore wind farm lowers the interest of using

increasingly large wind turbines in the wind farm. Accordingly, large wind turbines tend to be

used in offshore wind farms in order to explore a better wind source and meanwhile avoid

visual impact. A proper wind turbine size is site specific and chosen according to the available

wind data on site, which is normally determined in the preparatory phase of an offshore wind

farm.

The wind turbine size has been increasing year by year, from hub height of 40m, rotor

diameter of 24m in 1990, to nowadays hub height and rotor diameter both over 200m.

Accordingly, their nominal power has increased from 50kW in 1990 to 15MW in 2021, the

largest available offshore wind turbines nowadays. The table below includes a list of available

wind turbine types and their specifications from 3 world-leading wind turbine manufacturers,

Siemens Gamesa, Vestas and GE. Siemens Gamesa and Vestas are the first two players in

12
offshore market, with the largest installed wind capacity, largest number of orders, and taking

up the largest market share. The largest wind turbine commissioned is Vestas V164 9.5MW

in Borssele 3&4, newly commissioned in 2021. The largest wind turbine used in OWFs under

construction is GE Haliade-X 13MW in Dogger Bank A in UK, expected to commission in

2024. Additionally, Siemens Gamesa has received a firm order from RWE to supply wind

turbines for the 1.4 GW Sofia wind farm offshore the UK, with a 195km distance off the coast

of UK.

Table 2-1 Available wind turbines and specifications


Rotor
Wind turbine type Nominal power
diameter
SWT-3.6-120 3.6MW 120m
SWT-6.0-154 6.0MW 154m
Siemens SWT-7.0-154 7.0MW 154m
Gamesa SG 8.0-167DD 8MW 167m
SG 11.0-200DD 11MW 200m
SG 14-222DD 14MW (15MW with Power Boost) 222m
V112-3.45 MW 3.45 MW 112m
V117-4.2 MW 4.2MW 117m
V162-6.0MW 6.0MW 162m
V164-7.0MW 7.0MW 164m
Vestas V164-8.0MW 8.0MW 164m
V164-9.5 MW 9.5MW 164m
V164-10.0 MW 10MW 164m
V174-9.5 MW 9.5MW 174m
V236-15.0 MW 15MW 236m
Haliade150-6MW 6MW 150m
Haliade X-12MW 12MW 220m
GE
Haliade X-13MW 13MW 220m
Haliade X-14MW 14MW 220m

Future offshore wind farms are expected to be moved further from coast, into a much

deeper water depth. In order to capture the untapped abundant wind resources, which is

facilitated by the evolving floating foundation technology. One of the main differences

between onshore and offshore wind farm lies in the substructure. The substructure is a

mechanical support structure to transfer topside loads (wind turbine or substation) to the

seabed. An appropriate substructure type is supposed to be determined by a series of site-


13
specific environmental and seabed study, i.e., geophysical study, geotechnical study, and

metocean study, etc. Necessary environmental factors to consider include, but are not limited

to wind data, wave and current data, water level, seabed and soil description, geotechnical

characteristic data, etc. All of the above-mentioned studies are supposed to be carried out in

the preparatory phase of an offshore windfarm, i.e., in the macro-sitting study of offshore

wind farms, where optimal areas for an offshore wind farm is identified in a regional scale.

Offshore wind turbine substructure can be categorised into fixed foundations and

floating foundations. Fixed foundations are proven technology and widely applied in shallow

to transitional waters with depth up to 60 meters in offshore wind industry. Monopile, jacket,

gravity, self-elevation are the 4 basic types of fixed foundations, while tripods, bucket suction

are the combinations and transformations of the above basic foundations, all of which are

defined as fixed foundations. Fixed-foundation turbines are routinely installed in waters no

deeper than 60 meters. Sea depth is a stumbling block for fixed foundations to access deeper

waters since the heavier substructure required for deeper waters challenges the current

transport and construction capability. Accordingly, the majority of the currently

commissioned offshore wind farms only access wind resources from waters up to 60 meters.

Floating foundations are proposed to capture the plentiful untapped wind resources from

deeper waters. The floating turbines are still under pre-commercial phase and not up to utility-

scale application. As its name implies, wind turbines sit on buoyant substructures connected

to seabed by mooring cables instead of fixed foundations connected by steels. Floating wind

turbines facilitate developers to access untapped vast wind resources of water depth deeper

than 60 meters and up to 100 meters, which is projected to open up a new opportunity for

offshore wind industry. There are 3 main types of floating foundation: tension leg platform,

spar floater and semi-submersible foundation, all of which are on the verge of commercial

maturity and expected to be deployed at utility scale by 2024. With several conceptual

prototypes successful and being proven, several pilot floating wind farms have been setup, of

14
which WindFloat Atlantic project (with three 8MW wind turbines) has been fully

commissioned and supplying power to the utility grid of Portugal since June of 2020 [3].

Beyond its capability to capture enormous untapped better wind resource and adapt to deeper

sea waters, its relatively simple structure relieve the stress on installation of its fixed-bottom

counterparts. Accordingly, the decommission process is expected to be easier as well.

However, several key challenges are still to be overcome, one of which is turbine stability.

Substructure is designed to support the topside, i.e., wind turbines in this case, and minimise

the impact of wave and current motion on the rotating blades, which is still the same for

floating foundations. It is more challenging to stabilize turbines put on floaters rather on fixed

foundations.

Although floating turbine technology is currently not mature enough to utility-scale

application phase, floating substructure is still an enticing proposition to enable offshore wind

sites further from shore, with better winds. Additionally, its rapid falling costs has further

aided the development of this pipeline and promise to help to increase the share of offshore

wind in the power mix.

Both the large wind turbine size and new substructure design make it possible for

offshore wind farm developers to access a huge amount of abundant untapped wind resources

in open sea with a deeper water depth and further distance to coast. The wind farm size is

expected to reach up to giga-watt level, which not only challenges the offshore installation

for turbines but also the electric system design with a larger wind farm size. On the meanwhile,

better wind resources with consistent and higher wind speed is usually accessible at further

open waters rather than waters proximity to coast, which challenges the transmission of wind

power to onshore utility grid.

Hornsea wind farm is a 3-phase offshore wind farm project, of which the first phase,

Hornsea Project 1 is the largest currently commissioning offshore wind farms. Hornsea

Project 1 composes 2 subzones with rated capacity of 600MW each, located in areas with sea

15
depth of up to 30m and around 120km off the coast. 174 7MW wind turbine manufactured by

Siemens Gamesa are employed in this offshore windfarm, third of which are sit on suction

bucket due to its low cost and ease of installation. HVDC technique was initially considered

in the initial scoping report, in terms of its large capacity and long distance requirement.

However, HVAC is eventually employed in this project. It implies that an optimum

transmission option is based on dedicated analysis rather than empiricism. On the meanwhile,

where to put the offshore substation and to be integrated in the utility grid is also required to

be seriously considered. In this project, the onshore point is chosen at an existing 400 kV

National Grid substation at Killingholme. The second phase, named as Hornsea Project 2,

located adjacent to the phase 1 area, is under designed to utilize the existing export cable in

Hornsea Project 1, which is consistent with the protype proposed in chapter 3.

Offshore wind farms currently proposed and under construction have reached up to

gigawatt level. An electric system design model is required to accommodate the large scale

offshore wind farm with larger turbine size and number of wind turbines. Dogger Bank Wind

Farm is an example. It is expected to comprise 4 wind farms, called as Dogger Bank A, B and

C, Sofia Offshore Wind Farm, respectively, each of which has a proposed capacity of 1.2GW,

giving a total capacity of 4.8GW. It is worth noting that the group of offshore wind farms are

designated to sit in waters with depth of more than 60 meters. Floating foundation will be

employed in this project since fixed foundations fail to manage that sea depth. Haliade-X

turbines rated at 13 MW and 14MW provided by GE is employed in this project.

In conclusion, the offshore platform has advantages over its onshore counterpart by

good-quality wind resource, no noise and visual pollution. The higher capital cost and

challenge of installation contributes to the lagging installed capacity of offshore wind farm.

However, due to the recent technology advancement of offshore equipment, corresponding

cost has rapidly fallen, which is expected to further increase the share of offshore wind in the

power mix and make the price of offshore wind more competitive.

16
2.1.2 Electricity system configuration

The devices utilised in OWF can be separated into two categories: connected devices,

i.e., wind turbines, offshore substations equipped with step-up power transformers and

onshore point of common coupling (PCC), and connecting devices, i.e., medium voltage inter-

array cables and high voltage transmission cables. All abovementioned devices are involved

in the electric power system of an OWF and the electric system is functioning to collect the

generated wind power from each individual wind turbine, transmit the power via inter-array

cables to a “transfer terminal”, i.e. offshore substation, where uplifting the voltage level to

facilitate the far-distance bulk transmission, and convert the transmitted power to the proper

voltage and frequency according to grid code and then integrate that to the onshore PCC.

Accordingly, the offshore wind farm is composed of 3 parts: collector system, offshore

substation, and transmission system.

The system configuration of offshore wind farms can be defined in terms of whether

alternating current or direct current is used in its collector and transmission system. Most of

existing wind farms are operating in AC system. However, the choice of transmission option

is dependent on the rated capacity and distance to shore of offshore wind farm. If HVDC

transmission is utilized, the offshore wind farm has a AC/DC configuration. Otherwise, it is a

pure AC/AC configuration. In [4-8], some authors have proposed to extend the DC nature of

HVDC transmission to the internal collector system of offshore wind farm so as to formulate

a pure DC/DC wind farm. However, to the best knowledge of the researcher, the DC/DC wind

farm is still under research stage and not applied to real operation yet.

An optimal electric layout is of great importance for a wind farm to achieve an efficient,

economical, and reliable performance, thereby maximizing the benefit with least cost, which

is doubly true for offshore wind farm. Considering the higher capital cost for corresponding

equipment utilised in the offshore projects, the expense of an offshore wind farm on the

collector and transmission could reach up to 20% of the overall OWF project cost [16]. In

17
addition, due to the higher power capacity of an OWF than its onshore counterpart, an

appropriate electric system design inside the OWF becomes increasingly important since the

overall performance of OWF, i.e., efficiency and cost, is heavily dependent on the electric

system design.

To cater to the harsh offshore conditions, a careful protection design is required to

guarantee the safety operation of offshore wind farm due to the restricted accessibility.

However, the protection is dependent on the electric connection of the offshore wind farms

so that it is not discussed as a major issue in this thesis. The following is a review of several

standards applied in protection design. When a fault occurs in the system, the protection

system is designed to figure out the faulted location and disconnect the affected item as soon

as possible, in order to guarantee the safety operation of the remaining system and minimize

the negative impact caused by the fault. The protection system of offshore wind farm is similar

to that of onshore system, except some special protection equipment and protection design

are required for offshore system taken into consideration the harsh offshore conditions and

accessibility restrictions. The major equipment to protect in offshore platform comes from

offshore substation and corresponding transmission cables since the breakdown of these two

equipment contributes to an extreme impact on the system. Dual main protection with a back-

up device of equal performance is an effective method to deal with common mode failure of

major equipment. For instance, two independent tripping circuits are applied for transmission

cables in case one of them fails to clear the fault.

An appropriate protection design should have a proper grading of back-up protection.

The primary protection is supposed to always detect and trigger the circuit breakers rapidly

to clear the fault within the protection zone. Protections provided by adjacent equipment and

circuits (regarded as backup protection), can also detect the fault but a time delay is required

for them to react to the fault, in case of main protection failure to clear the fault. A good

coordination of main protection and backup protection is necessary to maintain the safety

18
operation of the offshore wind farm. Usually, system operators have specific grid code for

offshore wind farm to integrate in the utility grid. Fault clearance time for backup protection

at the connection point is usually required to be less than one second to limit the impact of

fault within offshore wind farm on the grid. A proper time grading for multiple-level offshore

system is required to be carefully carried out while satisfying the grid code.

A general instruction of protection scheme is provided following the normal power

flow direction, from wind turbines to shore. Initially, wind turbine transformer protection is

integrated by wind turbine manufacturers. The wind turbines are connected in string by inter-

array cables. The main protection for the array cables can be provided by overcurrent and

earth fault relays. Distance relays can also be used to protect the string. It is also required to

provide backup protection for wind turbines so that some special blocks are required to

integrate in the relays. Multiple inter-array cables are connected to the offshore substation via

a busbar so that busbar protection is required to clear phase-to-phase fault and earth fault.

With earthing transformers, the earth fault of busbar is limited, which poses difficulty to

differentiate the high income current from the low earth fault. Two alternatives proposed to

deal with the case are low impedance protection scheme and Reverse Interlocked Busbar

Protection (RIBBP). All of the relays installed on the MV switchboard are slaves of the master

relay, usually located between busbar and main transformer. The master relay collects the

information from all the relays to locate the fault, and then decide which circuit breakers are

open to clear the fault. Differential relays are usually employed to protect transformers, which

is the same story for offshore ones. For 2-windings transformers, a differential relay along

with HV REF and LV REF is provided, while a stand-alone relay providing restricted earth-

fault protection is compensated to the differential relay for 3-windings transformers. For the

LV switch panels, a directional overcurrent and earth fault relay is fitted, and can be an

integral part of the RIBBP busbar protection if applicable. It is able to provide backup

protection in the case of failure of the overcurrent protection on strings. The main protection

19
for transmission cables can be provided by differential protection if fibre optics is embedded

in the cables for communication use. Otherwise, distance protection will be utilised instead,

which can also be used as alternative option in case of communication loss. Backup protection

for transmission cable is provided by overcurrent and earth fault by a relay installed at the

onshore end. A short clearance time for earth fault can be obtained with the star/delta onshore

transformers, while a grading with transformer HV back-up is required for back-up

overcurrent fault.

MV and HV switchgears are major components including a series of relays and circuit

breaks, installed on the platform, to provide the protection to the system. GIS (Gas Insulated

Switchgear), with the insulating medium of SF6(Sulphur Hexafluoride), is commonly used in

high voltage switchgear on offshore platform. An appropriate configuration option for MV

and HV switchgear is supposed to be align with preliminary load flow analysis and fault level

analysis based on a specific electric layout design.

2.2 Collector system planning of offshore wind farms

2.2.1 Collector system topology

An offshore wind farm, as discussed before, can be decomposed into 3 systems,

collector system, substation and transmission system. In this chapter, a concise and

comprehensive review with regard to all aspects of collector system planning for an offshore

wind farm is presented. The collector system is simply responsible to collect all the generated

power from each individual wind turbines in the offshore wind farm to the substation in order

to facilitate the next-step transmission to shore.

The general schematic diagram for an OWF electric power system is shown in Fig. 2.3.

The devices utilised in OWF can be separated into two categories, in terms of their features:

connected devices, i.e., wind turbines, offshore substations equipped with step-up power

transformers or converters, and connecting devices, i.e., medium voltage inter-array cables.

20
All abovementioned devices constitute the electric power system of collector system in an

offshore wind farm.

Figure 2.3 Schematic diagram of a collector system


An offshore wind farm design is a complicated project, which necessitates an extensive

consideration of different perspectives of the offshore wind farm. Initially, a wider area to set

up the offshore wind farm should be determined, which is termed as Macro-sitting study, with

wind regime, seabed research, water depth and all other environmental impacts taken into

consideration [9-11]. The locations to sit wind turbines are necessary to be determined in the

next step to exploit the wind power potential as much as possible, and on the meanwhile, the

power defects such as wake effects is supposed to be included to better mimic the fluid

dynamic interactions between wind turbines [12]. How to connect these wind turbines and

transmit the generated power is the electric system design of an offshore wind farm. Due to

the higher investment cost of equipment used in offshore wind farms than its onshore

counterparts, the electric layout design is of more significance in offshore wind farms. In

addition, there are some other perspectives considered in offshore wind farm design, such as

providing ancillary services to the connected utility grid, and from the perspective of wind
21
farm owners, to optimize their power-selling strategy to maximize their profit, etc [13-15].

The electric system layout design of an offshore wind farm is intuitively mutually

affected by the micro-sitting study. To diminish power defect such as wake effect, wind

turbines tend to be sat in a sparse manner, with further distance in between, which, however,

poses a threat to boost enormously the cost of electric system layout design. In turn, a lower

electric system cost indicates a closer distribution of wind turbines in the wind farm, which,

in turn, lower the overall energy production of the wind farm. Usually, the micro-sitting study

aims to set up as many wind turbines as possible in a confined area, while the electric system

layout design of an offshore wind farm provides an optimal connection topology to connect

the wind turbines already fixed to PCC. Many literatures have simultaneously solved the

micro-sitting problem and electric system layout problem, most of which utilize a

decomposition strategy. However, the main objective of their study is to maximize the energy

production or generation cost, where the electric system layout design is not quite accurate

and required to be updated, such as avoiding cable crossings in the connection work [16].

Therefore, the main focus of the Chapter is about the electric system design of an offshore

wind farm.

Most of the offshore wind farms are operating in AC since the wind turbine generators

are operating in AC. There are several standard wind farm layout configurations: radial,

single-sided ring, double-sided ring, multi-ring and star topology, as shown in Fig.2.4 [17-20].

Radial configuration is the most widely used configuration in wind farm design, where wind

turbines are connected in series. Its benefit of lower cost enables it as the most widely used

configuration. However, it has the drawbacks of low reliability considering the generation

loss of the whole feeder just in case of any fault occurring near the busbar of substation. There

are two different configurations in ring topology: single-sided ring configuration and double-

22
sided ring configuration. In single-sided ring configuration, an alternative route connecting

the far-end wind turbine back to substation is added to the radial configuration in order to

improve the poor reliability in radial configuration at the expense of nearly doubling the cost

on cable connections [17]. The double-sided ring configuration provides a trade-off between

radial configuration and single-sided ring configuration, literally, lower the high expense in

single-sided ring configuration while improves the poor reliability in radial configuration,

where the ends of two neighbouring feeders are connected to formulate one feeder.

Correspondingly, the cable rate used in double-sided ring configuration is supposed to be

double of that used in radial configuration, considering the total number of wind turbines

connected on one string. However, the total cost of double-sided ring is still much lower than

single-sided double ring configuration. An updated version to further improve double-sided

ring is multi-ring configuration, where all the ends of feeders are connected. In case of any

fault occurring in any feeders, the generated power of the faulted feeder will be distributed

among the rest of feeders so that the rated capacity of cable is not necessarily double of that

in radial connection, which can correspondingly lower the system cost. The last is star

configuration, with a wind turbine set as centre and all other wind turbines connected

independently to that centre wind turbine. The reliability of star configuration is relatively

high since the affected wind turbine is only one in case of fault occurring at the connection

between centre wind turbine and other wind turbines. However, its cost is case-specific and

hard to determine regardless of specific layout. Considering the standard configurations have

their own merits and demerits, the optimal topology of a collector system is usually a

combination of the standard configurations instead of relying on one specific configuration.

23
Radial Single-sided Ring

Double-sided Ring Multi-ring

Star

Figure 2.4 Standard configuration for offshore wind farm

2.2.2 Submarine inter-array cable

The majority of currently operating offshore wind farms are operating at 33kV in their

collector systems of offshore wind parks, which is adequate for wind turbines with rated

capacity of below 6 MW. However, to fully exploit the abundant the offshore wind energy,

the offshore wind turbines have continued to grow in both power output and size, which has

24
exerted pressure on the original 33 kV inter-array cables. The largest offshore wind turbines,

launched by Siemens Gamesa in 2020, has already reached the rated capacity of 14MW [21].

The increasing rated capacity of wind turbines have forced to either improve the rated capacity

of offshore transformers or power converters in the substations or introduce extra substations

and platforms to adapt to the increased wind farm capacity. Although there is not an actual

offshore wind farm operating at 66 kV yet, wind farm developers and operators have been

prepared to switch from 33 kV to 66 kV, that three pilot projects, i.e., the Blyth Offshore

Demonstrator (UK), Nissum Bredning Vind (DK) and Aberdeen Bay (UK) wind parks are

utilising 66 kV cabling from Nexans and successfully connected to the grid and generating

power, from the first half of 2018 [22].

For offshore applications, submarine copper-core cross-linked polyethylene cables are

widely used. The total cost of the submarine cable can be defined as manufacturing cost and

transport and installation cost. Thereinto, the manufacturing cost of submarine cables,

represented per cable length, is usually case specific and hard to estimate in preliminary

design phase. Generally, it is dependent on two factors: rated current-carrying capacity

(usually represented by the cross-sectional area of the conductor and related to the amount of

conductor, i.e., copper or aluminium used in the cable) and rated voltage (determined by the

insulation material) [23]. The submarine cable manufacturing cost model is usually a function

of rated current or cross-sectional area with a set of coefficients provided respectively at

different rated voltages, as shown below.

Cac ,m = A + BeC S  € / km (2.1)

3U rated I rated
S= (2.2)
106

Where:

25
Cac ,m manufacturing cost of MV submarine inter-array cables [ € /km]

A , B ,C cost coefficients

I rated ,U rated , S rated current[A], rated voltage[V] and rated power [MVA]of the cable

Cost coefficients from [23-25] are listed in Table 2-2, for both 33kV and 66 kV.

Table 2-2 Cost coefficients for MV submarine cables


A [106] B [106] C

Ref. [23] [24] [25] [23] [24] [25] [23] [24] [25]

33kV 0.0458 0.0475 0.0538 0.0664 0.0688 0.0078 0.0410 0.0410 0.0054

66kV 0.0766 n/a 0.0900 0.0696 n/a 0.0817 0.0205 n/a 0.0035

There is an estimation based on well-known cable manufacturers prices that 66kV

submarine cables would be between 10% and 20% higher than for 33kV with same cross-

sectional areas.

The transportation and installation cost, despite represented per cable length, is highly

affected by the distance between the installed location and shores, the seabed type, which is

usually regarded as the highest uncertainty in total submarine cable cost. Table 2-3 presents

the estimated transportation and installation cost proposed by [24], [26-28].

Table 2-3 Installation and transportation cost of submarine cables [k € /km]


Ref. [24] [26] [27] [28]

Installation and
117 338 304 365
transportation [k € /km]

26
The submarine AC cables can be equivalently represented using the  model, shown
as Fig.2.5 below.

Figure 2.5 Equivalent model of AC cables


An estimation of power losses in submarine AC cables have been provided which is

represented by the equation below:

Sin2
Ploss = P0 l + C0 l 3 + Pk l (2.3)
Sn2

where P0 ,C0 , Pk are voltage-dependent parameters; Sin , Sn are the input power and rated

power of the cable, respectively; l represents the cable length.

2.2.3 Modelling choices

Planning objectives

The initial factor taken into consideration to formulate the optimal collector system is

the investment cost of all the involved components. Literally, the involved components in

collector system include inter-array submarine cables and possibly substation cost, which are

the only terms used to determine the optimal connection in the collector system, as in [24], `

The investment cost could be represented in an annualized form or total lifetime cost.

The overall collector system efficiency is also dependent on the electric system of a

collector system. An adequate electric system topology is critical to improve the efficiency

apart from guaranteeing the least overall investment cost. The efficiency is usually
27
represented by the cost of losses in the system. Therefore, the total objective in this case is the

investment cost plus the cost for operational losses in the collector system. In [17], [10],[32],

conclusions have been drawn that system topology has a strong impact on the overall system

efficiency. Authors in [32]-[34] have considered to involve the evaluation of system efficiency

into their model objective to formulate the optimal result, despite utilising different methods

to express losses.

Considering the harsh offshore conditions, the components utilized in an offshore wind

farm may have a higher danger of malfunction than those in the onshore wind farm. Therefore,

a rough estimation of maintenance cost covering routine inspection and corrective

maintenance is included in [34]. The maintenance cost is usually estimated as a specified ratio

to their investment cost.

A more accurate method to represent the impact of faulted components on the overall

design is to consider the cost associated with non-served energy in faulted conditions. Literally,

the reliability of the system is also concerned, in order to minimize the overall cost including

investment, operational cost and cost of energy losses. A N-1 criterion is used in [35] to

calculate impacted wind turbine generation under the cases of faulted components via a

discrete Markov process.

In some other literatures, the micro-sitting process of wind turbines are combined with

the optimal cabling work to formulate a two-stage optimization. In [36], it is formulated as a

two-objective problem with the first objective to maximize the annual energy production, and

the second objective to minimize the annual repayment for the cabling cost. The wind turbine

layout is completely separated from the cable layout design in [37], where an optimal wind

turbine layout is obtained in the first step and the optimal cable layout design is obtained on

the basis of least cable investment cost. A levelized production cost is set as the objective of

[38], which is the ratio between annual load repayment of cable cost [M$] and an average

energy production [MWh]. A net financial income in the cable connecting any two wind

28
turbines is used to evaluate the possibilities to set up a connection in between in [16], in which

the income is the cost for energy production, and the expenditure is the cost of the cable.

Power flow calculation

An optimal topology design is supposed to minimize the total cost, improving both

system efficiency and overall reliability, without violating the operational performance. The

cables selected for each connection should be large enough to withstand the power flow. On

the meanwhile, the operational cost of losses in the system is required to accurately predict

the profit. The power flow calculation involved in the collector system optimization problem

can be separated into 3 kinds: transportation model, direct current power flow and alternating

current power flow. In [34] and [40], transportation model has been used to represent the

power flow of the collector system, which satisfies the first Kirchoff’s law to balance the input

power and output power of a bus, as shown in the equation below.

P
p'  p
p' p + Pp ,generated −  ploss p' p =
p'  p
P
p'  p
pp' (2.4)

where the left-hand side of the equation is the power injected into the bus p, and the right-

hand side is the power flowing out of bus p. This is a generalised form of the transportation

power flow without consideration of any other loss of power due to faulted components or

system scenario with different wind speed.

The direct current load flow applies a set of inequality constraints on the power flow

and brings the power flow computation to the next-level complexity, which has been utilized

in [34], [38-42].

The generalized mathematical expression of full AC power flow equation is presented

below, which has been carried out in study [33], [38]. Pi g ,Qig are the net real power and

reactive power of bus i ; Vi ,i are the voltage magnitude and angle at bus i ; Gij , Bij are the

real and imaginary components of the branch ij .

29
Pi g = Vi V j Gij Gij cos ( i −  j ) + Bij sin ( i −  j ) 
j

Qig = Vi V j Gij Gij sin ( i −  j ) − Bij cos ( i −  j ) 


j (2.5)
max min
Vi  Vi  Vi
 imax   i   imin

Loss calculation

To incorporate the operational cost in the objective, the losses should be correctly

represented. The way of expressing losses in the submarine cables is dependent on the power

flow calculation method used in different models. As in transportation model or DCLF, the

losses incorporated in the model is either represented in a linear or quadratic form [32],[42].

The power flow on each branch can be figured out from the first Kirchoff’s law. The losses

can be estimated as below:

2
 Pp , p' 
loss p , p' =  R p , p' (2.6)
 3U
 rated 

This equation is still a estimation of the losses on the cable based on the power flow.

Actually, ACPF is the only way to get the accurate losses, which, however, introduces

nonlinearities to the problem and increases the overall complexity. On the meanwhile, the

cable parameters of each branch are determined by the layout design so that the problem has

to be decomposed into two steps with one master step to determine the topology and slave

problems to check the operational constraints of the collector system, which is time-

consuming. According to [16], cost of losses takes up 1%-3% of the overall cost. Given the

limited impact of losses on the electric system design, it is to be determined whether it is

worth of involving ACLF into the problem at the cost of computation time. Therefore, several

piecewise linearization and approximation of losses have been proposed [34],[43]. In [34], a

two-step iterative process of estimation has been involved to simplify the representation of

losses.

30
Stochasticity

In the collector system design of an offshore wind farm, the source of stochasticity

comes from the fluctuant wind speed and the equipment subject to failure. The most widely

used probability density function for the random wind speed distribution is Weibull functions,

as shown below.

k −1 k
ws
kw  −  
c 
f ( ws ) =  s  e 
(2.7)
c c 

where f ( ws ) is the probability of wind speed ws ; k indicates the shape parameter; c is


the scale parameter. If k=2, it is formulated as a Rayleigh function. k , c can be adjusted to

fit with a specific historical wind speed record. The stochasticity of wind speed can be simply

ignored and represented by an average wind speed based on the available wind speed on site

[40],[45], or discretized into a set of wind speed scenarios [33], [46], or simulated.

As discussed before, the failure rate of offshore components is higher than those used

onshore considering the harsh over-sea environmental conditions. To involve the reliability

of the overall system, system components subject to failure have to be considered. N-1

criterion [33] or N-2 criterion [47] is considered since the probability of failure in components

is relatively low. The transition of component state is represented by a discrete Markov

process in [33].

To involve the stochasticity of the problem, scenario enumeration method is used in

[33], [39]. In [33], a two-stage scenario tree is formulated to incorporate both the wind speed

stochasticity and the system component state. System states involving the stochasticity can

also be simulated by Monte Carlo method as in [49-50]. Both of these two methods can be

used to incorporate probabilistic description of the problem, where scenario enumeration

method seems more straightforward and easier to integrate in the mathematical formulation.

31
2.2.4 Solution method

As discussed in the last section, the optimization of collector system of an offshore

wind farm is an NP-hard problem due to its objective function and constraints. Initially, binary

or integer variables are involved in the mathematical model to represent the placement of

wind turbines and cable connections between them. The introduction of losses into the

objective function further increases the complexity of the problem and would introduce

nonlinearity and nonconvexity in the objective functions according to the proposed methods

of calculating losses. Similarly, secure system operation is supposed to be guaranteed so that

a set of power flow constraints are required to follow, which leads to the nonlinearity,

nonconvexity and multi-restriction in the constraints. In addition, power flow result is

dependent on the cable parameters, which is further determined by the result of electric system

layout design. Literally, one specific collector system layout design would lead to a different

power flow result and also a different operational cost. If substituted into the objective

functions, in turn, it will have an impact on the layout design. The involvement of non-served

power cost due to components failure has the same effect to the overall objective as system

efficiency. In conclusion, the system layout topology is mutually affected by its power flow

analysis and reliability analysis. The modelling choices of the discussed terms in the last

section have a decisive impact on the mathematical model of the electric system optimization

of collector system. The optimization of electric system design of an offshore wind farm is

considered as an NP-hard problem with the characteristics including nonlinearity,

nonconvexity, multivariable for both objective function and constraints, whose complexity is

linear with the number of wind turbines in the offshore wind farm.

According to the features of different proposed mathematical formulation, the collector

system optimization problem can be solved by two different kinds of optimization technique:

classical methods and non-classical methods.

Classical methods have a higher requirement on the expression of the mathematical

32
model. It can be characterized into 4 categories according to its variable types, objective or

constraint expressions: Linear programming (LP), Mixed integer programming (MIP), Mixed

integer quadratic programming (MIQP) and mixed integer nonlinear programming (MINLP).

Authors in [40] have exploited the impacts of wind farm capacity on the power flow by using

Linear programming (LP), which has the benefits of fast and efficient convergence, however,

at the cost of neglecting the discrete nature of the collector system planning problem. In MIP

and MIQP, integer variables are added to the problem in order to represent either the

placement of wind turbines or the cable connections between them. A mathematical model

taking stochastic wind and reliability of components into consideration is proposed as a MIP

in [39]. The reason why it is still a linear problem is because the loss of the system is not

considered in this paper. A combinatorial optimization model for wind turbine type and

placement is developed as a MIP in [35] and [45], to determine the wind turbine placement of

several specific choices. In [63], the objective function is to minimize the total cable length

in the collector system, so as to be formulated as a MIP. In [35], Bender’s decomposition and

a proposed Progressive Contingency Incorporation Algorithm is utilized to solve and speed

up the resolution of the model. The optimization problem of collector system proposed in [11]

is formulated as a MIQP to evaluate the objective functions of investment cost, system

efficiency and reliability cost, which is implemented in GAMS language, a high-level

modelling environment [50]. The losses in this paper are estimated using the transportation

model for power flow analysis, which leads to the quadratic term in its objective functions.

MINLP is rarely applied to optimize the collector system topology considering the complexity.

As for the range of the researcher’s knowledge, only authors in [34] has formulated the

optimization model as MINLP since they utilized a full AC power flow model to determine

the cable selection and secure operation of the overall wind farm, where Bender’s

decomposition has been applied to decompose the problem into master-slave mode, with

master level to determine the topology and slave level to determine the secure operation of

33
the specific topology, in order to reduce the complexity of the problem. However, its

ambiguous expression of losses is still a problem to be solved. All the above models using

classical optimization have a problem of applying a limit on the system topology to avoid

cable crossing as a set of constraints. Many of the simulation results still have cable crossings,

which is not acceptable in offshore wind farms since the crossed cables will impact each other

and it poses a challenge of installation. The explicit expression required in classical modelling

method fails to express this kind of limit applied on geometry.

On the contrary, non-classical method enables a more relaxed mathematical

formulation by decoupling the binary or integer layout decision from the following objective

functions calculation. Its feasibility and adaptability enable the non-classical methods to be

the most widely used methods in collector system optimization planning [51-62]. Non-

classical methods can be separated into: Heuristics methods, Metaheuristics methods and

Hybrids method.

Heuristics methods have been applied to several literatures with regard to offshore wind

farm, however, without a guarantee of optimality. An advanced heuristic search algorithm

similar to simulated annealing is proposed to optimize the annual profit of a wind park [57].

Authors in [64] apply a simple heuristic algorithm to simply optimize the location of the

substation in the offshore wind farm.

Genetic algorithms and particle-swarm optimization algorithms are the most widely

used Metaheuristics methods in wind farm planning.

Several optimization models based on genetic algorithms have been proposed for

optimizing the electric power system of OWFs [41] and [65-66]. A k-clustering based GA is

applied to cluster wind turbines and the cable connections are optimized via a minimum

spanning tree algorithm [67]. In [68], a multi-objective problem is proposed to maximize the

output from the offshore wind farm while minimize the capital investment as well as

operational cost. A novel method of using an adjacency matrix to represent the connection in

34
an OWF is applied in [30] to simply minimize total capital cost of the collector system, which

is solved by GA as well. No system efficiency has been included in this project. It is similar

in [69] that only investment cost of components are treated as the objective functions, where

the cabling network is treated as a MTSP problem solved by GA. Reliability assessment of

wind turbines are also included in [70-71] to determine the optimal number and size of

components in a wind farm project. Apart from offshore wind farm, there are several

literatures using GA to deal with the onshore wind farm planning. In [72], the wind turbine

sitting problem is affected by the negative impact of noise so that a multi-objective problem

is formulated to initially maximize the output and lower the noise impact to the local residents.

This project is solved by NSGA- II. A genetic algorithm-based method is proposed to

optimize an onshore wind farm cable design in terms of the minimum investment cost, power

losses and reliability cost [73]. Authors in [41] provided a review of performance of several

GA applied to the electric system layout optimization of OWFs and have concluded the

drawbacks of GA is premature convergence to local minima.

Particle-swarm optimization algorithm is widely used in micro-sitting problem of wind

turbines rather than internal electric system layout design [38], [53], [64], [74-76]. Authors of

[77] initially optimizes the sitting of wind turbines in offshore wind farm by the adaptive PSO,

and further optimize the cable layout using MST algorithms. A multi-objective problem is

proposed to maximize the energy output from a wind farm while minimizing the turbine cost,

solved by a multi-objective PSO algorithm [78]. Besides PSO algorithm, the wind turbine

layout design is encoded in a novel encoding mechanism and solved by differential

evolutionary algorithm [79].

A well-designed hybrids method can take advantages of several optimization algorithm.

A set of literatures have utilized to solve the collector system optimization in offshore wind

farms including wind turbine micro-sitting and cable connection optimization. A hybrid

genetic and immune algorithm has been applied to optimize the internal connection of an

35
offshore wind farm [43]. A two-echelon planning model is proposed in [80] to optimize the

locations of wind turbines in order to maximize the power output, where the random key

genetic algorithm is applied in the first echelon; particle-swarm optimization is applied in the

second echelon. The investment cost and losses cost have been taken into consideration to

optimize the cable connection work of collector system in [81]. A multi-objective problem

has been proposed to optimize the system topology of OWF, with one objective to maximize

the yearly power output solved by particle-swarm optimization algorithm, and the other to

minimize the investment cost on cabling solved by simulated annealing algorithm [36]. A

novel metaheuristic approach, as a hybridization of mathematical programming techniques

and heuristics algorithm, is proposed to optimize the internal cable connection topology of an

offshore windfarm considering potential obstacles [82]. A negative impact of noise and energy

production is simultaneously taken into consideration to determine the optimal witting of

wind turbines of an onshore wind farm [77]. An artificial intelligence technique with genetic

algorithm applied to optimize the wind turbine sitting and ant colony system algorithm is

applied in [9] to optimize wind turbine locations and line connection simultaneously.

2.3 Offshore substation options for offshore wind farms

2.3.1 Configuration

The function of collector system has been discussed in last section, that all the

generated power will be transmitted through the medium-voltage to the substation. In the

substation, the medium voltage level will be lifted to high voltage level to facilitate the long-

distance transmission finally to shore. The substation can be regarded as a “transfer terminal”

between transmission system and collector system to convert the transmitted power to the

proper voltage and frequency in order to satisfy the grid code so as to integrate into onshore

point of common coupling. In some pilot projects of offshore wind farms, no offshore

substations are set up, which literally indicates the generated wind power is fed into the grid

36
at medium voltage level, since the offshore wind farm has a relatively low rated capacity and

is usually located near coast (usually less than 10km). In some other cases, to fit into the

regional transmission network, a voltage step-up is still required. However, the substation is

located onshore instead of offshore, which is also suitable for some near-coast offshore wind

farms.

As discussed in the submarine inter-array cable section, collector systems in the

majority of offshore wind farms are operating at a voltage level of 33kV. Actually, the medium

voltage level within collector system is driven by the availability of switchgear and

transformers fit into the wind turbines, the maximum rated voltage of which are usually set

as 36kV under IEC standards [84]. Most of wind farm projects currently operating consist of

less than 100 wind turbines, whose rated capacity is between 3 and 5 MW, still subject to

rapid change as development progresses [84]. A maximum number of wind turbines

connected to one feeder is up to 10 considering the radial topology, the most widely used

connection type. If the distance of an offshore wind farm is further than 10km, one or multiple

substations will be set up to step up the nominal 33kV in collector system to 132 kV, 220 kV,

in order to facilitate the more efficient transmission to shore. However, to fully exploit the

abundant the offshore wind energy, the offshore wind turbines have continued to grow in both

power output and size, which has exerted pressure on the original 33 kV inter-array cables.

The increasing rated capacity of wind turbines have forced to either improve the rated capacity

of offshore transformers or power converters in the substations or introduce extra substations

and platforms to adapt to the increased wind farm capacity. Although there is not an actual

offshore wind farm operating at 66 kV yet, wind farm developers and operators have been

prepared to switch from 33 kV to 66 kV. Three pilot projects of offshore wind parks are

operating at 66kV and successfully operating and connected to shore [84].

The substation composes of power transformers (fit for further HVAC transmission) or

power converters (fit for further HVDC transmission), MV busbars, HV busbars, MV

37
switchgears, HV switchgears and platform. There are usually auxiliary diesel generators

equipped in the substation, which, however, is not considered in the researcher’s project. The

equipment used in offshore substations is not so special, but the size and isolation standard

should be able to fit in the compact space on platform and humidity environmental [84].

