Evaluation and optimisation of photovoltaic

(PV) plant designs

by

Tafadzwa L. Gurupira

Thesis presented in partial fullment of the requirements for

the degree of Master of Engineering (Electrical) in the
Faculty of Engineering at Stellenbosch University

Supervisor: Dr. A. J. Rix

March 2018
 Stellenbosch University https://scholar.sun.ac.za

Declaration

By submitting this thesis electronically, I declare that the entirety of the work contained therein
is my own, original work, that I am the sole author thereof (save to the extent explicitly
otherwise stated), that reproduction and publication thereof by Stellenbosch University will
not infringe any third party rights and that I have not previously in its entirety or in part
submitted it for obtaining any qualication.

March 2018
Date: .......................................

Copyright © 2018 Stellenbosch University

All rights reserved

i
 Stellenbosch University https://scholar.sun.ac.za

Abstract

Evaluation and optimisation of photovoltaic (PV) plant designs

T. L Gurupira

Department of Electrical and Electronic Engineering,

University of Stellenbosch,

Private Bag X1, Matieland 7602, South Africa.

Thesis: MEng (Electrical)

December 2017

Factors such as the increasing energy demand, the depletion of fossil fuels and the push
for more sustainable, less polluting energy sources have resulted in increased interest and
investments in renewable energy sources worldwide. Among these renewable energy sources
is solar photovoltaics (PV). PV is, however, riddled with several challenges of its own such as
a lack of an ecient, integrated and holistic approach to the PV system design optimisation.
Towards this end, the PV system design process is reviewed, and two optimisation frameworks
are formulated and presented as potential solutions. Of particular interest was how applicable
these frameworks would be in a South African context so two existing plants in the Northern
Cape province of South Africa were adopted as test cases; one for the xed tilt investigations
and the other for the single-axis tracking investigations. A constrained minimisation algorithm
was used to simultaneously evaluate various design variables to maximise the system total
annual energy output, maximise the system's performance ratio and minimise system losses
in the single objective optimisation problems. On the other hand, several design variables
and objectives were simultaneously evaluated in the multi-objective optimisation cases and a
multi-criteria decision making (MCDM) procedure was used to rank the various alternative
optimal PV system layouts to help select the most suitable layout for the given design scenario.
The successful integration of the optimisation algorithms and a reputable solar PV system
design tool led to faster execution of optimisation procedures and a more holistic approach to
system design while ensuring PV system modelling accuracy, expansiveness and adaptability
to several design scenarios.

ii
 Stellenbosch University https://scholar.sun.ac.za

Uittreksel

Evaluation and optimisation of photovoltaic (PV) plant designs

T. L Gurupira

Departement Elektriese en Elektroniese Ingenieurswese,

Universiteit van Stellenbosch,

Privaatsak X1, Matieland 7602, Suid Afrika.

Tesis: MIng (Elek)

Desember 2017

Daar is wêreldwyd 'n toenemende belangstelling in hernubare-energiebronne as gevolg van

'n hoër en stygende aanvraag na energie, asook die potensiële nansiële-winsgewendheid wat
hierdie bronne aanbied vir beleggers. Verder voorsien hernubare-energie bronne ook die geleenthied
om ons afhanklikheid vanaf besoedelende fossielbrandstowwe te verminder. Fotovoltaïese energie
is onder andere een van hierdie hernubare energie bronne. Die proses van energieverskang,
deur middel van fotovoltaïese aanlegte, bied egter nog na 'n paar uitdagings, soos byvoorbeeld 'n
gebrek aan 'n optimale ge- ïntegreerde en holistiese benadering tot die ontwerp van fotovoltaïese
stelsels. Met hierdie doel voor oë, word die ontwerp van die fotovoltaïese stelsel hersien, tesame
met die verdere toepassing van twee optimerings-metodes. Dit was verder van groot belang
om die toepassing van hierdie metodes in die konteks van Suid-Afrika te ondersoek. Daarom
is twee bestaande Fotovoltaïese aanlegte in die Noord-Kaap as toetsgevalle gebruik; een aanleg
is gebruik vir die ondersoek van 'n stasionêre-uitleg en die ander vir die ondersoek van 'n
enkel-as-volger-uitleg. 'n Beperkte optimerings-algoritme is gebruik om gelyktydig verskeie
ontwerpsveranderlikes te evalueer, sodat die totale jaarlikse energie gelewer, asook die stelsel
se prestasieverhouding gemaksimeer word. Hierdie doelwitte, tesame met die vermindering
van stelselverliese, is in ag geneem vir die enkel-doelwit-optimerings-probleme. Aan die ander
kant was daar verskeie veranderlikes en doelwitte gelyktydig geëvalueer in die multidoelwit-
optimerings-gevalle. 'n Multi-kriteria-besluitnemings-prosedure is gebruik om 'n rangorde toe
te ken aan die verskillende optimale Fotovoltaïese stelsel uitlegte, om die mees geskikte uitleg
vir die gegewe ontwerpsgevalle te vind. Die suksesvolle integrasie van die optimeringsalgoritmes,
asook 'n betroubare Fotovoltaïese-stelsel-ontwerps-program, het gelei tot die spoedige uitvoering
van die optimeringsproses, tesame met 'n meer holistiese benadering tot stelselontwerp. Die
doelwitte gestel is almal beriek, asook die eektiwiteit aangaande die ontwerp van die fotovoltaïese
modellering, skalering en aanpasbaarheid vir verskeie ontwerpsgevalle is aangespreek.