2.3.2 Power transformer

The estimation cost of a power transformer is a function of the rated power, which can

be generally represented in the equation below.

B
S 
CT = A   T ,max  (2.8)
 C
 

where A, B are cost coefficients estimated in different literatures; ST ,max is the rated power

of the transformer [MVA];. CT represents the transformer cost in M€. According to [23] and

[85], the cost coefficient is presented in Table 2-4. In [23], a more explicit cost estimation has

been proposed to onshore and offshore transformers, respectively, where offshore

transformers are obviously more costly than onshore transformers.

Table 2-4 Cost coefficient for power transformer


[23]
Ref [85]
Offshore Onshore

A 0.03327 0.12 0.03

B 0.7513 -0.413 -0.413

C 1 300 300

The equivalent circuit of a transformer is shown in Fig. 2.6. An approximation of losses

in transformer is proposed in [25], as shown below.

Pk 2
Ploss = S n P0 + Sin (2.9)
Sn

where Sn indicates the rated capacity [VA]; S in is the input power to the transformer; P0 , Pk
38
are per-unit no-load losses and load losses, respectively.

Figure 2.6 Equivalent circuit of a power transformer

2.3.3 Power converter

There are two types of converters, line commutated converters and voltage source

converters. Voltage source converters are more attractive than line commutated converters for

offshore wind farms. Considering the compact space of offshore platform, voltage source

converters enable a reduction of size. In addition, it enables the offshore wind farm to be

electrically decoupled from the onshore grid and operate at a frequency different from the

standard frequency (50 or 60 Hz). In [85], the author has proposed to remove the converters

installed at each individual wind turbines to save costs, and deploy the converters installed at

offshore substation to provide control over the wind turbines instead. The voltage source

converters do not rely on the AC system to provide support so as to facilitate the offshore

wind farms to be connected to a weak AC grid. Therefore, voltage source converters are

usually equipped in the offshore substations of an offshore wind farm to facilitate HVDC

transmission.

Similar to transformer cost, the estimated converter cost is a function of rated capacity

as well. The detailed equation is provided by [23], still separated into onshore converters and

offshore converters, where PN ,cov is the rated capacity, the only variable of this cost function.

Obviously, offshore converters is more expensive than onshore converters, considering the

39
higher installation and isolation standard offshore.

PN ,cov
Cdc ,off = 42 + 27 
300
M €  (2.10)

PN ,cov
Cdc ,on = 18 + 27 
300
M €  (2.11)

PL ,cov = 0.0042558  e1.1004171UF (2.12)

Eq. (2.12) provides an estimation of losses in converters, where UF represents the

utilization factor [86].

2.3.4. Single substation V.S. multiple substation

To fully exploit the abundant the offshore wind energy, the offshore wind turbines have

continued to grow in both power output and size. On the meanwhile, the number of wind

turbines in an offshore wind farm has increased as well. The two abovementioned factors have

forced to either improve the rated capacity of offshore transformers or power converters in

the substations or introduce extra substations and platforms to adapt to the increased wind

farm capacity.

Generally, in such a large offshore wind farm, there are hundreds of wind turbines

spreading across hundreds of square kilometres. If the sole-substation topology is still applied

in this case, the overall collector system cost to connect each wind turbine to the substation is

supposed to be considerably high from an economic perspective; the overall reliability is,

meanwhile, assumed to be poor considering the fact of losing all the generated power in the

case of an offshore substation breakdown or malfunction. Accordingly, the sole-substation

topology is not appropriate for a large-scale offshore wind farms in light of its poor

performance in both cost, reliability as well as secure operation. Multiple substations seem

superior to the conventional single substation method used in offshore wind projects.

However, the majority of literatures with regard to collector system planning simply

consider there is only one substation in their project, as in [17], [33] and [35]. Although

40
multiple substations is considered in [30], the substation location is predefined to limited

options. In [34] and [87], a full topology optimization of inner collector system layout and

substation sitting is provided instead of choosing among a limited number of choices.

2.4 Transmission and integration planning of offshore wind farms

2.4.1 Overview of transmission technique

The transmission system is responsible to transmit the collected power at substation

and integrate that into the regional transmission network. The transmission options taken into

consideration are medium voltage AC transmission, high voltage AC transmission, high

voltage DC transmission and multi-terminal high voltage DC network. MVAC transmission

option is only suitable for a small wind plant closer. Such wind power plants are usually

radially connected to PCCs. The majority of existing offshore wind farms, to date, are utilizing

high voltage ac transmission technique, which reduces the losses by elevating voltage at the

offshore substation [88-91]. However, the increasing offshore wind farm capacity and longer

to-shore distance is expected to challenge the most widely transmission option, given its

inadequate capacity due to its capacitive charging current [47]. Whilst the inadequacy of high

voltage ac transmission has been foreseen, high voltage dc transmission is making a stage

comeback given its benefits of facilitate bulk power transmission over a long distance. In

addition, VSC-HVDC transmission enables the offshore wind farm connected to a weak grid

[90]. HVDC transmission is usually considered more efficient than HVAC transmission given

the fact that its losses is much lower than that in HVAC transmission given the same amount

of transmitted power.

To date, most offshore wind farms are connected to shore via either HVAC or HVDC

transmission technique. However, its drawback cannot be neglected in consideration of the

overall reliability of the whole system. In case of any failure in the submarine transmission

cable, these two transmission options contribute to loss of generation from the whole offshore

41
windfarm. In addition, transmission system using HVAC or HVDC simply connect the

offshore wind farms in a one-to-one manner to one specific PCC without considering the

negative impact of integrating such a large amount of unstable renewable power on the grid.

A more reasonable solution to tackle the issue is to distribute the substantial wind power into

several PCCs among the onshore customers. Therefore, multi-terminal HVDC network is

introduced, as a compromise between cost and reliability.

2.4.2 HVAC submarine cable

For offshore applications, a three-phase submarine copper-core cross-linked

polyethylene cables are widely used. The total cost of the submarine cable can be defined as

manufacturing cost and transport and installation cost. Similar to MVAC submarine cable, the

manufacturing cost is dependent on two factors: rated current-carrying capacity and rated

voltage, which can be estimated using the same equation in Eq. (2.1).

Cost coefficients from [23] and [25] are listed in Table 2-5, for both 132 kV and 220

kV.

Table 2-5 Cost coefficients for HVAC submarine cables


A [106] B [106] C

Ref. [23] [25] [23] [25] [23] [25]

132kV 0.21 0.258 0.023 0.0273 0.0166 0.0028

220kV 0.354 0.416 0.012 0.0144 0.0116 0.0020

The transportation and installation cost, despite represented per cable length, is highly

affected by the distance between the installed location and shores, which has a similar cost

for submarine inter-array AC cables.

2.4.3 HVDC submarine cable

Similarly, the total cost of HVDC cable can be separated into manufacturing cost and

transport and installation cost. The manufacturing cost of HVDC cables, represented per cable

42
length, is dependent on two factors: rated capacity and rated voltage (determined by the

insulation material), which is estimated in [16] as a function of rated capacity with a set of

coefficients provided respectively at different rated voltages, as shown below.

Cac ,m = A + BP  € / km (2.13)

Where:

Cac ,m manufacturing cost of HVDC cables [ € /km]

A, B cost coefficients

P rated power [MVA]of the cable

Cost coefficients from [23] are listed in Table 2-6, for 80kV, 150kV and 320kV.

Table 2-6 Cost coefficients for HVDC cables


A [106] B [106]

80kV -0.028 0.0036

150kV -0.011 0.0018

320kV 0.032 0.0011

The transportation and installation cost for HVDC cables are assumed to be same as

HVAC cables and MVAC cables so as not to be provided here.

An estimation of power losses in HVDC cables have been provided in [25], which is

represented by the equation below:

Pin2
Ploss = Pk l (2.14)
Pn2

where Pk , Pin , PN are load parameter, input power and rated power of the HVDC cable,

respectively; l represents the cable length.


Usually, for offshore wind farm with larger rated capacity and longer distance to shore,

HVDC transmission tend to be considered. An empirical equation to calculate the break-even

distance of HVDC transmission is provided as below [91].


43
 
lDC = max 40 ,min 200 , 832.5  PN−0.4  (2.15)

PN is the rated capacity of the cable (either substation capacity or wind farm capacity).

The extra cost of setting up converters in HVDC transmission needs to be offset by the cost

savings of adopting HVDC transmission options with a cheaper unit cost than HVAC

transmission. If the transmission distance is larger than the critical distance lDC , HVDC

transmission option is more favourable. Otherwise, HVAC transmission will be chosen.

2.4.4 Multi-terminal SVC network

In one-to-one HVDC transmission for offshore wind farms, only two parties, i.e., wind

farm and utility grid, are involved. However, the introduction of multi-terminal network

enables a more flexible re-distribution of wind power among more than two parties in HVDC

transmission [92]. The merits for multi-terminal HVDC grid to integrate large-scale offshore

wind farms are provided in [93-99]. Extensive research has been done to the modelling,

control and operation of MTDC integration. The first mixed AC/DC transmission network

planning was begun in [100], where an analytical comparison between AC and DC

transmission was proposed. Mathematical models for MTDC with VSC are proposed in [101].

A novel control strategy for VSCs are presented with the consideration of wind speed

uncertainty in wind speed and failure of components [102]. An optimal hybrid AC/DC power

flow model is provided in [103]. However, the focus of most available literatures regarding

MTDC is about its control strategy to improve stability and power flow regulation, instead of

taking it as an alternative to integrate OWFs into associated RTN.

2.4.5 Deterministic V.S. Probabilistic planning

Many uncertainties are supposed to be considered in transmission and integration

planning of an offshore wind farm, such as load growth, offshore wind speed, availability of

new transmission links and so on. In theory, the transmission and integration planning

problem can be separated into two types: deterministic approach and probabilistic approach
44
[97]. In a deterministic approach, all the stochastic events are considered at the worst scenario,

while the estimates on the same events of how much, how long are considered and quantified

to obtain a more accurate expected outcome [104-105].

The uncertainties associated with the timing, location and severity of stochastic events,

have posed a challenge for probabilistic methods to carry out risk analysis on transmission

and integration planning, where the complexity is linear with the number of involved

components [47].

2.4.6 The impacts of integrating OWFs on the regional transmission network

From the viewpoint of offshore wind farm planner, the offshore wind farm planning is

considered under the assumption that however much the generated wind power is can be able

to be absorbed by the connected regional transmission network. In fact, the existing electrical

transmission systems have already been forced to approach their stability margin by

increasing power demand, retirement of conventional thermal power plants. The increasing

penetration of unstable renewable energy, such as wind power, would push the system stability

further to the limit and probably contributes to a cascading blackout in the worst case. System

stability should be considered to guarantee the secure and reliable operation of the regional

transmission network after integrating the offshore wind farms. Therefore, voltage stability

should be regarded as another metric besides total expenditure to characterize transmission

planning for OWFs.

In addition, the penetration of renewable resources would have a significant impact on

the original generation dispatch of the regional transmission network. Some measures are

supposed to be applied to accommodate the fast-changing generation insufficiency due to the

stochastic nature of renewable resources. Despite some novel prediction methods proposed to

deal with the stochastic nature of wind speed and load, the solution of the traditional OPF

may still contribute to system overflows or even system failure. The form of OPF with

renewable generation is proposed below.


45
min f ( x ,u ) (2.16)
u

where

ng

(
f ( x ,u ) =  ai pg2 ,i + bi pg2 ,i + ci ) (2.17)
i =1

where x indicates the system state variable; u is the set of control variables in the ACOPF
problem. ai , bi , ci the operation cost coefficient of generators. Therefore, the objective

function is still to optimize the economic power dispatch by minimizing the total generation

cost.

subject to

nb
pg ,i − pd ,i + pr ,i = Vi V j Gij Gij cos ( i −  j ) + Bij sin ( i −  j )  (2.18)
j =1

nb
qg ,i − qd ,i + qr ,i = Vi V j Gij Gij sin ( i −  j ) − Bij cos ( i −  j )  (2.19)
j =1

pgmin max
,i  pg ,i  pg ,i (2.20)

qgmin max
,i  qg ,i  qg ,i (2.21)

Vi min  Vi  Vi max (2.22)

ijmin  ij  ijmax (2.23)

pijmin  pij  pijmax (2.24)

There are two types of constraints: equality and inequality constraints. Eq. (2.18) and

(2.19) are equality constraints, representing the real and reactive power balance for each nodal

bus. The penetration of renewable energy resources is represented by pr ,i , qr ,i , both of which

are added to the left side of the power balance equations. Eq. (2.20) - (2.24) constitutes the

inequality constraints of ACOPF problem. The operation limits of each generator on bus i

46
should be satisfied, as shown in eq. (2.20) and (2.21). Voltage magnitude and angle of each

bus is supposed to be within the set bus limit in Eq. (2.22) and (2.23). The last inequality

constraints guarantee the cables between any two buses not to be overloaded.

2.5 Conclusion

The offshore wind farm can be decomposed according to their functions into 3 parts:

collector system, offshore substation and transmission system, the review of which is

presented in this chapter in the same order.

Initially, a background review of offshore wind farms is presented, where a comparative

analysis between onshore and offshore wind farm is carried out.

In collector system section, the merits and demerits of standard configurations of a

collector system are compared. A conclusion has been drawn that an optimal electric system

topology of a collector system is supposed to be a combination of the standard configurations,

so as to introduce the necessity of optimizing the collector system layout rather than following

the instructions of standard configurations. The key component in collector system planning

is the MV submarine inter-array cables, whose cost and loss model are both provided in this

section. The modelling choices of collector system planning have been introduced in 4 aspects:

modelling objectives, power flow calculation methods, expression of losses and involved

stochasticity, where the state-of-art literatures with regard to collector system planning have

been separated according to the proposed modelling choices. To fit with the modelling choices,

corresponding solution methods are reviewed to deal with the collector system problems.

In the substation section, the basic component in an offshore substation is described

initially, followed by the cost and losses model proposed for power transformers and power

converters. The increased wind turbine capacity and number of wind turbines in an offshore

wind farm forces the transition from single substation to multiple substations. The limitations

of current literatures with regard to different number and types of substations used in offshore

47
wind farm is analysed.

In transmission and integration section, an overview of possible transmission technique

has been provided. A comparative analysis is carried out for MVAC transmission, HVAC

transmission, HVDC transmission and MTDC network. In the transmission and integration

planning, probabilistic method is more accurate to fit in the cases given a set of uncertainties.

In addition, the impact of renewable resources on the existing electric system is investigated.

48
Chapter 3 AN ELECTRICAL SYSTEM LAYOUT

OPTIMIZATION FRAMEWORK FOR AN OFFSHORE WIND

FARM

In this Chapter, a novel system layout optimization framework has been proposed to

optimize simultaneously the collector and transmission system of an offshore wind farm given

a full set of submarine cable types. The location of substation is also optimized instead of

specifying beforehand.

3.1 Problem description

An optimal electric layout is of great importance for a wind farm to achieve an efficient,

economical, and reliable performance, thereby maximizing the benefit with least cost, which

is doubly true for offshore wind farm. Considering the higher capital cost for corresponding

equipment utilised in the offshore projects, the expense of an offshore wind farm on the

collector and transmission could reach up to 20% of the overall OWF project cost [17]. In

addition, due to the higher power capacity of an OWF than its onshore counterpart, an

appropriate electric system design inside the OWF becomes increasingly important since the

overall performance of OWF, i.e., efficiency and cost, is heavily dependent on the electric

system design. Hence, an optimal electric system design is necessary in order to minimize the

cost covering not only investment as well as operational cost, meanwhile following a set of

technical constraints like electric operational constraints and specific topology connection

constraints.

49
Figure 3.1 Schematic diagram of an OWF electric system
The general schematic diagram for an OWF electric power system is shown in Fig. 3.1.

In light of various functional features, the devices utilised in OWF can be separated into two

categories: connected devices, i.e., wind turbines, offshore substations equipped with step-up

power transformers and onshore point of common coupling (PCC), and connecting devices,

i.e., medium voltage inter-array cables and high voltage transmission cables. All

abovementioned devices are involved in the electric power system of an OWF and the electric

system is functioning to collect the generated wind power from each individual wind turbine,

transmit the power via inter-array cables to a “transfer terminal”, i.e. offshore substation,

where uplifting the voltage level to facilitate the far-distance bulk transmission, and convert

the transmitted power to the proper voltage and frequency according to grid code and then

integrate that to the onshore PCC. Likewise, according to the features of transmission, the

offshore windfarm can be separated into two sub-systems, collector system (collect and

accumulate the generated power) and transmission system (mainly transmit the accumulated

power to onshore power grid), respectively. The offshore substations can be regarded as the

electrical separator between the collector system and transmission system. Typically, medium

voltage level is utilised in collector system, while high voltage level is utilised in transmission

system.

The locations as well as the characteristics of each individual wind turbine are assumed

to be known, which is supposed to be the result of preliminary micro-sitting research to

maximize the capture of wind energy. Also, the position for point of common coupling (PCC)

50
is provided and in this chapter, only one PCC is considered to integrate the generated wind

power. AC electric power systems are mainly considered in collector system because the vast

majority of the OWF collector system in operation are running AC systems. Literally, electric

system design can be defined as the optimal placement of the electrical links and substations

(power transformers and auxiliary devices) both of appropriate size between turbines and PCC.

To be more precise, the OWF electric system layout optimization framework receives the

aforesaid information and returns the necessary decision variables to formulate the overall

OWF electric system as follows:

1) the connection topology within collector system, i.e., the number of WTs on each

“feeder”, the utilised cable type and connection scheme of each section on individual “feeder”,

the number of “feeders” connected to offshore substation;

2) the power capacity, and locations of the offshore substation;

3) the connection topology of the transmission system, i.e., the utilized cable type,

number of cables withhold the overall capacity, the connection scheme between the offshore

substation and PCC.

As for a small OWF with several wind turbines, all possible topologies combined with

available transmission technology can be enumerated using brute force. However, the

enumeration approach cannot be applied in the case of medium- and even large-size offshore

wind farms because the electric layout optimization framework of an OWF is an NP-hard

problem. The model is formulated as a mixed-integer nonlinear problem because both the

constraints and objective function have the characteristics such as nonlinearity, multivariable,

and nonconvexity. The complexity is proportional to the size of the wind farm i.e. the number

of WTs since the number of possible connections increases exponentially with the number of

wind turbines, which also enlarges the overall search region of the problem.

The framework takes into consideration two general layouts for the electric power

system design, dependent on whether an individual transmission network is used or not. In

51
practice, some offshore substations have already been set up offshore to facilitate further

connections from potential offshore wind farms under development and planning stage, which,

however, are owned and operated by other stakeholders, i.e. transmission company. In this

case, it would be a more economical approach for an OWF owner to connect their OWF to

this existing offshore substation rather than build up a new transmission line and “transfer

station” for their own use, which is termed layout 1, shown in Fig. 3.2. To the contrary, if

there is no such a “transfer station” around this OWF region, an individual offshore substation

equipped with transmission to PCC will need to be established, which is termed layout 2,

shown in Fig 3.3. Different colours and line widths in the figure represent different cross-

sectional sizes of the cables.

Figure 3.2 Schematic diagram for layout 1


The main contribution of the model is to propose a decision-making tool for offshore

52
wind farm planner to optimize the topology of collector system and transmission system at

the same time. Initially, a novel mathematical model to describe the problem has been

proposed. To simplify the complicated problem, a two-step rasterization method is used to

optimize the offshore substation location. A novel evolutionary algorithm has been applied to

efficiently solve the electric layout optimization problem.

Figure 3.3 Schematic diagram for layout 2

3.2 Mathematical modelling

As mentioned in last section, there are a lot of factors affecting the overall cost of an

offshore wind farm, i.e. various wind turbine model, no. of wind turbines, distance to shore,

submarine seabed type, voltage level utilised, cable size and length for each individual section,

connection scheme, etc, all of which are mutually coupled and related to each other. Therefore,

a large number of discrete variables are supposed to be introduced to represent the

abovementioned decision factors and they have to follow either linear or the nonlinear

relationships to mimic the natures of the electric layout system of an OWF.

53
In this section, the mathematical model for OWF electric system layout optimization is

given considering some justified assumptions, which is followed by the definition of the set

of input necessary to the problem. At the end of this section, the cost model for any

components involved in this model is provided in detail.

Figure 3.4 Connection configuration for network nodes


3.2.1 Hypothesis

As mentioned earlier, the electric system optimization of an OWF is considered as a

complex problem. Some reasonable hypotheses has been proposed, because of which this

complex problem, to some extent, can be simplified. The assumptions are listed below:

1) All the preliminary studies (including environmental analysis, seabed survey which

are necessary to determine the optimal location of WTs, offshore substations) have been

conducted. Namely, the locations for each individual wind turbine and feasible region to build

up potential substation are fixed.

2) Only one point of common coupling (PCC) is provided, thru which the generated

wind power from the OWF is injected into main grid.

3) The availability of WTs is assumed to be 100%.

4) Only one substation is set up to collect all the wind power generated from the OWF,

which is the only “bridge” between wind farm and main electric grid.

5) Wind speed is assumed to be the same for all WTs at any given time.

6) Voltage level within collector system is fixed and assumed to be either 33kV or 66kV,

54
both of which are the most widely used rated voltage level for submarine MVAC cables. The

abovementioned is true for HV voltage level as well, and 132kV, 220kV are the most widely

used rated voltage level in transmission system.

7) AC electric power systems are mainly considered because the vast majority of the

OWF collector system in operation are running AC systems.

8) Each pair of nodes (wind turbines or substation) can be connected in string, star and

the mixed string/star configuration, as shown in Fig. 3.4, while ring topology is not supported

in this project.

After performing the study of site and aerodynamic assessment, the number of wind

turbines and their optimal sitting are determined under the objective of maximizing wind

power generation by micro-sitting optimization process. Therefore, the first hypothesis can

be justified. Reliability is not considered so that electrical links are sit without redundancy

and during the operation period, all wind turbines are assumed to be always available, which,

therefore, supports the 3rd and 8th hypothesis. Considering the expensive cost to set up a

platform over the sea, it is reasonable to build up one offshore substation rather than more

substations catering to a medium-size OWF. The wind speed, over the smooth surface of sea,

is usually much smoother than that on land since the changing terrain and possible

constructions have a direct impact on the wind speed captured by the turbine blades. Therefore,

wind speed is regarded as the same for all wind turbines within the windfarm coverage, if

ignoring wake effect. In terms of wind speed and direction, both are normally considered in

the micro-sitting of wind turbines. The topology optimization framework proposed in this

chapter is based on the abovementioned conditions where all the turbines have been sat at

individual places. The objective is to find out the optimum electric system design for the

offshore wind farms. Additionally, studies have shown that the effects of power variance, or

literally wind speed variance, have the same effect on different topology connections in

offshore wind farm [17],[106]. The power characteristic are therefore considered the same for

55
all the wind turbines in offshore wind farms.

3.2.2 Inputs

According to the hypothesis listed in the last subsection, a series of data is supposed to

be provided as the input to the optimization framework. These data, which are collected from

public-accessible resources or empirical data, can be stored in a database. As some equipment,

i.e. cables utilised in each wind farm project is designated for individual project, the cable

cost offered from the manufacturer is probably different. In terms of the above reason, all the

data in the database can be easily changed catering to any various need. Necessary input for

this framework is listed in detail as below, regarding wind farm, wind turbine, any involved

equipment, and essential electrical data, respectively.

1) Wind farm basic information (after preliminary studies)

a. Wind speed distribution data (wind history)

b. Power output characteristic curve of WTs

c. No. of WTs

d. Lifetime, interest rate, cost of losses

2) Position of nodes in WF

a. Locations of WTs

b. Possible locations of offshore collector platforms

c. Locations of PCC (only 1 PCC in the model)

3) Components

a. Specifications of cables: rated voltage (MV, HV), AC or DC, Rated power,

cost, r, x, b

b. Specifications of transformers: Rated power, low voltage, high voltage, cost,

56
r, etc.

c. Specification of platform: cost only

4) Electrical data

Table 3-1 Possible voltage level for collector and transmission network

MV (kV) HV (kV)

33, 66 132, 155, 220

3.2.3 Problem formulation

A different connection scheme between any wind turbines and offshore collector

platform could produce a new layout, thereby initially leading to a different cable cost on

impacted sections, and finally a different total cost. The objective of this framework is to find

out the optimal layout for the OWF in order to minimize the investment cost along with

operational cost, while satisfying a set of technical constraints, i.e. equipment capacity limit

and connection constraints.

a. Initial investment cost model


According to the functional classification of an OWF as mentioned before, an OWF

can be separated into 3 parts: collector system, offshore substation, and transmission system.

Accordingly, the investment cost can be classified into these 3 parts as well.

Cinvestment = CCollect + COS + CTrans (3.1)

In collector system, the investment cost mainly refers to the cost on inter-array medium

voltage cables of different types to connect wind turbines to substation. The cost of MV inter-

array cables depends on the distance and the cable type used for each section, which is shown

as below.

CCollect = CMV _ Cable =  uij ,ia ja ,k  Dij ,ia ja  Cunit ( k )


ia ja ij k
(3.2)
+    uij ,i' j', f ,k  Dij ,i' j'  Cunit ( k )
f ij ,i' j' k

57
where the first term in (3.2) represents the terminal section directly connecting the nearest

wind turbine ij on this feeder to the offshore substation ia ja , while the second term

represents the connection between any two wind turbines in the inter-array cable system. Both

of these two terms are dependent on the binary decision variables uij ,ia ja ,k , uij ,i j , f ,k , distance
' '

of the cable sections and unit price of corresponding cable type k , Cunit ( k ) . The appropriate

cable type k is determined by the power flow, to be more precise, number of wind turbines

in the downstream of the feeder (under the consideration of the fact that the ratings of WTs

has been fixed). The unit cost is a function of rated capacity, given by [23]:

Cunit ( k ) = CSI +  +   exp(   10−8  Sk ) (3.3)

where C SI represents the shipping and installation cost, which is approximately 90 U.S.
dollars/m [24].  ,  , are voltage level dependent cost parameters (for MV voltage =33kv),

shown in Table 3-2. Sk represents the rated capacity of the MV submarine cables.

Table 3-2 Cost parameter for MV cables


  

4.21  10 3 0.053  106 4.1  106

As for offshore substation, the investment cost is typically composed of main power

transformer, auxiliary protect and control equipment for the transformer and offshore platform.

However, in this chapter, only one substation is assumed to be capable of collecting all the

generated wind power from the OWF. Additionally, the transformer cost is usually estimated

by its rated capacity, which is supposed to be capable of withholding the fully operating power

of the wind farm. Therefore, the transformer cost should be fixed for a specific OWF with

fixed rated capacity, which will not be changed with the connection topology in the collector

58
system. In terms of the platform, the abovementioned is the same since the currently used cost

for platform is typically an estimate as a constant. To sum up, both transformer and platform

cost are fixed and have little impact on the final connection result so that these two will not

be considered as a term to assess the outcome. However, that does not imply any connection

scheme will yield out the same offshore substation cost, which is justified by the protection

devices like switchgears. Switchgears is located at every feeder terminal and employed to

monitor, control, and isolate any faulted devices. Literally, the number of switchgears

equipped is the same as the number of feeders: number of MV switchgears corresponding to

number of MV arrays, while number of HV switchgears corresponding to number of HV

cables. Hence, the offshore substation cost considered in this chapter is merely the cost of

auxiliary protection devices while the cost of transformer and platform are neglected, as

shown below.

COS = CSGMV + CSGHV (3.4)

CSGMV = N MV  UCSGMV (3.5)

CSGHV = N HV  UCSGHV (3.6)

where the offshore substation cost COS considered in this chapter is merely composed of MV

switchgear cost CSGMV and HV switchgear cost CSGHV . N MV indicates the number of MV

feeders for collector system, while N HV indicates the number of HV transmission for

transmission system. UCSGMV and UCSGHV are the unit cost for MV switchgear and HV

switchgear, which are 0.473 M U.S.$ and 0.53 M U.S.$, respectively. [69]

In terms of transmission system cost, the investment cost is mainly composed of the

HV transmission cable cost, which is also dependent on the WF capacity and the distance

from the offshore substation and PCC. The cost equation is similar to that of collector system,

as shown below.

59
CTrans = CHV _ Cable =  ui * *  Di * *  Cunit ( h ) (3.7)
a ja ,i j ,h a ja ,i j
ia ja h

where ui ,i* j* ,h
is the decision binary variable to determine whether the substation ia ja is
a ja

connected to PCC via HVAC transmission cable type h ; Di ,i* j*


is the distance between
a ja

PCC and offshore substation ia ja ; Cunit ( h ) is the unit price of corresponding cable type h .

The appropriate cable type h is chosen based on the rule of cost minimization while

satisfying the power transmission capability for the whole OWF. The unit cost is a function

of rated capacity, given by [24]:

Cunit ( h ) = CSI +  +   exp(   10−8  Sh ) (3.8)

Sh = 3U rated I rated (3.9)

where CSI represents the shipping and installation cost, which is approximately 90 U.S.

dollars/m [24].  ,  , are voltage level dependent cost parameters (for MV voltage =33kv),

shown in Table 3-3. S h represents the rated capacity of the HV submarine cables.

Table 3-3 Cost parameter for HV cables

Voltage Level (kV)   

220 3.181  106 0.11  106 1.16  10 6

b. Operational cost model


The operational cost mainly refers to the losses in both MV inter-array cables and HV

transmission cables, as shown below.

Cop = Closs _ MV + Closs _ HV (3.10)

60

Closs _ MV =   ( I sf )
2
 Rk  Dij ,ia ja
 ia ja f k
(3.11)

 (I )
2
+ fc  Rk  Dij ,i' j'   t  Closs
f ij ,i' j',c =iji' j' k 

Closs _ HV = ( I s )  Rh  Di
2
,i* j*
 t  Closs (3.12)
a ja
ia ja h

Cop _ MV and Cop _ HV are cable losses for MV collector system and HV transmission

system, respectively. Same as investment cost for MV cables, the loss in collector system is

classified into terminal connection and WT interconnection, therefore, I sf and I fc are the

corresponding current of these two connections. The losses are classified into 2 parts simply

because of the parameter definition and the losses, in nature, are the same. I s represents the

current flow on transmission cable. Rk and Rh are unit resistance for MV cable type k

and h . t represents the available time and Closs is the cost of unserved energy.

Besides operational losses in cable, there is supposed to be losses in the offshore

substation, to be more specific, the main power transformer as well, which, however, has been

neglected. The reason is same as mentioned in investment cost for substation, that only one

substation is established in this topology catering to medium-size OWF. As the only power

transformer is responsible for all the generated wind power, its capacity has been fixed to the

rated capacity of the OWF. Therefore, the losses in the transformer should be approximately

the same since the transformer type has been fixed, which has justified why the loss in the

transformer is not included in this model.

c. Mathematical formulation
Considering the above cost model, the electric system layout optimization framework

can be formulated as below.

Objective: min Cinvestment + Cop (3.13)

61
Subject to:

Sij ,i' j'  uij ,i' j', f ,k  Sk


Sij ,ia ja  uij ,ia ja f ,k  Sk (3.14)
Sia ja ,i* j*  uia ja ,i* j* h  Sh

pia ja  POS (3.15)

k K
(3.16)
h H

N f  N max
f (3.17)

max
nWT , f  NWT ,f (3.18)

fi  f j =  , fi , f j  F (3.19)

Initially, the objective is composed of investment cost and operational cost, both of

which have been described in the previous sections. Sij ,i' j' , Sij ,ia ja and Sia ja ,i* j* are the

apparent power flow on MV array cables between wind turbines ij , i' j' and between wind

turbine ij , ia ja ,and on transmission cable between offshore substation ia ja , and onshore

PCC i* j * , all of which are not supposed to exceed the rated capacity of corresponding cable

type S k or S h . Namely, the proper cable size for both MV array cables and HV transmission

cable depends on the power. pia ja refers to the 2-dimensional coordinate for offshore substation

and its location is confined to a feasible region POS . Equation (3.l6) indicates the

standardization of cable type. The available cross-sectional area of MV cable K(mm2) = [50,

70, 95, 120, 150, 185, 240, 300, 400, 500, 630, 800, 1000], while that of HV cable H(mm2) =

[150, 185, 240, 300, 400, 500, 630, 800, 1000,1200, 1400, 1600, 1800]. N f and N max
f

represents the actual number of feeders accessible to the substation and the maximum number

of feeders permitted to connect with the substation. Therefore, equation (3.17) indicates the

limit on total number of feeders connected to the substation. Similarly, equation (3.18) limits

62
the total number of wind turbines located on the feeder based on the permissible voltage drop

and cable capacity.

Besides the above constraints relating equipment selection to system security, the

technical constraints between wind turbines, inter-array feeders and offshore substation are

detailed from equation (3.20) to (3.25).

The total number of feeders equipped at substation should not exceed the maximum

max
permitted number of feeders N f .

0   z ifa ja  N max
f ,ia ja (3.20)
f

Wind turbine i is clustered to feeder f , only if feeder f has already been chosen;

otherwise no wind turbine can be connected to feeder f . Meanwhile, the number of wind

turbines connected to a feeder f cannot exceed NWT


max
,f .

z ifa ja   xij , f  NWT


max ia ja
,f  z f (3.21)
ij

There is a feeder f through wind turbine ij and i' j' , only if these two wind turbines

are both clustered to feeder f . If wind turbine ij and i' j' are connected, there is only one

cable type between these two wind turbines.

2 uij ,i' j', f ,k  xij , f + xi' j', f (3.22)


k

u
k
ij ,i' j', f ,k 1 (3.23)

To avoid crossing, one wind turbine must be clustered to one and only one feeder.

x
f
ij , f =1 (3.24)

No. of inter-array cables connected to a collector platform is determined by both the

number of WTs connected to it and the number of interconnections between them, beneficial

to eliminate closed loop between branches or ring topology, however, still facilitate a full
63
optimization of the topology considering string and mixed star and string topology.

z
f
f = N −    uij ,i ' j ', f ,k
ij ,i ' j ' f F k
(3.25)

It is necessary that no wind turbine should be isolated so that there must be one and

only one cable from that wind turbine,

  u
f i' j'N wt \{ ij } k
ij ,i' j', f ,k =1 (3.26)

3.3 Electric system layout optimization framework

The mathematical model presented in the last section justify the optimization of electric

system layout for an OWF is a mixed-integer nonlinear problem. The complexity is

proportional to the size of the wind farm, i.e. the number of WTs since the number of possible

connections increases exponentially with the number of wind turbines, which also enlarges

the overall search region of the problem. As for a small OWF with several wind turbines, all

possible topologies combined with available transmission technology can be enumerated

using brute force. However, the enumeration approach cannot be applied in the case of

medium- and even large-size offshore wind farms since the involved decision variables and

number of devices to be determined increases with the size of the wind farm. To obtain an

optimal topology as well as picking out the most appropriate devices, the optimization

framework is developed, which is detailed in this section.

3.3.1 Data input and pre-processing

Initially, any necessary user-defined or fixed input are defined to specify specific case.

Table 3-4 covers the set of all the necessary data input.

The data input has been categorized into 4 classes: economic data, typically used as

constant in the objective functions; WF specifications, where the equivalent full-load period

is used in loss evaluation overall the wind farm; cable specifications; and locations for all
64
nodes in the OWF. However, the location for offshore substation should be noted. As

mentioned in 3.1, there are two general layouts considered for the OWF, where the difference

between these two layouts lies in the existence of pre-specified offshore substation. If there

has been an offshore substation, the location of the offshore substation is specified, and the

electric layout optimization merely matters the inter-array connection. If the offshore

substation is not pre-specified, a permissible region is proposed regarding the preparation

process like environmental analysis, seabed survey.

Table 3-4 Data input for the optimization framework


Economic data WF specifications

• Expected lifetime • WT related: rated capacity, no. of WTs


• Energy price • Electric specifications: collector
system voltage level, transmission
• Interest rate
system voltage level
• Equivalent full-load period
Cable specifications Location

• Cross-sectional area (for both MV and • Geometric coordinate for point of


HV cables) common coupling (PCC)
• Current-carrying capability (for both • Geometric coordinates for OWF
MV and HV cables)
• Permissible region for offshore
• Unit cable cost (for both MV and HV substation or specified selection for
cables) offshore substation
• Resistance per length (for both MV
and HV cables)

After specifications of the case, pre-processing of the set of data will be executed, and

the result is shown in Table 3-5. Initially, as for layout 2, without specifications of offshore

substation, a set of candidate centers can be proposed based on the geometric coordinate for

all involved nodes in OWF, which is elaborated in the next section. For layout 1, the location

for offshore substation is pre-specified, without any need to further optimize. According to

the coordinates provided, a distance matrix can be obtained between any wind turbines in the

65
OWF. Meanwhile, the distance from the proposed candidate offshore substation to PCC and

each individual wind turbines can be worked out, from which several WTs with the least

distance can be screened out as candidate feeders for corresponding candidate offshore

substation. The wind turbine rating along with collector system voltage level can be used to

derive current flow for each wind turbine under full-loaded condition. Therefore, the

maximum number of wind turbines for each cable type capable to connect with can be derived.

The same process can be executed for transmission cables. The maximum number of wind

turbines can be connected to one feeder is decided by the maximum cable size. As the number

of wind turbines is already known, the minimum required number of feeders can be worked

out as well.

Table 3-5 Data preprocessing for the optimization framework


Center Distance

• Geometric coordinate for candidate • Distance matrix including distance


centres (for layout 2) between any two wind turbines (for
collector system)
• Possible feeders for each candidate
centres (for layout 1 and 2) and pre- • Distance from the PCC to candidate
specified centre (for layout 2) centres (layout 2) and pre-specified
centre (layout 1)

Cable capability Feeder

• No. of wind turbines can be connected • Maximum number of wind turbines


capable to be connected
• for each MV cable type
• Minimum number of feeders required
• No. of wind turbines can be connected
for the specified wind farm
• for each HV cable type
• No. of HV cables and corresponding
cable type to withstand the wind farm
rated capacity

3.3.2 Discretization of Candidate Centre Location

Distance is the most important variables in the electrical layout optimization problem,

which is necessary to evaluate the cable cost as well as losses in the layout. The distance

between any two wind turbines, or between wind turbine and offshore substation, or between
66
substation and PCC are all involved in this model and represented by Dij ,i' j' , Dij ,ia ja ,

Di ,i* j*
, respectively. As for distance between wind turbines, the calculation is actually
a ja

simple as below since the 2-dimensional geometric coordinate of all wind turbines are

predefined. Literally, the distance matrix between wind turbines is fixed once the wind farm

has been specified.

Dij ,i' j' = pij − pi' j' (3.27)

where pij and pi' j' are vectors indicating the 2-dimensional cartesian coordinate for any

two wind turbines; Dij ,i' j' refers to the Euclidean distance between these two wind turbines.

However, when the distance involves the offshore substation, the case is not the same.

The location of offshore substation is denoted by pia ja , which is also an optimized variable

in the model. Following the same procedure, Dij ,ia ja , Di ,i* j*


can be mathematically denoted
a ja

as below.