iii
 Stellenbosch University https://scholar.sun.ac.za

Acknowledgements

I would like to express my sincerest gratitude to the following people and organisations:

ˆ God almighty for the opportunity, the gift of life and blessings that have seen me get this
far,

ˆ Dr Arnold Rix, my supervisor, for his assistance and guidance,

ˆ Scatec Solar for the funding my studies,

ˆ My family and friends for their support, prayers and encouragement,

ˆ My friends and fellow oce mates: Brian De Beer, JP Botha and the rest of the PV
and Media Lab guys for their wisdom, the memories we made and the endless banter
that made the masters work more bearable. A special mention out of this group goes to
Warren Farmer and Armand Du Plessis for their help with the translation of this thesis'
abstract to Afrikaans.

iv
 Stellenbosch University https://scholar.sun.ac.za

Dedications

I would like to dedicate this thesis to my family. You did all you could to see me get this far
and I will continue to do all I can to represent you and make you proud.

v
 Stellenbosch University https://scholar.sun.ac.za

Contents

Declaration i
Abstract ii
Uittreksel iii
Acknowledgements iv
Dedications v
Contents vi
List of Figures viii
List of Tables x
Nomenclature xiii
1 Introduction 1
1.1 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Project Description . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.3 Thesis Outline . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8

2 Theoretical Background 10
2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
2.2 Solar Resource and Site Selection . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.3 PV System Components . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
2.4 PV System Performance Parameters . . . . . . . . . . . . . . . . . . . . . . . . 22
2.5 PV System Simulation Tools . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
2.6 Plant Design Fundamentals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
2.7 Chapter Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34

3 Preliminary Design Optimisation 36

3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
3.2 Literature Review . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
3.3 Optimisation Criteria . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
3.4 Optimisation tools and strategies: Case Studies . . . . . . . . . . . . . . . . . . 38
3.5 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60

4 Optimisation Strategies 61
4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
4.2 Evolutionary Algorithms (EAs) in the design of grid-connected PV systems . . . 61

vi
 Stellenbosch University https://scholar.sun.ac.za

CONTENTS vii

4.3 Optimisation Methodology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62

4.4 Single Objective Optimisation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
4.5 Multi-objective Optimisation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
4.6 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72

5 Results and Analyses 73

5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
5.2 Single objective optimisation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
5.3 Multi-objective optimisation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
5.4 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99

6 Conclusions 104
6.1 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104
6.2 Recommendations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106

Appendices 108
A Canadian Solar PV Module Datasheet 109
B Kyocera PV Module Datasheet 112
C First Solar PV Module Datasheet 115
List of References 118
 Stellenbosch University https://scholar.sun.ac.za

List of Figures

1.1 Carbon intensity of electricity generation technologies given in grams of carbon

emitted per kWh of electricity generated from the respective technology [1]. . . . . 2
1.2 Renewable energy sources. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.3 Evolution of PV installations from the year 2000 to the year 2016 [2]. . . . . . . . . 4
1.4 Solar resource distribution in South Africa [3]. . . . . . . . . . . . . . . . . . . . . . 4
1.5 Map showing the distribution of 79 renewable energy power plants from the rst
four rounds of the REIPPP in South Africa [4]. . . . . . . . . . . . . . . . . . . . . 5

2.1 Illustration of the sun's altitude and solar azimuth angles β and φs , respectively. . . 12
2.2 Illustration of the solar declination angle. . . . . . . . . . . . . . . . . . . . . . . . 13
2.3 Terrestrial solar irradiation characteristics . . . . . . . . . . . . . . . . . . . . . . . 14
2.4 PV Cell Technologies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
2.5 Types of PV modules . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
2.6 Mounting methods as illustrated in the System Advisor Model documentation[5]. . 23
2.7 Variation of module eciency with irradiance at dierent cell temperature levels . . 26
2.8 Variation of module eciency with temperature at dierent irradiance levels . . . . 27
2.9 Module congurations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
2.10 Eect of changing the azimuth on the power output prole of a PV system (in the
southern hemisphere) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
2.11 Reference gure for shading loss calculations . . . . . . . . . . . . . . . . . . . . . . 33
2.12 Shading derate factors for dierent PV module congurations adapted from [6] . . . 33
2.13 System sizing algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35

3.1 Renewable energy power plants in South Africa courtesy of [7] . . . . . . . . . . . . 39

3.2 A screenshot of the SAM application user interface . . . . . . . . . . . . . . . . . . 40
3.3 A screenshot of the PVSyst application user interface . . . . . . . . . . . . . . . . . 41
3.4 A screenshot of the PVLib-python toolbox, running on Jupyter . . . . . . . . . . . 42
3.5 Total energy yield for the year 2014 . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
3.6 Clear sky and cloudy day energy proles . . . . . . . . . . . . . . . . . . . . . . . . 48
3.7 Algorithm for iterative tilt angle optimisation . . . . . . . . . . . . . . . . . . . . . 50
3.8 Monthly Energy Yield comparison of various tilting and mounting methods . . . . . 52
3.9 Total Energy Yield Comparison for the various tilt and mounting methods . . . . . 52
3.10 Graph of variation of system performance with inter-row spacing . . . . . . . . . . 55
3.11 Graph showing variation of shading losses with inter-row spacing . . . . . . . . . . 56
3.12 A pictoral representation of the east-west orientation. . . . . . . . . . . . . . . . . 56
3.13 Depiction of the apparent path of the sun as the day and year passes. . . . . . . . . 57
3.14 Comparison of installation capacity showing the increase in the number of PV
modules that can t into the site due to the east-west orientation. . . . . . . . . . 58
3.15 Comparison of energy output of the PV systems indicating the increases in the
energy production due to the east-west orientation. . . . . . . . . . . . . . . . . . . 59

viii
 Stellenbosch University https://scholar.sun.ac.za

LIST OF FIGURES ix

3.16 Comparison of the system performances showing a reduction the percentage drop
in production due to the east-west orientation. . . . . . . . . . . . . . . . . . . . . 59

4.1 SAM simulation process ow . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63

4.2 Illustration of a Pareto optimal solution set from a 2 dimensional performance space 67
4.3 Genetic algorithm ow chart . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
4.4 Illustration of the tournament selection operation . . . . . . . . . . . . . . . . . . . 69
4.5 An illustration of the crossover operation . . . . . . . . . . . . . . . . . . . . . . . . 69
4.6 2-dimensional performance space from a bi-objective optimisation problem . . . . . 70

5.1 Comparison of the performance of PolySi module technologies with respect to each
objective for the xed tilt case, Case 1 . . . . . . . . . . . . . . . . . . . . . . . . . 78

6.1 Illustration of source circuits connected using the daisy chain and leapfrog wiring
methods. Figures adopted from [8] . . . . . . . . . . . . . . . . . . . . . . . . . . . 106
 Stellenbosch University https://scholar.sun.ac.za

List of Tables

2.1 Decomposition model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15

2.2 Transposition models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
2.3 Sources of solar resource datasets . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.4 Soiling Loss Scale . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26

3.1 Sub-system conguration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39

3.2 Summary of plant losses . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
3.3 2014 Total Annual Yield Comparison . . . . . . . . . . . . . . . . . . . . . . . . . . 47
3.4 Table of season calendar dates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
3.5 Table with the calculated monthly, seasonal and biannual tilt angles . . . . . . . . . 51

4.1 Decision matrix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71

5.1 fmincon stopping criteria . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74

5.2 Inequality constraint parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76
5.3 Inequality constraint parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76
5.4 Lower and upper bounds of solar PV system design variables . . . . . . . . . . . . . 76
5.5 Fixed-tilt total annual energy output, Qout , optimisation results . . . . . . . . . . . 76
5.6 Optimisation results for maximum PR in MonoSi, PolySi and Thin Film PV module
technologies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
5.7 Optimisation results for minimum system losses in MonoSi, PolySi and Thin Film
PV module technologies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
5.8 Inequality constraint parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
5.9 Lower and upper bounds on solar PV system design variables . . . . . . . . . . . . 80
5.10 Single-axis tracking total annual energy output, Qout , optimisation results . . . . . . 80
5.11 Single-axis tracking performance ratio, PR, optimisation results . . . . . . . . . . . 80
5.12 Single-axis tracking performance ratio, PR, optimisation results . . . . . . . . . . . 81
5.13 MOGA stopping criteria . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
5.14 Criteria and weights for the TOPSIS method . . . . . . . . . . . . . . . . . . . . . . 83
5.15 Lower and upper bounds of solar PV system design variables . . . . . . . . . . . . . 84
5.16 Fixed Tilt MonoSi Pareto front . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
5.17 Fixed Tilt PolySi Pareto front . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
5.18 Fixed Tilt Thin Film Pareto Front . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
5.19 Fixed Tilt Ranking of the best congurations from the 3 dierent technologies . . . 88
5.20 Criteria and weights for the TOPSIS method . . . . . . . . . . . . . . . . . . . . . . 89
5.21 Lower and upper bounds of solar PV system design variables . . . . . . . . . . . . . 90
5.22 MonoSi Pareto Front - Tracking . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
5.23 Evaluation of PolySi modules in a xed tilt with dierent weights . . . . . . . . . . 92
5.24 Thin Film Pareto Front - Tracking . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
5.25 Ranking of the best congurations from the 3 dierent technologies under investigation
after applying TOPSIS. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94

x
 Stellenbosch University https://scholar.sun.ac.za

LIST OF TABLES xi

5.26 Criteria and weights for the TOPSIS method . . . . . . . . . . . . . . . . . . . . . . 95

5.27 Evaluation of MonoSi modules in a xed tilt system with dierent weights. . . . . . 96
5.28 Evaluation of PolySi modules in a xed tilt with dierent weights . . . . . . . . . . 97
5.29 Evaluation of Thin Film modules in a xed tilt system with dierent weights . . . . 98
5.30 Criteria and weights for the TOPSIS method . . . . . . . . . . . . . . . . . . . . . . 99
5.31 Evaluation of layouts of MonoSi modules in a single-axis tracking system with
objectives with dierent weights. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100
5.32 Evaluation of PolySi modules in a single-axis tracking system with dierent weights. 101
5.33 Evaluation of Thin Film modules in a single axis tracking with dierent weights. . . 102
 Stellenbosch University https://scholar.sun.ac.za