(x ) +(y )
2 2
Dij ,ia ja = pia ja − pij = ia ja − xij ia ja − yij (3.28)

Di * * = pia ja − pi* j* = ( xia ja − xi* j* ) 2 + ( yia ja − yi* j* ) 2 (3.29)


a ja , i j

where pia ja and pi* j* are vectors indicating the 2-dimensional cartesian coordinate

(xia ja )( )
, yia ja , xi* j* , yi* j* for offshore substation and PCC, respectively; Dij ,ia ja ,and Di
a ja ,i* j*

refer to the Euclidean distance for the terminal section connected to substation, and

transmission cable length. Equation (3.15) confines a permissible coordinate region for

substation, implying the sitting of the offshore substation is a continuous problem. From the

above equation (3.28) and (3.29), it can be concluded that the term of distance has been

already nonlinear to the location variable (x ia ja )


, yia ja , let alone substituting them into

equation (3.2), (3.7), (3,11), (3.12) and being multiplied with a set of binary variables. To sum

67
up, the sitting of offshore substation in a continuous region contributes to the nonlinearity and

nonconvexity of problem. To simplify and linearize the model, discretising the permissible

region into a set of candidate choices is initially carried out. The continuous sitting problem

of offshore substation has been turned into a discrete optimal location selection problem. With

a specified cartesian coordinate (xia ja )


, yia ja , the distance from the specific substation to both

wind turbines and PCC can be worked out directly from the above equation (3.28) and (3.29),

which is also part of data pre-processing step.

Cell centre

Figure 3.5 Raster map of a permissible region


Rasterization method is proposed and applied to transform the continuous offshore

substation sitting problem to a discrete sitting problem. The permissible region is rasterized

into a raster map with nr rows, nc columns of cells. Each of the cells is represented by the

coordinate of its corresponding centre. Fig. 3.5. is shown as an illustrative example for the

rasterization method. The shaded region represents the permissible region for offshore

substation, which has been rasterized into 21 cells with 7 columns and 3 rows. Each of the

cells is denoted by the center node coordinate. After the rasterization, the continuous feasible

region has been discretized into the 21 separate areas, and the choice among the 21 candidate

centers is determined by the set of binary variables bia ja .

The raster size is chosen dependent on the size of the permissible region and an

appropriate size could be able to gain the trade-off between accuracy and computational
68
efforts. It is obvious that the smaller the raster size is set, the larger nr and nc , thereby the

larger number of binary variables bia ja required. Accordingly, the result will be more

accurate at the expense of longer execution time and heavier computational efforts.

The introduction of discrete placement of offshore substation brings in a set of binary

variables bia ja , where bia ja = 1 indicates the offshore substation is placed at the

corresponding coordinate (x ia ja )
, yia ja ; otherwise, this place is not chosen. As for the

topology considered in this chapter, only one substation is considered. Therefore, some rules

have been set regarding to the new variable bia ja .

b ia ja
ia ja =1 (3.30)

bia ja   z ifa ja  N max


f  bia ja (3.31)
f

Equation (3.30) rules out that only one candidate substation will be chosen for any

topology, which is in accordance with the assumption presented. Equation (3.31) indicates

that if substation ia ja is not chosen, the corresponding feeder for this substation will not

exist, either. If this offshore substation has been chosen, i.e. bia ja = 1 the total number of

feeders should be less than the maximum number of feeders permissible to connect with one

max
substation N f .

It takes a lot of computational effort to directly raster the permissible region under a

relatively high resolutions of raster map. Instead, we can imitate the adjustment of microscope,

with initially coarse adjustment and followed by fine adjustment. Similarly, we could obtain

a fine enough result for the offshore substation sitting, based on which we can narrow the

search region to a much smaller one with area of 2 times the row raster size for rows, and 2

times the column raster size for the column. This coarse-fine adjustment helps to accelerate

the speed to locate the best placement for offshore substation and relieve the nonlinearity led

69
by continuous offshore substation sitting problem.

Take Fig. 3.5 as an illustrative example for the two-step-coarse-fine adjustment. If the

central point (red-colored node in Fig. 3.6) in Fig. 3.5. is chosen as the optimum result for

offshore substation sitting in the coarse adjustment step, then the red-shaded area is the

narrower search region to be further rasterized in the fine adjustment step, which is illustrated

in Fig. 3.6. The search region is centered at exactly the optimum node, result of the coarse

adjustment, with

Figure 3.6 Illustrative example for coarse-fine rasterization for a permissible


region

length of 2 times the original column raster map size and width of 2 times the original row

raster map size. In the fine adjustment step, a much smaller raster map size can be applied to

rasterize the narrower search region, thus making it much easier to get an accurate result.

The raster map method is only applied for layout 2, with no specified offshore

substation. As for layout 1, this step can be skipped since the exact coordinate of offshore

substation has been specified, instead of a permissible region in layout 2.

3.3.3 Clustering wind turbines into feeders

After rasterizing the permissible region for offshore substation into discrete center node

representing each raster cell, the candidate center can be individually represented by its

70
coordinate. The distance from each candidate center to every wind turbine in the OWF can be

i j
worked out as well. z fa a , shown in the nomenclature section, indicates whether feeder f

is used to connect with offshore substation ia ja . In this framework, feeder f can be

deduced for a fixed offshore substation ia ja by screening out the first several number of

wind turbines closest to ia ja . Literally, a set of candidate feeders can be picked out according

to its distance to corresponding offshore substation ia ja , thereby formulating a matrix with

number of rows equal to number of raster cells (also substations) and number of columns

max
equal to the maximum feeder number N f . The method of defining closest WTs as the

feeders can be illustrated through the connection scheme and cable cost. As for the connection

scheme, the wind turbines are assumed to be connected in a radial topology. The offshore

substation is the origin, to be more specific, functioning as the “root” of all the connection,

while each of the feeder is similar to the “branch” and increasingly diverging as stretching out

to the far-end wind turbine. Therefore, the wind turbines closest to the connection, located at

upstream of the feeder, are inclined to represent the feeder instead of any other wind turbines

in the string. In terms of cable cost, wind turbines located in the upstream, is responsible to

gather all the generated power from the downstream wind turbines, thereby requiring a cable

type with larger cross-sectional area, and obviously larger unit cost. The method of reduction

to absurdity can be applied here to justify that the closest WT is supposed to be directly

connected to the offshore substation. If another wind turbine, a little bit further from the

offshore substation is used to connect with the offshore substation, there must be a more

optimum connection scheme with less cost, by replacing that further wind turbine with a

closer wind turbine. The justification for defining the set of closest wind turbines as feeders

has been justified by the above two argumentations.

Candidate feeders can be deduced after comparing distance from each wind turbine to

specific candidate centres. The number of candidate feeders is limited by the maximum
71
max
number of feeders permitted to connect with offshore substation, N f . Thereby, in this wind

max
turbine layer, the problem can be defined as clustering N wind turbines into N f feeder

strings. The binary variables xij , f is introduced to denote the clustering or not of wind

turbine ij to a feeder f .

i j
According to the binary variable z fa a and xij , f , the wind turbine can be clustered into

feeders, followed by the submarine cable connection scheme. Focusing on each wind turbine

on individual feeders, the objective is to connect all the wind turbines in a manner of cost

minimization, which can be regarded as a variation of the minimum spanning tree problem.

Prim’s algorithm is typically applied to solve the above-mentioned problem with a connected,

weighted, and undirected graph [107]. It is a greedy algorithm to guarantee that the tree

topology is updated at each turn with a new connection with least cost, thereby achieving an

overall optimization of cost. In this model, distance is regarded as the weight for each branch

since both cable investment cost and cable loss cost is linear with cable length of each cable

section. Each “branch” in the tree represents the real connection between any two wind

turbines. The MSTs can be not only formulated in a pure string connection, but also star

connection. As a whole, the MSTs strength out in a radial manner. Prim’s algorithm is applied

in this model to locally optimize the submarine cable connection scheme, which is regarded

as a reasonable simplification for introducing large number of variables uij ,i' j ' , f ,k and uij ,ia ja ,k .

The result, MSTs, provides the information regarding connection scheme between wind

turbines on individual feeders. Additionally, based on the number of wind turbines located in

the downstream of each point, appropriate cable type can be determined for each cable section

based on its current-carrying capacity. Accordingly, the inter-array cable cost for the overall

collector system can be worked out. After adding the switchgear cost and HV transmission

cable cost, the investment cost can be obtained. Besides, the power flow through each cable

72
section can be estimated based on the wind turbine ratings and the topology. The operational

cost, regarded as cost of losses, can be obtained as well. Summation of investment cost and

operational cost is taken to evaluate the objective fitness.

3.3.4 Avoiding cable crossing

Each feeder has been connected and formulated as MSTs by prim’s algorithm in the

last section. However, crossings between feeders, which is not acceptable for electrical system

layout for an OWF, are still likely to exist and required to be eliminated.

Analytic geometric can be applied here to detect any crossings in the topology, shown

as below. The cartesian coordinate of any wind turbine has been pre-defined and the

connection scheme all over the collection system has been specified based on the preceding

steps. Therefore, the cable crossing problem for any connection within a real WF topology

can be abstracted to the determination for crossing between any line segments in a 2-

dimensional cartesian coordinate. For instance, there is two feeders connecting WT ij and

i' j' , pq and p' q' . The discriminant, shown below, determines whether the line segment

( ij , i' j' ) crosses with line segment ( pq , p' q' ).

( y pq − fij ( x pq ))  ( y p' q' − fij ( x p' q' ))  0 (3.32)

( yij − f pq ( xij ))  ( yi' j' − f pq ( xi' j' ))  0 (3.33)

where xij , yij , xi' j' , yi' j' , x pq , y pq , x p' q' , y p' q' are the x or y coordinate of the four

endpoint of the two segments. f ij and f pq , respectively, indicates the linear equation

passing through these two line segments. In a nutshell, it is essentially substituting cartesian

coordinate of the two endpoints into another line function and comparing the result with its

original y-coordinate. If there is one endpoint with larger result than its original value but one

with less result than its original value, it implies there is a crossing between these two

segments, literally, unacceptable cable crossing in the electric layout. Fig. 3.7 gives an

73
intuitive and visual justification of the above determinant.

If any cable crossing detected between feeders, the penalty function will be applied to

the specific crossed sections. The penalty function is positively correlated with number of

crossings in the specific topology. After adding the penalty function, the fitness evaluation

has been updated as below.

UC ( ij,i' j' ) = PC   CC ( ij,i' j' ) (3.34)


f ij ,i' j'

Figure 3.7 Illustrative display of cable crossing determinant


Penalty of cable crossing is only applied to the cable investment cost instead of

covering operational cost as well. Therefore, collector system cost model is the only impacted

factor, where CC represents the function to calculate the investment cable cost between wind

turbines ij and i' j' , while UC represents the updated cable cost after considering cable

crossing. PC is a constant larger than 1, representing the penalty multiplier applied for crossed

cables. If two cable sections have been determined as crossing, their original investment cost
74
will be multiplied by this penalty factor, greater than 1, thereby leading to a larger investment

cable cost. Obviously, the connection topology with crossing is not the optimum topology and

there must be an alternative with better fitness evaluation. Therefore, the algorithm will screen

out the costlier topology with penalty for crossing and eventually evolve to an optimum

topology without any crossings between the submarine cables.

3.3.5 Natural Aggregation Algorithm

The electrical system layout optimization of an OWF is formulated as a mixed integer

nonlinear problem. As a NP-hard problem, classical optimization method usually imposes too

many requirements for the model to follow so that the proposed layout optimization problem

tends to be solved by heuristic optimization method, which can be easily adapted in this

difficult model. Natural aggregation algorithm, a newly proposed evolutionary algorithm in

our lab group, is applied to solve the problem, which hereinafter is referred to NAA for

convenience and saving space. As its name implies, NAA is inspired by the self-aggregation

behaviors of group-living animals [108]. According to numerous biological researches,

group-living animals tend to aggregate into several groups because living in such groups is

beneficial for the animals to share resources like food and shelters. However, if the living

group is overcrowded, more competition for the limited resources will occur among the group,

which leads to the group less attractive to the swarm members. Researchers have found out

living-group animals have the intelligence to decide whether to join or leave the group.

Therefore, the formation of a group is completely dependent on the decision-making behavior

of individual animals in and outside of the group. The autonomously individual decision-

making ability can be imitated in the optimization process towards the global optima.

Evolutionary algorithm often has challenge through the optimization process on

determining whether to exploit deeper around the currently best result or to explore in a wider

area to discover other fair results. Literally, a balance between exploitation and exploration

needs to be reached, which can be solved by mimicking the autonomous aggregation


75
behaviors of group-living animals. Similar to other EA, NAA maintains a population, which

can be further divided into multiple sub-population, denoted as “shelter”. Each individual in

the population has its own determination to stay or leave the “shelter”, so that the sub-

population is fully self-optimized by a stochastic migration model. Individuals located in the

“shelters” are responsible for exploiting around the good results, while individuals not

belonging to any “shelters” are playing the roles of exploring wider area to avoid being

trapped to local optima. The abovementioned procedures are adopting two different search

strategy for individuals located in and outside of the group. Accordingly, NAA can achieve a

good balance between exploitation and exploration.

Like other population based EAs, a population of individuals is maintained in each turn

in NAA. Each individual represents a possible solution to the problem. For a general

mathematical model shown below,

objective min f(x)


s.t. bli  xi  bui i = 1, 2 ,..., D

 
where f ( x ) is the objective function, x = x1 , x2 ,..., xD represents the optimized variable

with a dimension of D; bli and bui are the lower and upper bound set for the i th variable.

This model aims to find out a solution vector x to minimize the objective function f ( x )

under the bounded constraints. For a population of N individuals in NAA, each individual can

be represented in a D-dimension vector. The i th individual in the population, Ri , can be

denoted as below:

Ri =  xi1 , xi2 ,..., xiD  i = 1,..., N

In our electric layout optimization model, the binary variables z fa


i ja
and xij , f

correspond to the vector x , and the dimensionality of the model, referred as D, is dependent

on the total number of variables in our problem, which is highly related to number of wind

76
i j
turbines in the OWF. Each Ri is a permutation of z fa a and xij , f within the bounded range

and yields out its own objective values. At each generation, a stochastic migration module is

initially applied to re-distribute the population, which mimics the self-organisation behaviour

of each the group-living animals. Afterwards, two different search modules, similar to the

mutation step in conventional EAs, is adopted to update the individuals to realize the

exploitation and exploration towards better solution. The better solution represents a better

distribution of resources like food or shelter for the group-living animals. All the updated

individuals will be reassessed and sorted in a descending order by their fitness values,

according to which the shelters can be updated. The process iterates until satisfying the

termination conditions, including objective values falling into acceptable region and reaching

the predefined maximum generation. The termination mimics the fact that all the groups of

this species have arrived at an equilibrium when the total resources have been perfectly

distributed among them.

3.3.6 Flowchart for the layout optimization framework

All the above-mentioned methodologies are combined to formulate as an electrical

system layout optimization framework to achieve the cost minimization topology. The

framework can be separated according to functionality of different parts in OWF into offshore

substation selection, wind turbine clustered into feeders and submarine cable type

determination. Initially, in the preparation step, all the necessary data related to the electrical

system of an OWF has been input and pre-processed, through which the appropriate raster

size for the first coarse adjustment can be determined. Then, the continuous permissible region

of offshore substation has been discretized by the rough raster map size and denoted by a set

of center coordinates using rasterization method. NAA is utilized in this framework to find

the optimum wind turbine cluster. The initial generation have been randomly generated but

still complying the set of constraints presented in this section. The information regarding how

77
to cluster the wind turbines into different feeders has been stored in the species. How the wind

turbines are connected in a feeder is similar to the minimum spanning tree problem, which is

solved by prim’s algorithm under the rules of minimum distance. The optimal placement and

type determination for submarine inter-array cables are determined, upon to which crossing

determinant has been carried out for each cable sections and then formulate the total cable

cost considering penalty cost. Similarly, losses evaluation can be estimated with specified

cable type. To sum up all the involved cost, all the species can be evaluated and ranked upon

their fitness value. Later, the algorithms converge to a global optimum under coarse raster

map size. If more accurate result required, smaller raster map size is applied to the obtained

new search region to carry out the fine-tuning rasterization.

The process of the electrical system layout optimization framework is described as

below.

1) Obtain all the necessary data regarding the electrical system, including economic
data, wind farm specifications, location of all the nodes, available cable
specifications

2) Preprocess the input and determine the appropriate coarse raster map size
according to the permissible region for offshore substation

3) Raster the continuous offshore substation region into discrete raster cells and
calculate the number of variables required for the problem

4) Initialize NAA algorithm and randomly generate the initial population complying
the proposed constraints.

5) Using Prim’s algorithm to figure out the connection between wind turbines on
one feeder

6) Fill in the power matrix S

7) Determine whether there is any cable crossing in the population

8) Get the appropriate cable type for each section of feeders

9) Work out the cable cost (including penalty cost for cable crossing) as well as
operational cost (i.e., losses)
78
10) Add cost of additional devices like switchgears and transmission cable
cost/losses cost and obtain the final fitness value for each individual in the
population

11) Judge whether terminal condition set for NAA has been satisfied, (usually set as
number of generation), if satisfied, come to step 12), otherwise, perform the
generation of new population and go back to step 5).

Figure 3.8 Flow chart of the layout optimization framework


79
12) Judge whether fine adjustment of raster map is required, if yes, update the raster
map size and search region for offshore substation. Then go back to step 3) to
start fine coarse rasterization method; if no, stop the framework and output the
optimal design.

As mentioned in the section 3.1, there are two types of topologies considered in this

framework, where the difference lies in the existence of offshore substation, i.e., layout 1,

with specified substation location, while layout 2, with offshore substation confined in a

continuous region. The above procedure is formulated for layout 2. As for layout 1, the steps

regarding coarse and fine rasterization can be skipped, while all the other steps remain the

same as layout 2.

3.4 Case studies

The electric layout optimization framework has been verified on a specific case in this

section. The case has been modified to cater to the two different layout method presented in

this chapter, one with a pre-defined substation, while one with no pre-defined substation but

a specified point of common coupling (PCC) to connect with the main grid. The reminder of

this section is organized as follows. Initially, a brief case description regarding the offshore

wind farm has been introduced in the first subsection. Then, according to a set of input data,

proper pre-processing is presented as mentioned in section 3.3.1. Finally, the optimization

result for layout 1 and layout 2 is, respectively, drawn in subsection 3 and 4.

3.4.1 Case description

The proposed optimization framework is applied to a particular OWF to validate its

effect. The OWF is composed of 40 wind turbines, each with a rated generating capacity of

6MW, which, thereby, has a total rated generating capacity of 240MW. As mentioned in the

hypothesis, the optimal placement of wind turbines is another research topic and has been

determined by carrying out the macro-sitting and micro-sitting research. Accordingly,

locations of all the wind turbines are already known. The turbine layout of this virtual OWF
80
is represented in Fig. 3.9, where each of the black dots represents one individual wind turbine.

The rectangular site of OWF covers an area of about 28 square kilometers. According to the

figure, 40 wind turbines are distributed in 8 rows and each of the rows has 5 wind turbines.

Both row and column are separated by a mean distance of about 1 km, approximately equal

to seven times of the length of wind turbine diameter, which is also the common practice to

sit between any pairs of wind turbines.

The red dot remarked in Fig. 3.9 represents the location of the specified center, which

has a geometric coordinate of (13, 8.2). This is designed to cater to layout 1, with pre-specified

offshore substation to collect the generated wind power. This coordinate of the substation is

just one example used in the case study. It does not imply the substation is confined to this

specific coordinate. Instead, it can be any user-defined values catering to any specific case.

The point of common coupling (PCC) is assumed to be located at the coordinates (-20, -20),

which is designated for layout 2, without a fixed offshore substation. However, PCC is not

marked in the layout of OWF to improve the readability of the layout figure and mainly focus

within wind farm region since the distance from the PCC to the OWF is much larger than that

between wind turbines.

81
Figure 3.9 Wind turbine sitting in the OWF
The collector system voltage level is assumed to be 33kV, while it is assumed to be 220

kV for the transmission system. Reliability is not covered in this chapter so that all the devices

involved such as wind turbines and MV or HV cables are assumed to be 100% available. The

price of non-served energy is set as 80$/MWh, used to evaluate the cost of losses.

3.4.2 Input data and pre-processing for the optimization

This framework, by nature, can be divided into three steps, substation locating, wind

turbine clustered into feeders, submarine cable connections. With the consideration of the

wind farm cable specifications, the clusters of wind turbines in one feeder, the type of

equipment used in MV collector system, the connection scheme between wind turbines can

be simplified and optimized based on some pre-processing step of the input data.

Initially, cable parameters are a group of input required to be processed. As for medium

voltage submarine cables, a set of three-core copper conductor medium voltage submarine

cables are utilised, whose cross-sectional areas vary between 50mm2 and 1000mm2. The cable

parameters are obtained from a manufacturer [69], and the unit cost is obtained by substituting
82
the parameters into the cost equation (3.3) listed in section 3.2.4. All parameters with regards

to the submarine MV cables are shown in Table 3-6.

Table 3-6 MV cable parameters and unit cost


Cross- Conductor Current
section carrying Voltage Sk No. of Unit price
resistance
area capacity (kV) (MW) WTs (K US$/km)
(  / km )
(mm2) (A)
50 0.641 185 33 10.57 1 175.9401
70 0.443 215 33 12.29 2 181.8912
95 0.320 255 33 14.58 2 190.5057
120 0.253 300 33 17.15 2 201.213
150 0.206 335 33 19.15 3 210.3576
185 0.164 370 33 21.15 3 220.2839
240 0.125 430 33 24.58 4 239.3144
300 0.100 480 33 27.44 4 257.3502
400 0.0778 550 33 31.44 5 286.4283
500 0.0605 630 33 36.01 6 326.0597
630 0.0469 745 33 42.58 7 397.7658
800 0.0367 850 33 48.58 8 482.446
1000 0.0291 950 33 54.30 9 584.9665
The rated electric generating capacity of WT in this case is assumed to be 6MW. With

the consideration of MV voltage set as 33kV, the current from a wind turbine to such a

collector system, is calculated as below:

6 MW  ( 3  33kV )  104 A

Namely, a fully operated wind turbine with the same ratings will inject 104A to the

feeder which it is connected with. Based on that current flow, it is able to deduct the number

of wind turbines that can be connected for each MV cable type. For instance, the current

carrying capacity for MV submarine cable with cross-sectional area of 1000 mm2 is 950A as

shown in Table 3-6. The number of wind turbines able to be connected with this cable can be

calculated as below:

950 A / 104 A  9 wind turbines

Therefore, the maximum allowable current for the cross-sectional area of 1000mm2 is

83
greater than the sum of the currents from 9 wind turbines, under full load conditions with

6MW capacity. The same procedure can be applied to any other cable types to obtain the

number of wind turbines connected for the corresponding cross-sectional areas, which is also

included in the table. As shown in the table, the choice of cross-sectional areas for MV

submarine cables is in fact, dependent on the number of wind turbines connected to it. A

higher number of wind turbines connected on one feeder requires a cable type with larger

cross-sectional area in the upstream of the feeder. Some of the MV cable types yield out the

same number of connected WTs. To find out an optimal result with least cost, cheaper cable

with less cross-sectional area will obviously, be chosen among the set of cables when they

share the same number of connected WTs. Therefore, the bolded cable types listed in the table

are chosen as the possible cable types and actually utilised in this case. For example, MV

cable types with cross-sectional area of 70, 95 and 120 mm2 can withhold, at most, 2 wind

turbines in the downstream. Obviously, with the consideration of connecting 2 wind turbines,

cable type with 70 mm2 cross-sectional area will be used due to its cheapest unit price.

Meanwhile, since 1000 mm2 is the maximum cable cross-sectional area, it implies that

the maximum number of wind turbines connected with one feeder is 9. If the number of wind

turbines on one feeder is greater than 9, then there is no submarine cable type suitable for this

case and it must be separated into 2 feeders. With the consideration of totally 40 wind turbines

in this case, the minimum number of feeders required to be equipped to connect all of the 40

wind turbines to the substation can be obtained:

40  9  5 feeders

The maximum number of feeders considered in this case is set to be 10, 2 times of the

minimum number of feeders. This setting is reasonable and can be verified in the simulation

result. Therefore, 10 wind turbines which are closest to the substation will be screened out to

represent the candidate feeder.

Similarly, in terms of HV transmission submarine cables, a set of three-core copper

84
conductor high voltage submarine cables are utilised, whose cross-sectional areas vary

between 150mm2 and 1800mm2. The cable parameters are obtained from a manufacturer [69],

and the unit cost is obtained by substituting the parameters into the cost equation (3.7) listed

in section 3.2.4. All parameters with regards to the submarine HV cables are shown in Table

3-7.

Table 3-7 HVAC cable parameters and unit cost


Cross- Conductor Current
section
resistance carrying
Voltage Sk Unit price
area (kV) (MW) (K US$/km)
(  / km ) capacity (A)
(mm2)
150 0.206 335 220 127.6521 415.6765
185 0.164 370 220 140.9889 422.8669
240 0.125 430 220 163.852 438.1024
300 0.100 480 220 182.9046 454.2783
400 0.0778 550 220 209.5781 483.8602
500 0.0605 630 220 240.0622 542.2504
630 0.0469 745 220 283.8831 635.8967
800 0.0367 850 220 323.8935 791.3435
1000 0.0291 950 220 361.9986 1024.044
1200 0.0220 1030 220 392.4827 1300.348
1400 0.0167 1118 220 426.0152 1741.417
1600 0.0104 1195 220 455.3562 2296.357
1800 0.0043 1271 220 484.316 3064.373
2000 0.0010 1330 220 506.798 3866.375
The rated wind farm capacity is 240MW as specified in the previous section. Referring

to Table 3-7, HVAC cable type with cross-sectional area of 500 mm2 is appropriate to transmit

the collected power from offshore substation to onshore PCC since its rated capacity is just

larger than 240MW. Therefore, the number of HVAC cable to transmit the power is one in

this case study.

All of the above-mentioned cost equations are referenced from public-accessible data,

obtained from several publications. However, any other user-defined cost data can be put into

this model to cater to different market or geographical conditions.

According to the coordinate of all the wind turbines, we can find out that the wind farm

85
covers a region with x-coordinate from 8 to 15, and y-coordinate from 4.9 to 11.5. The PCC

is located at (-20, -20). We assume the offshore substation can be located at anywhere between

PCC and OWF. Therefore, the permissible region to locate the offshore substation is confined

to a region with x coordinate from -20 to 15, and y coordinate from -20 to 11.5.

All the experiments are carried out by MATLAB version 2019a on a 4-core, 64-bit

DELL pc with Intel Core i5-2400 CPU and RAM 4 Giga byte.

3.4.3 Optimization result and analysis for layout 1

The optimization on Layout 1 has been carried out in this subsection. As mentioned

earlier in this section, the offshore substation has been pre-specified and fixed at coordinate

(13, 8.2), so that the optimization of offshore substation sitting and further transmission to

onshore grid are skipped in this case. The inner system topology optimization, along with

optimal cable sitting between wind turbines is the main concern in this case. Literally, the

main concern is about how to connect these 40 wind turbines to the substation with high cost

performance while satisfying a set of operational constraints. To better illustrate the advantage

of the proposed algorithm, a layout optimization without consideration of system operational

losses is implemented here as a benchmark for a comparison with the proposed

comprehensive optimization framework.

a. Case 1: Layout 1 without the consideration of operational cost


Investment cost of all the involved devices is the only benchmark on objective in this

section. The operational cost is neglected to observe the final result of electric system

topology of the collector system, which is thus regarded as a benchmark system topology.

Initially, Natural Aggregation Algorithm, as an evolutionary algorithm, is utilized as the solver

to solve the layout optimization problem, where the optimization process mimics the

evolution of insect crawling track in order to evolve to the optima. The evolution of the best

86
fitness value throughout the overall optimization process is presented in Fig. 3.10, where the

objective function value has been converged and stabilized to the optimal result at the end of

the generation. 3000th generation can be regarded as the turning point of objective functions,

before which there is relatively large variation in the fitness values, while after which the

fitness value has slightly changed. The evolution finally evolves to the optima at

approximately 8000th generation.

The optimal electric layout topology within the offshore wind farm (layout 1) is shown

in Fig. 3.11. As layout 1 has no concern about how to transmit the generated power from

substation to shore, the topology presented in the figure only covers the internal part within

the offshore wind farm. It can be observed that the connections between wind turbines, and

connections between wind turbines and offshore substation have no crossings between each

other, which justify the effectiveness of avoiding cable crossing criterion.

As calculated in the previous section, the number of feeders is confined between 5 and

10 as a constraint to withhold this 40-wind-turbine wind farm. The total number of feeders in

the optimal topology is 6, satisfying the constraint. Each of the feeders is represented in

different color to differentiate one from another. The wind turbines nearest to the substation

located at each feeder are chosen from 10 candidate feeders and used to denote the feeder. In

this figure, the red dot denotes where the offshore substation is located, and the green dots

represent such wind turbines acting as representative of corresponding feeders, whose

distance from substation is indeed the least from the substation. That justifies the validity of

limiting the direct connection between closest wind turbines and substation. Such wind

turbines near the substation are denoted as located at the “upstream” of the feeder, while, on

the contrary, wind turbines located far from the substation are assumed to be located at the

“downstream” of the feeder.

87
Figure 3.10 Evolution of the best individual in each generation (without losses)

Figure 3.11 Electric system topology of wind farm collector system (without
considering losses)
The cable type suitable for each section depends on the power injected from the

downstream of the feeder. According to the optimization result, the number of cable sections

are summed up as below: 11 branches of cable of size 3*50 mm2, 8 branches of cable of size

3*70 mm2, 5 branches of cable of size 3*150 mm2, 8 branches of cable of size 3*240 mm2, 4

branches of cable of size 3*400 mm2, 4 branches of cable of size 3*500 mm2, one branch of

cable of size 3*600 mm2, 2 branches of cable of size 3*800 mm2. String and mixed star and
88
string connection schemes are utilized to connect the wind turbines in the collector system.

The topology is fully optimized according to the input data regarding location and prices, and

the connection scheme is different from the common practice to connect wind turbines in one

row or column radially to the substation.

The detailed cost for different devices in this case study is presented in Table 3-8. It is

assumed there is one transformer per substation, so that the transformer cost is fixed as long

as the wind farm capacity has been pre-specified. Accordingly, the factor of transformer cost

has no impact on the system topology. The only cost in offshore substation that perhaps affects

the connection topology lies in the MW switchgear cost. 6 feeders are required, which implies

6 MV switchgears need to be equipped on the lower voltage side of the substation. As shown

in this table, the investment cost of this case can breakdown into two parts: investment cost

on cables and MV switchgears. Thereinto, MV switchgear cost represents 22.5% of the total

investment, while MV cable cost represents the remaining 77.5%.

Table 3-8 Optimization result for layout 1 (without the consideration of losses)
Optimization variable Optimized value
Number of MV feeders 6
Number of WTs per feeder Between 5 and 8
Type of connection scheme String, mixed string and star
Cost of MV switchgear [M$] 2.838
Cost of MV submarine cable [M$] 9.7857
Total investment cost [M$] 12.6237
MV cable cost breakdown
Cross-sectional area Number of branches Length [km] Cost [M$]
[mm2]
50 11 11.60 2.0404
70 8 8.35 1.5189
150 5 5.18 1.0899
240 5 5.33 1.2755
400 4 4.69 1.3432
500 4 4.73 1.5425
630 1 0.64 0.2546
800 2 1.50 0.7236

89
b. Case 2: Layout 1 with the consideration of operational cost
Both investment cost of all involved devices and operational cost are considered in the

optimization of the 40-WT offshore wind farm in this section. The evolution of the best fitness

value throughout the overall optimization process is presented in Fig. 3.12, where the

objective function value has been converged and stabilized to the optimal result at the end of

the generation. 3000th generation can be regarded as the turning point of objective functions,

before which there is relatively large variation in the fitness values, while after which the

fitness value has slightly changed. The evolution finally evolves to the optima at

approximately 8000th generation.

The optimal electric layout topology with the consideration of losses for layout 1 is

shown in Fig. 3.13. Similar to case 1, the topology presented in the figure only covers the

internal part within the offshore wind farm and has no concerns regarding the connection

between offshore substation and onshore grid. Also, there is still no crossings between wind

turbines even if adding the operational cost, which justifies the adaptability of the non-

crossing criterion.

As shown in the topology figure, the total number of feeders in the optimal topology is

5, satisfying the constraint. Each of the feeders is represented in different to color to

differentiate one from another. The wind turbines nearest to the substation, marked as green

dots are chosen to represent the feeder. The connections scheme between wind turbines is still

in string and mixed star in the collector system. However, adding operational cost contributes

to some change in the number of cable sections shown as below: 14 branches of cable of size

3*50 mm2, 6 branches of cable of size 3*70 mm2, 4 branches of cable of size 3*150 mm2, 4

branches of cable of size 3*240 mm2, 3 branches of cable of size 3*400 mm2, 2 branches of

cable of size 3*500 mm2, 3 branches of cable of size 3*600 mm2, 2 branches of cable of size

3*800 mm2,2 branch of cable of size 3*1000 mm2.

90
Figure 3.12 Evolution of the best individual in each generation (with losses)

Figure 3.13 Electric system topology of wind farm collector system (with losses)
The optimized result and corresponding cost in this case study is presented in Table 3-

9. The number of MV feeders is 5, which implies 5 MV switchgears need to be equipped on

the lower voltage side of the substation. As shown in this table, the total cost is composed of

operational cost and investment cost, which can be further divided into two parts: investment

cost on cables and MV switchgears. Thereinto, investment cost takes up 52% of the total cost,

while operational cost takes up the remaining. Additionally, MV switchgear cost is $2.365

million, representing 18.6% of the total investment, while MV cable cost represents the

remaining 81.4%.
91
Compared with the result without losses, it is obvious to find out the introduction of

losses actually contributes to a different topology. If the result obtained in Case 1 is substituted

into this objective function, corresponding cost breakdown can be obtained as well, which is

listed in the last column and named “Cost of Case 1”. Initially, the total number of feeders are

Table 3-9 Optimization result for layout 1 (with losses)


Optimization variable Optimized value Cost of Case 1
Number of MV feeders 5 6
Number of WTs per feeder Between 6 and 9 Between 5 and 8
Type of connection scheme String, mixed string and String, mixed string and
star star
Cost of MV switchgear [M$] 2.365 2.838
Cost of MV submarine cable [M$] 10.3375 9.7857
Total investment cost [M$] 12.7025 12.6237
Operational cost [M$] 11.745 12.4008
Total cost [M$] 24.4479 25.0245
MV cable cost breakdown
Cross-sectional Number of Length [km] Cost [M$]
area [mm2] branches
50 14 14.86 2.6141
70 6 6.23 1.1327
150 4 4.17 0.8765
240 4 4.13 0.9871
400 3 3.21 0.9196
500 2 2.64 0.8601
630 3 3.42 1.3587
800 2 1.48 0.7135
1000 2 1.50 0.8752

less in case 2 with consideration of losses, thus leading to the reduced expense in MV

switchgear. Accordingly, the number of WTs in case 2 is on average larger than that in case 1,

which implies the improvement of utilization of individual feeders in case 2. The unit

resistance of each cable type is not simply linear to the cable cost. In terms of the cable

investment cost, there is a little increase. However, the increase in investment cost can be

offset in operational cost since the operational cost of case 2 is $11.745 million, 5% less than
92
that of result from Case 1. This model tries to figure out the balance between cable investment

cost and its associated operational cost. The result approves the necessity of involving

operational cost into the wind farm topology design phase.

3.4.4 Optimization result and analysis for layout 2

In this section, the result of applying the optimization framework on layout 2 is

presented. As mentioned in the project description, layout 2 does not only concern about the

internal topology connection of an OWF but also takes the optimization of transmission

system into consideration. Literally, the main concern is about how to connect these 40 wind

turbines to the onshore PCC via an offshore substation properly sited with high cost

performance while satisfying a set of operational constraints. Thereinto, the optimal position

of sitting the offshore substation is top priority in the transmission system optimization. In

this case, the substation location is also an optimized variable instead of being pre-specified

beforehand. As for the collector system, the optimized variables are same as those in layout

1, including the inner system topology optimization, along with optimal cable sitting between

wind turbines. To better illustrate the advantage of the proposed algorithm, a layout

optimization without consideration of system operational losses are implemented here as a

benchmark for a comparison with the proposed comprehensive optimization framework.

a. Case 3: Layout 2 without the consideration of operational cost


Investment cost is the only benchmark to assess the objective values in this section,

while the operational cost is neglected, in order to observe the impact of investment on the

final result of electric system topology of the collector system. The result is regarded as a

benchmark, to be further compared with case 4. Also, the two-step-coarse-fine raster map

adjusting is clearly demonstrated in this case.

According to the coordinate of all the wind turbines, it can be found outthat the wind

farm covers a region with x-coordinate from 8 to 15, and y-coordinate from 4 to 11.5. The

PCC is located at (-20, -20). According to the assumption, the offshore substation can be

93
located at anywhere between PCC and OWF. Therefore, the permissible region to locate the

offshore substation is confined to a region with x coordinate from -20 to 15, and y coordinate

from -20 to 11.5. In fact, the optimal site of the offshore substation tends to approach the wind

farm site rather than the PCC site. Intuitively, the unit cost of HVAC transmission cable is

much higher than that of MVAC inter-array cables, and the distance between the offshore

substation and PCC is much longer than one single section between wind turbines, which is

usually about 1km. Accordingly, the transmission cost takes up more in the total cost so that

a site closer to coast seems to be more economic. However, moving the offshore substation

towards PCC has a much stronger impact on the collector system cost than on the transmission

cost. Obviously, the less distance of transmission cables can lessen the transmission cost. On

the meanwhile, with the consideration of collector system, all of the upstream cable sections

on all the feeders have been prelonged. As a whole, the decrease in transmission cost can be

easily offset by the increase in the inter-array cable cost. Accordingly, a conclusion can be

drawn that the optimal site to sit an offshore substation is much closer to the OWF site than

the onshore PCC site, which can also be verified by the further optimization result in this case

study.

According to the conclusion drawn, the search region of offshore substation can be

shrinked. In this case, the shrinked area is set to start at one third of the total permissble region

from the edge of the offshore wind farm. For example, the permissible region covers an area

with x from -20 to 15, and y from -20 to 11.5. The furthest wind turbine from PCC in the

OWF has the coordinate of (15, 11.5), which has been identified as the edge of the OWF. This

permissible region streches over 35km along x coordinate and 31.5km along y coordinate.

The shrinked search region covers one third of the original region, namely, 35km/3  11.7km

along x axis, 31.5km/3  10.5km along y axis. The new shrinked search region can be deducted

by substracting this new distance from the obtained edge in the OWF, which, accordingly,

stretches along the x axis from 3.3 to 15, while along y axis from 1 to 11.5. Further rounded

94
towards the OWF, the new search region can be obtained as x from 4 to 15, and y from 1 to

11. Therefore, it covers an area of 11km (along x axis) * 10km (along y axis). The two-step

coarse-fine tuning raster map has been applied to avoid the high computational effort of high

resolution. Initially, in the coarse raster map step, a raster cell size of 2.5km*2.5km is applied

so that it raster the search region into 6*5 cells. Every raster cell is denoted by the center

coordinate. The raster centers have x coordinate at [4, 6.5, 9, 11.5, 14, 15], while have y

coordinate located at [1, 3.5, 6, 8.5, 11]. The center coordinate of each square raster cell is the

permutation between these x and y coordinates.