NOMENCLATURE xiii

Nomenclature

AC Alternating Current

AHP Analytical Hierarchical Processes

AM Air Mass

Ambtemp Ambient temperature

AOI Angle Of Incidence

ASCI American Standard Code for Information Interchange

BoS Balance of system

BW Bid Window

CdTe Cadmium Telluride

CF Capacity Factor

CIS Copper Indium Silinum

CM Circuit Mismatch

CSP Concentrated Solar Power

CSV Comma Separated Values

DC Direct Current

DHI Diuse Horizontal Irradiance

DISC Direct Insolation Simulation Code

DNI Direct Normal Irradiance

EQT Equation Of Time

E-W East-West

EMF Electromotive Force

GCR Ground Cover Ratio

GHI Global Horizontal Irradiance

Ginc Global Incident Irradiance

GW Gigawatt

GWp Gigawatt peak

IAM Incidence Angle Modier

IPP Independent Power Producer

ISO International Organisation of Standardisation

 Stellenbosch University https://scholar.sun.ac.za

NOMENCLATURE xiv

LCOE Levelized Cost Of Energy

LID Light Induced Degradation

LST Local Solar Time

MCDM Multiple Criteria Decision Making

MLPE Module Level Power Electronics

MPPT Maximum Power Point Tracking

MW Megawatt

MWp Megawatt peak

MWh Megawatt hour

NASA National Aeronautics and Space Administration

NIS Negative Ideal Solution

NOD Number Of Days

NREL National Renewable Energy Laboratory

N-S North-South

PIS Positive Ideal Solution

POA Plane Of Array

PR Performance Ratio

PV Photovoltaic

PVGCS Photovoltaic Grid Connected System

REFIT Renewable Energy Feed in Tari

REIPPPP Renewable Energy Independent Power Producer Procurement Program

SAW Simple Additive Weighting

SAM System Advisor Model

SAURAN South African Universities Radiometric Network

SQP Sequential quadratic programming

SODA Solar radiation data

STC Standard Test Conditions

TMA Total Module Area

TOPSIS Technique for Order Preference by Similarity to the Ideal Solution

WindVel Wind Velocity

 Stellenbosch University https://scholar.sun.ac.za

Chapter 1
Introduction

1.1 Background
There is a strong case being made worldwide for the adoption of renewable energy sources. The
global energy sector is grappling to manage the increasing energy demand. This increase in
demand is mainly driven by global population growth and the economic development of some of
the world's emerging economies. To make matters worse, fossils fuels, which curently account
for the majority of the global primary energy production are facing depletion due to their
nite nature[9],[10] and [11]. Renewable energy sources are, therefore, ideal in this situation
as they will be able to provide a more sustainable and diversied energy supply. Adoption
of renewable energy sources over fossil fuels is also vital to the reduction of carbon emissions
globally [12]. These emissions, from the burning of fossil fuels, trap heat and contribute to the
global warming phenomenon which has harmful eects on our climate, environment and health
[13]. According to a special report compiled by the International Panel on Climate Change
(IPCC), the life-cycle global carbon emissions associated with renewable energy are minimal
compared to those produced by fossil fuels as shown in Figure 1.1.

Some of the common renewable energy sources, as shown in Figure 1.2, include:

ˆ solar, which harnesses energy from sunlight,

ˆ wind, where energy is captured by turbines as the wind blows,

ˆ biomass, which turns organic matter into electricity or transportation fuel,

ˆ geothermal, which uses the earth's internal heat for a wide range of uses such as electricity
production and heating and cooling,

ˆ ocean, which can be harnessed as mechanical energy from the tides and the waves or even
thermal when the sun's heat meets the water and

ˆ hydro-power, where owing water is harnessed using turbines to generate electrical energy.

1
 Stellenbosch University https://scholar.sun.ac.za

CHAPTER 1. INTRODUCTION 2

Carbon intensity by technology

1001
Carbon intensity (g CO2eq/kWh)

840

469

48

45

22

18

16

12

4
Figure 1.1: Carbon intensity of electricity generation technologies given in grams of carbon
emitted per kWh of electricity generated from the respective technology [1].

Wind Ocean Geothermal

Hydropower Solar Biomass

Figure 1.2: Renewable energy sources.

 Stellenbosch University https://scholar.sun.ac.za

CHAPTER 1. INTRODUCTION 3

1.1.1 Solar Energy

Solar energy can be further broken down into several technologies:

ˆ Photovoltaics (PV) in which absorbed solar irradiation is directly converted into electrical
energy using collectors. These collectors are usually made of semi-conductor material
which operates on the principle of the photoelectric eect to generate electricity. There is
also a dierent type of photovoltaics known as Concentrated Photovoltaic (CPV) which
uses a large amount of solar irradiation focussed onto a highly ecient multi-junction
solar cell using lenses and curved mirrors. This technology is usually used in areas with
high Direct Normal Irradiance (DNI).

ˆ Concentrating Solar Power (CSP), where sunlight is reected and concentrated onto
receivers and the solar energy is converted into heat through a medium, usually water,
and the steam produced is used to turn turbines. This technology is also ideal in regions
with high DNI.

ˆ Solar Thermal technologies which are used for water heating, space heating and space
cooling, as well as process heat generation.

ˆ Solar energy for architecture and urban planning, where buildings are oriented towards
the sun's azimuth to provide light and warmth.

1.1.1.1 Solar Energy Worldwide

Several sources, including the 'Snapshot of Global PV Markets :2016' [14], state that the total
installed PV capacity at the end of 2016 was at least 303 GW. 265 GW of the 303 GW were
contributed by the 25 International Energy Agency Photovoltaic Power Supply (IEA PVPS)
countries alone. This was a direct result of a global expansion of about 75 GW in 2016 which
equals a growth rate of about 50% from the 50.9 GW installed globally in 2015. Figure 1.3
shows the progression of global PV installations, in gigawatts, from the year 2000 to the year
2016. An additional 85 GW is projected to be installed by the end of 2017 and the growth in
the photovoltaic industry is anticipated to continue. There is an estimated capacity of between
3000 GW and 10000 GW expected to be installed by the year 2030 [2]. Most of this expansion
is not only linked to the urgency in trying to solve the global energy and pollution crisis, but
it is also directly linked to the technological advancements and industry re-structuring in the
PV industry over the past couple of years. The consequent price reductions and expansion of
the knowledge base in the industry will potentially also continue to inuence the cumulative
number of solar installations worldwide.

On the PV technology side of things, a look at the price and cost evolutions related to solar
cells shows that the average price per watt has dropped over the past few decades. The price of
crystalline silicon solar cells, for instance, has dropped from approximately $77 per watt in 1977
to as low as $0.27 per watt as gazetted by the Energy Trend Company [15]. Since PV modules
are more expensive than any other PV system component in the Balance of System (BoS),
most current PV system designs are done for better value for money by optimally installing
te modules. Recent research has, however, come to light indicating that optimising the BoS
costs is becoming increasingly important due to decreasing module prices. This, in turn, is also
beginning to inuence PV system design decisions [16].
 Stellenbosch University https://scholar.sun.ac.za

CHAPTER 1. INTRODUCTION 4

Evolution of PV Installations

303
Installed capacity (GWPeak,DC )

229
178
138
101
70
40
23
16
9.2
6.7
5.1
3.7
2.6
2.1
1.6
1.3

20 00 200 1 200 2 200 3 2004 2 005 20 0 6 200 7 20 0 8 20 0 9 2010 2011 2012 2013 2014 2015 2016

Figure 1.3: Evolution of PV installations from the year 2000 to the year 2016 [2].

1.1.1.2 Solar Energy in South Africa

Solar energy is one of the most readily accessible renewable energy resources in South Africa.
This can be seen in Figures 1.4a and 1.4b showing the distribution of the solar resource in
South Africa. The Northern Cape Province oers the most favourable levels of solar irradiation
and this explains why, as shown in Figure 1.5, most of the solar PV plants are located in this
province.

(a) Distribution of GHI (b) Distribution of DNI

Figure 1.4: Solar resource distribution in South Africa [3].

 Stellenbosch University https://scholar.sun.ac.za

CHAPTER 1. INTRODUCTION 5

Figure 1.5: Map showing the distribution of 79 renewable energy power plants from the rst
four rounds of the REIPPP in South Africa [4].

Over the years South Africa, like the rest of the world, has also seen growth in the number
of PV installations. This is believed to be due to the rising costs of electricity, constraints in
the electricity supply system, technological advancements and price structure improvements in
the alternative energy sector in the country [17]. In addition to that, the adoption of solar
energy sources has also been expediated by government and municipal incentives. For instance,
a tax incentive was implemented by the government in January 2016 through the South African
Revenue Service (SARS) for renewable energy systems under the amended Income Tax Act No.
58 of 1962. This scal incentive essentially allows for a reduction on movable assets used in
renewable energy production on a 50|30|20 basis over the course of three years, for systems
greater than 1 MWp, if the asset is being put to use for the rst time by the taxpaying entity
[18]. For systems smaller or equal to 1 MWp, a discount of 28% is granted to the taxpaying
entityin year 1 [19]. Several metropolitan municipalities have implemented feed-in-taris and
net-metering as incentives for residential and commercial customers. These include the City of
Johannesburg Metropolitan Municipality [20], City of Cape Town Metropolitan Municipality
[21], and eThikwini Metropolitan Municipality [22] with feed-in-tari programs and Nelson
Mandela Bay Metropolitan Municipality with an already established net metering guideline
document [23].

Eorts are being made on several other fronts to unlock more of South Africa's solar potential
such as developing and growing more credible solar resource databases. SAURAN [24], which
stands for Southern African Universities Radiometric Network, is one such initiative. SAURAN
is spearheaded by the Centre for Renewable and Sustainable Energy Studies (CRSES) at the
University of Stellenbosch and the Group for Solar Energy Thermodynamics (GSET) at the
University of KwaZulu-Natal. Its aim is to provide high-resolution ground-based radiometric
 Stellenbosch University https://scholar.sun.ac.za

CHAPTER 1. INTRODUCTION 6

data from stations in locations in the Southern African region. Furthermore, the Department
of Energy (DOE) and the Department of Science and Technology have a joint initiative known
as the Solar Energy Technology Roadmap (SETRM). STERM's main goal is to foster the
development of solar energy technologies locally, within the connes of relevant policies and
national initiatives [17]. These are but a few of the examples showing the growth and the
development of the solar PV market in South Africa.