The evolution of the best fitness value in coarse tuning throughout the overall

optimization process is presented in Fig. 3.14, where the objective function value has been

converged and stabilized to the optimal result at the end of the generation. 2000th generation

can be regarded as the turning point of objective functions, before which there is relatively

large variation in the fitness values, while after which the fitness value has slightly changed.

The evolution finally evolves to the optima at approximately 5000th generation.

Figure 3.14 Evolution of the best individual in each generation for coarse tuning
in layout 2(without losses)

95
The optimized topology for the OWF has been shown in Fig. 3.15. To improve the

resolution and magnify details of the inner topology, the PCC coordinate is not shown in the

figure. There is still no crossings between any wind turbine pairs. However, the only crossing

lies between the HV transmission cable and one MV inter-array feeder, which can also be

avoided by setting up a detour like in case 1 and 2 for transmission cable route.

As shown in the topology figure, the total number of feeders in the optimal topology is

5, satisfying the constraint set for feeders. The number of feeders has decreased from 6 in

layout 1 to 5 in layout 2, which implies the higher usage rate of feeders in this case.

Additionally, the possible reason is assumed to lie in the fact that the pre-specified center

location in case 1 deviates to the right from the geometric center of the wind farm. The

increase in feeders can help avoid the cable crossings. Each of the feeders is represented in

different colors to differentiate one from another. The connections scheme between wind

turbines is still in string and mixed star in the collector system.

Figure 3.15 Electric system topology of an OWF under coarse tuning (without
losses)
The optimized result and corresponding cost in this coarse tuning step are presented in

Table 3-10. The optimal coordinate of offshore substation is (11.5, 8.5), located in the center
96
of the permissible region preset in the data processing period, with x coordinate from x axis

from 3.3 to 15, y axis from 1 to 11.5, which has justified the validity of shrinking the

permissible region of OS. All of the 40 wind turbines are located on 5 MV feeders, each of

which has 7 to 9 wind turbines. The connection scheme between wind turbines has been fully

optimized, formulated in different forms of string, mixed string and star. Only investment cost

has been considered as the metric to evaluate the topology solution in this benchmark case

with no operational cost taken into consideration. In layout 2, it not only determines the

connections within collector system, but the optimal sitting of substation so as to determine

the transmission cost. Therefore, the total cost is composed of investment cost divided into 4

parts: investment cost on HV cables, MV cables, HV and MV switchgears. To be noted, the

cost of transformer is not provided in the table. It is because the transformer cost has been

fixed as long as the offshore wind farm has a fixed rated capacity considering a single-

substation topology. Literally, the cost of transformer is neglected in the optimization process

since it has no effect on the internal topology of the case according to the hypothesis.

Thereinto, HV transmission cable cost takes up about 63.1% of total investment cost, while

MV cable cost 28.8% of the total investment cost. Additionally, the cost on switchgear makes

up the remaining 8%.

One of the most important result obtained in this coarse-tuning step is the coordinate

of offshore substation. If more precise location is further required, a further rasterization can

be made on the substation coordinate obtained in the coarse raster map step. In this case, the

obtained coordinate is (11.5, 8.5), with previous raster size of 2.5*2.5, which implies the

selected search region for the offshore substation has been further shrunk to an area with x

from 9 to 14, y from 6 to 11. In the fine-adjusting raster map step, a raster cell size of 1*1 has

been chosen to further rasterize the search area into a 7*7 raster cells, with x coordinate {9,

10, 11, 11.5, 12, 13, 14} and y coordinate {6, 7, 8, 8.5, 9, 10, 11}. Accordingly, the optimized

97
topology of this fine-adjusting step is shown in Fig. 3.16 and its detailed breakdown of cost

is presented in Table 3-10.

Figure 3.16 Electric system topology of an OWF under fine tuning (without
losses)

Initially, to achieve a more precise location to sit offshore substation, the center location

has been moved from initial (11.5, 8.5) to (11, 8) in the fine-tuning step. In both coarse-tuning

and fine-tuning, there are 5 feeders to sit all the 40 wind turbines. Therefore, the cost on MV

switchgear, dependent on number of MV feeders, is same for both cases. As described in case

introduction, the OWF has a rated capacity of 240MW. A 500mm2 3-core copper HVAC

submarine cable is fully capable of transmitting such a 240MW generated power. The number

of HV submarine cable is supposed to be 1 and the cost for HV switchgear is $0.53 million

same for both cases. Its corresponding transmission cost should be the unit cost of 3*500 mm2

HVAC cable multiplied by its distance between substation and onshore

98
Table 3-10 Optimization result for layout 2 (raster cell 1*1/without losses)
Optimization variable Optimized value Optimized value
(coarse-tuning) (fine-tuning)
Offshore substation coordinate (11.5, 8.5) (11, 8)
Number of MV feeders 5 5
Number of WTs per feeder Between 7 and 9 Between 7 and 9
Type of connection scheme String, mixed string String, mixed string
and star and star
Cost of MV switchgear [M$] 2.365 2.365
Cost of MV submarine cable [M$] 10.1663 10.4888
Cost of HV switchgear [M$] 0.53 0.53
Cost of HV submarine cable [M$] 22.557 22.182
Total investment cost [M$] 35.6182 35.5658
Final MV cable cost breakdown in fine raster map
Cross-sectional area Number of branches Length [km] Cost [M$]
[mm ]2

50 14 14.61 2.5705
70 6 6.33 1.1521
150 5 5.28 1.1119
240 2 2.05 0.4911
400 3 3.22 0.9233
500 3 3.18 1.0361
630 3 3.25 1.2929
800 2 2.09 1.0101
1000 2 1.54 0.9019
HV cable cost breakdown
Cross-sectional area Number of cables Length [km] Cost [M$]
[mm2]
500 1 41.7732 22.182
PCC, which finally should be $22.182 million. Compared with coarse-tuning optimal result,

the center has moved closer to onshore PCC. Accordingly, the cost on transmission cables has

decreased by about $0.4 million in the final optimal result of fine-tuning step. On the contrary,

the new location of substation leads to increase of cost on MV inter-array cables, which has

increased from $10.3156 to $ 10.4888. However, from an overall view of cost breakdown, the

increase in MVAC inter-array submarine cables has been offset by the decrease in HVAC

transmission cables. The optimal cost in fine-tuning step has decreased by 0.17%. Although

the saving in this case seems not so obvious compared with the total cost, it will be more

obvious when applying to a larger system. In conclusion, this two-step coarse-fine tuning

raster map method is effective to evaluate the collector system cost and transmission system

cost, so as to achieve an optimum equilibrium between these two main costs.


99
b. Case 4: Layout 2 with the consideration of operational cost
Both investment cost and operational cost are considered in the optimization of the 40-WT

offshore wind farm in this section. Similar to case 3, the optimization can be divided into 2

steps, initially rasterized with small resolution and followed with large resolution. The process

of rasterization is same as case 3: the total permissible region for offshore substation is

initially divided by a raster map size of 2.5*2.5, the topology of this coarse-tuning

rasterization is shown in Fig. 3.17, where the cell center located at (11.5, 8.5) has been chosen

as the optimal substation location. With the coarse raster cell size of 2.5*2.5, the search region

of offshore substation has been narrowed into an area with x-coordinate from 9 to 14, y-

coordinate from 6 to 11. To further refine the location of substation, a rasterization to this area

will be applied by a smaller raster size 1*1. The final optimal topology of the OWF is shown

in Fig. 3.18.

Figure 3.17 Electric system topology of an OWF for case 4 (coarse raster map)

100
Figure 3.18 Electric system topology of an OWF for case 4 (fine raster map)

Figure 3.19 Evolution of the best individual in layout 2(with losses)


The evolution of the best fitness value throughout the overall optimization process in

fine adjusting is presented in Fig.3.19, where the objective function value has been converged

and stabilized to the optimal result at the end of the generation. 4000th generation can be

regarded as the turning point of objective functions, before which there is relatively large

variation in the fitness values, while after which the fitness value has slightly changed. The

evolution finally evolves to the optima at approximately 14000th generation. The total

generation required to stabilize the output is larger than that in coarse-tuning step since when

101
further rasterizing the map, the gap between individual raster cells is smaller and harder to

differentiate. Accordingly, larger number of generations are required to evolve to the optimal

result.

The optimal electric layout topology obtained in both the initial coarse tuning step and

fine- tuning step are shown in Fig.3.17 and Fig.3.18. Similar to case 3, the topology presented

in the figure covers both the internal part within the offshore wind farm and the transmission

system between offshore substation and onshore grid. Feeders are differentiated from another

by different colors. There is still no crossing within collector system, justifying the validity of

the cable crossing avoidance criterion. However, crossings occur between the HV

transmission cable and MV cables, which can be avoided by replacing the directly connected

HV transmission cables by the dotted line to make a detour around the inter-array cables.

The optimized result and corresponding cost in both coarse and fine-tuning step of this

case study is presented in Table 3-11. The upper half part compares all the necessary

information regarding the OWF in the 1st and 2nd step. Initially, the coordinate of offshore

substation is not the same: with 2.5*2.5 raster size, the raster cell of (11.5, 8.5) is chosen to

sit the substation, while with further rasterization of precision of around 1 km, (11, 8) is chosen

to sit the substation. Both topologies have the same number of feeders, leading to the equal

cost in MV switchgears. The optimized connection schemes are both string, mixed string and

star. The cost in HV switchgear is also same in both steps, since one HV 500 mm2 cable is

enough to withhold the total capacity of the OWF. The difference lies in the MV submarine

cable and HV submarine cable. In coarse-tuning step, the investment cost of MV cables is

lower, while the transmission investment cost is higher. The situation in fine tuning step is

just the opposite. The reason is because the center location has been moved towards the

onshore PCC (-20, -20) in the further rasterization step. Accordingly, the reduced distance

102
Table 3-11 Optimization result for layout 2 (raster cell 1*1/with losses)

Optimization variable Optimized value Optimized value


(coarse-tuning) (fine-tuning)
Offshore substation coordinate (11.5, 8.5) (11, 8)
Number of MV feeders 5 5
Number of WTs per feeder Between 7 and 9 Between 7 and 9
Type of connection scheme String, mixed string String, mixed string
and star and star
Cost of MV switchgear [M$] 2.365 2.365
Cost of MV submarine cable [M$] 10.3156 10.4888
Cost of HV switchgear [M$] 0.53 0.53
Cost of HV submarine cable [M$] 22.557 22.182
Total investment cost [M$] 35.7676 35.5658
MV cable losses cost [M$] 11.414 11.8053
HV cable losses cost [M$] 17.1473 16.8622
Total lifelong operational cost [M$] 28.5613 28.6675
Total cost [M$] 64.3289 64.2334
Final MV cable cost breakdown in fine raster map
Cross-sectional area Number of branches Length [km] Cost [M$]
[mm2]
50 14 14.61 2.5705
70 6 6.33 1.1521
150 5 5.28 1.1119
240 2 2.05 0.4911
400 3 3.22 0.9233
500 3 3.18 1.0361
630 3 3.25 1.2929
800 2 2.09 1.0101
1000 2 1.54 0.9019
HV cable cost breakdown
Cross-sectional area Number of cables Length [km] Cost [M$]
[mm2]
500 1 41.7732 22.182

between offshore substation and PCC contributes to the lower investment cost as well as the

cost of losses in transmission cable. But the distance between the offshore substation and each

WT in the OWF has increased, leading to the increase in investment cost as well as the cost

of losses in inter-array submarine cable. Overall, the increase in MV inter-array cables has

been set off in the decrease in HV transmission cables. The total cost of fine-tuning step is

64.2334 million$, 0.15% less than that of coarse tuning step.

In terms of the final optimal result of fine raster map, the total cost is composed of

operational cost and investment cost, which can be further divided into 4 parts: investment
103
cost on MV and HV cables, and MV and HV switchgears. Thereinto, total investment cost is

$35.5658 million, which takes up 55.4% of the total cost, while lifelong operational cost,

$28.6675 million, takes up the remaining 44.6%. With respect to investment cost, the cost of

HVAC cable is $22.182 million (take up 62.2% of total investment cost), much higher than

that of MV inter-array cables (take up 29.5% of total investment cost). The switchgear cost is

$2.895 million, only taking up the 8.1% of the investment cost. MV switchgear cost is $2.838

million, representing 22.4% of the total investment, while HV switchgear cost represents the

remaining 77.6%. In terms of operational cost, transmission system pays more with $16.8622

million for lifelong, compared with $11.8053 million in MV collector system.

The lower half of the table summarizes the MVAC cable used for different sections.

According to the optimization result, the number of cable sections are summed up as below:

14 branches of cable of size 3*50 mm2, 6 branches of cable of size 3*70 mm2, 5 branches of

cable of size 3*150 mm2, 2 branches of cable of size 3*240 mm2, 3 branches of cable of size

3*400 mm2, 3 branches of cable of size 3*500 mm2, 3 branches of cable of size 3*600 mm2,

2 branches of cable of size 3*800 mm2, 2 branches of cable of size 3*1000 mm2.

Compared with the result without losses, the optimal topology obtained in the second

step, fine-tuning , is same as that in benchmark case without losses. However, in the initial

coarse-tuning step, the introduction of losses still contributes to a different topology. That is

because the involvement of losses is not simply linear with cable length of cable sections. The

result still approves the necessity of involving operational cost into the wind farm topology

design phase. In the benchmark case without losses, the objective is to find an optimum

topology with the least cable cost expenditure, while in the real optimization model, the

overall topology is optimized from a realistic lifetime long view, covering both initial

investment and operational cost.

The proposed framework provides an optimum electric system topology for the specific

offshore wind farm based on some justified simplifications, which can be used as a guide for

104
wind farm developers to set up the offshore wind farm with an accurate cost predication. After

the electric layout design, several studies are required to carry out, including load analysis,

fault analysis, etc. Thereinto, fault analysis is essential to carry out the following protection

design: how to distribute the relays, which kind of relay is proper, how to coordinate and set

up these multi-level relays, in order to provide effective and efficient reaction in case of

contingencies. A serious protection design is especially vital to offshore wind farm given the

restricted access to fault on site due to the harsh offshore conditions. However, the protection

devices considered in the model is simply abstracted to switchgears with a constant unit cost.

It is obvious that the cost used to represent protection devices in the system is relatively simple

than that in reality, though which has little effect on the electric system design.

3.5 Conclusion

The proposed electric system optimization framework performs a comprehensive

optimization of collector and transmission system of an OWF in order to minimize the

investment and operational cost. It receives the specifications of wind farm, available types

of devices, as well as the geometric location of individual wind turbines, onshore PCC,

existing location or permissible region for offshore substation. In return, it provides the layout,

the proper cable type for each individual connection and appropriate location to sit the

offshore substation. The feasibility of the framework has been verified by a virtual OWF with

40 wind turbines by checking technical constraints and feasibility such as cable crossings.

The optimization results are based on publicly accessible data. This framework can be used

as a useful decision-making tool for offshore wind farm planners to find out the best

economical and technically feasible topology after the initial wind turbine micro-sitting study.

Although the planning method presents a comprehensive electric system design for OWFs,

however, it may not be adaptable to a large-scale OWF, which probably accommodates

hundreds of wind turbines across hundreds of square kilometers. An electric layout design

105
with distributing the large number of wind turbines into multiple substations are more viable

and reliable for such an offshore wind farm, In the next chapter, a planning method for a multi-

substation offshore wind farm will be presented .

106
Chapter 4 HIERARCHICAL OPTIMIZATION FOR

THE COLLECTOR SYSTEM PLANNING OF LARGE

OFFSHORE WIND FARMS WITH MULTIPLE

SUBSTATIONS

To fully exploit the offshore wind resources, offshore wind farm will have a dramatic

increase in its coverage area and rated capacity, which makes the single-substation electric

system less adequate to manage. A more reliable layout optimization with multiple substations

is proposed and discussed for a large-scale offshore wind farm in this chapter.

4.1 Problem Description

An optimal electrical layout has been verified of great importance for an OWF to

achieve a balance between economy and efficiency. The problem how to collect and transmit

generated power from each individual wind turbine to shore in a cost-minimization manner

has been proposed and properly resolved in the last chapter. However, the assumption

proposed for the model has confined that there is a sole substation in the electric topology.

Literally, the generated power from all the wind turbines is collected to the sole substation,

which is responsible to uplift the voltage and adjust the electric frequency according to the

grid code to integrate the OWF. This sole offshore substation takes the role of electric interface

between the OWF and onshore PCC. This sole-substation topology has been widely applied

in current operating OWF, whose capacity is moderate and distance to shore is relatively short.

The rated capacity of large-scale offshore wind farms (OWFs), currently, has reached

up to gigawatt level and the distance from the offshore-shore wind farm to the coast has been

up to hundred kilometres [109]. Generally, in such a large offshore wind farm, there are

hundreds of wind turbines spreading across hundreds of square kilometres. If the sole-
107
substation topology is still applied in this case, the overall collector system cost to connect

each wind turbine to the substation is supposed to be considerably high from an economic

perspective; the overall reliability is, meanwhile, assumed to be poor considering the fact of

losing all the generated power in the case of an offshore substation breakdown or malfunction.

Accordingly, the sole-substation topology is not appropriate for a large-scale offshore wind

farms in light of its poor performance in both cost, reliability as well as secure operation.

Dividing the large number of offshore wind turbines into different partitions is an efficient

way to improve the above-mentioned two drawbacks of sole-substation topology and multi-

substation topology has to be considered in case of large-scale offshore wind farms.

Additionally, the majority of offshore wind farms in commission have utilized high

voltage alternating current link to transmit their generated power to onshore grid. Accordingly,

high-voltage AC submarine cables have been considered as the only method for the

transmission system in the sole-substation topology, which is reasonable regarding to its

moderate capacity and off-coast distance. However, it is no longer the one and only choice in

large-scale offshore wind farms. High voltage direct current technology tends to attract more

interests in offshore wind farms with increasingly larger size and longer off-coast distance in

light of its relatively low expense and high efficiency for bulk-power long-distance

transmission. Several studies have proposed a concept “critical distance” (usually 60-100 km

depending on the ratings of OWFs), above which HVDC transmission is superior to HVAC

transmission. Apart from the abovementioned benefits, voltage-source-converter HVDC has

the following advantages: facilitates to electrically decouple the offshore wind farm from the

onshore main grid; enables the connection to weak AC network; has the capability to control

the power injected from each VSC converter terminal. To take the advantages of DC

transmission, several researchers have been investigating to extend the DC transmission to

collector system to formulate a pure DC offshore wind farm [4-8]. In this chapter, medium

voltage alternating current is the only method considered in collector system since AC

108
transmission is most widely used in medium voltage network. If HVDC technology is

assumed the most appropriate way, the offshore wind farm will be formulated as a hybrid

AC/DC wind farm, with converters sitting on the offshore platform and acting as the interface

between OWF and onshore main grid. On the contrary, if HVAC is still considered as cost

saving, the OWF is a pure AC offshore wind farm.

In conclusion, to cater to the features of large-scale offshore wind farm, the structure

can be defined as a multi-substation OWF with correspondingly multiple collector systems.

Also, the choice of substation (equipped with either voltage-source power converters or power

transformer) and utilised transmission technology (either HVDC or HVAC) are dependent on

the final optimum result. The general schematic diagram for a large-scale OWF electric power

system is shown in Fig. 4.1.

Figure 4.1 General schematic diagram for a large-scale OWF


Similar to that in the last chapter, there are 3 functional factors in the OWF structure,

which are collector systems (collect and accumulate the generated power), transmission

systems (mainly transmit the accumulated power to onshore power grid) and substations (act

as interfaces between collector systems and transmission systems). Thereinto, the

abovementioned features have little impact on collector system since it still utilizes MVAC

transmission and its connection topology is limited to radial connection. However, the
109
differences lie in the substations and transmission systems in the OWF. As for substations,

the most obvious change is the total number of substations, which indicates the fact of

partitioning wind turbines into multiple partitions, or substations in the context. Besides, with

both HVAC and HVDC technology considered, the substation can be equipped with either

power transformers or power converters. Accordingly, the following transmission method

should be either HVAC or HVDC, respectively.

It is worth noting that the optimum electric layout is to consider how to connect all the

wind turbines via multiple substations to the destination PCC in a cost minimization way. The

highlight is to properly partition wind turbines into a proper number of clusters whose centres

are also properly sit in an optimum location. Once the partition process finished, the following

optimization is to connect the wind turbines already clustered to the cluster to the cluster

centre, which is same as the research done in last chapter. With known cluster centre location

and all the wind turbine locations, the inner collector system optimization of different

collector systems is independent from each other. Clearly, the overall OWF electric layout

model can be divided into 3 layers: offshore substation optimization layer, inter-array feeder

optimization layer and submarine cable section optimization layer, to formulate a hierarchical

optimization framework for large-scale OWF layout planning, as shown in Fig. 4.2.

Sub 1 Sub 2

Substation Layer

Inter-array feeder layer

Inner collector system

Submarine cable section


layer

Figure 4.2 Illustrative diagram for multi-layer in offshore substation


110
In the substation layer, it is to determine how to partition wind turbines into proper

number of partitions, where Fuzzy clustering technique, a widely-used partition technique,

has been applied to divide the wind turbines and simplify the overall complicated, nonlinear,

and nonconvex layout optimization process. The inter-array feeder layer and submarine cable

section layer are a summarization of inner collector system layout optimization. In inter-array

feeder layer, it is to properly locate each wind turbine into a feeder, while in submarine cable

section layer, it is to determine the most economic connection scheme to connect wind

turbines on one feeder. Nature aggregation algorithm and Prim’s algorithm have been applied

in inter-array feeder layer and submarine cable section layer, respectively. Overall, fuzzy

clustering algorithm, Nature aggregation algorithm and Prim’s algorithm are integrated and

formulated the hierarchical optimization framework for large-scale OWF layout planning.

A comparison of the proposed technique to others is provided in the table below, which

justifies its competitiveness and comprehensiveness. Initially, the proposed framework has a

more thorough consideration of involved equipment in the offshore system. To get a more

accurate lifecycle cost, annual maintenance cost and operational cost are both included in the

cost model. The electric system layout in collector system is fully optimized in radial

connection scheme without any unacceptable crossings between cables. On the meanwhile,

the submarine cables for both transmission and collector system is chosen among a full set of

cable types instead of simply one or two predefined cable types. Both HVAC and HVDC are

considered as transmission alternatives in transmission system. The proposed framework can

screen out the superior transmission options based on specific wind farm information.

Additionally, multiple substations are usually required to accommodate the increasing size of

offshore wind farm. This framework helps to determine the optimal placement of the offshore

platform instead of fixing them to several provided positions as in [30], [32], [34] and [82].

111
Table 4-1 Comparison of proposed framework with other techniques

This chapter proposes a hierarchical optimization framework for large-scale OWF

layout planning, which can effectively reduce the dimensionality and complexity of problem

by applying fuzzy clustering technique to cluster wind turbines into proper clusters. The inner

collector system layout can be optimized independently and simultaneously by parallel

optimization tools to speed up process time.

This chapter is organized as follows: Section 4.2 illustrates mathematical problem

formulation, including how to model different cost categories for involved components.

Section 4.3 presents a detailed explanation for the hierarchical optimization framework: fuzzy

clustering algorithm for substation layer, NAA and Prim’s algorithm for the other two layers

(considered as inner collector system optimization). The hierarchical framework has been

verified on two case studies in section 4.4, followed by which the conclusion is drawn in the

final section.

4.2 Problem formulation

The overall cost of a large-scale offshore wind farm can be affected by a lot of factors
112
i.e., various wind turbine model, no. of wind turbines, distance to shore, submarine seabed

type, voltage level utilised, cable size and length for each individual section, connection

scheme, etc. All of the above factors are mutually coupled, same as discussed in Chapter 3.

In this multi-substation topology model, the number of clusters and number of wind turbines

in each cluster are mutually coupled. Therefore, a large number of discrete variables are

supposed to be introduced to represent the abovementioned decision factors and they have to

follow either the linear or the nonlinear relationships mimic the natures of the electric layout

system of an OWF.

In this section, the mathematical model for OWF electric system layout optimization is

provided. Some justified assumptions are proposed initially, which is followed by a detail

description of the mathematical model.

4.2.1 Hypothesis

Some reasonable hypotheses have been proposed to simplify the complicated problem,

as listed below.

1. All the preliminary studies (including environmental analysis, seabed survey which are

necessary to determine the optimal location of WTs, offshore substations) have been

conducted. The wake effect is generally considered in the preliminary study to

determine the optimal placement of wind turbines, which is not considered in this

model.

2. All wind turbine locations have been known.

3. The operating limits of wind turbines on active power P and reactive Q are neglected

since both of these two parameters can be fully adjusted by control strategy embedded

in wind turbines.

4. Only one point of common coupling (PCC) is provided, whose location is provided.

5. Wind speed variance has been considered the same for all the wind turbines. Also, the

wind turbine power characteristics are considered the same for all the wind turbines.
113
6. MVAC electric power systems are mainly considered in collector system because the

vast majority of the OWF collector system in operation are running AC systems. Both

HVAC and HVDC are transmission alternatives in the transmission system, depending

on the substation rated capacity and its distance to shore.

7. Each pair of nodes (wind turbines or substation) can be connected in string, star and the

mixed string/star configuration, while ring topology is not supported in this project.

8. All the substations are connected with PCC independently. Literally, no meshed

networks are considered for transmission systems.

9. The number of transformer or converter for each substation is limited to 1. For example,

with the cluster rate of 100MW, the transformer or converter for this cluster should

fulfil that specific rated capacity, 100MW. The case of dividing the 100MW to 2

transformers (or 2 converters) is not considered.

10. The transmission technique for all the substations in one OWF should be the same. To

be more specific, if HVAC transmission technique has been chosen, all the transmission

systems utilize HVAC transmission in their corresponding transmission systems, and

vice versa. It is not possible that one substation utilizes HVAC, while the other

substations utilize HVDC. In that case, it can be hardly determined whether to build a

transformer substation or a converter station at onshore PCC.

11. The cables buried in the same connection corridor are assumed to be of same type.

According to the hypothesis, all the locations including wind turbines and onshore PCC

are provided; the available equipment parameters are provided. Hypothesis 5 assumes the

wind speed variance is considered the same for all the wind turbines, similar to that in Chapter

3, given the limited effects of wind speed variance on system configuration. For a specific

wind turbine model chosen in the model, it can be abstracted as a generator with equivalent

rated capacity. To optimize the OWF layout is to determine the placement of necessary

equipment such as appropriate cables and offshore platform to provide a complete route to

114
“destination”, onshore PCC, for every individual wind turbine in the OWF. Also, multi-

substation solution requires an appropriate number of partitions, wind turbine partitions and

substations located in appropriate cluster centres. Obviously, a large number of variables have

to be introduced to represent the connection scheme between every two nodes in the network.

The complexity of the problem increases with the number of wind turbines in the OWF.

4.2.2 Mathematical model

The objective function of OWF topology optimization is to minimize the investment

cost, maintenance cost as well as operational cost, while satisfying a set of technical

constraints. The offshore windfarm can be partitioned according to its individual function into

transmission system, offshore substation system and collector system. To accommodate a

large-scale OWF, the OWF has multiple offshore substations and voltage levels. Therefore,

the difference of optimization of single-substation OWF and multi-substation OWF primarily

lies in determination of each individual wind turbine clustered among the several substations.

The traditional scheme of a multi-substation OWF is shown in Fig. 4.1, which is consistent

with the functional partition abovementioned.

The total cost can be broken down according to the functional partition: collector

system, offshore substation, and transmission system (no mesh connection in transmission),

which can be expressed as below:

Total Cost = CCollector + CSUB + CTrans (4.1)

As for collector system, it is functioning to collect the generated power from each

individual wind turbine located in the OWF and then transmit it to the “transfer station”,

offshore substation. The total cost of a collector system includes its investment cost, related

maintenance cost and operational cost. Thereinto, the investment cost of a collector system is

by nature the cost of medium voltage inter-array submarine cables, which is determined by

the cable type chosen for each section and the distance between two connected wind turbines.

The investment is assumed to be made in the first year and paid off in the following years
115
throughout the lifetime of OWF. Maintenance cost is an estimated cost of routine service for

the submarine cables. Operational cost is regarded as the cost of losses on the inter-array

cables in the collector system. Detailed expression of cost in collector system is provided as

below. (r = 4%, y = 20)


ns

i =1
(
CColloctor =  CCollector
Investment
i
Ma int ain
+ CCollectori
Operation
+ CCollectori
) (4.2)

r (1 + r )
y
Investment
CCollector =  UMCc  d s  u psfc (4.3)
(1 + r )
y
− 1 p , s , f ,c
i

Ma int ain
CCollectori
= kc 
p , s , f ,c
UMCc  d s  u psfc (4.4)

 d s  ( I sf )
2
Operation
CCollectori
=  MR
s , f ,c
c  t  Closs (4.5)

Initially, this model takes multi-substation scheme of OWF into consideration. The total

cost of collector system, thereby, covers the abovementioned 3 kinds of costs of each

individual collector system, which can be differentiated from its subscript i . CCollector
Investment
i
,

Ma int ain
CCollectori
and CCollector
Operation
i
denote the investment cost, maintenance cost and operational cost of

corresponding collector system i , respectively. Equation(4.3) express the formation of

investment cost, where UMCc is the unit cost (US$/km) of MVAC cable of type c ; d s

represents the distance of section s ; u psfc is a binary decision variable denoting the

connection scheme of wind turbine p , located at section s on feeder f via a MVAC

cable type c . Maintenance cost, in Equation (4.4) is expressed as proportional to the

investment cost, where kc is a constant representing the pre-defined maintenance s s factor.

Operation
CCollectori
is the operational cost, an estimation of losses in the collector system, where MRc

is the resistance of MVAC cable of type c , source from specific cable manufacturer; I sf

is the current flow in this section s of feeder f ; Closs is exactly the price of power sold

116
to the utility; t represents the equivalent available time of OWF.

The offshore substation is responsible to uplift the voltage level to facilitate further

transmission to shore. In this case, both high voltage alternating current technology and high

voltage direct current technology are taken into consideration. The offshore substation is

composed of either a main power transformer (for HVAC transmission) or a power converter

(for HVDC transmission), platform, some auxiliary protection and control devices. The total

cost of offshore substation can also be defined as the summation of investment cost of devices,

maintenance cost and estimated operational cost, which is expressed as below.


ns
CSUB =  (C
i =1
Investment
SUBi
Ma int ain
+ CSUB i
Operation
+ CSUBi
) (4.6)

r (1 + r )
y

C Investment
SUBi = iAC Ctr ( Si ) + iDC Csvc ( Si ) + C p ( Si ) + CSG ,i  (4.7)
(1 + r ) − 1
y

Ma int ain
CSUB i
= ks iAC Ctr ( Si ) + iDC Csvc ( Si ) + C p ( Si ) + CSG ,i  (4.8)

Operation
CSUBi
= (1 −  )  Si  t  Closs (4.9)

The investment cost of substation i is expressed in Eq. (4.7), which is composed of

main power transformer cost Ctr ( Si ) , power converter cost Csvc ( Si ) , offshore platform cost

C p ( Si ) and the cost of protection devices CSG ,i . A binary decision variable iAC has been

introduced to determine whether a power transformer is installed on the substation i ;

iAC = 1 means a transformer has been chosen and it is HVAC transmission that is utilized in

the following transmission systems; otherwise, iAC = 0 indicates the power transformer is

not utilized. Similarly, iDC is the binary decision variable to denote whether a power

converter is installed in this substation i . It is obvious that power transformer and converter

cannot exist at the same time, otherwise there would be conflicts for further transmission

technique. The cost of transformer (or converter) and platform is estimated as a function of

117
the substation capacity Si , as shown in the equations below. [23],[30]

Ctr ( Si ) = 0.03327  Si0.7513 (4.10)

Si
Csvc ( Si ) = ( 42 + 27  )  106 (4.11)
300

C p ( Si ) = 2.14 + 0.0747  Si (4.12)

Switchgear is the protective devices considered in this model, whose cost is expressed

as below.

CSG ,i = CMVSG ,i + CHVSG ,i (4.13)

CMVSG ,i = n f ,i  UCMVSG (4.14)

CHVSG ,i = nHV ,i  UCHVSG (4.15)

The investment cost of switchgear can be broken down into MV switchgear cost and

HV switchgear cost, both of which are positively correlated to the number of cables connected

to the substation i , n f ,i and HV side nHV ,i . The unit cost of MV switchgear UCMVSG and

HV switchgear UCHVSG are constants and assumed to be 0.473m$ and 0.53m$, respectively

[30].

The maintenance cost of substation, expressed in eq. (4.8) covers the routine

maintenance of all involved devices of offshore substation, which is also assumed to be

proportional to the total investment cost. ks denotes the specific proportion between

maintenance and investment for substation.

The operational cost of offshore substation is mainly referred to its transformer losses

or converter losses. For transformers, the losses can be technically separated to core losses

generated in its magnetic material, and copper losses generated in its windings, both of which

are hard to figure out. For converters, the losses are generated in power electronic devices. To

simplify the operational losses in power transformer and power converter, the efficiency 

118
is introduced to roughly estimate the possible losses in substation.

In terms of the transmission system, it is functioning to transmit the collected and

transformed power from offshore substation to onshore PCC. The transmission system

considered in this model has a one-to-one correspondence to the offshore substation. Literally,

meshed network is not covered in the transmission system. Each substation is independently

connected to shore via its own transmission system. One reasonable assumption has been

proposed that the cable type in one transmission system is same. The main components in

transmission system are transmission cables, which could be either HVAC or HVDC cables,

depending on cluster ratings and offshore distances. Similar to collector system, the cost of

transmission system mainly comes from its transmission cable cost and can also be defined

in investment cost, maintenance cost and operational cost.


ns
CTrans =  (C
i =1
Investment
Transi
Ma int ain
+ CTransi
Operation
+ CTransi
) (4.16)

r (1 + r ) d i
y

C Investment
Transi = iAC nHV
AC
,iUHCc
AC
+ iDC nHV
DC DC
,iUHCc 
 (4.17)
(1 + r )
y
−1

Ma int ain
CTransi
= kc di iAC nHV
AC
,iUHCc
AC
+ iDC nHV
DC DC
,iUHCc 
 (4.18)

2
I 
C Operation
= n AC AC AC
HR di  i AC  tCloss
Transi i HV ,i c
 n HV ,i 
2
(4.19)
 Ii 
+ iDC nHV
DC DC
,i HRc d i  DC  tCloss
 nHV ,i 

The transmission technique utilized in transmission system must be consistent with the

choice for substation, so that the same binary decision variables, iAC and iDC , are

introduced in Eq. (4.17) to determine the utilization of transmission technique. The expression

of cost is almost the same for HVAC and HVDC cables besides the difference in superscript,

where AC indicates the case of utilizing HVAC transmission cables; DC indicates utilizing

HVDC transmission cables. To simplify the explanation and save space, the definition below

119
provides a general introduction of the expression without differentiating HVDC cases from

HVAC cases.

Investment
The investment cost of transmission system connected with substation i , CTransi , is

determined by the cable type c with unit cable cost denoted as UHCc , and the distance

from its offshore substation i to shore, d i . nHV ,i is the number of cables of type c

required to withstand the power capacity of substation i . The maintenance cost is still

proportional to the investment cost. The operational cost, as shown in Eq. (4.19), is estimated

as the cost on losses on the transmission cables, where HRc represents the unit resistance

(ohm/km) of HVAC cable of type c ; I i is the current flow out of the offshore substation i .

In terms of large-scale offshore wind farm, multiple transmission cables are possibly required

to satisfy the large rated capacity of each cluster, namely nHV ,i  2 . The current I i is

Ii
supposed to be shared equally on the transmission cables, thereby being , since the
nHV ,i

same cable type is assumed in transmission system for one substation.

Considering all the above specifications, the electric system layout design model for a

large-scale offshore wind farm can be summarized as below.

min Total Cost =CCollector + CSUB + CTrans


s.t.
S sf  MCc
Si / nHV ,i  HCc
Ci  C j =  , Ci ,C j  C ,Ci  C j (4.20)
Ci  C j  ...  Ck = I , Ci ,C j ,..., Ck  C
f i  f j =  , f i , f j  F
iAC + iDC = 1, Ci  C
iAC =  jAC , Ci , C j  C , Ci  C j

where the objective of the model is to minimize the levelized cost of OWF, including all the

expense for all the equipment. Thereinto, the apparent power flow in every cable section of
120
the OWF should not exceed the rated capacity of the chosen cable type, to guarantee the secure

and reliable operation. Ci indicates the set of wind turbines in substation i . Wind turbine

clustered into substation i cannot be clustered into any other substation j . There is no

overlap between any clusters. The union of all the clusters should include all the wind turbines.

f i represents the set of wind turbine located on feeder f i . Similarly, no wind turbine can

be put on two or more feeders, otherwise there would be intersections between feeders, which

is not allowed in offshore wind farm cable layout design. For each substation i , either HVAC

or HVDC cable is utilized to connect with onshore PCC. The last constraint is consistent with

the 10th hypothesis that the transmission technique for all the substations in one OWF should

be the same.

4.3 Hierarchical layout optimization framework

The offshore wind farm layout optimization can be visually divided into 3 layers,

offshore substation optimization layer, inter-array feeder optimization layer and submarine

cable section optimization layer. Thereinto, offshore substation layer concerns clustering

individual wind turbines to appropriate offshore substation; inter-array feeder layer is

regarded as further clustering the wind turbines which has been already clustered to this

specific substation into appropriate feeders; while submarine cable section layer is primarily

related to connect wind turbines clustered into one feeder in an economical and practical

connection scheme. Accordingly, this multi-layer optimization can return a comprehensive

optimal connection topology from the wind turbine to onshore PCC.

4.3.1 Offshore substation optimization layer

Given a large-scale offshore wind farm, its capacity can reach out several hundreds of

megawatts, where hundreds of wind turbines spread across hundreds of square kilometres.

With a view of higher failure rate and longer repair time of devices off the shore, it is not

economic to set up only one offshore substation to collect all the generated power from the
121
OWF. Setting up multiple substations is beneficial to increase the overall reliability of OWF

and mitigate risks of potential offshore substation malfunction, which justifies the method of

distributing offshore wind turbines among multiple substations.

The objective of offshore substation optimization layer primarily concerns clustering

the wind turbines with known locations to several appropriate offshore substation whose

locations to be determined. Thereby, the subordination of each individual wind turbines to the

substation, along with substation locations, need to be optimized. Clustering techniques have

been widely used in industrial applications with requirement of clustering according to the

specific attribute of research objects, thereby reaching the aim of simplifying the problem.

Clustering techniques are generally classified into two classes: crisp clustering with explicit

boundaries between clusters and fuzzy clustering without such a boundary and allowing

overlaps between clusters [110]. Fuzzy clustering is obviously more appropriate to be applied

in partitioning an OWF into several undefined clusters without specific pre-defined cluster

boundaries. Unlike ref [34], where wind turbines are partitioned merely according to their

distance, other impacts on the partition result including cable cost connecting each WT to the

substation, substation cost and transmission cost are also involved in order to

comprehensively update the centre location as well as optimize the partition. The reason to

involve the other impact factors is that all the involved cost cannot be expressed as linear to

distance only. The fuzzy clustering model is formulated as below.

Min . obj =
ns N
(4.21)
 (  ) ( d ) + c ( CCollectorij ) + s ( CSUBij ) + t ( CTransij ) 
m 2 2 2 2

i =1 j =1
ij ij


dij = p j − PCi (4.22)

CSUBij = CSUBi / N (4.23)

CTransij = CTransi / N (4.24)

122
CTransi  nHV ,iUHCc PCC − PCi (4.25)

s .t . ij 0 ,1 (4.26)

ns


i =1
ij =1 (4.27)

N
0<  ij <N (4.28)
j =1

where ns is the number of clusters, also referred to as number of offshore substations, which

is required to be predefined; N is the total number of wind turbines; PWT = P1 , P2 , P3 , ..., Pn 

constitutes the date set to be clustered, where each data sample (or WT) is denoted by its

location p j , and defined as a 2-dimensional coordinate  x j , y j  ; Pc =  PC1 , PC2 , ..., PCc 

is the cluster centre set, for example, centre i is denoted by its location PCi , defined as a

2-dimensional coordinate  xi , yi  ;  ij represents the membership degree of wind turbine

j to substation i ; m denotes the exponential index, 1  m   ; given the coordinate of

both ends, the distance between centre (or substation) and each data sample (or wind turbine)

can be calculated, which is expressed as d ij in Eq. (4.21).

CCollector ij is defined as the total cost of connecting wind turbine j to substation i ,

which has already been detailed provided in Eq. (4.2). It covers all the involved cost rather

than the cable investment cost only in ref [88]. To save space, it is not added in the equations

above. If wind turbine j is not clustered to substation i in the last iteration, wind turbine

j is assumed to be connected with its nearest wind turbine clustered to substation i , thus

CCollector ij defined as the corresponding cost of connecting it to substation i . The detailed

process for insertion of non-clustered wind turbine into other substation is provided in the

section 4.3.4.

CSUBij and CTransij quantize the impact of substation and transmission cost after taking
123
wind turbine j into cluster i , which are related to the calculated cost of substation i ,

CSUBi and corresponding transmission cost CTransi . The impact on each wind turbine is

assumed to be the same so that both of these two components are equally shared among all

the wind turbines. Thereinto, the detailed equation of calculating CTransi is provided in Eq.

(4.16), which is linear to the distance between the substation i and PCC. PCC is also a 2-

dimensional coordinate vector, denoted as  PCC1 , PCC2  . c , t , s are constants, set as

regulation indices to scale down the collector system cost, substation cost and transmission

cost to the same magnitude of distance.

Take the summation of wind turbines in Eq. (4.21) out, the objective function can be

rewritten as below.

n
 ns 2 
Min . obj =    ( ij ) ( dij ) + c ( CCollectorij ) + s ( CSUBij ) + t ( CTransij )   (4.29)
m 2 2 2

j =1  i =1

  

Given the constraint in Eq. (4.27), the Lagrange multiplier function of the objective

function can be provided as below.

ns
 ns

L =  ( ij ) ( dij ) + c ( CCollectorij ) + s ( CSUBij ) + t ( CTransij )  +   1 −  ij  (4.30)
m 2 2 2 2


 

i =1  i =1 

The variable to be determined in the objective is  ij and  , so that taking the partial

derivative of L with respect to these variables, the following equation can be obtained.

L
= mij m−1 ( dij ) + c ( CCollectorij ) + s ( CSUBij ) + t ( CTransij )  −  , i 1, 2 ,..., c (4.31)
2 2 2 2

ij  

ns
L
= 1 −  ij (4.32)
 i =1

Extremum usually occurs when the partial derivative is equal to zero. Let Eq. (4.31)

and (4.32) be equal to 0. It can be obtained,

124
1
  m −1
  
(4.33)
ij =  2 
 m ( dij ) + c ( CCollector ij ) + s ( CSUBij ) + t ( CTransij )  
 
2 2 2

  

Take (4.33) into (4.32),  can be obtained,

m
= m −1 (4.34)
 ns 
 1 
 i' = 1 ( d ) 2 +  ( C
Collectori' j ) + s ( CSUBi' j ) + t ( CTransi' j ) 
2 2 2

 i' j c

Take (4.34) into (4.33),  ij can be worked out,

−1
 1

 ns ( dij ) + c ( CCollector ij ) + s ( CSUBij ) + t ( CTransij )
 
2 2 2 2 m −1

 
ij =     (4.35)
 i' =1 1

 ( di' j ) + c ( CCollector i' j ) + s ( CSUBi' j ) + t ( CTransi' j )  m −1
2 2 2

   

dij2 = ( xi − x j ) + ( yi − y j )
2 2
If substituting the distance and

di2 = ( xi − PCC1 ) + ( yi − PCC2 ) into the objective function, similarly, the coordinate of
2 2

each substation i can be updated in each iteration,

1 n  t  coef 2 UHCc2 
( )
m
xi = ij xj + 2
PCC1 
n
 t  coef 2  UHCc2  n
( )  
m j =1
ij 1 + 
j =1  n2 
(4.36)
1 n  t  coef 2 UHCc2 
( )
m
yi = ij y
 j + 2
PCC2 
n
 t  coef 2  UHCc2  n
( )  
m j =1
ij 1 + 
j =1  n2 

where coef referred to the proportion between total transmission cost and investment cost.

It is also related to the cable type chosen for transmission.

According to Eq. (4.35) and (4.36), with the distance and cost obtained from the last

clustering iteration, membership degree matrix can be updated accordingly. Similarly, the

substation location can also be updated. If the change in the objective value is smaller than

tolerance, the result refers to the optimal partition of the OWF into appropriate offshore

125
substation. Literally, after clustering step, the cluster centre locations have been determined

and the membership degree between each wind turbine and any cluster can be provided.

4.3.2 Wind turbine allocation

The previous fuzzy clustering merely provides the subordination membership degree

matrix and centre location. An automatic wind turbine allocation step should be proposed to

further determine the specific subordination of each WT to each substation, with the objective

of maximizing the total membership degree overall the OWF while satisfying the substation

capacity constraints, or rather the maximum rated capacity of transmission cables. According

to the proposed hypothesis, the rated capacity of transformers or converters equipped at each

substation is only determined by the number of wind turbines and their rated operating power.

In fact, the maximum rated capacity of each cluster is confined by the maximum rated current

carrying capacity of the available high voltage submarine cable types (HVAC or HVDC).

A binary variable z ij is introduced to determine the subordination of wind turbine j

to collector substation i , where z ij = 1 means wind turbine j clustered to substation i ;

otherwise, z ij = 0 , not clustered to substation i .Therefore, an automatic wind turbine

allocation model is formulated as below.

ns N

 
i =1 j =1
ij z ij (4.37)

s .t .
ns

z i =1
i
j =1 (4.38)

z
j =1
i
j  N smax (4.39)

It is formulated as a binary linear programming problem, with known membership

degree as the coefficient. Each WT can only be connected to one substation and there is no

126
overlap of the subordination of different substations, as shown in Eq. (4.38). It is the

realization of the constraint between clusters in Eq. (4.20), Ci  C j =  . Meanwhile, the total

number of wind turbines connected to one substation should satisfy the maximum substation

capacity, as shown in Eq. (4.39).

By this wind turbine allocation method, all the wind turbines can be directed to the

most appropriate cluster without violating the maximum rated capacity set for each cluster.

To be more specific, the overall complicated and messy layout design work for a large-scale

OWF has been simplified and segmented into several smaller clusters with smaller number of

wind turbines after the offshore substation optimization layer.

4.3.3 Transmission system optimization

Once finishing the clustering of wind turbines, the number of wind turbines in each

partition has been fixed thereby specifying the rated capacity required for transmission cables

of each cluster. Accordingly, the transmission cable selection is only determined by the rated

capacity of each cluster and has no concern of the inner topology in each cluster (or collector

system), so that it can be optimized in the substation level. As discussed in Chapter 2, normally

HVAC and HVDC are two transmission alternatives considered for offshore wind farm

transmission. If given a set of both HVAC and HVDC cables with different cable sizes, a

comparative analysis to select an appropriate and economic transmission cable type is

required. One reminder to be noticed is that the maximum rated capacity of each cluster is

determined by the largest available HVAC and HVDC cable rates, thereby probability

contributing to different number of clusters. One precondition of fuzzy clustering is a

specified number of clusters. Given the common practice that HVAC rated capacity is usually

lower than that of HVDC, the required number of clusters with HVAC cables is accordingly

larger than that with HVDC cables, which further indicates the introduction of extra platforms

possibly uplift the total cost for choosing HVAC cables. On the other hand, converters to

facilitate HVDC transmission are much higher than power transformer for HVAC
127
transmission. Literally, the choice of transmission option is not only associated with the

transmission cable cost itself, but also concerned of substation cost. If only given several

transmission options, it is simple to work out by brute force. A more general comparative

optimization model is presented here to determine the best choice for transmission system.

nsac
obj. min 1 −  ( DC
)  (C (C ) + C (C ))P S ac tr S
S ac =1
nsac
+ n c ,S ac
 UHCc  d S ac
S ac =1 cC ac
(4.40)
nsdc
+ DC
 (C (C ) + C (C ))
P S dc svc S dc
S dc =1
nsdc
+ n c ,S dc
 UHCc  d S dc
S dc =1 cC dc

s.t.
nsdc
 DC
  n
dc dc
c ,S dc
 M  DC (4.41)
S =1 cC

(1 −  )  C
DC
S ac
 n c ,S ac
 HCc (4.42)
cC ac

nc ,S ac  nc',S ac = 0 , c,c'  C ac , S ac (4.43)

 DC  CS  dc n  HCc (4.44)
c ,S dc
cC dc

nc ,S dc  nc',S dc = 0 , c,c'  C dc , S dc (4.45)

The objective is the total cost of substation and transmission cost, C P , Ctr and CSVC

are cost functions of offshore platform, power transformer, and SVC converters, as listed in

eq.(4.10)-eq(4.12), all dependent on the rated capacity of each cluster. CS ac and CS dc are

the fixed capacity of clusters after clustering considering HVAC and HVDC, respectively. In

addition, the transmission cable cost is involved as a permutation and combination of

available cable types. The objective aims to find out the most economic transmission method

to connect the obtained cluster centre in the last step with onshore PCC, and help to determine

128
which is more preferable between HVAC and HVDC, which cable type is the most appropriate

one used. Accordingly,  DC , nc ,S ac , nc ,S dc are the integer decision variables to be optimized in

this model, denoting whether HVDC transmission is used, number of HVAC cable type c

used for substation S in AC options, number of HVDC cable of type c used for substation

S in DC options. UHCc represents the unit cost for transmission cable of type c . d S is

the distance to shore for substations of HVAC and HVDC cases. nsac and nsdc are number

of clusters for AC and DC options, fixed in clustering step.

The first constraint relates  DC to nc ,S dc , if no HVDC cables are chosen

nsdc
(  n
dc dc
c ,S dc
= 0 ), indicates HVDC transmission technique is not chosen, namely  DC = 0 ,
S =1 cC

vice versa. The following constraints guarantee: (1)the selected cable type is capable for the

substation ratings; (2) the cable type used for one substation is limited to 1. Literally, for a

400MW substation, it is not acceptable to use one 300MW cable with one 100MW cable.

Accordingly, the power flow is assumed to be equal for the transmission cables given cases

of multiple cables applied. Thereinto, eq.(4.42) and eq.(4.43) is related to HVAC transmission,

while eq.(4.44) and eq.(4.45) is related to HVDC transmission. Take HVAC transmission for

an illustrative example. Eq.(4.42) requires the installed HVAC cable is capable to withhold

the rated capacity of substation S under AC assumption, where the left-hand side of the

equation represents the rated capacity, and the right represents the total installed cable capacity.

Eq.(4.42) confines the cable type installed in one substation should be the same. nc ,S ac is a

integer variable to represent the number of cable type c for substation s . if nc ,S ac = 1

indicates this cable type c is chosen, other cable type c' is forced to be zero according to

this constraint. It is formulated as a mixed-integer linear problem, which can be solved by a

great deal of solvers.

129
4.3.4 Inner-cluster layout optimization

After applying Fuzzy clustering technique in offshore substation optimization layer, the

whole offshore wind farm has been segmented in an appropriate manner (maximize the

overall membership degree between wind turbines and each offshore substation), where the

location of each cluster and the final cluster result of each individual wind turbine has been

provided. Namely, all the location information available for both wind turbines and the cluster

centre is on hand. What is to be considered in the next step is to optimize the connection route

for each wind turbine to the cluster centre located at a fixed location, which is similar to the

research work done in Chapter 3. The overall connection within each cluster is summarized

as inner-cluster layout optimization. Each individual cluster can be regarded as an

independent small-scale offshore wind farm so that its inner collector system layout can be

optimized independently and simultaneously.

The general aim of the hierarchical layout optimization framework is to decompose the

complicated layout optimization problem step-by-step. The offshore substation optimization

layer is to simplify the problem to optimization within each individual offshore substation,

which can be further optimized in the following two layers. The concept of simplification is

consistent in the overall optimization. Generally, the collector system layout can be

considered from the view of each individual wind turbine. For each wind turbine, it is

necessary to determine which wind turbine injects power to the specific wind turbine (the

downstream neighbour) and which wind turbine absorbs power from the specific wind turbine

(the upstream neighbour). Similarly, the same procedure should be done on the upstream

neighbour and the upstream neighbour of the upstream neighbour, until it comes to the

destination, “substation”. Literally, if the downstream and upstream neighbour WT of any

WTs in the cluster have been identified, the inner-cluster layout of the specific cluster can be

drawn. However, it is quite a complicated problem to involve a large number of binary

variables to represent the on/off of the connection route between any two wind turbines.

130
To simplify the inner-cluster layout optimization, the collector system layout is

simplified by two layers: inter-array feeder optimization layer and submarine cable section

optimization layer. To imitate the clustering technique in substation layer, the complete

collector layout problem can be decomposed into two steps: initially, to determine which

feeder is appropriate to put the wind turbines; the connections within the feeder. Thereinto,

the inter-array feeder optimization layer further divides WTs in the same cluster into different

feeder clusters and aims to minimize the total cost of all the feeders. After properly locating

wind turbines into feeders, the way how to connect WTs put in one feeder is determined in

submarine cable section optimization layer. Instead of considering the possible connections

between one WT and all the other WTs in the cluster, only WTs in the same feeder are

considered as alternatives for upstream and downstream WTs, which reduces a large number

of variables and simplifies the problem.

As for inter-array feeder optimization layer, the principal is to determine the

subordination of wind turbines to feeders. Initially, some candidate feeders have been selected

for each cluster. The principal to screen out feeders has been explained and verified in Chapter


3. The set F = f1 , f 2 ,..., f n f ,s
 represents the candidate feeders for one cluster s , where

n f ,s is an integer variable to represent the number of feeders in cluster s ; the set


P = p1 , p2 ,..., pnWT ,s  represents all the wind turbines clustered to cluster s after offshore

substation optimization layer, where nWT ,s denotes the number of wind turbines in cluster

s . The number of candidate feeders is usually larger than the required minimum number of

feeders in the optimum result. Therefore, a binary variable z sf is introduced to determine

which one of the candidate feeders is chosen to wind turbines. Besides, x p , f is a binary

decision variable to determine whether wind turbine p is clustered to feeder f . To realize

the constraints regarding inner-cluster topology in Eq. (4.20), a set of constraints are applied

131
to the above two sets of variables, as shown below.

x f
p, f = 1, p  P (4.46)

z sf   x p , f  NWT
max s
, f  z f , f  F (4.47)
p

nWT , f =  x p , f , f  F (4.48)
p

n f ,s =  z sf (4.49)
f

where Eq. (4.46) guarantees each individual wind turbine can be put on only one feeder,

consistent with the constraint fi  f j =  in Eq. (4.20). Besides, if a feeder f is not

chosen, no WTs can be put on that feeder. From Eq. (4.47), z f = 0 forces
s
x
p
p, f =0 ,

s
exactly same as the abovementioned constraints. On the meanwhile, if z f = 1 , the number of

wind turbines should be larger than 1, otherwise, it is not necessary to set up this feeder; the

number of wind turbines should be smaller than the maximum required number of WTs on

feeders, satisfying the maximum capacity of MVAC submarine cables. In fact, the direct

variable of this inter-array feeder layer is x p , f . All the other variables are intermediate

s
variables including the choice of feeders, z f , the number of wind turbines on feeder f ,

nWT , f , the number of feeders in the cluster, n f ,s . Similar to Chapter 3, Nature Aggregation

Algorithm, a newly proposed evolutionary algorithm, has been applied to solve the problem.

Similar to other EAs, a population of individuals are maintained and upgraded in each

iteration towards the global best optima. Each individual is a complete answer to the inter-

array feeder optimization layer, coded based on binary coding, as shown in Fig. 4.3. The string

of codes is separated into feeder sections, where the total number of codes in each feeder

section is equal to the number of wind turbines in the specific cluster.

132
Figure 4.3 Inter-array feeder layer encoding

With each individual representing the feeder information, the specific connection

scheme between wind turbines on this feeder is still to be figured out, in order to work out the

collector system cost of each cluster, which refers to the submarine cable section optimization

layer. The problem is to connect a set of wind turbines radially to the substation node in a cost

minimization manner, which can be easily solved by Prim’s algorithm. In this layer, the

appropriate cable type is determined for each section on the feeder, to avoid overload and

guarantee the secure and reliable operation, which is consistent with the constraint S sf  MCc

in Eq. (4.20). Literally, the submarine cable section optimization layer determines the

placement of appropriate cable type in each section of a feeder. The specific feeder cost can

be worked out with corresponding cable cost and section length. Accordingly, if taking each

individual feeder cost into the individual, the collector system cost of the cluster (with feeder

cluster same as the individual) can be worked out as well, which can be regarded as the fitness

of each individual in the population. The fitness of each individual can help the population

evolve the global optima, literally best feeder partitions for wind turbines in one cluster.

133
4.3.5 Cost definitions between wind turbines and cluster centres

In the offshore substation optimization layer, the wind turbines are divided into several

clusters. Wind turbines are usually partitioned merely according to their distance to each

cluster centre when fuzzy clustering technique is used to partition wind turbines into clusters.

However, the clustering result can only guarantee the sum of “distance to sub” of each wind

turbine, rather than to minimize the overall cost of offshore wind farm. Additionally, the

involved cable cost cannot be verified as linear to distance only. In this hierarchical layout

optimization framework, the cost of collector system, transmission system and substation

system are involved in the objective besides distance to evaluate the wind farm clustering.

Distance between wind turbines and cluster centres is not the only metric to update the

subordination membership of wind turbines.

CCollectorij is defined as the total cost of connecting wind turbine j to substation i ,

where substation i could be any substation i  S . The collector cost between wind turbine

j to substation i covers the levelized investment cost, annual maintenance cost and annual

operational cost and the way of calculating CCollectorij follows the expression in Eq. (4.2)- Eq.

(4.5). To update the membership matrix U and cluster location, the collector system cost of

a wind turbine to any cluster centres, including the cluster which it belongs to and also which

it does not belong to, is supposed to be calculated. For the case that substation i is the cluster

of wind turbine j , CCollectorij is exactly the total cost between the specific wind turbine and

substation along the feeder in the optimum result. All connected sections between the target

wind turbine and the substation on the feeder are taken into consideration. An illustrative

example of collector system cost between wind turbines and its clustered substation has been

shown in Fig. 4.4, where an optimized feeder connection is provided. In Fig. 4.4(a), the target

WT located at 2 sections away from the centre, the cost of red-coloured sections is the

collector system cost. In Fig. 4.4(b), the affected sections are coloured in red, while the circled
134
branch is not included since it is not part of the route between WT and substation.

Figure 4.4 Illustrative example for collector cost between Node j and its own
Substation
For the case that substation i is not the cluster of wind turbine j , the collector cost

calculation strategy is supposed to be modified since the route between wind turbine j to

substation i is not connected. A fictive connection should be made up to obtain a complete

route. In this framework, a search of WTs clustered in substation i is executed in order to

find out the WT j' closest to wind turbine j . A fictious connection between WT j' and

wind turbine j is added in order to obtain a through path. Fig. 4.5 provides an illustrative

example for the case when substation i is not the cluster which wind turbine j belongs to.

The dotted line indicates the fictious connection between wind turbine j and its closest wind

turbine j' in substation i . The collector cost in this case should cover the red-coloured

path and the fictious path.

Sub i

Node j

Node j (Sub i

Figure 4.5 Illustrative example for collector cost between Node j and other
Substations
CSUBij and CTransij quantize the impact of substation and transmission cost between

wind turbine j and cluster i , which are related to the calculated cost of substation i ,

135
CSUBi and corresponding transmission cost CTransi . Thereinto, the detailed equation of

calculating CTransi is provided in Eq. (4.16) to Eq. (4.19), while the detailed equation of

calculating CSUBi is provided in Eq. (4.6) to Eq. (4.9). The impacts of transmission cost and

substation cost on each wind turbine are assumed to be the same so that both of these two

components are equally shared among all the wind turbines as shown in Eq. (4.23) and Eq.

(4.24).

4.3.6 Flow chart for the hierarchical optimization framework

The detailed process of the proposed hierarchical optimization framework is described


as below, and its flow chart diagram is shown in Fig.4.6.

1) Pre-define the proper number of clusters

2) To accelerate the speed of convergence, the initial membership degree matrix and centre
location are obtained by applying a fuzzy clustering only based on the distance data
between wind turbine and potential substation. It is designed to be solved by the fuzzy
logic toolbox in MATLAB.

3) With the membership degree between wind turbines and substations, automatically
allocate the wind turbines to appropriate substation to maximize the total membership
degree of the overall OWF.

4) Transmission option for multiple substations are determined.

5) After clustering WTs into specific substation, it is required to further cluster the WTs
into appropriate feeders, which is the inter-array feeder clustering optimization.
According to the provided centre location, several nearest WTs are picked out as
candidate feeder. Thereby, it is formulated as a problem of clustering WTs of one
substation into proper feeders/candidate WTs. When coming to the feeder level, the
optimal cable type chosen for each section on this feeder is supposed to be determined.
Accordingly, cost of each individual functional part can be calculated.

6) Evaluate the real cost obtained in this iteration. Also, calculate the virtual cost CCollectorij ,

CSUBij and CTransij according to the given rules.

7) Substitute these cost factors to the iterative formula to update membership degree matrix
136
and substation location. Determine whether the change in substation location is smaller
than the pre-set tolerance, if yes, the optimal partition of WT into offshore substation
with proper location is obtained; otherwise, return to step 3

Start

Collect necessary input


data for OWF

Determine the number of clusters

Substation Layer Execute the Fuzzy clustering to


update U and Centre location

WT allocation

Transmission option
selection

Inter-array feeder layer Submarine cable section layer


Initialize the
population

Evaluation of fitness Cable connection & cable type


value and constraints determination

No Perform NAA to update


the population Prim s algorithm

If the terminal
No For each individual
conditions are met?

Yes

Output Collector cost/


Substation cost/ Trans cost

If the terminal
conditions are met?

Yes
Optimal system
topology

Figure 4.6 Flow chart of the Hierarchical optimization framework for OWF
layout design
4.4 Case studies

The hierarchical optimization framework for large OWFs with multiple substations has

been verified on two specific cases in this section. To verify its effectiveness and applicability,

two cases catering to different scales of OWF have been taken into consideration, one with

moderate number of offshore WTs (80 WTs) and the other with a large number of offshore

WTs (150 WTs), both of which can be separated into proper clusters by applying fuzzy

clustering method to minimize the total levelized cost. Additionally, both HVAC and HVDC
137
technology have been covered in the model, which is also a choice able to be decided by this

optimization framework. These two cases are described in detail in the following sections,

respectively, both of which are organized as follows. Initially, a brief case description

regarding the offshore wind farm has been introduced in the first subsection. Then, according

to a set of input data, proper pre-processing is presented, followed by the optimization result

as well as analysis for both cases. Finally, the conclusion of this chapter is drawn in section

4.5.

4.4.1 Case 1: moderate-scale OWF

a. Case description
The proposed framework has been applied to a real OWF, called “Banc de Guèrande”

to verify its effect, which has been built in 2015. The OWF is composed of 80 wind turbines,

each with a rated generating capacity of 6MW, which, thereby, has a total rated generating

capacity of 480MW. Similarly, the optimal sitting of each individual WTs is the result of the

preliminary macro-sitting and micro-sitting research work, thereby not included in this case

study. Accordingly, locations of all the wind turbines are already known. The turbine layout

of this OWF is represented in Fig.4.7, where each of the black dots represents one individual

wind turbine. The irregular site of OWF covers an area of about 80 square kilometres.

According to the figure, 80 wind turbines are distributed in 7 rows and each of the rows has

7-13 wind turbines. Both row and column are separated by a mean distance of about 1 km,

approximately equal to seven times of the length of wind turbine diameter, which is also the

common practice to sit between any pairs of wind turbines. It is important to mention that the

coordinates of WTs used in this case study is not the actual wind farm since it is not open to

public access. Therefore, the researcher has estimated the coordinates for all the WTs

according to the accessible published report for the project “Banc de Guèrande”.

138
Figure 4.7 Wind turbine sitting in the OWF “Banc de Guèrande”

The point of common coupling (PCC) is assumed to be located at the coordinates (-20,

-20). However, PCC is not marked in Fig.4.7 to improve the readability of the layout figure,

which mainly focuses the internal topology of the offshore wind farm region, since the

relatively long distance between the PCC and the OWF makes the detailed topology between

wind turbines harder to figure out.

The collector system voltage level is assumed to be 33kV, while it is assumed to be 220

kV for the transmission system. Reliability is not covered in this chapter so that all the devices

involved such as wind turbines and MV or HV cables are assumed to be 100% available. The

price of non-served energy is set as 80$/MWh, used to evaluate the cost of losses.

With all the available, necessary location information and price parameter, the main

concern is literally about how to transmit the generated wind power from these 80 wind

139
turbines to PCC with high cost performance while satisfying a set of operational constraints.

More importantly, with limited capacity on transmission cables and substations, it matters

how to properly cluster all WTs into different clusters, or substation in this context. The

location of each cluster center, also referred to as substation in this context, can be optimized

as well.

b. Input data and pre-processing for the optimization


The input data in regard to the basic information of this windfarm is listed in Table 4-2

as below, all of which has been mentioned in the project description in section a.

Table 4-2 Parameters of necessary input of the OWF


Parameter Value
PCC coordinate (-20, -20)
Number of wind turbines 80
Rated power of wind turbine (MW) 6MW
Rated capacity of wind farm (MW) 480MW
Rated voltage level in collector system(kV) 33kV
Rated voltage level in transmission system(kV) 220kV
Lifetime of the wind farm(years) 25
Interest rate 4%

This hierarchical framework can be literally separated into 3 layers: substation layer (to

determine how to cluster wind turbines into substations), feeder layer (to allocate wind

turbines to proper feeders) and submarine cable connection layer (to connect wind turbines in

one feeder with a proper connection scheme in proper cable size). Correspondingly, the

available high voltage transmission cable types and medium voltage array cable types pose a

direct limit on maximum number of wind turbines in one partition, and maximum number of

wind turbines in one array cable, thereby indirectly affecting number of partitions, and number

of feeders.

Take this case for example. A set of three-core copper conductor submarine cables are

utilised, whose cross-sectional areas vary between 50mm2 and 1000mm2, listed in Table 4-3

as below.

140
For the submarine cables with maximum cross-section area (1000 mm2), the current

carrying capacity is 950A. With pre-defined high voltage level of 220 kV, its capacity can be

calculated:

3  220kV  950 A  362MW

The rated power of wind turbine used in this wind farm is 6MW. The number of wind

turbines this cable is able to carry can be worked out as below.

362  6  60wind turbines

Table 4-3 Cable parameters of the utilised three-core copper conductor medium
voltage submarine cables
Cross-section area Conductor resistance Current carrying
(mm2) (  / km ) capacity (A)

50 0.641 185
70 0.443 215
95 0.320 255
120 0.253 300
150 0.206 335
185 0.164 370
240 0.125 430
300 0.100 480
400 0.0778 550
500 0.0605 630
630 0.0469 745
800 0.0367 850
1000 0.0291 950

With 80 wind turbines in this wind farm, the partition number must be no less than 2.

Literally, this wind farm is supposed to be divided into two clusters based on the available

cable parameters. It will lead to poor reliability of the wind farm if all 80 wind turbines are

allocated to one single substation. A breakdown in the substation contributes to the generation

loss of the whole OWF. Meanwhile, the MV submarine cable cost increases since the total

cable length increases. Instead, a larger number of partitions can uplift the overall reliability

and help reduce the MV submarine cable cost, however, at a sacrifice of increase in substation
141
cost and transmission cost. More substations lower the impact of substation breakdown on

generation loss and meanwhile reduce distance from each wind turbine to substation, both of

which are benefits of more partitions. However, one more partition refers to one more

platform as well as one more electric power substation (either transformer or converters, and

switchgears). Additionally, no mesh topology is considered in transmission network so that

all the substations are connected to PCC in a “one-to-one” manner. One extra substation also

contributes to an independent transmission cable between partitions and onshore PCC. Both

of the two abovementioned factors lower the interest on larger partition numbers. In brief, this

wind farm is partitioned into 2 clusters in order to achieve a balance between collector system

cost and substation/transmission cost. The maximum number of wind turbines in one partition

is 60, satisfying the maximum transmission cable capacity.

To be noted, the choice between HVAC and HVDC cable in transmission system has

been simplified and easy to figure out since the high converter cost reduces the possibility of

applying HVDC transmission in this case. As is well known, HVDC has the strength of lower

cost in bulk power transmission, however, at the cost of higher cost on converters. Namely,

the utilise of HVDC transmission only occurs when the cost of converters can be offset by the

savings on HVDC cables. According to the empirical law, the higher rated capacity and the

longer distance to shore, the more possibilities of choosing HVDC as transmission method.

In this case, if HVAC applied, the transformer cost of both clusters (with each capacity of

about 240MW) should be approx. $4 million, while it reaches up to approx. $120 million for

2 partitions and approx. $80 million for 1 partition in HVDC transmission. The gap of

substation cost between HVAC transmission and HVDC is at least $76 million. As for

transmission cable costs, 500 mm2 3-core cables are utilised, whose unit cost is $0.531 million

per km, while 2000 mm2 HVDC cable is required considering single pole configuration,

whose unit cost is about $1.7 million per km. Accordingly, the unit cable cost of the used

HVDC transmission is higher than that of HVAC transmission, let alone possible to offset the

142
cost in converters. In conclusion, HVAC transmission is applied in this case study, rather than

HVDC transmission.

According to the flow chart, the initial membership degree matrix U 0 and centre

location PC0 are obtained by applying a fuzzy clustering only based on the distance data

between wind turbine and potential substation. Some necessary parameters worked from the

input and initial input of fuzzy clustering are shown in Table 4-4.

The initial cluster centre coordinates are listed in the table, whose distances to shore

can also be worked out. The total 80 wind turbines have been clustered into two partitions,

one with 38 wind turbines and the other with 42 wind turbines, both of which are less than 60,

the maximum number of WTs in clusters. The partition result is shown in Fig. 4.8.

Figure 4.8 Initial Wind farm partition result

143
Table 4-4 Parameter values worked out from input
Parameter Value
Number of partitions 2
Maximum number of wind turbines in each partition 60

Coordinate of initial cluster Cluster 1 (9.246, 9.664)


centres Cluster 2 (16.239, 7.777)
Number of WTs partitioned in Cluster 1 38
the initial partition Cluster 2 42
Rated capacity of clusters Cluster 1 228MW
Cluster 2 252MW
Initial distance between cluster Cluster 1 41.65km
centres and PCC Cluster 2 45.66km
Transmission cable type Cluster 1 500mm2
Cluster 2 500mm2
Current from a fully operated 6MW wind turbine 104A
Maximum number of WTs on an array cable 9
Minimum number of array Cluster 1 5
cables required for each
Cluster 2 5
partition
Minimum number of array cables 9
Maximum number of array Cluster 1 10
cables allowed for each Cluster 2 10
partition

After the initial partition, the transmission option selected for two substations are both

HVAC cable with cross-sectional area of 500mm2.

Similar to Chapter 3, the maximum number of WTs on each MV array cable can be

worked out. According to the rated power of the utilised wind turbine type and chosen rated

medium voltage level, the current from such a fully operated wind turbine can be calculated

as below:

6MW  ( 3  33kV )  104 A

Namely, a fully operated wind turbine with the same ratings will inject 104A to the

feeder where it is connected with. Referred to Table 4-3, the 1000 mm2 cable have the largest

current carrying capacity, 950A. The maximum number of wind turbines connected to one

feeder can be calculated as below:

950A / 104A  9 wind turbines


144
Under the partition of 39 and 41 wind turbines, the minimum number of MV array

cables for each cluster should be:

39  9 = 5
41  9 = 5

where both results need round up to an integer. Hence, both partitions are supposed to have at

least 5 array cables to accommodate their wind turbines. Similarly, 9 feeders are required to

accommodate the total 80 wind turbines. The maximum number of array cables in each

partition have been set as twice of their minimum value, accordingly 10 in both partitions. All

of these parameters affecting the internal topology result are listed in table 4-4.

All of the cost equations are referenced from public-accessible data, obtained from

several publications. However, any other user-defined cost data can be put into this model to

cater to different market or geographical conditions.

All the experiments are carried out by MATLAB version 2019a on a 4-core, 64-bit DELL

pc with Intel Core i5-2400 CPU and RAM 4 Giga byte.

c. Simulation result and analysis


The optimal electric layout topology of the offshore wind farm is shown in Fig.4.9. It

can be observed that the connections between wind turbines, and connections between wind

turbines and offshore substation have no crossings between each other, which justify the

effectiveness of avoiding cable crossing criterion. The only crossings occur between array

cables and transmission cables, which, however, can be easily avoided through a detour on

transmission cables.

Initially, the cluster result, listed in Table 4-5, has been updated thru multiple iterations,

compared with the initial settings in Table 4-4. The essential information of the final cluster

result is listed in the table below. The number of partitions of fuzzy cluster has been set as 2,

apparently same as the initial value in Table 4-4. Both centers have been moved a little bit

from their initial center location, where cluster 1 center has moved further from PCC and

cluster 2 center has moved closer to PCC. The obvious difference through fuzzy iterations lies
145
in the new distribution of wind turbines among the 2 clusters. Previously, the 80 wind turbines

have been divided into 38/42, while they have been evenly distributed into 2 clusters, with

40/40 in both clusters.

Figure 4.9 Optimal connection topology of the wind farm network.


Table 4-5 Optimal cluster result
Parameter Value
Number of partitions 2
Maximum number of wind turbines in each 60
partition
Coordinate of initial cluster Cluster 1 (10.00, 9.37)
centres Cluster 2 (15.67, 7.87)
Number of WTs partitioned in Cluster 1 40
the initial partition Cluster 2 40
Rated capacity of clusters Cluster 1 240MW
Cluster 2 240MW
Distance between cluster Cluster 1 41.98km
centres and PCC Cluster 2 45.27km

146
The detailed cost for MV inter-array cables, substation and HV transmission cables in

this case study are presented in Table 4-6, Table 4-7 and Table 4-8, respectively.

In terms of MV collector system, the total number of feeders in the optimal topology is

10, larger than 9, the minimum number of feeders calculated in the last section. According to

the partition of 40/40, both clusters are supposed to have at least 5 feeders, which is exactly

same as the actual feeder numbers in the optimal result. Each of the feeders is represented in

different color to differentiate one from another. 10 wind turbines closest to the corresponding

substations are directly connected to cluster center and used to denote the feeder.

The cable type suitable for each section depends on the power injected from the

downstream of the feeder. According to the optimization result, the number of cable sections

are summed up as below: 31 branches of cable of size 3*50 mm2, 11 branches of cable of size

3*70 mm2, 9 branches of cable of size 3*150 mm2, 7 branches of cable of size 3*240 mm2, 2

branches of cable of size 3*400 mm2, 7 branches of cable of size 3*500 mm2, 4 branches of

cable of size 3*630 mm2, 6 branches of cable of size 3*800 mm2 and 3 branches of cable of

size 3*1000 mm2. String and mixed star and string connection schemes are utilized to connect

the wind turbines in the collector system. The topology is fully optimized according to the

input data regarding location and prices, and the connection scheme is different from the

common practice to connect wind turbines in one row or column radially to the substation.

The investment cost of MV submarine cables accounts for 46.3% of the total MV cable

cost. The remaining are annual maintenance cost and estimated operational cost, taking up

28.5% and 25.3%, respectively.

147
Table 4-6 MV submarine cable cost breakdown
Optimization variable Optimized value
Number of MV feeders 10
Number of WTs per feeder Between 6 and 9
Number of feeders in different clusters (5,5)
Type of connection scheme String, mixed string and star
Levelized investment cost[M$] 1.5150
Annual maintenance cost [M$] 0.9327
Operational cost [M$] 0.8274
Total MV submarine cable cost [M$] 3.2751
MV cable cost breakdown
Cross-sectional Number of Length [km] Levelized investment
area [mm2] branches Cost [M$]
50 31 32.0691 0.4174
70 11 11.7160 0.1577
150 9 9.5609 0.1489
240 7 7.5049 0.1329
400 2 2.2625 0.048
500 7 7.6866 0.1855
630 4 4.8659 0.1432
800 6 4.8398 0.1728
1000 3 2.5094 0.1086
In terms of substation, it is composed of platform, power transmission (HVAC

transmission applied in this case) and auxiliary protection equipment. In accordance with the

proposed hypothesis, there is one transformer per substation, so that the transformer cost is

fixed as long as the wind farm capacity has been pre-specified. In the optimal result, the

partition in the OWF is 40/40 and both clusters have the rated capacity of 40*6=240MW.

Accordingly, the costs on power transformers are the same for both substations. Also, it is

assumed that both substations are connected with PCC independently. The number of HVSG

required should be 1 for both substations (equal to the quantity of transmission cables). The

number of MVSG is same as number of feeders in both clusters, which is 5 for both substation

1 and 2. All the above-mentioned costs are same for both substations, which contributes to

the same total substation cost.

148
Table 4-7 Substation cost breakdown
Substation 1
Substation 1 coordinate (10.00, 9.37)
Number of WTs 40
Transformer capacity [MW] 240
Transformer cost [M$] 0.1572
Platform cost [M$] 1.485
Number of MVSG 5
MVSG cost [M$] 0.175
HVSG cost [M$] 0.0392
Total levelized investment cost [M$] 1.8505
Annual maintenance cost [M$] 0.7502
Estimated operational cost [M$] 6.7277
Total cost for substation 1 [M$] 9.3283
Substation 2
Substation 1 coordinate (15.67, 7.87)
Transformer capacity [MW] 240
Transformer cost [M$] 0.1572
Platform cost [M$] 1.485
Number of MVSG 5
MVSG cost [M$] 0.175
HVSG cost [M$] 0.0392
Total levelized investment cost [M$] 1.8505
Annual maintenance cost [M$] 0.7502
Estimated operational cost [M$] 6.7277
Total cost for substation 2 [M$] 9.3283
Total substation cost [M$] 18.6566
The operational cost of substation accounts for the largest portion of the total substation

cost, 72.1%. The remaining are the total levelized investment cost and maintenance cost,

taking up 19.8% and 8.1%, respectively.