1.2 Project Description

1.2.1 Problem Statement

Despite all the notable progress on the national and international front, there is still a lot of
work to be done towards attaining the South Africa's sustainability goals [25] and tackling
some of the drawbacks in the global solar industry. One of the biggest of these drawbacks
is the high capital costs of PV systems and it is a problem too big for any single entity to
try to solve on its own. Therefore, while the government, municipalities and policy makers
play their role in facilitating the process of nding solutions to this, PV system designers also
have an important role to play in their own rite. This includes researching and implementing
optimisation techniques that will enable them to deploy solar projects at lower costs without
sacricing plant performance [26] and [27]. Optimisation can be dened as the process of
making a design or system more eective or functional [28]. With regards to the PV system
design in our context, it means revisiting the design process to nd areas of improvement that
could result in higher energy production, better system performance and lower system losses
and operation costs.

PV system design tools play a signicant role in the design optimisation process. For designers
to have an idea of how the PV system they will be designing is likely to perform in a specic
location under specic conditions, they make use of simulation software. The software takes
weather and site data as inputs and runs simulations and reports back the expected plant yield
and performance based on the proposed PV system design layout. This is an indispensable
step in the design phase as important decisions will be made based on the simulation results so
it requires software that can accurately model PV system behaviour and enable the desisn of
a PV project that is bankable. Bankability is essentially an index of the nancial strength and
stability of a project. Bankability is inuenced by factors such as the quality of the PV system
components in the PV system, which can be directly linked to the manufacturing companies[29]
and [30]. Other important factors taken into account are the reliability of the simulation results,
the simulation software and the datasets used as well as the experience of the system designers.
This therefore means that even the selection of the simulation software cannot be taken lightly.
Simulations and optimisation process may require several iterations and evaluations based on
a number of plant performance parameters so robustness, modelling exibility and the extent
of functionality of the software also need to be taken into account. The unfortunate limitation
with most PV system software tools is that their focus is mostly on:

ˆ PV module selection, interconnection and behaviour in the operating conditions,

ˆ inverter selection, interconnection and behaviour and

ˆ the inuence of mounting structures.

However, cables and other BoS components are seldom taken into account, save for accounting
for when their losses are estimated.
 Stellenbosch University https://scholar.sun.ac.za

CHAPTER 1. INTRODUCTION 7

Moreover, each PV project comes with its own constraints and objectives that sets it apart
from other projects. As such, there is hardly ever a one-size ts all design procedure that can
cater for all types of projects [26] and [31]. There are also, often, numerous design parameters
and objectives that need to be designed for and this could lead to:

ˆ numerous iterations being needed to reach a decision,

ˆ some objectives being incommensurate and therefore dicult to compare with each other,
and

ˆ some objectives and parameters being antagonistic, therefore necessitating some sort of
compromise or trade-o for the selection of the optimal layout selection.

1.2.2 Goals and Objectives

The goal with this project was to investigate the optimisation strategy currently being implemented
by some plant designers in order to identify areas of improvement. The current method involves
calculating the ideal pitch and tilt angle of a PV system in a particular location, varying these
two parameters together with other design variables till the system is optimised for maximum
energy output and then implementing nancial modelling to optimise for levelised cost of energy.
This is done for the three dierent module technologies, monocrystalline, polycrystalline and
thin lm, and the technology and layout that makes the most nancial sense is selected. This
method is therefore investigated for improvement potential.

The rst step was to understand and evaluate the current trends and then formulate optimisation
frameworks that would be more eective and benecial to the designers. Particular attention
was paid to ensure the relevance and applicability of the optimisation frameworks to the South
African context. Three key desired outcomes were identied and laid out as follows:

(a) The frameworks would have to be suitable for constrained optimisation, as each design
space is dened by certain constraints that the system designers have to adhere to. In
addition, the optimisation frameworks would have to be easily adaptable to diverse design
objectives and parameters since as stated before, dierent projects come with dierent
goals and requirements.

(b) The number of iterations needed to nd an optimal layout from the design space becomes
increasingly large when the search space is widened and/or when the design parameters or
objectives to be investigated simultaneously are increased. Robust mathematical models
that can handle such high computational requirements would, therefore, be required.
These models would also have to come with or be easily incorporated with reliable PV
system design tools to ensure system modelling accuracy and bankability.

(c) Multi-objective optimisation was identied as a possible approach to the design process.
This would, however, possibly lead to designers having to compare and contrast incommensurable
parameters and dealing with antagonistic objectives. As such, methods for dealing with
parameters in dierent units and decision making schemes suitable for multiple criteria
would have to be implemented.