In terms of transmission system, both clusters are assumed to connect with onshore

PCC in a “one-to-one” manner. To withhold 40 wind turbines in both clusters, the transmission

cable should have at least 240MW rated capacity. 500 mm2 HVAC 3-core copper submarine

cables whose capacity is 3  220kV  630 A = 240MW , are applied to connect both

substations with onshore PCC. The different distance between substations to shore contributes

to the different investment cost on transmission cables, $1.6499 million for sub 1 and $1.7788

million for sub 2. Overall, investment cost, annual maintenance cost and operational cost

amount up 52.1%, 21.1% and 26.8%.

149
Table 4-8 Transmission cost breakdown
Substation 1
Cross-sectional area [mm2] 500
Cable length [km] 41.9885
Levelized investment cost [M$] 1.6499
Annual maintenance cost [M$] 0.6689
Operational cost [M$] 0.8475
Total transport cable cost (substation 1) [M$] 3.1663
Substation 2
Cross-sectional area [mm2] 500
Cable length [km] 45.2681
Levelized investment cost [M$] 1.7788
Annual maintenance cost [M$] 0.7211
Operational cost [M$] 0.9136
Total transport cable cost (substation 1) [M$] 3.4136
Total transport cable cost [M$] 6.5799

As explained in the last section, HVAC cables have been utilised in this case. It can

also be verified via the empirical equation of the break-even distance of HVDC technology.

 
lDC = max 40 ,min 200 , 832.5  PN−0.4 
If the OWF is also partitioned into 40/40, PN , referring to the rated capacity, is 240MW.

Take 240MW into the above equation,

 
lDC = max 40 ,min 200 , 832.5  240 −0.4  = 92.96km
The distances between clusters and PCC are 41 and 45 km, both less than the critical

HVDC distance, which also verifies that HVAC rather than HVDC should be used to transmit

the power from both partitions to shore.

Also, if all wind turbines in the OWF is not partitioned, PN , referring to the rated

capacity, is 480MW. Take 480MW into the above equation,

 
lDC = max 40 ,min 200 , 832.5  480 −0.4  = 70.45km
According to the centre coordinate of 2 partitions, the distance between the centre and

PCC are supposed to be 42km, less than the critical HVDC distance. Consequently, it is also

the HVAC cable that connects the whole OWF to onshore PCC.

150
Overall, the levelized cost of this OWF is $28.5166 million, where collector system,

substation and transmission system take up 11.5%, 65.4% and 23.1%. Obviously, substation

becomes the costliest part due to the expensive power transformer and platform to sit the

transformer. Transmission system takes the second costliest place, in light of high unit cost of

high voltage submarine cables and long transmission distance. Collector system, in turn, is

composed only of MV submarine cables, whose unit cost is relatively low, and length is short,

usually about 1km so that it is the least expensive part of overall OWF cost.

4.4.2 Case 2: large-scale OWF

a. Case description
In this case, the hierarchical optimization framework is applied to a 150-WT windfarm

to verify its efficiency to deal with large-scale systems. The OWF is composed of 150 wind

turbines, each with a rated generating capacity of 10 MW, which, thereby, has a total rated

generating capacity of 1500 MW. Similarly, turbine layouts are the result of preliminary result

and regarded as an input of this study, which has been shown in Fig. 4.10. The OWF covers

an area of about 150 square kilometres. According to the figure, 150 wind turbines are

distributed in 10 rows and each of the rows has 11-19 wind turbines. The distance between

different rows and columns are set approximately equal to seven times of the length of wind

turbine diameter, same as in Case 1.

The point of common coupling (PCC) is assumed to be located at the coordinates (12,

-45). Similarly, PCC is not marked in the layout figure to improve the readability of the layout

figure, in order to mainly focus on the internal turbine layout topology of the offshore wind

farm region. According to the wind turbine coordinate, the WT nearest to shore in the wind

farm is still 68km away from the onshore PCC. Literally, this wind farm is a large-scale and

long-distance-to-shore OWF.

The collector system voltage level is assumed to be 66kV, while the transmission

system voltage is dependent on the transmission cables selected for the transmission system.

151
All the other cost parameters like cost of energy are still same as in Case 1, so as not to be

repeated here.

With all the available, necessary location information and price parameter, it needs to

be determined how to transmit the generated wind power from these 150 wind turbines to

PCC with high cost performance while satisfying a set of operational constraints. It is

obviously not practical to collect from all the wind turbines into one single substation, either

Figure 4.10 Turbine layouts of the 150-WT wind farm


from the view of economic and reliability. It matters how many clusters to cluster these wind

turbines as well as how to properly cluster all WTs into clusters, or substation in this context.

Meanwhile, the maximum capacity of the cables is supposed not to be exceeded to guarantee

the secure operation of the whole system. The location of each cluster center, also referred to

as substation in this context, can be optimized as well. With multiple cable types included in

the case study, the choice between AC and DC transmission in transmission system can also
152
be clarified through the optimization process.

b. Input data and pre-processing for the optimization


The input data in regard to the basic information of this windfarm is listed in Table 4-9

as below, all of which has been mentioned in the project description in section a.

Similar to case 1, the hierarchical framework can be literally separated into substation

layer, feeder layer and submarine cable connection layer, where all of these layers are

mutually affected and also limited by corresponding equipment capacity.

Table 4-9 Parameters of necessary input of the OWF


Parameter Value
PCC coordinate (12, -45)
Number of wind turbines 150
Rated power of wind turbine (MW) 10MW
Rated capacity of wind farm (MW) 1500MW
Rated voltage level in collector system(kV) 66kV
Lifetime of the wind farm(years) 25
Interest rate 4%

In terms of substation layer, the cluster size (also referred as number of WTs with

specified rated capacity) is limited by power ratings of the transmission cables. The number

of clusters is mutually affected by each cluster size. In this case study, both HVAC and HVDC

cables are transmission alternatives, whose parameters are listed in Table 4-10 and Table 4-

11.

Table 4-10 Cable parameters of submarine HVAC cables

Cross-section Rated voltage Conductor resistance Current carrying


area (mm2) (kV) (  / km ) capacity (A)

1000 220 0.0291 950


1200 220 0.022 1030
1400 220 0.0167 1118
1600 220 0.0104 1195
1800 220 0.0043 1271

153
Table 4-11 Cable parameters of submarine HVDC cables

Conductor
Cross-section Rated voltage resistance Current carrying
area (mm2) (kV) capacity (A)
( m / km )
630 320 31.4 836
800 320 28.4 1207
900 320 25.4 1425
1000 320 22.4 1644
1200 320 19.2 1791
1400 320 16.5 1962

Both HVAC and HVDC cables are included in the optimization. Initially, for the

submarine HVAC cables with maximum cross-section area (1800 mm2), the current carrying

capacity is 1271A. With pre-defined high voltage level of 220 kV, its capacity can be

calculated:

3  220kV  1271A  484MW

The rated power of wind turbine used in this wind farm is 10MW. The number of wind

turbines this cable is able to carry can be worked out as below.

484  10  48wind turbines

With 150 wind turbines in this wind farm, the partition number for HVAC transmission

must be no less than 4. Literally, this wind farm is supposed to be divided into 4 clusters based

on the available HVAC cable parameters. The maximum number of wind turbines in one

partition is 48, satisfying the maximum transmission cable capacity.

If HVDC transmission is considered in this case study, a single polar configuration is

assumed in the case study. The maximum current carrying capacity listed in the table is 1962A.

Under the rated voltage of 320kV, its capacity can be calculated:

320kV  1962A  627MW

The number of wind turbines this cable is able to carry can be worked out as below.

154
627  10  62 wind turbines

With 150 wind turbines in this wind farm, the partition number for HVDC transmission

must be no less than 3. Literally, this wind farm is supposed to be divided into 3 clusters based

on the available HVDC cable parameters. The maximum number of wind turbines in one

partition is 62, satisfying the maximum transmission cable capacity.

To sum up, HVAC and HVDC cables contribute to different number of clusters: 3

clusters in HVDC transmission and 4 clusters in HVAC transmission. After the clustering

result has been fixed, the transmission option selection provides the best optimal transmission

option for this case, which is HVDC transmission with one 1000mm2 HVDC cable for all

three substations. The comparison result using HVAC is shown in the neighbouring column.

To verify the optimal selection of transmission cables, the following will be carried on under

two cases: one still with optimal HVDC option, the other only considering HVAC option.

Figure 4.11 Initial Wind farm partition result for HVAC transmission

155
Figure 4.12 Initial Wind farm partition result (HVDC)
Accordingly, the initial membership degree matrix U 0 and centre location PC0 are

obtained by applying a fuzzy clustering only based on the distance data between wind turbine

and potential substation. Some necessary parameters worked from the input and initial input

of fuzzy clustering are shown in Table 4-12. The initial cluster centre coordinates are listed in

the table, whose distances to shore can also be worked out. The total 150 wind turbines have

been clustered into 3 or 4 partitions, where the number of WTs does not exceed the maximum

number of WTs required in clusters. The partition result for HVAC and HVDC are shown in

Fig. 4.11 and Fig. 4.12.

Similar to Case 1, the maximum number of WTs on each MV array cable can be worked

out. According to the rated power of the utilised wind turbine type and chosen rated medium

voltage level, the current from such a fully operated wind turbine can be calculated as below:

10MW  ( 3  66kV )  87.5A

Namely, a fully operated wind turbine with the same ratings will inject 87.5A to the

feeder where it is connected with. Referred to Table 4-3, the 1000 mm2 cable have the largest

156
current carrying capacity, 950A. The maximum number of wind turbines connected to one

feeder can be calculated as below:

950A / 87.5A  10 wind turbines

Take HVDC transmission for example, the whole wind farm has been partitioned into

52, 51 and 47 wind turbines. The minimum number of MV array cables for each cluster should

be:

52  10 = 6
51  10 = 6
47  10 = 5

where both results need round up to an integer. Hence, these 3 partitions should have at least

6, 6 and 5 array feeders to accommodate the allocated wind turbines. Similarly, 17 feeders, at

least, are required to accommodate the total 150 wind turbines. The maximum number of

array cables in each partition have been set as twice of their minimum value, shown in the

table below. All of these parameters affecting the internal topology result are listed in Table

4-12.

Table 4-12 Parameter values worked out from input


Parameter Optimal Value Comparison Value
Number of partitions 3 4
Maximum number of wind turbines in
62 48
each partition
(9.378,32.26)
(9.41, 31.09)
(11.67,26.59)
Coordinate of initial cluster centres (14.89, 29.47)
(16.33,30.21)
(17.95, 24.76)
(18.62,24.54)
Number of WTs partitioned in the initial
52/51/47 40/35/35/40
partition

Rated capacity of clusters 520/510/470MW 400/350/350/400MW

77.30km
76.13km
Initial distance between cluster centres 71.59km
74.53km
and PCC 75.33km
70.01km
69.85km
1*1000mm2 1*1400mm2
Transmission option selected 1*1000mm2 1*1000mm2
1*1000mm2 1*1000mm2
157
1*1400mm2

Current from a fully operated 10 MW


87.5A 87.5A
wind turbine
Maximum number of WTs on an array
10 10
cable
Minimum number of array cables
6/6/5 4/4/4/4
required for each partition
Minimum number of array cables 17 16
Maximum number of array cables
12/12/10 8/8/8/8
allowed for each partition

All of the cost equations are referenced from public-accessible data, obtained from

several publications. However, any other user-defined cost data can be put into this model to

cater to different market or geographical conditions.

All the experiments are carried out by MATLAB version 2019a on a 4-core, 64-bit

DELL pc with Intel Core i5-2400 CPU and RAM 4 Giga byte.

c. Simulation result and analysis

With both HVAC and HVDC considered in this case study, the optimal electric layout

topology of the offshore wind farm is shown in Fig.4.13 and Fig.4.14. It can be observed that

the connections between wind turbines, and connections between wind turbines and offshore

substation have no crossings between each other, which justify the effectiveness of avoiding

cable crossing criterion. The only crossings occur between array cables and transmission

cables, which, however, can be easily avoided through a detour on transmission cables.

158
Figure 4.13 Optimal connection topology of the wind farm network (HVAC)

Figure 4.14 Optimal connection topology of the wind farm network (HVDC)
159
Initially, the cluster result has been updated thru multiple iterations, compared with the

initial settings in Table 4-12. The essential information of the final cluster result is listed in

the table below. The number of partitions of fuzzy cluster has been set as 3 in HVDC, and 4

in HVAC apparently same as the initial value in Table 4-12. As for HVDC case, obviously,

the partitions of wind turbines in the wind farm has been changed, with wind turbines

distributed much more evenly than initial cluster result. From the boundary shape,previously,

the 150 wind turbines have been divided into 52/51/47, while the number of wind turbines in

each cluster is 52/50/48 in the optimum result. All centers have been moved a little bit from

their initial center location, where cluster 1 & 2 center has moved closer to PCC and cluster

3 center has moved further from PCC. As for HVAC case, the partitions of the 4 clusters have

been changed and also their centers have moved apart from their initial set cluster coordinate.

The detailed cost for MV inter-array cables, substation and HV transmission cables in

this case study are presented in Table 4-13- Table 4-16, respectively.

Firstly, the optimum collector systems for both HVAC and HVDC cases are compared.

According to Fig. 4.13 and Fig. 4.14, the collector system topologies in both cases are both

fully optimized according to the input data regarding location and prices, and the connection

scheme is different from the common practice to connect wind turbines in one row or column

radially to the substation. The type of connection schemes in both cases cover string and

mixed star and string. Generally, each individual wind turbine is connected radially from its

substation. Each individual feeder in both cases is represented in different color to

differentiate one from another. Cable type for each connection is dependent on the power flow

in the corresponding section.

However, different predefined number of partitions contributes to completely different

connection topology in collector systems in these two cases. In the case of utilizing HVDC

transmission, the total number of feeders in HVDC case is 19, larger than 17, the minimum

160
Table 4-13 Cost and volume of overall collector system
Optimization variable Optimized value Comparison
Number of partitions 3 4
Transmission technique HVDC HVAC
Number of MV feeders 19 20
Number of WTs per feeder Between 4 and 10 Between 3 and 10
Number of feeders in different
6/6/7 5/5/5/5
clusters
String, mixed string String, mixed string and
Type of connection scheme
and star star
Levelized investment cost[M$] 3.5184 3.0084
Annual maintenance cost [M$] 0.0018 1.2196
Operational cost [M$] 3.5863 1.9263
Total MV submarine cable cost
7.1066 6.1543
[M$]

number of feeders calculated in the last section. According to the partition of 52/50/48, the

minimum required number of feeders in each cluster are supposed to be 6/5/5 feeders, where

the actual number of feeders satisfy the constraint. In comparison, the total number of feeders

in case of HVAC transmission is 20, similar to that in HVDC case. To accommodate the

partitions of 35/40/40/35 wind turbines, the minimum number of feeders for corresponding

partitions is 4. The number of feeders in all the cluster satisfies the constraint. Correspondingly,

with a larger number of feeders in HVAC case, the wind turbines seem distributed more

dispersedly, which can also be seen from the number of WTs on feeders. Generally, the total

cost for collector system in HVAC seems more attractive, 13.4% less than that in HVDC case.

It is consistent with our assumptions that a less collector system cost will be contributed by a

larger number of partitions since each wind turbine in the cluster is closer to their substation.

Intuitively, with regard to the cost breakdown of MVAC cables, cable types with larger rated

capacity are more frequently used in HVDC case than in HVAC case, which should be the

direct reason of higher collector system cost in HVDC. That is because there are more wind

turbines in each cluster in HVDC case.

161
Table 4-14 MV submarine cable cost breakdown
MV cable cost breakdown
Cross-
sectional HVDC HVAC
area [mm2]
Levelized Number Levelized
Number of Length Length
investment of investment
branches [km] [km]
Cost [M$] branches Cost [M$]
50 72 76.66 1.1975 83 85.74 1.3393
120 18 19.61 0.3504 15 16.37 0.2925
185 13 14.24 0.2784 14 15.36 0.3005
300 7 8.15 0.1863 10 10.97 0.2507
400 10 11.59 0.2948 6 6.04 0.1537
500 8 9.72 0.2813 10 10.62 0.3074
630 10 9.78 0.3455 7 5.97 0.2109
800 8 8.64 0.3700 4 2.68 0.1150
1000 4 4.12 0.2142 1 0.74 0.0384

In terms of substation, it is composed of platform, power converter and auxiliary

protection equipment for HVDC transmission, while the AC/DC converter will be replaced

by power transformer with HVAC transmission.

The quantity of either transformer or converter is assumed to be 1 per substation, so

that the transformer cost is only dependent on the rated capacity of each cluster. For example,

in the case of HVDC transmission, the 150 wind turbines have been partitioned into 52/48/50,

so that the rated capacity of each cluster is supposed to be 520/480/500 MW. Accordingly, the

rated capacity of converters installed on the corresponding substation is 520MW, 480MW,

500MW, respectively.

Similarly, in the case of HVAC transmission, the 150 wind turbines have been

partitioned into 35/40/40/35, so that the rated capacity of each cluster is supposed to be

350/400/400/350 MW, exactly the transformer ratings listed in the table below. Generally, the

unit cost of converters is much higher than that of transformers. Accordingly, the total

levelized investment cost for all the 3 substations is $19.314 million in HVDC case , while

that of all the 3 substations is $10.6274 million in HVAC case. The equipment cost in HVDC
162
case is almost twice of that in HVAC case. Correspondingly, the maintenance cost is higher

in HVDC since it is linear with its initial investment cost. However, the huge cost difference

in investment has been offset by the operational cost, set as estimated losses cost in

transformer or converter. The operational cost in HVDC and HVAC cases are $2.9434 million

and $42.048 million, respectively. The converter has a better efficiency than transformer,

which could facilitate a great deal of saving during the lifetime of offshore wind farm.

Table 4-15 Substation cost breakdown


Optimization
Optimized value Comparison
variable
Number of partitions 3 4
(18.1937, 24.8290)

(15.9374, 29.5557)
(10.2301, 30.9728)
Substation
(14.2351, 28.4413)
coordinate (12.0006, 27.1637)
(17.5094, 25.4730)
(9.8405, 31.7026)

Number of WTs in
52/48/50 35/40/40/35
each partition
Transformer
520/480/500 350/400/400/350
capacity [MW]
Levelized
0.2007/0.2219/0.2219/0.2007
transformer/converte 6.5712/6.3048/6.4380
r cost [M$]
Levelized Platform 2.0931/2.3695/2.3695/2.0931
3.0328/2.8117/2.9223
cost [M$]
Number of MVSG
6/6/7 5/5/5/5
in each partition
MVSG cost in each 0.1750/0.1750/0.1750/0.1750
0.2100/0.2100/0.2450
partition [M$]
HVSG cost in each 0.0392/0.0392/0.0392/0.0392
0.0392/0.0392/0.0392
partition [M$]
Levelized
2.5081/2.8056/2.8056/2.5081
investment cost 9.8532/9.3657/9.6445
[M$]
Total levelized
investment cost 28.8634 10.6274
[M$]
Annual maintenance 1.0168/1.1374/1.1374/1.0168
3.9946/3.7969/3.9099
cost [M$]
Total annual
maintenance cost 11.7014 4.3084
[M$]
163
9.8112/11.2128/
Estimated
operational cost 1.0204/0.9419/0.9811
11.2128/9.8112
[M$]
Total operational
2.9434 42.048
cost [M$]
13.3360/15.1559/
Cost for substation
14.8682/14.1045/14.5355
[M$] 15.1559/13.3360

Total substation
43.50 56.9838
cost [M$]
Owing to the high cost in converters, it is found out that the investment cost of

substations with converters accounts for the largest portion of the total substation cost, 66.4%.

However, in substations with power transformers, the operational cost of substation accounts

for the largest portion of the total substation cost, 73.8%. Taking all the cost involved for

substations into consideration, the total substation costs are $43.5 million and $56.9838

million for HVDC and HVAC cases, respectively. In conclusion, HVDC technique seems

more attractive with regard to substation cost.

In terms of transmission system, all the clusters are assumed to connect with onshore

PCC in a “one-to-one” manner. As for HVDC case, with approximate 50 wind turbines in all

the clusters, the transmission cable should have at least 500MW rated capacity. 1000mm2

HVDC submarine cables whose capacity is:

320kV  1644A = 526MW  500MW

are applied to connect substations with onshore PCC.

As for HVAC case, with approximate 40 wind turbines in all the clusters, the transmission

cable should have at least 400MW rated capacity. 1400mm2 3-core cooper HVAC submarine

cables whose capacity is:

3  220  1118A = 426MW  400MW

are applied to connect substations with onshore PCC.

Levelized investment cost in the case of HVDC transmission is $22.9093 million, while

levelized investment cost in the case of HVAC transmission is $30.0474 million. There is a
164
saving of $7.1381 million with utilizing HVDC transmission, the reason of which comes from

the low total transmission length and low unit cost in HVDC. Besides, the low resistance in

HVDC submarine cables contributes to the lower operational cost in HVDC transmission.

Taking all the cost involved for transmission systems, a saving of $11.9542 is generated

between HVDC case ($35.5785 million) and HVAC case ($47.5327), which also justifies the

utilization of HVDC in this 150-wind-turbine wind farm. Overall, for transmission system,

investment cost, annual maintenance cost and operational cost amount up 64.4%, 26.1% and

9.5% in HVDC case, while they amount up 63.2%, 25.6% and 11.2% in HVAC case. In

conclusion, HVDC transmission is superior to HVAC from the transmission system analysis.

Table 4-16 Transmission cost breakdown


Optimization
Optimized value Comparison Value
variable
Transmission
HVDC HVAC
technique
Cross-sectional
1000/1000/1000 1000/1400/1400/1000
area [mm2]
70.1031/74.6596/
Cable length
75.9934/73.4753/70.6880
[km] 72.1637/76.7330

5.3123/9.6210/
Levelized
investment cost 7.9078/7.6458/7.3557
9.2994/5.8147
[M$]
Total investment
22.9093 30.0474
cost [M$]
2.1537/3.9004/
Annual
maintenance cost 3.2059/3.0996/2.9820
3.7700/2.3573
[M$]
Total
maintenance cost 9.2875 12.1814
[M$]
1.4474/1.1554/
Operational cost
1.2600/1.0381/1.0836
[M$] 1.1168/1.5842

Total operational
3.3817 5.3038
cost [M$]
Total transport 8.9134/14.6768/
12.3737/11.7834/11.4214
cable cost [M$]
165
14.1862/9.7563

Total
transmission 35.5785 47.5327
cost [M$]

In conclusion, although the high converter cost drives up the total electric system cost

in HVDC transmission, the total cost is compensated by utilizing HVDC cables in

transmission systems. Therefore, the optimum cluster result for the 150-WT offshore wind

farm is partitioned into 3 clusters with 52, 48 and 50 wind turbines in each cluster, respectively.

The cheaper total cost for HVDC option also verifies the validity of the transmission option

selection process.

On the other hand, the result can be also verified through the empirical equation of the

break-even distance of HVDC technology. For HVDC case, the rated capacity for each cluster

is approximately 500MW ( PN = 500MW )

 
lDC = max 40 ,min 200 , 832.5  PN−0.4  = 69.31km
The distances between clusters and PCC are 75.9934, 73.4753 and 70.6880 km, all of

which are greater than the critical HVDC distance. Literally, it verifies that HVDC rather than

HVAC should be used to transmit the power from both partitions to shore.

Overall, the levelized cost of this OWF in HVDC case is $86.1851 million, where

collector system, substation and transmission system take up 8.2%, 50.5% and 41.3%.

Obviously, substation becomes the costliest part due to the expensive converters and platforms.

Transmission system takes the second costliest place, in light of high unit cost of high voltage

submarine cables and long transmission distance. Collector system, in turn, is composed only

of MV submarine cables, whose unit cost is relatively low, and length is short, usually about

1km so that it is the least expensive part of overall OWF cost.

The proposed hierarchical framework provides an optimum and complete electric

system topology design from far-end wind turbines to onshore integration point for an
166
offshore wind farm based on some justified simplifications, which can be used as a guide for

wind farm developers to set up the offshore wind farm in a cost minimization manner. In

reality, cost may be not the only metric to determine the electric system design. A series of

studies should be performed to guarantee the viability of offshore wind farm under all the

possible situations, which includes but is not limited to short circuit study, harmonics study

transient study, etc. All of these are required to satisfy to guarantee the safe operation while

complying with the grid code. For instance, VSC converters are applied in the proposed model,

whose rated capacity and installed location can be obtained in the outcome. However, with

the installed converters, protection coordination study is required to coordinate the integrated

protection in converters with other protection techniques in the downstream and upstream of

the system. The setup of offshore wind farm is a complicated process, involving

interdisciplinary teams and usually carried out in several phases. The electric system design

is optimized in the manner of cost minimization in this chapter. It is also possible to optimize

the system design by maximizing total energy production, and overall reliability, all of which

give a different view of the problem. The final decision is usually a tradeoff between all the

available options.

4.5 Conclusion

A comprehensive cost model is developed to take all the necessary equipment in a

large-scale offshore wind farm with multiple substations into consideration. The choice of

HVDC and HVAC technique in transmission systems can also be determined in this proposed

cost model. To simplify the complicated layout design process, a hierarchical optimization

framework including offshore substation optimization layer, inter-array feeder optimization

layer and submarine cable section optimization layer, is proposed based on the features of

different parts of offshore wind farms. The fuzzy clustering algorithm is applied in the

offshore substation optimization layer, to properly partition all the wind turbines into a

167
number of clusters with center locating offshore substations. The following inter-array feeder

layer and submarine cable section layer are integrated to optimize the inner system layout of

a cluster. Thereinto, NAA algorithm is applied in inter-array feeder optimization layer to

further partition wind turbines into a feeder, while Prim’s algorithm is utilized to optimize the

connections within a feeder. Finally, the proposed hierarchical framework is validated on two

OWFs, one with 80 wind turbines and the other with 150 wind turbines. This framework has

been verified as an efficient optimization tool for offshore wind farm planners to optimize the

electric system layout design for large-scale offshore wind farms with multiple substations.

Although the planning method considers multiple substations, however, the risk associated

with transmission network still exists, since every substation is connected independently to

shore via its own transmission cables. Any fault occurring on that transmission cable poses a

danger of generation loss from wind turbines belonged to that substation, which is detrimental

to the overall reliability of the offshore wind farm. In the next chapter, a planning method for

a more reliable transmission option in offshore wind farm will be presented .

168
Chapter 5 COMPREHENSIVE INTEGRATION
PLANNING MODEL FOR MULTIPLE REMOTE WIND
FARMS
Offshore wind farms are usually connected to shore via an independent HVAC or

HVDC transmission cable, which is detrimental to the overall reliability of the offshore wind

farm. In this chapter, a more reliable transmission and integration option for multiple offshore

wind farms are discussed and optimized.

5.1 Problem description

Offshore windfarms have attracted global interests due to its better available wind

resources off the coast for governments to achieve their ambitious targets to cut down carbon

emission [111]. These offshore windfarms are typically regarded as remote wind farms located

far away from coast, which poses a challenge to transmit their generated power to the onshore

demand side. Generally, transmission network of offshore windfarms has been classified into

3 categories, multiple HVAC transmission, multiple HVDC transmission and multi-terminal

HVDC network, all shown in Fig. 5.1. As discussed in Chapter 3 and Chapter 4, offshore

windfarms are connected individually to shore via either high-voltage alternating current

transmission (HVAC) or high-voltage direct current transmission (HVDC). The choice of

HVAC or HVDC is dependent on the rated capacity and distance from shore of the

corresponding windfarm. The transmission method considered in Chapter 3 and Chapter 4 is

consistent with the first two categories, multiple HVAC and multiple HVDC, respectively

shown in Fig. 5.1(a) and Fig. 5.1(b). The prominent feature of these two transmission options

is that each individual transmission between offshore windfarms (or clusters in one large-

scale windfarms) is completely independent from each other. Literally, the transmission

systems of different offshore windfarms (or clusters) can be regarded as independent

169
individuals so that the following analysis like power flow analysis can be treated

independently. These two transmission options are actually simple to deal with since it is

simply a “one-to-one” connection between the offshore windfarm and the proposed PCC. The

only variable to be determined is to select an appropriate high voltage submarine cable type,

as discussed in the previous two chapters. However, its drawback cannot be neglected in

consideration of the overall reliability of the whole system. In case of any failure in the

submarine transmission cable, these two transmission options contribute to loss of generation

from the whole offshore windfarm. Additionally, the submarine cable fault cannot be

guaranteed to be removed immediately after being detected in view of the severe weather

condition over the sea. Therefore, how to improve the reliability of transmission options is of

great importance to deal with.

Fig. 5.1(a)
Offshore Wind Farm I VSC Station 1 VSC Station 2
VSC
AC DC AC Grid
VSC 1 VSC 2
DC AC

Offshore Wind Farm II VSC Station 3 VSC Station 4


VSC
AC DC AC Grid
VSC 1 VSC 2
DC AC

Fig. 5.1(b)
Offshore Wind Farm I VSC Station 1 VSC Station 2
VSC
AC DC AC Grid
VSC 1 VSC 2
DC AC

VSC
AC DC AC Grid
VSC 3 VSC 4
DC AC
Offshore Wind Farm II
VSC Station 3 VSC Station 4

Fig. 5.1(c)

Figure 5.1 Illustrative transmission options for Offshore windfarms (a) Multiple
HVAC transmission (b) Multiple HVDC transmission (c) A simplified MTDC network
170
Habitual methods to improve transmission reliability is to set up a redundant cable in

some major connections in case of the failure in its primary cable. It is not realistic to add

redundant cables to all the transmission paths given the high expenditure on HV submarine

cables. Therefore, multi-terminal HVDC network is introduced, as a compromise between

cost and reliability. As shown in Fig. 5.1(c), a bypass path is set up to bring the independent

transmission of two offshore windfarms together to formulate the most simplified 4-terminal

MTDC network, where these two transmission cables are no longer independent from each

other. Intuitively, if the transmission cable between Offshore wind farm I and PCC breaks

down, the generated power can be transmitted to the onshore network through the bypass and

the transmission cable between Offshore wind farm II and its PCC. Literally, the overall

reliability can be increased by providing alternative paths in MTDC network. Besides, the

stability and reliability performance of the connected grid can be improved by power flow

regulation services from MTDC grids.

14
12 13

11

10
6
G
G
1
9
5

4 8
2
G

G
3

G
Offshore Wind Farm I Offshore Wind Farm II

Figure 5.2 Illustrative example of integrating OWFs into an onshore utility grid

171
The advantages of MTDC network in OWF transmission pose a strong incentive to

research on OWFs’ integration via MTDC grid. Most of existing literatures regarding offshore

wind farm transmission planning, as in [33-35], still consider independent HVAC and HVDC

as alternatives to connect OWFs to shore. All of the existing literature with regard to topology

planning of Offshore windfarms mainly focus on its collector system topology. The

transmission network is only confined to connect one wind farm to an existing onshore PCC

whose location is provided beforehand. The difference of this project is to connect multiple

wind farms (rather than one) to an existing utility grid via several unknown PCCs whose

locations are yet to be determined. An illustrative diagram is provided in Fig. 5.2, where two

wind farms are expected to connect with an onshore utility grid, represented in the form of

14-bus network. It is to be determined how these two wind farms are connected with the

onshore grid and which point of the utility grid is the optimal PCC to integrate the wind farms.

In this example, these two Offshore windfarms can be integrated to the grid via all the existing

buses instead of one single point. In another word, 14 buses are all considered as potential

PCCs to integrate the generated power from these two OWFs.

The initial expenditure is not the only factor considered in selection of optimal topology.

The corresponding operational costs in both MTDC grid and onshore utility grid are added to

the MTDC integration model for multiple remote wind farms. Literally, the impact of WT

integration on utility grid is also taken into consideration. The MTDC integration is expected

to provide support to the local grid without violating stability and secure operation constraint.

To better imitate the real operation process and get an accurate result, the stochastic features

of both wind speed and power demand are covered in the optimization model. It is different

from Chapter 3 and Chapter 4, where wind speed is assumed to be always constant to simplify

the optimization model.

Besides, stability is another metric considered in the model, instead of regarding costs

as the one and only factor to evaluate the transmission topology. The previous research

172
assumes PCC can absorb however much wind power is generated from offshore wind farm.

In fact, the uncertain and stochastic wind power has imposed a burden on the stability of

connected regional transmission network thereby affecting the actual output from offshore

wind farm. Voltage stability should be taken into consideration so as to improve wind power

penetration while maintaining system stability margin. Wind turbines are unstable and

sensitive to disturbance, making it different from conventional synchronous generators.

Extensive research have been done to the modelling, control, and operation of MTDC

integration. Mathematical models for MTDC with VSC are proposed in [101]. A novel control

strategy for VSCs are presented with the consideration of wind speed uncertainty in wind

speed and failure of components [102]. An optimal hybrid AC/DC power flow model is

provided in [103]. However, the focus of most available literatures regarding MTDC is about

its control strategy to improve stability and power flow regulation, instead of taking it as an

alternative to integrate OWFs into associated RTN.

MTDC grid is considered to integrate multiple OWFs into associated RTN. In this

chapter, a comprehensive decision model is proposed to optimize OWF integration via MTDC

network, where 3 key factors charactering OWF integration is taken into consideration:

investment cost, operational cost in both RTN and MTDC grid, and evaluation of stability.

The locations for grid-side MTDC terminals are optimized instead of specifying in other

literatures. The uncertainty of wind speed and power demand are considered to better fit into

the real operation.

The rest of this chapter is structured as follows. Section 5.2 illustrates mathematical

problem formulation including the detailed definition of parameters and variables applied in

the mathematical model. The integrated transmission planning method of the optimization

problem is proposed in Section 5.3, including a modified mathematical model after

introducing uncertainty of wind speed and power demand, stability. The proposed integration

planning model has been verified on a case study in section 5.4, followed by which the

173
conclusion is drawn in the final section.

5.2 Cost Model Description

This model aims to automatically identify the optimal integration point to locate MTDC

grid (grid-side terminals) in an existing RTN and return converters of proper size, with the

aim to minimize the levelized investment cost and operational costs for both MTDC grid and

the associated RTN, while following a set of operational constraints.

In this section, a detailed model description is provided for the integration planning of

multiple remote wind farms. Some justified assumptions are initially proposed to simplify the

complicated problem. At the end of this section, the cost model for any components involved

in this model is provided in detail.

5.2.1 Hypothesis

To represent the topology of MTDC integration, a lot of discrete variables are involved

so as to formulate a mixed integer nonlinear problem. Besides, the operational cost is mutually

affected by the topology. Literally, the topology determines the operational cost, while in turn,

the change in operational cost leads to change in the objective so as to further change the

topology. Actually, a lot of factors can contribute to a different optimum integration result.

Accordingly, a reasonable simplification of the complicated model can be realized by

proposing several hypotheses, which is also necessary to focus on the impact of actual

research factors. Some reasonable hypotheses have been proposed, because of which this

complex problem, to some extent, can be simplified. The assumptions are listed below:

Hypothesis (1): All buses in RTN share the same change tendency of power demand.

Literally, in heavy loaded period, power demand at all the buses is high, while in lightly

loaded period, power demand at all the buses is low.

Explanation: the assumption is actually consistent with the real-world power consumption.

Usually, the power demand reaches its peak in the morning and evening, while it reaches its

174
minimum at midnight. This assumption is reasonable to avoid the messy combinations of load

level on different buses.

Hypothesis (2): Wake effect is ignored in this model so that wind speed at all the wind turbines

is assumed to be always the same.

Explanation: wake effect is usually considered in the optimal sitting of wind turbines.

However, it is the transmission planning of multiple offshore wind farms that is concerned in

this model. All the wind farms are abstracted to a synchronous generator with a rating equal

to the superposition of N identical wind turbines. The inner topology of offshore windfarm is

not considered so that wake effect can be ignored legitimately.

Hypothesis (3): Power characteristic curve of all wind turbines are the same.

Explanation: it is consistent with the real-world application that the wind turbines in the same

OWF is of the same type.

Hypothesis (4): There is a long distance between OWFs and onshore RTN so that the offshore

wind speed is assumed to make no difference on the onshore power consumption.

Explanation: as the name of “remote wind farm” implies, all the offshore wind farms are

located in high sea, over 50 km from coast. Therefore, the windy weather in offshore wind

farms will not change the power consumption activities of onshore utility grid.

Hypothesis (5): All bus locations are known, for both ac buses in onshore utility grid and

offshore wind farms.

Explanation: the aim of this project is to optimize the integration of multiple wind farms into

onshore grid. Therefore, the locations of these two objects are supposed to be provided as

known conditions.

Hypothesis (6): All the AC buses in the utility grid can be regarded as candidates to sit power

converters and integrate MTDC grid with the rest of the utility grid.

Explanation: instead of specifying a PCC, offshore wind farms are integrated to a whole utility

grid, where all the buses is the potential PCCs to connect with. Obviously, this condition is a

175
full optimization of the problem. The set of selected buses can be specified and confined to

several buses instead of all the buses, which can be seen as a simplified version of this model.

Hypothesis (7): The power flow in the MTDC grid is assumed to be from windfarms to the

utility grid. The case of feeding back from the main grid is not considered in this model.

Explanation: the objective to set up the MTDC grid is to transmit the generated power from

OWF. Therefore, the power flow should be from the OWFs to the grid.

5.2.2 Inputs

According to the proposed hypotheses, some minor parameters have been ignored and

all necessary input for the integration model is listed below, categorised into 4 research objects:

offshore wind farms, onshore grid, transmission cable and others.

1. Cable specifications: rated voltage, rated capacity, unit cable cost, unit resistance

2. Wind Farm: wind speed, power vs wind speed characteristic curve, cut in and cut off
speed, rated power capacity

3. Onshore utility grid: historical power demand data, topology, cable specifications of
each connection in the network, existing thermal power generating units (and its
operation cost parameters)

4. Others: distance between all known buses in the system, i.e., distance between OWFs,
distance between OWFs and each individual buses of onshore main grid, distance
between buses of onshore main grid; cost of non-served energy

5.2.3 Problem Formulation

This model aims to minimize total annualized cost F1 , including investment expenses

and operation cost for both MTDC grid and AC grid. Thereinto, the operational cost in AC

grid is to quantize the support of MTDC grid after integrating OWFs into the grid. To improve

readability of model formulation, all indices, except those for sets, follow the same naming

rules that superinduces refer to system scenario, while subindices represent locations and refer

to bus number of both MTDC and AC system.