1.2.3 Signicance of Research

Solving this problem will lead to frameworks that:

ˆ replace or complement the current arduous and redundant plant design procedure,
 Stellenbosch University https://scholar.sun.ac.za

CHAPTER 1. INTRODUCTION 8

ˆ are easily adaptable to various design scenarios, with regards to constraints, parameters
and objectives,

ˆ provide a faster and well-rounded approach to the PV system design process and

ˆ yield accurate and bankable PV system designs.

This will, as a result, be a step towards improving plant productivity, reducing system costs
and ensuring investor condence and reducing investment risks. In the long run, this could
lead to more protability for the independent power producers (IPPs) while at the same time
improving the prospects of adequate and aordable electrical energy for the consumers in South
Africa.

1.3 Thesis Outline

The researcher aims at coming up with frameworks for optimally designing utility-scale grid
connected PV systems taking multiple variables into account simultaneously, in a constrained
design space. The organisation of the rest of this thesis is as follows:

ˆ Chapter 2: In this chapter, the fundamentals behind solar PV systems are presented.
Conditions favourable for solar PV projects are discussed rst, followed by a breakdown
of the PV system as a whole, with particular focus being drawn to the modules, the
inverters and the mounting systems. Thereafter, system performance parameters are
briey examined and the basics of PV system design are discussed, together with the role
of simulation software in the design and evaluation process.

ˆ Chapter 3: The preliminary design optimisation process is discussed in this chapter.

First, the current conventional design optimisation procedure that is supposedly in need
of improvement is presented, and thereafter, a literature review is conducted to discuss
other optimisation procedures that are also being implemented elsewhere in the PV
industry. Criteria for the optimisation process in the form of design parameters and design
constraints are touched on as well as several investigations into some of the optimisation
strategies found in literature. This then forms the basis on which our nal optimisation
frameworks are formulated.

ˆ Chapter 4: In this chapter, the optimisation frameworks are presented and discussed
in more detail. First a literature review is performed to analyse the algorithms that
have been used in PV system design to date; their strengths, their weaknesses and their
applicability to this context. The optimisation methodology implemented in this project
is then broken down into the software used for the PV system design aspect of the process
and the algorithms implemented for the optimisation process and how the two processes
are integrated. In subsequent sections the algorithms used to perform the constrained
multi-variable optimisation for single and multiple design objectives are presented and
then conclusions drawn from the design methodology presented.

ˆ Chapter 5: In this penultimate chapter, the formulations of the optimisation problems

are presented together with the optimisation results generated from the implementation
of the optimisation frameworks discussed in Chapter 4. The problems are presented
as cases representing the xed tilt design scenarios as well as the single-axis tracking
scenarios and for each case, three investigations are undertaken to perform the same
optimisation for the three PV module technologies - monocrystalline, polycrystalline and
thin lm. Comparisons are then drawn from the trends in each individual PV technology's
 Stellenbosch University https://scholar.sun.ac.za

CHAPTER 1. INTRODUCTION 9

investigation as well as across all three technologies to nd the best technology and its
corresponding optimal layout.

ˆ Chapter 6: In this nal chapter, all the conclusions drawn from each of the former
chapters are brought together and nal conclusions and relevant recommendations are
made.
 Stellenbosch University https://scholar.sun.ac.za

Chapter 2
Theoretical Background

2.1 Introduction
PV systems can be generally classied as grid-connected or stand-alone systems based on how
the system is connected to electrical loads and, in addition to that, the conguration of their
components and their operational requirements [32] and [33]. The main dierences are found
in:

ˆ whether they provide either alternating current (AC) or direct current (DC) power to the
loads,

ˆ whether they are connected to energy storage media such as batteries or not and

ˆ whether they are connected to the grid or operate independent of it.

2.1.1 Stand-alone PV Systems

This type of PV system operates independent of the electric utility grid and that is why it
is often referred to as the o-grid system. Stand-alone systems come in two types: one that
makes use of battery storage and one that is directly coupled to an electrical load.

ˆ Directly coupled systems are usually very simple in nature. The PV module is directly
connected to a load that matches the current and voltage output of the module. This
type of stand-alone system is typically used on devices and appliances that need to run
during the day since it only produces electricity when the sun is up. Common uses of
such systems are ventilation and water-pumping.

ˆ Stand-alone systems with battery technology can be connected directly to the load
and battery storage to ensure that excess energy is stored and made available for use in
times when the sun is not shining. This type of system is expected to supply all the energy
needed by the load and other sources of energy are sometimes used to supplement the
solar energy such as fuel generators and wind or hydro turbines to form what is referred
to as hybrid PV systems [32], [34] and [33].

2.1.2 Grid-connected PV Systems

This type of system is also referred to as grid-tied, inertied or utility-interactive system. As
the name suggests, grid-connected systems are meant to operate in parallel with and tied
with the electric utility grid. The primary component in this type of system is the DC-to-AC
inverter. The inverter converts the DC power produced by the PV modules into AC power

10
 Stellenbosch University https://scholar.sun.ac.za

CHAPTER 2. THEORETICAL BACKGROUND 11

that is synchronous with the voltage and power quality requirements of the utility grid. The
parameters to be satised to ensure safe and secure functioning of the electric power system
are supplied in the particular country or regions's grid code [32], [34] and [33]. A grid code
is a document that species the technical requirements that a facility, such as an electricity
generating plant, has to meet in order to connect to a public electric network [35].