176
min F1 = C1 + C2 + C3 (5.1)

y  
r  (1 + r ) 
C1 =   , C ij ij ,c W M i 
C  d   + C y (5.2)
(1 + r ) − 1  c ii , j
y
i
j 

C2 = Closs  T    p ss PMTDC
ss
,i (5.3)
ss i B

ng

(
C3 = T   p ss  ai PGiss2 + bi PGiss + ci ) (5.4)
ss i =1

C1 is the annualized investment cost of both HVDC cables and converters. CC and

CW are unit cost of HVDC cable and converters, respectively; d ij refers to the distance

between bus i and j in the network. In this case, the new cables are established to

configure the MTDC grid. Therefore, it includes the possible connection between windfarm

terminals, the possible connections between grid-side terminals and the possible connections

between windfarms and grid-side terminals; yi is an integer decision variable indicating the

rated capacity of converters installed at terminal i . It is worth noting that this terminal i

mentioned here only refers to the grid-side terminals in MTDC grid. Namely, the cost of

converters installed as the interface of onshore grid is considered instead of all the MTDC

terminals. The reason is because the converters installed in the windfarm side has a fixed rated

capacity equal to the rated capacity of corresponding wind farms. Therefore, the cost for such

converters is fixed and has no impact on the following connection topology so as to be

neglected in the objective of this model. In this case, the cost equation of converters applied

in this model is linear with the capacity of converters, where CW is the unit cost of converters

[m$/MW].

C2 refers to operational cost in MTDC grid, literally cost on power losses in MTDC

grid. Closs is a constant referred to as cost of non-served energy; T is the annual operation

ss
time; PMTDC ,i denotes the real power into terminal i in MTDC grid. It is assumed that

177
ss
PMTDC ,i is positive when the power is injected into the MTDC grid. Fig. 5.3 shows a diagram

Figure 5.3 A simplified MTDC diagram


of one MTDC grid with both wind farm sides and grid sides. The arrow marked in the

diagram indicates the positive power flow for each individual terminal. It is obvious that the

ss
active power in the windfarm sides, PMTDC ,i , are always positive since the aim of setting up

ss
the MTDC grid is to transmit the generated power from OWFs to shore. However, PMTDC ,i

in the grid-side terminals are always negative. The direction of power flow is in accordance

with the proposed hypothesis (7). From a general view, the OWFs act as several generators,

while the grid side terminals act as some infinite load who absorb however much power

generated from the OWFs. Therefore, the internal losses in the MTDC grid are for account of

s
the OWFs. The sum of PMTDC ,i for all the terminals in MTDC grid should be the internal

losses in MTDC grid.


MTDC
Ploss = P ss
MTDCi
i B

p ss represents the probability of system state ss . Accordingly, the weighted sum of

ss
PMTDCi with corresponding probability denotes the total internal losses of MTDC grid under

all system states.

C3 is the operation cost considering all the system scenarios in the onshore utility grid.

178
ai ,bi , ci are cost coefficients of the generator installed on bus i ; PGss,i indicates the real

power generation of the generator on bus i under system state scenario ss . p ss is the

(
probability of system state scenario ss . Therefore, p ss  ai PGiss2 + bi PGiss + ci ) is the cost of

operating generator i under system state scenario ss . To take all the generators and system

states into consideration, the operation cost in local AC utility grid can be obtained.

5.3 Integrated Transmission Planning for Multiple Large-scale

Remote Wind Farms

The mathematical model proposed in the last section is a general description of the

problem for integrating multiple remote wind farms via MTDC network into the local utility

grid. One of the highlights in this project is quantizing the impact of integrating OWFs via

MTDC on the local utility grid instead of merely evaluating the investment cost. It is expected

that the added wind power facilitates less dependence on existing thermal generators so as to

increase the penetration of renewable energy and reduce the operational cost in the grid. To

accord with the real applications as much as possible, the stochasticity of wind speed and

power demand is considered. The importance of measuring system stability is proposed.

Given the two above modifications, the mathematical model has been updated in section 3.

The detailed implementation of the proposed integrated transmission planning model is given

in the following sections.

5.3.1 Stochasticity

The term “system state”, denoted by superscript ss in the cost model, is important to

work out the operational cost in both MTDC grid and local utility grid. In fact, it is the wind

speed at the offshore sites and power demand in the onshore grid that contribute to the

different system state in this model. Initially, the wind speed has a direct impact on the

generated wind power from offshore windfarms, which act as generators in the MTDC grid
179
and accordingly, have a direct impact on the losses in MTDC grid, literally operational cost

in MTDC grid. On the meanwhile, the operational cost in the onshore grid is the cost of

operating the existing thermal generators, which is affected by two factors, i.e., local power

demand and wind generation. Literally, the uncertainty of offshore wind speed and power

demand in RTN results in the source of stochasticity in the model. Therefore, the system state

s can be defined as wind speed scenarios ws and the power demand scenarios ls .

The wind speed and power demand data used in this chapter is from [113] and [114].

According to the statistics of provided wind speed data and power demand data, their

probability density functions are both continuous. Discretization of probability density

function is applied so that the continuous probability density function for both wind speed

and power demand from given data, can be divided and represented by a specified number of

intervals. Each interval is usually represented by its average value, and its probability can be

calculated by the integral of PDF in the corresponding interval. Given the power characteristic

curve of wind turbines, the discretized wind speed scenario can be transformed to power

generation scenario. According to the proposed hypotheses (2) and (3), the actual power

injected to MTDC grid from each wind farm at each wind speed scenario is supposed to

generation of one single wind turbine in the corresponding wind speed scenario multiplied by

number of turbines operating in the OWF. By comparison, the discretized power demand is

easy to deal with, which can be directly substituted to the optimal power flow in each system

state.

As assumed in Hypothesis(4), there is no relations between uncertainty in wind speed

and power demand. The system state can be decomposed in a two-stage tree structure to

facilitate both wind speed and power demand, as shown in Fig. 5.4.

180
Figure 5.4 Two-stage stochastic model of wind speed and power demand
The precondition to apply the two-stage stochastic optimization model is to guarantee

the two stage stochastic variables are independent from each other. In this model, the

stochastic variables involved are wind speed and power demand, which are assumed to be

independent in the hypothesis (4), which satisfies the precondition of decomposing the

stochastic model into a two-step model. The system state ss in the mathematical model can

be replaced by wind speed scenario ws and power demand scenario ls . The overall

mathematical model is supposed to be rewritten, which is detailed in section 5.3.3. System

states can be regarded as the permutation and combination of different wind speed scenarios

and power demand scenarios. There is one thing to be noted that the total number of system

states is dependent on the number specified to discretize the PDF of wind speed and power

demand. It is obvious that the larger number of discretization is applied, the more accurately

the system state is represented.

5.3.2 Voltage Stability

Apart from considering the operational cost of utility grid after integrating MTDC

network, the overall stability of the utility grid is necessary to re-evaluate in order to minimize

the negative impact of integrating the flexible wind power on the connected grid, especially

in the case of a weak connected grid. Integrating such a large amount of renewable energy

usually poses a challenge to the local utility grid.

The analysis of static voltage stability essentially evaluates the proximity of the static

operation to voltage collapse point. For better description of the concept, the PV curve of an
181
arbitrary bus of system is depicted in Fig. 5.5. Point A indicates the current operation point,

while point B represents the maximum power it can withstand before voltage collapse.

Figure 5.5 PV curve of an arbitrary bus


Voltage Collapse Proximity Indicator (VCPI), based on maximum power transfer

theory, is applied to evaluate voltage stability in this model [115]. VCPI performs better than

loadability margin (LM) in adapting to planning research since its system-dependent nature

makes it possible to evaluate voltage stability without specifying an incremental direction for

generation and load [116].

,ls V2 cos 
Pl ws
max = (5.5)
Zl 4 cos (( −  ) / 2)
2

V2 sin 
Qlwsmax,ls = (5.6)
Zl 4 cos (( −  ) / 2)
2

Pl ws ,ls Qlws ,ls


VCPI lws ,ls = ,ls
= ws ,ls (5.7)
Pl ws
max Ql max

(
F2 = VCPI max = max VCPIlws ,ls ) (5.8)

where V , Pl ,Ql , respectively, represent voltage magnitude, real power and reactive power at

the receiving end of branch l from power flow analysis; Pl ws ,ls ws ,ls
max and Ql max denote the

maximum real power and reactive power transfer capability of branch l . These two terms

are obviously affected by the stochastic variables, i.e., wind speed and power demand, so that

182
the superscript ws and ls are added to represent the corresponding system state. zl

represents the line impedance; and  is line impedance angle;  is load impedance angle.

According to the equation above, VCPI index is exactly the ratio between real power

and its maximum power transfer capability, which is supposed to fall in 0 to 1. As VCPI

approaches 1, the branch is approaching its transfer capability, which can be considered more

unstable. If the power flow on the branch keeps increasing and surpasses the limit, voltage

collapse occurs. All branches in the system are supposed to be considered to overall examine

the whole system. In this model, the overall stability of the system is measured by the

maximum VCPI in all branches considering system scenarios, as shown in Eq. (5.8), where

the objective of stability evaluation is simply to minimize VCPI index on all branches.

5.3.3 Multi-Objective and Normalization Process

As discussed in section 5.3.1, the system state is defined by two stochastic variables

wind speed scenario ws and power demand scenario ls . Therefore, the formulation of the

total annualized cost in Eq. (5.1) to Eq. (5.4) is supposed to be rewritten to specify the general

system state, which is shown as below in Eq. (5.9) to Eq. (5.12).

min F1 = C1 + C2 + C3 (5.9)

y  
r  (1 + r ) 
C1 =   , CC  dij ij ,c + CW M yi  (5.10)
(1 + r ) − 1  c ii , j
y
i
j 

C2 = Closs  T    p ws PMTDC
ws
,i (5.11)
ws i B

ng

(
C3 = T   p ws  pls  ai PGiws ,ls2 + bi PGiws ,ls + ci ) (5.12)
ws ls i =1

The change mainly lies in Eq. (5.11) and Eq. (5.12) since both operational costs are

affected by the system state. The operational cost in MTDC grid, C2 , is only affected by

ws
wind speed. PMTDCi represents the real power from terminal i to MTDC grid under wind

183
speed scenario ws . The probability of wind speed scenario ws is represented by p ws . The

operational cost in utility grid is affected by both wind speed and power demand. The system

state p ss in the original equation has been replaced by the multiplication of probability of

wind speed scenario ws and probability of power demand ls , p ws  pls , which is

consistent with the assumption that these two variables are treated independently. PGws,i ,ls is

real power output of generator installed at bus i under system scenario of ws and ls .

After taking voltage stability index into consideration, it is formulated as a multi-

objective model. According to the detailed objective function in previous sections, a compact

mathematical model is generalized and presented here.

Objective:

 min F1 = C1 + C2 + C3
F : ws ,ls
(5.13)
 min F2 = VCPI max = max(VCPI l )

s.t.

Ui  Ui  Ui , i   B
max max
− I MTDC  I MTDC ,ij  I MTDC , i, j   B (5.14)
ws
PMTDC ,i  yi , i   M

 PMTDC ,i + PG ,i − PD ,i = Vi  j =1V j ( Gij cos  ij + Bij sin  ij )


 ws ws ,ls ls nb


QGws,i ,ls − QDls,i = Vi  jb=1V j ( Gij cos  ij − Bij sin  ij )
n

PG ,i  PGws,i ,ls  PG ,i , i   N
QG ,i  QGws,i ,ls  QG ,i , i   N (5.15)
Vi  Vi  Vi , i   N
Fij  Fij  Fij , i , j   N

It is formulated as a multi-objective problem, to minimize the total system cost as well

as to guarantee the stability of the overall system.

Eq. (5.14) and Eq. (5.15) provide a set of constraints with regard to the integration

planning of windfarms via MTDC grid. Initially, in both MTDC grid and connected RTN, bus
184
voltage is supposed to be within secure operation limit and power flow on both HVDC cables

and transmission cables in RTN should not exceed their cable ratings. Equipment such as

converters and generators in RTN should be guaranteed not to be overloaded.

The two objectives are measured in different scale, which are normalized in this model

by Fuzzy membership function [117] to improve the comparability between multiple

objectives, as shown below,

Fi ( x, y ) − Fi worst
f i ( x, y ) = for i=1,2
Fi worst − Fi best

i is a subscript representing the i th objective function; Fi ( x , y ) is a general form to

represent the ith objective function; Fi worst and Fi best ,respectively, represent the worst and

best results of the ith objective function. According to the above equation, f i ( x, y )

measures the relative distance of objective function to its worst result, which should be always

between 0 and 1. With a fi ( x, y ) closer to 1, the result is approaching its optimum and

regarded as a better result than others.

5.3.4 MTDC Power Flow Model

MTDC grid has an advantage of capability to regulate power flow from MTDC

terminals into associated grid by simply adjusting terminal voltage and control mode of

converter stations. A dynamic MTDC power flow algorithm proposed in [118] is applied to

involve such dynamic feature of MTDC into power flow analysis. It is assumed that nW

OWFs are integrated to a RTN via nm grid-side interfaces of a MTDC grid. The set of

ws ws
variables, PMTDC ,1 , PMTDC ,2 ,,,
ws
PMTDC ,nm , PMTDC ,nm +1 ,,,
ws ws
PMTDC ,nm + nW , represent the real

power into MTDC grid from its terminals. As for windfarm side, with lossless converters, the

power from OWFs into MTDC terminals is exactly the generated output power from OWFs,
ws ws
PMTDCi = PWFi (5.16)

185
The active power from RTN to MTDC can be viewed as a linear function of terminal

voltage, exactly named as dc voltage droop control. K i represents the droop gain of terminal

i in MTDC grid, U i0 and PMTDC


0
,i
denote reference terminal voltage and reference injected

power, respectively,

ws
PMTDC 0 0
,i = PMTDC ,i + ki U i − U i ( ) (5.17)

Based on Kirchhoff’s first law, the nodal input current is exactly same as the sum of

current outside of the node,

I = YU (5.18)

Substitute (5.18) into Ohm’s law, real power into terminal i of MTDC grid is

presented as below,

ws
PMTDCi = U i  YijU j (5.19)
j

Substitute (5.16) and (5.17) into (5.19), a power flow matrix equation of MTDC grid is

formulated,

Y U 2 + Y1,2U 2U1 + + Y1,nm + nwU nm +nwU1 − K1 (U1 − U10 ) 


 P0   1,1 1 
 − K 2 (U 2 − U )
MTDC ,1
 Y U U + Y2,2U 2 2
+ + Y2,nm +nwU nm +nwU 2 0 
 PMTDC
0
,2
  2,1 1 2 2

   
   
 0   
P
 MTDC ,n m  = Ynm ,1U1U nm + Ynm ,2U 2U nm + + Ynm ,nm +nwU nm +nwU nm (
− K nm U nm − U n0m ) (5.20)

 ws   
 PWFI  Ynm +1,1U1U nm +1 + Ynm +1,2U 2U nm +1 + + Ynm +1,nm +nwU nm +nwU nm +1 
   
   
 PWF  
ws

  Y + Ynm + nw ,nm + nwU n2m + nw
 nm +nw ,1U1U nm +nw + Ynm + nw ,2U 2U nm + nw + 
nw

P = J DC U

where, P denotes a vector of real power into MTDC grid; while U represents a vector of

MTDC terminal voltage; the Jacobian matrix of the MTDC grid is termed as J DC , relating

T
U ( ) = U 1( ) ,U 2( ) ,...,U n(m)+ nW 
i i i i
real power of MTDC terminal to MTDC terminal voltage.  

P ( ) =  PMTDC () ()
i ws iws i
represents the terminal voltage of ith iteration, while
 ,1 , PMTDC ,2 ,...,

186
T
ws ()
i

PMTDC ,nm + nW  represents the real power into MTDC grid of all the terminals in the i
th

iteration. Calculate the derivative of both sides of equation (5.20),

P
(i +1 )
J DC = U =U (i)
(5.21)
U

Assuming tiny voltage variance around the operating point U ( i ) , terminal voltage of

the next iteration can be represented as below,

( ) ( P( )
-1
U(
i+1 )
= U ( ) + J DC
( i+1 ) i+1 )
- P( )
i i
(5.22)

The terminal voltage can be updated following Eq.(5.22) in each iteration until its

variance is less than the tolerance. The MTDC power flow model can provide the terminal

voltage in all the buses of MTDC grid so that the power flow on each path in MTDC can be

worked as well. Therefore, the operation cost in MTDC grid , C2 , can be worked out.

5.3.5 Simplified mathematical model

A simplification can be made to the proposed mathematical model. The decision

variable  ij ,c represents the connection or not between bus i and bus j , which is

multiplied with a distance term d ij in eq. (5.10). In fact, locations of all the buses are known

beforehand, literally, the distances between which can be worked out. A distance matrix D

is formulated as a square matrix size of nW + nb .

0 d 1,2 ... ... d1,nb d1,nb +1 ... ... d1,nb +nw 


 
d 2,1 0 ... ... d 2,nb d 2,nb +1 ... ... d 2,nb +nw 
 
 ... ... ... 

D = d nb ,1 d nb ,2 ... ...0 d nb ,nb +1 ... ... d nb ,nb +nw  (5.23)
 
d nb +1,1 d nb +1,2 ... d nb +1,nb 0 ... ... ... d nb +1,nb +nw 
 
 ... ... ... 
d 
 nb +nw ,1 d nb +nw ,2 ...d nb +nw ,nb d nb +nw ,nb +1 ... 0 

where the first nb rows correspond to AC buses in the utility grid and the last nW rows

187
correspond to the windfarms. Each individual element in the D matrix, d ij is a known

parameter. For a specified MTDC topology, what is required to figure out is the buses where

MTDC grid-side terminals can be located. If an integer variable xi is introduced to denote

the ac bus number integrating terminal i of MTDC grid, the connections within MTDC grid

can be regarded as picking out one element from the distance matrix D . Generally, the

connections in MTDC grid can be categorized into 3 kinds, connections between OWFs,

connections between OWFs and grid-side terminals, and connections between grid-side

terminals. To represent the distance between OWF i and j , i , j  , D ( i , j ) is just the

answer since both buses are known. To represent the distance between OWF i and potential

PCC point of terminal j , x j indicates the corresponding bus number so that the distance

( )
in between can be represented by D i , x j . In terms of connections between grid-side

terminals i and j , the specific bus number of the two ends are both represented by the

( )
integer variable xi , x j and the distance is formulated as D xi , x j . The cable types used for

each individual path in MTDC grid is required to withstand the case of fully-operating OWFs,

easy to be determined. According to the above explanation, the initial cost equation can be

modified as below:

r  (1 + r ) y
C1 = [CC (  D(i, j) +  D(i,x j )
(1 + r ) y − 1 i , jWF iWF , j M
(5.24)
+  D(xi ,x j )) + Cw  y] i
i , j M i M

Correspondingly, there are a set of general boundaries set for the new decision variables

xi :

1  xi  nb , i  M (5.25)

xi − x j  0 , i , j  M (5.26)

188
xi , an integer variable which denotes the chosen bus number for terminal i , should be less

than total number of buses in RTN, nb ; Eq. (5.26) is to guarantee that the grid-side terminals

are placed on different AC buses in RTN. Otherwise, the topology of MTDC grid will be

changed with less terminals installed at the same AC bus.

5.3.6 Implementation of the Model

It is formulated as a multi-objective MINLP, which is initially solved by NSGA-II, an

enhanced version of NSGA designated to solve multi-objective problem. In NSGA-II, each

chromosome includes a complete solution to the optimization problem.

Coding rules: to avoid conflicts and ambiguity, coding rules are important to

evolutionary algorithm. There are two set of integer decision variables involved in this model:

X represents a vector of integer variables xi (integrated AC bus number) and Y represents a

vector of integer variables yi (capacity required for each terminal). The detailed format of

chromosome is shown in Fig.5.6.

Figure 5.6 Encoding format of chromosome

Selection of the best compromise solution: A pareto set will be obtained when the

optimization process terminates. Initially, normalization process will be applied to the optimal

set of results, as described in section 5.3.3. In this model, min-max method is used to screen

out the best compromise result among the set of optimal results. The best result is regarded as

(
the maximum of min values of the normalized objective, denoted as max min ( f1 , f 2 ) . )

189
5.3.7 Flow Chart for the Integrated Transmission Planning Model

The integrated transmission planning model includes the following steps, also

represented in a flow chart as shown in Fig. 5.7.

Input: import necessary system parameters, historic wind speed data and power demand

data.

Scenario generation: apply curve fitting to the available wind and load data, discretize

the continuous PDF of both these two uncertainty values and formulate system scenarios.

Initialization: randomly generate the first generation by simply following the set

boundary for each gene.

Main loop procedure: a bionic loop similar to general GA integrating crossover,

mutation and replacement procedure to produce offspring.

Candidate fitness calculation: given all the system scenarios, both annualized cost and

steady-state voltage stability assessment VCPI are calculated for each individual in the

population pool.

Termination: the optimization process will stop and return the best individual

(representing the best solution to the problem) if the termination criterion has been satisfied,

the methodology procedure returns with the optimal result and terminates; otherwise, it will

go back to step of Main loop procedure and continue.

190
Figure 5.7 Flowchart of methodology
5.4 Case Studies

The comprehensive integration planning model for multiple remote offshore windfarms

via MTDC network is verified in this section. Multiple cases have been designated to justify

the necessity of involving stochastic optimization and voltage stability into the comprehensive

model. The reminder of this section is organized as follows. Initially, a brief case description

regarding the offshore windfarms and onshore utility grid has been introduced in the first

subsection, including several necessary input data. In the following 3 sections, the simulation

results of 3 cases are provided and analyzed, respectively. Finally, the comparison and

analysis are drawn in the last subsection.

5.4.1 Case description

Simulations are performed on the IEEE 14-bus benchmark ac system, which consists

191
of 14 buses, 20 branches and 5 generators, as shown in Fig. 5.8. Bus 1 is regarded as slack

bus [119]. This 14-bus AC system is regarded as the onshore utility grid to integrate the remote

offshore windfarms. According to the hypothesis (6), all these 14 buses can be potential

candidate buses to install grid-side VSCs as the grid-side terminals of the MTDC grid.

2 OWFs, located about 100km away from shore are expected to connect to a 14-bus

benchmark AC utility grid via 3 interface buses of a ±300 kV 5-terminal MTDC grid, as

shown in Fig. 5.9. These two wind farms consist of 130 3-MW and 160 3-MW wind turbines,

respectively. The rated capacity of the remote windfarms are, accordingly, 390MW and 480

MW. With the fixed rated capacity of windfarms, the power converters installed at the offshore

wind farm side are supposed to be large enough to withstand the generated power when the

wind turbines are operating at full load. In this case, all converters are assumed to be 100%

available. Therefore, the capacity of the wind farm side converters should be equal to the rated

capacity of the corresponding wind farms. The cost of converters is assumed to be linear with

its capacity. Fixed capacity indicates the costs of such windfarm-side converters are fixed as

well, which justifies the reason why the investment cost of converters in the objective function

only refers to the grid-side terminal converters. On the contrary, the capacity of grid-side

power converters is dependent on the topology so that the initial cost of such converters is

added in the objective function to have an overall evaluation of the annualized investment.

192
Figure 5.8 14-bus RTN

Figure 5.9 5-terminal MTDC grid topology

In this case study, the internal topology of offshore windfarms are not the priority and

the OWFs are abstracted to a synchronous generator with rated capacity of 390MW and

480MW, respectively. The Vestas V112 3.0MW offshore wind turbine model is used, whose

cut-in, cut-off, and rated speeds are 3m/s, 25m/s and 12m/s. The power output characteristics

curve of the wind turbine is presented in Fig. 5.10.


193
Figure 5.10 Power output characteristic curve of wind turbine
The historic wind speed record in one year is obtained from a wind observation station,

published by the Australian Bureau of Meteorology [113]. The hourly wind speed of a year

according to the observations is all shown in Fig. 5.11, where the wind speed is extremely

unstable and fluctuant all the time. After applying the approach of discretizing the probability

density function of the available wind speed data, the obtained density function is shown in

Fig. 5.12, where the variable wind speed has been discretized into 14 scenarios.

Figure 5.11 Hourly wind speed of a year


194
Figure 5.12 Discretization of the probability density function of wind speed

The hourly power demand in one year is shown in Fig.5.13. The probability density of

power demand data is discretized into 14 scenarios. To sum up, these two stochastic variables

contribute to 14*15=320 system cases.

Figure 5.13 Hourly load percentage to the peak load of one year

195
Some other inputs of the test model are shown below:

• Unit cost of converter is $0.1175 million/MW [28].

• A life expectancy of the system is assumed to be 20 years.

• Interest rate is set 4%.

• Cost of loss is set $90/MWh.

• The cable type applied in case study has the cross-sectional area of 1000 mm2, the

unit cost of which is $0.6 million/km [120].

• Droop gains K1, K2 and K3 are set equal and negative under the assumption of average

power output from MTDC terminals.

Various cases have been designated and investigated to clearly classify the

effectiveness and validity of the proposed model. All of these 3 cases are dealt with the

proposed comprehensive integration planning model and the results are analyzed and

compared in the following sections.

Case A: Deterministic optimization without considering stochastic wind speed and power

demand.

Case B: Stochastic optimization with multiple objectives (both cost and voltage stability)

considering stochastic wind speed and power demand.

Case C: Stochastic optimization with single objective (only cost) considering stochastic wind

speed and power demand.

5.4.2 Case A: deterministic optimization without uncertainty

In case A, the two stochastic variables, wind speed and power demand are both assumed

to set as their average value. Average wind speed is reasonable since the highest wind speed

cannot be always available during the operation period.

In case of multiple-objective model, pareto front is derived for IEEE 14-bus system

without considering uncertainty of wind speed and demand. It is not a single optimum

obtained as in single-objective model. Instead, a set of optimum results are obtained and
196
formulates the pareto front.

Table 5-1 summarizes the optimal Pareto solution for case A. f1 and f2 ,

respectively, denote the annualized cost and voltage stability index VCPI after normalization.

It is easy to figure out that the best voltage stability f 2 is always contradictory to the

minimum annualized cost f 1 . For instance, solution #9 has the best performance on overall

system stability ( f1 = 0 ) but its annualized investment is the highest ( f 2 = 1) . On the

contrary, solution #1 is completely opposite ( f 1 = 1) , with least cost but worst stability

performance ( f 2 = 0 ) . All of the solutions presented in the table are the optimum result of
the case study, however, with different weights applied to the two objectives. It provides a

series of solutions for the decision-makers to choose according to their own situation. For

instance, if voltage stability is the primary decision factors, solution #9 will be chosen. Pareto

front is depicted in Fig. 5.14, which presents the pareto front according to the derived pareto

optimal set shown in Table 5-1. Solution #6 is chosen as the best compromise solution by

using the min-max method, of which the corresponding decision variables are given in Table

5-2. The terminal 1, 2 and 3 of MTDC grid is integrated with AC bus number 13, 3 and 9 in

the 14-bus system. The rated capacity of the corresponding converters are 262MW, 303MW

and 300MW. The optimum topology according to this best compromise solution is provided

in Fig. 5.15.

Table 5-1 Pareto optimal solution of deterministic optimization (Case A)


Solution # f1 f2 min ( f 1 , f 2 )
1 1 0 0
2 0.837 0.317 0.317
3 0.827 0.341 0.341
4 0.739 0.561 0.561
5 0.711 0.569 0.569
6 0.577 0.949 0.577
7 0.548 0.958 0.548
8 0.206 0.965 0.206
9 0 1 0
197
Table 5-2 Optimal decision variables for the best compromise solution (Case A)
Terminal Integrated ac Converter
No. bus number Capacity (MW)
Terminal 1 Bus 13 262
Terminal 2 Bus 3 303
Terminal 3 Bus 9 300

Figure 5.14 Pareto front of Case A: Deterministic Optimization

14
13 Terminal 1
12
11

6 10

G
G
1

5 Terminal 3
9

2
4

G 3 8
Terminal 2
G
G

Terminal 4

W
Terminal 5
W

Figure 5.15 Optimal topology of Case A: Deterministic Optimization

198
5.4.3 Case B: stochastic optimization with multiple objectives

Uncertainties of demand and wind speed are considered in this case and the optimal

Pareto solutions is presented in Table 5-3. By using the min-max method, solution #4 is the

best compromise result in the optimal set. Solution #1 corresponds to cost minimization

solution ( f 1 = 1) , whereas solution #7 provides the highest annualized cost ( f1 = 0 ) .

Solution #7 has the best voltage stability performance ( f 2 = 1) , whereas solution #1

performs worst in voltage stability ( f 2 = 0 ) . The attained Pareto front is shown in Figure 7.
The corresponding pareto front of this case is depicted in Fig.16.

The decision variables derived from best compromise solution is detailed depicted in

Table 5-4. The terminal 1, 2 and 3 of MTDC grid is integrated with AC bus number 5, 4 and

2 in the 14-bus system. The rated capacity of the corresponding converters are 266MW,

302MW and 299MW. The optimum topology according to this best compromise solution is

provided in Fig. 5.17.

Table 5-3 Pareto optimal solution of stochastic optimization (Case B)


Solution # f1 f2 min ( f 1 , f 2 )
1 1 0 0
2 0.92 0.51 0.51
3 0.71 0.58 0.58
4 0.67 0.67 0.67
5 0.63 0.76 0.63
6 0.48 0.95 0.48
7 0 1 0

Table 5-4 Optimal decision variables for the best compromise solution (Case B)
Terminal Integrated Converter
No. ac bus number Capacity (MW)
Terminal 1 Bus 5 266
Terminal 2 Bus 4 302
Terminal 3 Bus 2 299

199
Figure 5.16 Pareto front of Case B: Stochastic Optimization

14
13
12
11

6 10

G
G
1

Terminal 1 5
9

Terminal 3 2 Terminal 2 4

G 3 8

G
G

Terminal 4

W
Terminal 5
W

Figure 5.17 Optimal topology of Case B: Stochastic Optimization

200
5.4.4 Case C: stochastic optimization with single objective

Uncertainties of demand and wind speed are still considered in this case. However, the

only difference lies in the objective, where annualized investment cost is the only way to

evaluate the result. The model has been transformed from multi-objective problem to single-

objective problem. Accordingly, the optimum result will be one optimal solution instead of a

pareto set as in Case A and B.

The decision variables of Case C is detailed depicted in Table 5-5. The terminal 1, 2

and 3 of MTDC grid is integrated with AC bus number 2, 8 and 4 in the 14-bus system. The

rated capacity of the corresponding converters are 271MW, 299MW and 297MW. The

optimum topology according to this best compromise solution is provided in Fig.5.18.

To analyze the cost breakdown of the optimal result for Case C, the investment cost

and operation cost in utility grid take up the two largest proportion of the total cost, 60.4%

and 39.0%, respectively. The losses in the MTDC grid is because of the low resistance in the

HVDC cables.

Table 5-5 Optimal solution for Case C

Total Cost (M$) 48.76


Total cost breakdown
Operation cost C2 OPF Cost C3
Investment cost C1 (M$)
(M$) (M$)
29.44 0.30 19.02
Proposed integration method
Converter Capacity
Terminal No. Integrated ac bus number
(MW)
Terminal 1 Bus 2 271
Terminal 2 Bus 8 299
Terminal 3 Bus 4 297

201
14
13
12
11

6 10

G
G
1

5
9

Terminal 1 2 Terminal 3 4

G 8
3 Terminal 2
G
G

Terminal 4

W
Terminal 5
W

Figure 5.18 Optimal topology of Case C: Stochastic optimization with single


objective
5.4.5 Analysis

A comparison of optimal objective values of case A, B and C is presented in Table 5-6.

For case A and B with multiple objectives, only the best compromise solution is presented,

while as for case C with single objectives, its stability index VCPI is calculated according to

the obtained optimal topology in this case.

Table 5-6 VCPI index for Case A, Case B and Case C


Case A Case B Case C
Total Cost
F1
53.8 49.07 48.76
VCPI max 0.167 0.187 0.186
Initially, comparing the final optimal topology for case A, B and C, it is easy to figure

out that their solutions are different for all the three cases with different objectives and

different forms of expressing uncertainties. Accordingly, it justifies the validity of

introduction of stochasticity of both wind speed and power demand, as well as adding voltage

stability as a supplementary objective in offshore wind farm integration via MTDC grid.

As for annualized cost, the cost in Case B and C are less than that of Case A. The
202
discrepancy in annualized cost is due to the larger operational cost in RTN since OWFs are

assumed to always operate at average wind speed in Case A. Accordingly, power demand in

case A is expected to rely more on generators in RTN than in case B and C, so as to drive up

the operational cost C3 in case A.

Additionally, it is apparent that after considering the uncertainty of wind speed and

demand, the VCPI index of Case B and C has increased and is larger than that of Case A,

which indicates Case B and C appear to be less stable. However, the result is, in fact,

consistent with the real case of integrating unstable wind generation into the system. On the

meanwhile, the similar VCPI index of Case B and C seems to imply it is not necessary to add

voltage stability as a supplementary objective to characterize the integrated transmission

planning of OWFs. However, only the best compromise result of case B, which assigns a same

weight to both cost and stability by using min-max method, is shown in the table. If it is

required more on stability performance, solution #5, #6 and #7 in Table 5-3 has the VCPI of

0.174, 0.176 and 0.183, which could be considered to satisfy the requirements at the sacrifice

of investment increase.

To validate the necessity of considering the stochasticity of wind speed and power

demand, a 24 hour generation and power consumption curve is drawn on different wind speed

level and load level as shown in Fig. 5.19.

(a) (b)

203
(c) (d)

Figure 5.19 Generation versus demand curve for a case with different wind
speed level and high demand (a) low wind speed (b) moderate wind speed (c) high wind
speed (d)extremely high wind speed
According to the cost analysis, C3 of Case C, the operation cost of AC system

accounts for 39.0% of total annualized cost, next only to investment cost. The reason comes

from the stochastic features of wind speed and power demand. In fact, the installed capacity

of offshore windfarms (870MW) is far larger than local power demand (259MW) so that there

is no need to turn up local generators to satisfy the load during the fully operated period of

offshore windfarms. However, the changing wind speed and power demand cannot guarantee

the wind generated to be always sufficient to supply power demand in AC system.

Accordingly, local generators are required to turn up to compensate the discrepancy between

generation and power demand. A 24-hour generation and demand curve of a case with

different wind speed and high demand is screened out and shown in Fig.5.19. The red and

black curves indicate power demand and total generated power in AC system, respectively,

while curves of other colors indicate injected power from individual MTDC terminals.

According to Fig. 5.19, output power from MTDC terminals are almost identical since droop

gain of converters are set equal to evenly distribute wind power. The tiny discrepancy is

caused by the size difference of offshore windfarms. Also, it could be observed that

throughout this 24 hours, local ac generators have to be turned up to compensate the difference

204
between wind power generation and power demand when solely relying on wind power is not

enough. In low wind speed level, there is little wind power generation from the OWF so that

the local demand almost relies on the thermal generation units in the AC system. In moderate

wind speed level, partial of the power demand can be compensated by the OWF generation.

However, it is not enough so that the generators should be operating during the day. As for

high wind speed, in some high wind hours, the total wind power generation exceeds the local

demand, which indicates the excessive wind power can be transmitted to the main grid

through the slack bus. In both (c) and (d), the thermal generators only need turning up when

the power demand cannot be met by the wind power generation.

Although there is little attention paid to the protection and control strategies in MTDC

grid, a fast and reliable protection and control system is necessary for safety operations of

MTDC grids. The fully controllable conduction pattern of HVDC transmission thanks to

HVDC converters is essentially regarded as the strengths over its AC counterpart, which,

however, challenges the traditional protection and control strategies widely used for AC grid

in industry. In terms of control strategy, multiple VSC converters are involved in the MTDC

grid, which contributes to a problem of coordination among these converters. It is a tradition

to apply the master-slave control strategy given multiple devices. However, the distance

between converters in a multi-terminal grid tends to be up to 100km or more. The signal

transmission duration cannot be neglected given such a long distance between the central

converter to the remaining. Consequently, local control should be used and integrated in the

corresponding converters to minimize the interactions between multiple converters. Voltage

droop control is typically used in onshore converters, same as in this chapter, which requires

these converters to collaborate to regulate voltage and maintain power flow. The voltage

control duties in the MTDC grid is therefore shared between multiple converters rather than

relied on one fixed master converter. The droop control is to adjust active power to DC voltage,

similar to the frequency control in AC system, so that a universal measurement of system load

205
is efficient to maintain the voltage while keeping interactions to a minimum. Therefore, the

droop control strategy is considered for the onshore converters in this chapter. As for the

offshore ones interfacing offshore wind farms, converters are set to maintain the AC voltage

and frequency in the offshore wind farm. A coordinated control system matters for MTDC

grid and there are a great many existing coordinated control strategies proposed for MTDC

converters, as in [121-123], all of which have provided a strong support for considering

integration planning of offshore wind farms via a MTDC grid, the main topic of this chapter.

In terms of protection, given the larger fault current and faster transients for DC fault

in MTDC grids, a fast and dynamic fault detection and clearing process is essential to

differentiate the fault and take appropriate actions to remove the fault, in order to minimize

disturbances caused by contingencies. Currently, it is normal practice to remove the entire DC

grid in case of any contingencies due to the lack of appropriate equipment qualified to provide

such a fast response. The hybrid DC breaker developed by ABB is regarded as the key

technology to realize protection for large MTDC grids with a total clearing time less than 5ms.

Besides, many different fault detection strategies have been proposed to locate fault by voltage,

current or other combined measurements [124-126]. Also, ABB has developed a fast and

reliable control and protection system, called MACH, for HVDC system to guarantee the

safety operation of HVDC systems [127].

Both protection and control system are important to secure the operation of MTDC

grids. Existing studies and developed devices have been proposed to deal with the related

issues. The related techniques are regarded as a background to realize the idea to take MTDC

grid into transmission planning for multiple offshore wind farms.

The proposed integration planning model for multiple remote wind farms via MTDC

grid is to optimize the integration options for offshore wind farms with the consideration of

all the involved systems based on some justified simplifications. It helps to determine the

optimum integration point instead of one fixed PCC with a fixed coordinate in past literatures.

206
Both load flow analysis and cost assessment have been taken into consideration. However,

the integration for wind farms are more complicated to deal with, where several studies such

as short circuit study, harmonics study transient study are required to carry out to optimize the

OWF integration from an operational view. Besides, the VSC converters are assumed lossless,

which is represented by a droop model in the chapter. Using this simple model can be regarded

as a justified simplification of the system complexities since the internal control and

protection strategy has little change on the outcomes. In real application, the actual

configuration and operation of VSC converters is farm more complicated than the simplified

model used in this chapter, although it is out of the author’s research scope.

5.5 Conclusion

A stability-constrained offshore transmission planning for multiple remote windfarms

via multi-terminal VSC-HVDC grid is proposed in this chapter. It is concerned about the

locations where to integrate the OWFs and the optimal sizing for converters installed at

appropriate terminals. Investment cost is not the only objective function to characterize the

optimization model. Instead, the impact of offshore wind power injection on RTN is also

included due to two factors: the integration of offshore wind power is optimized to reduce the

dependency on conventional thermal power plant; the overall system stability is examined to

make sure the integration will not cause stability problem. To better fit into real operation, the

stochasticity of wind speed and power demand is taken into consideration. The optimization

is validated in a case study to integrate 2 OWFs to a 14-bus system via a 5 terminal MTDC

grid under 3 different conditions. It can also be easily applied to solve a larger OWFs

integration and transmission problem with more OWFs connected to a larger RTN via a more

complicated MTDC grid.

207
Chapter 6 CONCLUSIONS AND FUTURE WORK
This research thesis covers the key aspects regarding offshore wind farm planning from

its inner collector system design, substation sitting problem, transmission planning to the

novel integration planning by MTDC grid. The main achievements of this research are:

• Propose a comprehensive and explicit cost model for offshore wind farm planning.

• Fully optimize the electric system layout of an offshore wind farm rather than
relying on one simple standard configuration.

• Simplify the complexity of collector system optimization model by proposing


reasonable reduction of variables thereby reducing the search space and improving
the computational efficiency.

• Provide a guideline optimization tool adaptable for large-scale offshore wind


farms, which receives the locations and capacity of wind turbines and PCCs and
returns an optimal connection topology with proper cable types and substation
types placed at the optimal location

• Propose a decision tool to determine the choice between HVAC and HVDC for
OWF transmission planning.

• Research on the impact of OWF integrations on the connected utility grid and
involve stability performance to determine the best integration location for
multiple wind farms

In this PhD project, the researcher initially propose a comprehensive electric system

optimization framework of collector and transmission system of an OWF in order to minimize

the investment and operational cost. The continuous substation sitting problem is simplified

and discretized by a 2-step rasterization method. The main objective is to reduce the number

of variables representing the topology and reduce the search space by introducing the concept

of allocating wind turbines to feeders, and further optimizing the connections within feeders

by minimum spanning tree algorithm, in order to make sure each generated feeder is

reasonable and not internally crossed. A cost adjacency matrix is introduced to realize the
208
penalty mechanism for crossing between feeders. The optimization framework can be

regarded as a guideline tool for electric system layout in offshore wind farm, which receives

the specifications of wind farm, available type of devices, as well as the geometric location

of individual wind turbines, onshore PCC, existing location or permissible region for offshore

substation. In return, it provides the layout, the proper cable type for each individual

connection and appropriate location to sit the offshore substation. The feasibility of the

framework has been verified by a virtual OWF with 40 wind turbines by checking technical

constraints and feasibility such as cable crossings.

For large-scale wind farm, the introduction of multiple offshore substations brings the

complexity of the electric system design of an offshore wind farm to the next level. Initially,

a comprehensive cost model is developed in Chapter 4 to take all the necessary equipment in

a large-scale offshore wind farm with multiple substations into consideration. To simplify the

complicated layout design process, a multi-level optimization framework including offshore

substation optimization layer, inter-array feeder optimization layer and submarine cable

section optimization layer, is proposed based on the features of different parts of offshore

wind farms. The fuzzy clustering algorithm with both distance and cost as objective functions

is applied in the offshore substation optimization layer, to properly partition all the wind

turbines into a number of clusters with center locating offshore substations. An adaptive wind

turbine allocation method is proposed to allocate wind turbines to substations, maximizing

the overall membership degree while satisfying the capacity limit. The choice of HVDC and

HVAC technique in transmission systems can also be determined in the substation level. The

following inter-array feeder layer and submarine cable section layer are integrated to optimize

the inner system layout of a cluster. Thereinto, NAA algorithm is applied in inter-array feeder

optimization layer to further partition wind turbines into a feeder, while Prim’s algorithm is

utilized to optimize the connections and determine the cable types within a feeder. Finally, the

proposed hierarchical framework is validated on two OWFs, one with 80 wind turbines and

209
the other with 150 wind turbines. This framework has been verified as an efficient

optimization tool for offshore wind farm planners to optimize the electric system layout

design for large-scale offshore wind farms with multi-station.

In Chapter 5, a stability-constrained offshore transmission planning for multiple remote

windfarms via multi-terminal VSC-HVDC grid is proposed in this chapter. It is concerned

about the locations where to integrate the OWFs and the optimal sizing for converters installed

at appropriate terminals. Investment cost is not the only objective function to characterize the

optimization model. Instead, the impact of offshore wind power injection on RTN is also

included due to two factors: the integration of offshore wind power is optimized to reduce the

dependency on conventional thermal power plant; the overall system stability is examined to

make sure the integration will not cause stability problem. To better fit into real operation, the

stochasticity of wind speed and power demand is taken into consideration. The optimization

is validated in a case study to integrate 2 OWFs to a 14-bus system via a 5 terminal MTDC

grid under 3 different conditions. It can also be easily applied to solve a larger OWFs

integration and transmission problem with more OWFs connected to a larger RTN via a more

complicated MTDC grid.

Future works can be carried out in different directions. Although the cumulative

installed offshore wind capacity still cannot match with that of onshore wind, offshore wind

actually has been evolving up to 30 years, with the first offshore wind project, Vindeby

Offshore Wind Farm erected in 1991 in Denmark. The first batch of offshore wind farms

usually have taken up the best off-coast places with plentiful wind resources, but their ratings

are far more less than the currently constructed OWFs. In the coming years, many inchoate

offshore wind farm will be expected to decommission since the life expectancy of OWFs are

designed as 20 years. It is possible to replace the retired wind turbines with currently advanced

wind turbines so that the existing transmission system can still take effect and reduce the

expenditure to set up a completely new transmission cable. Meanwhile, the newly added wind

210
turbines can take advantage of the best wind resources to generate electricity. It is possible

that there is a mix of old and new wind turbines in one offshore wind farm in the coming

years. How to sit the new higher-rating wind turbines in the existing OWF is worth discussing

since it is expected to have little impact on the existing wind turbines (considering wake

effects) and maximize the total generation with little investment on setting up new lines.

Another interesting research topic is the impact of new mega wind turbines. The largest wind

turbine currently launched has a rated capacity up to 14MW with a 222-meter rotor diameter.

Will the new large wind turbines pose challenge to the associative electric system component

in offshore wind farm? If it is positioned in an existing old offshore wind farm, the impact of

its large rotor on the existing wind turbines is supposed to be investigated. In addition, to

follow the power system operators’ instructions, many offshore wind farms have to cut off

their excessive wind power, which implies waste of wind energy. It is possible to take

advantage of the excessive wind power in seawater electrolysis to produce hydrogen and

further to produce ammonia. It will realize the transformation of energy type from electric

energy to chemical energy, which is easy to transmit by sea transportation instead of setting

up another expensive transmission lines to transmit the electric power to shore. How to utilize

this hydrogen technology in offshore wind farm is worth discussing. The distribution of

generated electricity between direct transmission to shore and water electrolysis needs

optimized to maximize the profit of offshore wind farm owners and reduce the extra cost for

setting up new cables.

211
BIBLIOGRAPHY

[1] J.Lee, F.Zhao, “Global Wind Report 2019,” Global Wind Energy Council (GWEC),
Belgium, Mar.25, 2020.

[2] L.Ramirez, D.Fraile, G.Brindley, “Offshore Wind in Europe: Key Trends and Statistics
2019”, WindEurope, February, 2020.

[3] A. Larson, Floating Platform Are an Offshore Wind Gamechanger, POWER, Sep.1,
2020. [Online] Available: https://www.powermag.com/floating-platforms-are-an-
offshore-wind-gamechanger/

[4] N. Holtsmark, H. J. Bahirat, M. Molinas, B. A. Mork and H. K. Hoidalen, "An all-DC


offshore wind farm with series-connected turbines: an alternative to the classical
parallel AC Model?," in IEEE Transactions on Industrial Electronics, vol. 60, no. 6, pp.
2420-2428, June, 2013.

[5] D. Valcan, P. C. Kjær, L. Helle, S. Sahukari, M. Haj-Maharsi and S. Singh, "Cost of


energy assessment methodology for offshore AC and DC wind power plants," in 2012
13th International Conference on Optimization of Electrical and Electronic Equipment
(OPTIM), Romania, 2012, pp. 919-928.

[6] M. A. Parker and O. Anaya-Lara, “Cost and losses associated with offshore wind farm
collection networks which centralise the turbine power electronic converters,” IET
Renew. Power Gener., vol. 7, no. 4, pp. 390–400, 2013.

[7] M. D. Prada-Gil et al., “Feasiblity analysis of DC collector grid for offshore wind power
plants,” Renewable Energy., vol. 78, pp. 467-477, 2015.

[8] G. Stamatiou, “Techno-economical analysis of DC collection grid for offshore wind


parks,” Master's thesis, ,University of Nottingham, Nottingham, U.K., 2010.

[9] D.Vagiona, “Sustainable site selection for offshore wind farms in the South Aegean—
Greece,” Sustainability., vol.10, no.3, March, 2018. [Online]. Available doi:
10.3390/su10030749.

[10] A.Chaouachi, C.F.Covrig, M.Ardelean, “Multi-criteria selection of offshore wind


farms: case study for the Baltic States,” Energy Policy, vol.103, pp.179-192, Jan.2017.
[Online]. Available: https://doi.org/10.1016/j.enpol.2017.01.018.

[11] S.Lundberg, “Evaluation of wind farm layouts,” EPE Journal, vol.16, pp. 14-21, 2006.
[Online]. Available doi: 10.1080/09398368.2006.11463608

[12] M. Mohseni, S. M. Islam, "A review of enabling technologies for large wind power
plants to comply with recent grid codes," 2011 IEEE PES Innovative Smart Grid
212
Technologies, pp. 1-6, 2011. [Online]. Available doi:10.1109/ISGT-
Asia.2011.6167126.

[13] L. Yu, R. Li, L. Xu and G. P. Adam, "Analysis and Control of Offshore Wind Farms
Connected With Diode Rectifier-Based HVDC System," IEEE Transactions on Power
Delivery, vol. 35, no. 4, pp. 2049-2059, Aug. 2020, doi:
10.1109/TPWRD.2019.2960405.

[14] S. M. Muyeen, R. Takahashi and J. Tamura, "Operation and Control of HVDC-


Connected Offshore Wind Farm," IEEE Transactions on Sustainable Energy, vol. 1,
no. 1, pp. 30-37, Apr., 2010, doi: 10.1109/TSTE.2010.2041561.

[15] M. S. Mahmoud and M. O. Oyedeji, "Adaptive and predictive control strategies for
wind turbine systems: a survey," in IEEE/CAA Journal of Automatica Sinica, vol. 6,
no. 2, pp. 364-378, Mar., 2019, doi: 10.1109/JAS.2019.1911375

[16] Y. Wu, C. Lee, C. Chen, K. Hsu and H. Tseng, "Optimization of the wind turbine
layout and transmission system planning for a large-Scale offshore wind farm by AI
technology," IEEE Transactions on Industry Applications, vol. 50, no. 3, pp. 2071-
2080, May-Jun., 2014, doi: 10.1109/TIA.2013.2283219.

[17] G. Quinonez-Varela, G. W. Ault, O. Anaya-Lara, and J. R. McDonald, “Electrical


collector system options for large offshore wind farms,” IET Renew. Power Gener.,
vol. 1, no. 2, pp. 107–114, Jun. 2007.

[18] E. Apostolaki-Iosifidou, R. Mccormack, W. Kempton, P. Mccoy and D. Ozkan,


"Transmission design and analysis for large-scale offshore wind energy
development," IEEE Power and Energy Technology Systems Journal, vol. 6, no. 1,
pp. 22-31, Mar., 2019, doi: 10.1109/JPETS.2019.2898688.

[19] L. P. Garces, A. J. Conejo, R. Garcia-Bertrand and R. Romero, "A bi-level approach


to transmission expansion planning within a market environment," IEEE
Transactions on Power Systems, vol. 24, no. 3, pp. 1513-1522, Aug. 2009, doi:
10.1109/TPWRS.2009.2021230.

[20] A. Prasai, J. Yim, D. Divan, A. Bendre and S. Sul, "A new rchitecture for offshore
wind farms," IEEE Transactions on Power Electronics, vol. 23, no. 3, pp. 1198-1204,
May 2008, doi: 10.1109/TPEL.2008.921194.

[21] J.Parnell, “Siemens Gamesa Launches 14MW Offshore Wind Turbine, World’s
Largest,” GreenTech Media, May. 19, 2020. [Online]. Available:
https://www.greentechmedia.com/articles/read/siemens-gamesa-takes-worlds-
largest-turbine-title

[22] A.Afanoukoe, K.Kanareva, “Cable supplies for the world's first 66 KV offshore
windfarms,” Nexans, Sep. 26, 2018. [Online]. Available:
https://www.nexans.com/newsroom/news/details/2018/09/Supply-of-cables-for-the-
213
first-66kv-offshore-windfarms.html

[23] H. Ergun, D. Van Hertem and R. Belmans, "Transmission system topology


optimization for large-scale offshore wind integration," IEEE Transactions on
Sustainable Energy, vol. 3, no. 4, pp. 908-917, Oct. 2012, doi:
10.1109/TSTE.2012.2199341.

[24] O. Dahmani, S. Bourguet, M. Machmoum, P. Guerin, P. Rhein and L. Josse,


"Optimization and reliability evaluation of an offshore wind farm architecture," IEEE
Transactions on Sustainable Energy, vol. 8, no. 2, pp. 542-550, April 2017. doi:
10.1109/TSTE.2016.2609283

[25] S. Lundberg, “Performance Comparison of Wind Park Congurations," Dept. of


Electric Power Engineering, Chalmers University of Technology, G¨oteborg, Sweden,
Report No.30R, 2003.

[26] L. Xuan, S. Qiang, L. Wenhua and M. Yulong, "Study on fault ride-through capability
of wind farm integration using MMC-HVDC," 2014 International Conference on
Power System Technology, 2014, pp. 2596-2601, doi:
10.1109/POWERCON.2014.6993531.

[27] A.Guldbrand, "Earth faults in Extensive Cable Networks," Licentiate Thesis, Lund
University, Sweden, 2009.

[28] M. Dicorato, G. Forte, M. Pisani, M. Trovato,"Guidelines for assessment of


investment cost for offshore wind generation", Renewable Energy, Volume 36, Issue
8, pp. 2043–2051, August 2011.

[29] A.Dordevic and Z.Durisic, "Mathematical model for the optimal determination of
voltage level and PCC for large wind farms connection to transmission network," IET
Renewable Power Generation,Vol. 13,pp. 2240-2250, 2018.

[30] O. Dahmani, S. Bourguet, M. Machmoum, P. Guérin, P. Rhein and L. Jossé,


"Optimization of the connection topology of an offshore wind farm network," in IEEE
Systems Journal, Vol. 9, no. 4, pp. 1519-1528, Dec. 2015, doi:
10.1109/JSYST.2014.2330064.

[31] Z. Li, M. Zhao, and Z. Chen, “Efficiency evaluation for offshore wind farms,” in Proc.
Int. Conf. Power System Technology, pp. 1–6. 2006.

[32] P. Bresesti, W. L. Kling, R. L. Hendriks, and R. Vailati, “HVDC connection of


offshore wind farms to the transmission system,” IEEE Trans. Energy Convers., vol.
22, no. 1, pp. 37–43, Mar. 2007.

[33] M. Banzo and A. Ramos, "Stochastic optimization model for electric power system
planning of offshore wind farms," in IEEE Transactions on Power Systems, vol. 26,
no. 3, pp. 1338-1348, Aug. 2011, doi: 10.1109/TPWRS.2010.2075944.
214
[34] Y. Chen, Z. Y. Dong, K. Meng, F. Luo, Z. Xu and K. P. Wong, "Collector system
layout optimization framework for large-scale offshore wind farms," in IEEE
Transactions on Sustainable Energy, vol. 7, no. 4, pp. 1398-1407, Oct. 2016, doi:
10.1109/TSTE.2016.2549602.

[35] S. Lumbreras and A. Ramos, "Optimal design of the electrical layout of an offshore
wind farm applying decomposition strategies," in IEEE Transactions on Power
Systems, vol. 28, no. 2, pp. 1434-1441, May 2013.

[36] S. Paul and Z. H. Rather, "A new bi-Level planning approach to find economic and
reliable layout for large-scale wind farm," in IEEE Systems Journal, vol. 13, no. 3,
pp. 3080-3090, Sept. 2019, doi: 10.1109/JSYST.2019.2891996.

[37] Y. Wu, P. Su, Y. Su, T. Wu and W. Tan, "Economics- and reliability-based design for
an offshore wind farm," in IEEE Transactions on Industry Applications, vol. 53, no.
6, pp. 5139-5149, Nov.-Dec. 2017, doi: 10.1109/TIA.2017.2737399.

[38] P. Hou, W. Hu, M. Soltani and Z. Chen, "Optimized placement of wind turbines in
large-scale offshore wind farm using particle swarm optimization algorithm," in IEEE
Transactions on Sustainable Energy, vol. 6, no. 4, pp. 1272-1282, Oct. 2015, doi:
10.1109/TSTE.2015.2429912.

[39] S. Lumbreras and A. Ramos, "A Benders' decomposition approach for optimizing the
electric system of offshore wind farms," 2011 IEEE Trondheim PowerTech, 2011, pp.
1-8, doi: 10.1109/PTC.2011.6019371.

[40] J.O.G.Tande et al. "Impact of TradeWind offshore wind power capacity scenarios on
power flows in the European HV network," Wind Power : Alternative Energy Source.
7th International Workshop on Large Scale Integration of Wind Power and on
Transmission Networks for Offshore Wind Farms, Icfai University Press, 2009,
pp.145-160.

[41] M.Zhao, Z.Chen, F.Blaabjerg, " Optimisation of electrical system for offshore wind
farms via genetic algorithm ," IET Renewable Power Generation, vol.3, pp. 205–216,
Jun. 2009. doi: 10.1049/iet-rpg:20070112.

[42] M. Zhao, Z. Chen and J. Hjerrild, "Analysis of the behaviour of genetic algorithm
applied in optimization of electrical system design for offshore wind farms," in
IECON 2006, Paris, France, 2006, pp. 2335-2340, doi: 10.1109/IECON.2006.347333.

[43] D.D. Li, C.He and Y. Fu, "Optimization of internal electric connection system of large
offshore wind farm with hybrid genetic and immune algorithm," in 2008 Third
International Conference on Electric Utility Deregulation and Restructuring and
Power Technologies, Nanjing, 2008, pp. 2476-2481, doi:
10.1109/DRPT.2008.4523827.

[44] P. Sanchez-Martin, A. Ramos, and J. F. Alonso, “Probabilistic midterm transmission


215
planning in a liberalized market,” IEEE Trans. Power Syst., vol. 20, pp. 2135–2142,
2005.

[45] I. Mustakerov and D. Borissova, “Wind turbines type and number choice using
combinatorial optimization,” Renew. Energy, vol. 35, pp.1887–1894, 2010.

[46] A. A.B.Rújula and R. Martínez, "A new tool for the optimal design of electrical cables
in wind farms, " Renewable Energy and Power Quality J., vol.1, pp. 344-349, 2005.

[47] K. Meng, W. Zhang, J. Qiu, Y. Zheng and Z. Y. Dong, "Offshore transmission Network
planning for wind integration considering AC and DC transmission options," in IEEE
Transactions on Power Systems, vol. 34, no. 6, pp. 4258-4268, Nov. 2019, doi:
10.1109/TPWRS.2019.2912414.

[48] N. Barberis Negra, O. Holmstrom, B. Bak-Jensen and P. Sorensen, "Aspects of


relevance in offshore wind farm reliability assessment," in IEEE Transactions on
Energy Conversion, vol. 22, no. 1, pp. 159-166, Mar. 2007, doi:
10.1109/TEC.2006.889610.

[49] A. Sannino, H. Breder and E. K. Nielsen, "Reliability of collection grids for large
offshore wind parks," 2006 International Conference on Probabilistic Methods
Applied to Power Systems, 2006, pp. 1-6, doi: 10.1109/PMAPS.2006.360415.

[50] A. Brooke, D. Kendrick, A. Meeraus, R. Raman, and R. E. Rosenthal, "GAMS—A


User’s Guide," GAMS Development Corporation, Washington, DC, USA, 2008.
[Online]. Available: http://www.gams.com/dd/docs/bigdocs/GAMSUsers-Guide.pdf.

[51] P. Sood, V. Winstead and P. Steevens, "Optimal placement of wind turbines: A Monte
Carlo approach with large historical data set," 2010 IEEE International Conference
on Electro/Information Technology, 2010, pp. 1-5, doi: 10.1109/EIT.2010.5612130.

[52] C.N.Elkinton, J.Manwell, J.G. McGowan, “Offshore wind farm layout optimization
(OWFLO) project: an introduction,” FME Transactions, vol.38, pp.107–114, 2005.

[53] C.Wan, J.Wang, G.Yang, X.Zhang, “Optimal micro-siting of wind farms by particle
swarm optimization,” in International Conference in Swarm Intelligence, Berlin,
2010, pp. 198–205.

[54] C.Szafron, “Offshore windfarm layout optimization,” in 9th International Conference


on Environment and Electrical Engineering (EEEIC), Prague, Czech Republic, 2010,
pp. 542–545.

[55] B.Rasuo, A.Bengin, “Optimization of wind farm layout,” FME Transactions, vol. 38,
pp. 107–114, 2010.

[56] H. S. Huang and C. Yun, “Distributed genetic algorithm for optimization of wind farm

216
annual profits,” in Int.Conf. Intelligent Systems Applications Power Systems, Niigata,
Japan, Nov. 5–8, 2007.

[57] M. Bilbao and E. Alba, “Simulated annealing for optimization of wind farm annual
profit,” in 2nd Int. Conf. Logistic Industrial Informatics, Linz, Austria, Sep. 10–12,
2009.

[58] F. Wang, D. Liu, and L. Zeng, “Modeling and simulation of optimal wind turbine
configurations in wind farms,” in Int. Conf. World Non-Grid-Connected Wind Power
Energy Conference, Nanjing, China, Sep. 24–26, 2009.

[59] R. Rahmani, A. Khairuddin, S. M. Cherati, and H. A. M. Pesaran, “A novel method


for optimal placing wind turbines in a wind farm using particle swarm optimization
(PSO),” in Proc. IPEC, Oct. 27–29, 2010, pp. 134–139.

[60] C. Wan, J. Wang, G. Yang, and X. Zhang, “Particle swarm optimization based on
Gaussian mutation and its application to wind farm micro-siting,” in Proc. 49th IEEE
Decision Control, Atlanta, GA, USA, Dec. 15–17, 2010, pp. 2227–2232.

[61] L. Ningsu, “Analysis of offshore support structure dynamics and vibration control of
floating wind turbines,” in Proc. 31st Chin. Control Conf., Jul.25–27, 2012, pp. 6692–
6697.

[62] L. Sethuraman, V. Venugopae, and M. Mueller, “Drive-train configurations for


floating wind turbines, " in 8th Int. Conf. Ecological Vehicles Renewable Energies,
Monte Carlo, Monaco, Mar. 27–30, 2013.

[63] A.Klein et al.,"An integer programming model for branching cable layouts in offshore
wind farms," Adv. Intell. Syst. Comput., 2015, 359, pp. 27–36.

[64] P. D. Hopewell, F. Castro-Sayas and D. I. Bailey, "Optimising the design of offshore


wind farm collection networks," in Proceedings of the 41st International Universities
Power Engineering Conference, Newcastle-upon-Tyne, 2006, pp. 84-88, doi:
10.1109/UPEC.2006.367720.

[65] H. Lingling, F. Yang and G. Xiaoming, "Optimization of electrical connection scheme


for large offshore wind farm with genetic algorithm," 2009 International Conference
on Sustainable Power Generation and Supply, 2009, pp. 1-4, doi:
10.1109/SUPERGEN.2009.5348118.

[66] Y. Chen, H. Li, B. He, P. Wang, and K. Jin, “Multi-objective genetic algorithm based
innovative wind farm layout optimization method,” Energy Convers. Manag., vol.
105, pp. 1318–1327, 2015.

[67] J. S. Shin and J. O. Kim, “Optimal design for offshore wind farm considering inner
grid layout and offshore substation location,” IEEE Trans. Power Syst., vol. 32, no.
3, pp. 2041–2048, May 2017.
217
[68] S. Rodrigues, P. Bauer, and P. A. Bosman, “Multi-objective optimization of wind farm
layouts- complexity, constraints handling and scalability,” Renew. Sustain. Energy
Rev., vol. 65, pp. 587–609, 2016.

[69] F. M. Gonzalez-Longatt, P. Wall, P. Regulski, and V. Terzija, “Optimal electric


network design for a large offshore wind farm based on a modified genetic algorithm
approach,” IEEE Syst. J., vol. 6, no. 1, pp. 164–172, Mar. 2012.

[70] X. Han,Y. Qu, P.Wang, and J.Yang, “Four-dimensional wind speed model for
adequacy assessment of power systems with wind farms,” IEEE Trans. Power Syst.,
vol. 28, no. 3, pp. 2978–2985, Aug. 2013.

[71] A. Dolatabadi, B. Mohammadi-Ivatloo, M. Abapour, and S. Tohidi, “Optimal


stochastic design of wind integrated energy hub,” IEEE Trans. Ind. Inform., vol. 13,
no. 5, pp. 2379–2388, Oct. 2017.

[72] W. Y. Kwong, P. Y. Zhang, D. Romero, J. Moran, M. Morgenroth, and C. Amon,


“Wind farm layout optimization considering energy generation and noise
propagation,” in Proc. ASME Int. Des. Eng. Tech. Conf. Comput. Inf. Eng. Conf., 2012,
pp. 323–332.

[73] A.M.Pemberton, T.D.Daly, N.Ertugrul, "On-shore wind farm cable network


optimization utilizing a multi-objective genetic algorithm," WindEng., pp. 659–673,
2013. [Online]. Available doi: org/10.1260/0309-524X.37.6.659.

[74] C.Wan, J.Wang, G. Yang, H. Gu, and X. Zhang, “Wind farm micro-siting by Gaussian
particle swarm optimization with local search strategy,” Renew. Energy, vol. 48, pp.
276–286, 2012.

[75] H. Yang, K. Xie, H. Tai and Y. Chai, "Wind Farm Layout Optimization and Its
Application to Power System Reliability Analysis," in IEEE Transactions on Power
Systems, vol. 31, no. 3, pp. 2135-2143, May 2016, doi:
10.1109/TPWRS.2015.2452920.

[76] P. Hou, W. Hu, Z. Chen, "Optimization for offshore wind farm cable connection layout
using adaptive particle swarm optimization minimum spanning tree method," IET
Renew. Power Gener., vol.10, pp. 694–702, 2016.

[77] P. Hou, W. Hu, Z. Chen, "Optimization of offshore wind farm cable connection layout
considering levelised production cost using dynamic minimum spanning tree
algorithm," IET Renew. Power Gener., vol.10, pp. 175–183, 2016.

[78] K. Veeramachaneni, M. Wagner, U.-M. O’Reilly, and F. Neumann, “Optimizing


energy output and layout costs for large wind farms using particle swarm
optimization,” in Proc. IEEE Congr. Evol. Comput. (CEC), Jun. 2012, pp. 1–7.

[79] Y.Wang, H. Liu, H. Long, Z. Zhang, and S. Yang, “Differential evolution with a new
218
encoding mechanism for optimizing wind farm layout,” IEEE Trans. Ind. Inform., vol.
14, no. 3, pp. 1040–1054, Mar. 2018.

[80] H. Long and Z. Zhang, “A two-echelon wind farm layout planning model,” IEEE
Trans. Sustain. Energy, vol. 6, no. 3, pp. 863–871, Jul. 2015.

[81] A. Cerveira, A. de Sousa, E. S. Pires, and J. Baptista, “Optimal cable design of wind
farms: the infrastructure and losses cost minimization case,” IEEE Trans. Power Syst.,
vol. 31, no. 6, pp. 4319–4329, Nov. 2016.

[82] M. Fischetti and D. Pisinger, "Inter-array cable routing optimization for big wind
parks with obstacles," 2016 European Control Conference (ECC), Aalborg, 2016, pp.
617-622, doi: 10.1109/ECC.2016.7810357.

[83] P. Mittal and K. Mitra, "Energy-noise trade-off to optimize the total number and the
placement of wind turbines on wind farms: a hybrid approach," 2017 Indian Control
Conference (ICC), Guwahati, 2017, pp. 129-136, doi:
10.1109/INDIANCC.2017.7846464.

[84] J.S.Finn et al, “Guidelines for the Design and Construction of AC Offshore
Substations for Wind Power Plants”, Cigre, Nov. 2011.

[85] M. de Prada, L. Igualada, C. Corchero, O. Gomis-Bellmunt and A. Sumper, "Hybrid


AC-DC offshore wind power plant topology: optimal design," in IEEE Transactions
on Power Systems, vol. 30, no. 4, pp. 1868-1876, Jul. 2015, doi:
10.1109/TPWRS.2014.2354457.

[86] B. Railing, G. Moreau, L. Ronstrom, J. Miller, P. Bard, J. Lindberg, and P. Steckley,


“Cross sound cable project—second generation VSC technology for HVDC,” in Proc.
Cigre Session, Paris, France, 2004.

[87] S. Wei, L. Zhang, Y. Xu, Y. Fu and F. Li, "Hierarchical Optimization for the Double-
Sided Ring Structure of the Collector System Planning of Large Offshore Wind
Farms," in IEEE Transactions on Sustainable Energy, vol. 8, no. 3, pp. 1029-1039,
Jul. 2017, doi: 10.1109/TSTE.2016.2646061.

[88] S. Lundberg, “Wind farm configuration and energy efficiency studies—series DC


versus AC layouts,” Ph.D. dissertation, Dept. Energy and Environment, Chalmers
Universty of Technology, Goteborg, Sweden, 2006. [Online] Available:
https://core.ac.uk/download/pdf/70568175.pdf

[89] “Global Offshore Wind Farm Database, 2011”, 4C offshore, 2011. [Online]. Available:
http://www.4coffshore.com/offshorewind/.

[90] F. F. Da Silva and R. Castro, “Power flow analysis of HVAC and HVDC transmission
systems for offshore wind parks,” Int. J. Emerg. Elect. Power Syst., vol. 10, pp. 1–13,
2009.
219
[91] R. Liu, "Progress of long-distance DC electrical power transmission," in 1st
International Conference on Electrical Materials and Power Equipment (ICEMPE),
Xi'an, China, 2017, pp. 93-96, doi: 10.1109/ICEMPE.2017.7982153.

[92] B. Sfurtoc, R. da Silva and S. Chaudhary, "A MTDC system layout review based on
system revenue a Kriegers Flak case study," in 4th International Conference on
Power Engineering, Energy and Electrical Drives, Istanbul, 2013, pp. 793-800, doi:
10.1109/PowerEng.2013.6635711.

[93] I. Martínez Sanz, B. Chaudhuri and G. Strbac, "Inertial response from offshore wind
farms connected through DC grids," in IEEE Transactions on Power Systems, vol. 30,
no. 3, pp. 1518-1527, May 2015, doi: 10.1109/TPWRS.2014.2349739.

[94] Prieto-Araujo, E., Bianchi, F. D., Junyent-Ferre, A., and Gomis- Bellmunt, O.,
“Methodology for droop control dynamic analysis of multiterminal VSC-HVDC
grids for offshore wind farms,” IEEE Trans. Power Del., Vol. 26, pp. 2476–2485, Oct.
2011.

[95] J. Beerten, S. Cole and R. Belmans, "Modeling of multi-terminal VSC HVDC systems
with distributed DC voltage control," in IEEE Transactions on Power Systems, vol.
29, no. 1, pp. 34-42, Jan. 2014, doi: 10.1109/TPWRS.2013.2279268.

[96] K. Rouzbehi, A. Miranian, J. I. Candela, A. Luna and P. Rodriguez, "A generalized


voltage droop strategy for control of multiterminal DC grids," in IEEE Transactions
on Industry Applications, vol. 51, no. 1, pp. 607-618, Jan.-Feb. 2015, doi:
10.1109/TIA.2014.2332814.

[97] T. M. Haileselassie and K. Uhlen, "Impact of DC line voltage drops on power flow of
MTDC using droop control," in IEEE Transactions on Power Systems, vol. 27, no. 3,
pp. 1441-1449, Aug. 2012, doi: 10.1109/TPWRS.2012.2186988.

[98] Cao, J., Du, W. J., Wang, H. F. F., and Bu, S. Q., “Minimization of transmission loss in
meshed AC/DC grids with VSC-MTDC networks,” IEEE Trans. Power Syst., Vol. 28,
pp. 3047–3055, Aug. 2013.

[99] C. Gavriluta, I. Candela, A. Luna, A. Gomez-Exposito and P. Rodriguez, "Hierarchical


control of HV-MTDC systems with droop-based primary and OPF-based secondary,"
in IEEE Transactions on Smart Grid, vol. 6, no. 3, pp. 1502-1510, May 2015, doi:
10.1109/TSG.2014.2365854.

[100] R. M. Edgell and R. S. Bayless, "AC and DC transmission comparison for


Kaiparowits coal-fired generation plant," in IEEE Transactions on Power Apparatus
and Systems, vol. 95, no. 4, pp. 1123-1135, Jul. 1976, doi: 10.1109/T-PAS.1976.32205.

[101] X. Zhang, “Multiterminal voltage-sourced converter-based HVDC models for power


flow analysis,” IEEE Trans. Power Syst., vol. 19, no. 4, pp. 1877–1884, Nov. 2004.

220
[102] X. Chen, H. Sun, J. Wen, et al., “Integrating wind farm to the grid using hybrid
multiterminal hvdc technology,” IEEE Trans. Ind. Appl., vol. 47, no. 2, pp. 965–972,
Mar./Apr. 2011.

[103] W. Feng, L. A. Tuan, L. B. Tjernberg, et al., "A new approach for benefit evaluation
of multiterminal VSC–HVDC using a proposed mixed AC/DC optimal power flow,"
IEEE Transactions on Power Delivery, vol. 29, no. 1, pp. 432-443, Feb. 2014.

[103] R. Billinton, G. Yi, and R. Karki, “Application of a joint deterministicprobabilistic


criterion to wind integrated bulk power system planning,” IEEE Trans. Power Syst.,
vol. 25, no. 3, pp. 1384–1392, Aug. 2010.

[105] W. Li and P. Choudhury, “Probabilistic transmission planning,” IEEE Power Energy


Mag., vol. 5, no. 5, pp. 46–53, Sep./Oct. 2007.

[106] R. C. Prim, “Shortest connection networks and some generalizations,” Bell Syst. Tech.
J., vol. 36, no. 6, pp. 1389–1401, May 1957.

[107] F. Luo, Z. Y. Dong, Yingying Chen and J. Zhao, "Natural aggregation algorithm: A
new efficient metaheuristic tool for power system optimizations," in 2016 IEEE
International Conference on Smart Grid Communications (SmartGridComm),
Sydney, NSW, 2016, pp. 186-192, doi: 10.1109/SmartGridComm.2016.7778759.

[108] C. S. Seo et al., “Offshore wind power planning in Korea,” in Proc. Eur. Conf. Power
Electron. Appl., 2013, pp. 1–6.

[109] R. O. Duda and P. E. Hart, Pattern Classification and Scene Analysis, Hoboken, NJ,
USA: Wiley, 1974.

[110] D. Jayaweera and S. Islam, “Security of energy supply with change in weather
conditions and dynamic thermal limits,” IEEE Trans. Smart Grid, vol. 5, no. 5, pp.
2246–2254, Sep. 2014.

[111] Y.V. Makarov, P.V. Etingov, J. Ma, et al., “Incorporating uncertainty of wind power
generation forecast into power system operation, dispatch, and unit commitment
procedures,” IEEE Trans. Sustain. Energy, vol. 2, no. 4, pp. 433–442, Oct. 2011.

[112] Australian Bureau of Meteorology Website. [Online]. Available: http://www.


bom.gov.au/.

[113] C. Grigg, et al., "The IEEE Reliability Test System-1996. A Report Prepared by the
Reliability Test System Task Force of the Application of Probability Methods
Subcommittee," IEEE Transactions on Power Systems, vol. 14, no. 3, pp. 1010-1020,
Aug. 1999.

[114] A.Dordevic and Z.Durisic, " General mathematical model for the calculation of

221
economic cross sections of cables for wind farms collector systems," IET Renewable
Power Generation, Vol. 12, pp. 901-909, 2018.

[115] M. Moghavvemi and O. Faruque, “Real-time contingency evaluation and ranking


technique,” IEEE Proc. Gener. Transm. Distrib, vol. 145, pp. 517–524, Sep.1998.

[116] S.Nikkhah, and A.Rabiee, “Voltage stability constrained multi ‐ objective


optimisation model for long‐term expansion planning of large‐scale wind farms,”
IET Gener. Transm. Distrib., vol.12, pp. 548-555, 2017. doi:org/10.1049/iet-
gtd.2017.0763.

[117] S. M. Mohseni-Bonab and A. Rabiee, "Optimal reactive power dispatch: a review,


and a new stochastic voltage stability constrained multi-objective model at the
presence of uncertain wind power generation," in IET Generation, Transmission &
Distribution, vol. 11, no. 4, pp. 815-829, 2017.

[118] K. Meng, et al., "Hierarchical SCOPF considering wind energy integration through
multiterminal VSC-HVDC grids," IEEE Transactions on Power Systems, vol. 32, no.
6, pp. 4211-4221, Nov.2017.

[119] P.Razdan, P.Garrett, “Life Cycle Assessment of Electricity Production From a Vestas
V112 Turbine Wind Plant”, Vestas Wind Systems A/S, Dec.2015. [Online]. Available:
https://www.vestas.com/~/media/vestas/about/sustainability/pdfs/lcav11020mw1812
15.pdf

[120] R.D. Zimmerman and C.E. Murillo-Sanchez, “Matpower 5.1 – User’s Manual”,
Power Systems Engineering Research Center, Mar. 2015.

[121] C.Yuan, X.Yang, D.Yao, C.Y, “The thyristor based hybrid multiterminal HVDC
System,” CSEE HVDC&PE Committee Annual Conference 2015, Shanghai, China,
Oct. 2015

[122] N. M. Kangwa, C. Venugopal and I. E. Davidson, "A review of the performance of


VSC-HVDC and MTDC systems," 2017 IEEE PES PowerAfrica, 2017, pp. 267-273,
doi: 10.1109/PowerAfrica.2017.7991235.

[123] M. Barnes, D. Van Hertem, S. P. Teeuwsen and M. Callavik, "HVDC systems in Smart
Grids," in Proceedings of the IEEE, vol. 105, no. 11, pp. 2082-2098, Nov. 2017, doi:
10.1109/JPROC.2017.2672879.

[124] Y. M. Yeap, N. Geddada, K. Satpathi and A. Ukil, "Time- and frequency-domain fault
detection in a VSC-interfaced experimental DC test system," in IEEE Transactions
on Industrial Informatics, vol. 14, no. 10, pp. 4353-4364, Oct. 2018, doi:
10.1109/TII.2018.2796068.

[125] C. Li, A. M. Gole and C. Zhao, "A fast DC fault detection method using DC reactor
voltages in HVDC grids," in IEEE Transactions on Power Delivery, vol. 33, no. 5,
222
pp. 2254-2264, Oct. 2018, doi: 10.1109/TPWRD.2018.2825779.

[126] R. Li, L. Xu and L. Yao, "DC fault detection and location in meshed multiterminal
HVDC Systems based on DC reactor voltage change rate," in IEEE Transactions on
Power Delivery, vol. 32, no. 3, pp. 1516-1526, June 2017, doi:
10.1109/TPWRD.2016.2590501.\

[127] C.Rytoft, et al., "ABB Review Special Report: 60 years of HVDC, " ABB Group R&D
and Technology, Zurich, Switzerland, Jul. 2014. [Online] www.abb.com/abbreview.

223

You might also like