A variation of this grid-tied system is one that comes with battery backup in order to power
selected backup loads when the grid is down. This type of system usually comes with a
bi-directional interface between the AC output circuits and the utility network. This interface
enables excess power to be fed back into the grid when the AC power output is more than the
load demand and enables the grid to supplement the PV system output when the load demand
is greater. As can be noted from the extra features, this type of grid-tied system is the more
complex and less cost-eective [36].

2.2 Solar Resource and Site Selection

2.2.1 Chapter Overview

In this section the focus will focus on the basics of PV system design, particularly that of the
grid-connected battery-less type. The layout of the sections will be as follows: in Section 2.2 the
solar resource and site selection process are described, in Section 2.3 PV system components
are discussed, in Section 2.4 system performance parameters are discussed, in Section 2.5 PV
system design tools are described and in Section 2.6 plant design fundamentals are discussed.

Solar resource collectively refers to the various solar radiation parameters which together
with other meteorological parameters such as ambient temperature and wind speed can be
used to characterise a site. The solar and weather data along with the sun's position, relative
to the earth, can be used to model system performance and perform energy yield predictions.
PV system performance estimation is a crucial step in determining whether a potential solar
PV project will be feasible and viable in a particular location and it is heavily dependent on
the accuracy of these three elements that are used in the modelling process.

The solar resource and weather conditions of a place are often intermittent and prone to
vary from one year to another so the datasets need to be compiled over a number of years,
typically ten, to create historical site data with a reasonable degree of condence for use in
yield prediction. The position of the sun, however, is predictable and sets of equations and
algorithms are usually used to pin-point the sun's position relative to the earth at any given
time.

2.2.2 The sun's position relative to the earth

The sun's position can be described in terms of two main angles, the altitude (elevation) angle,
β , and the azimuth angle, φs as shown in Figure 2.1. Equations 2.1 and 2.2 allow us to compute
the solar altitude and solar azimuth angles from the latitude, L, day of the year, represented
by the declination angle, δ, and time of the day, represented by the hour angle, H. All the
angles involved can be described as:

ˆ Solar azimuth angle, φS , - the angle between the north-south axis and the direction of
the sun.
 Stellenbosch University https://scholar.sun.ac.za

CHAPTER 2. THEORETICAL BACKGROUND 12

ˆ Solar altitude angle, β , - the angle between the horizontal plane and the geometric
centre of the sun. Its complement is called the zenith angle and it features in several
other solar related calculations.

ˆ Declination angle, δ , - the angle drawn between the plane of the equator and the line
between the geometric centre of the sun and the centre of the earth. This angle changes
throughout the year as a result of the earth's orbit and the sun's apparent up and down
movement through the seasons as shown in Figure 2.2 (see Equation 2.24).

ˆ Hour angle, H , - the angular representation of time where time is quantied as the
number of degrees the earth must rotate to be directly over the line of longitude which
happens at solar noon (see Equation 2.26).

β = sin−1 (cosLcosHcosδ + sinLsinδ) (2.1)

sinHcosδ
φs = sin−1 (2.2)
cosβ

Figure 2.1: Illustration of the sun's altitude and solar azimuth angles β and φs , respectively.
 Stellenbosch University https://scholar.sun.ac.za

CHAPTER 2. THEORETICAL BACKGROUND 13

Figure 2.2: Illustration of the solar declination angle.

2.2.3 Solar radiation parameters

The solar resource of a location is usually dened using three main measured terrestrial solar
radiation (solar radiation within the earth's atmosphere) parameters. These are:

ˆ Global Horizontal Irradiance (GHI)

ˆ Direct Normal Irradiance (DNI)

ˆ Diuse Horizontal Irradiance (DHI)

2.2.3.1 Global Horizontal Irradiance (GHI)

Global Horizontal Irradiance (GHI) refers to the total amount of solar power that is received
by a unit area of a horizontal surface. GHI is made up of diuse and direct beam components
of solar irradiation which are explained in the next subsections. Figure 2.3 shows some of these
solar irradiation characteristics.

2.2.3.2 Direct Normal Irradiance (DNI)

The Direct Normal Irradiance (DNI) is also referred to as the beam component because it is
the beam power component received on a unit area of a surface normal to the direction of the
sun. Solar PV tracking systems and concentrating solar power technologies mainly seek out to
maximise their collectors' reception of this component of the solar irradiation. If (measured)
values of the DHI and GHI are available, equation 2.3 can be used to estimate the values of
the DNI.
G = Gdif f + Gdir cos(θZ ), (2.3)

where:
G is the GHI,
 Stellenbosch University https://scholar.sun.ac.za

CHAPTER 2. THEORETICAL BACKGROUND 14

Figure 2.3: Terrestrial solar irradiation characteristics

Gdif f is the DHI,

Gdir is the DNI and
θZ is the zenith angle of the sun.

Fitted models known as decomposition models can also be used to approximate the values
of the DNI of a particular area by approximating the relationship between the clearness index,
kt , and the diuse fraction, kd . kt is the ratio of the GHI to the extraterrestrial irradiance,
Gextra :
G
kt = , (2.4)
Gextra
while kd is given as the ratio between the DHI and the GHI:

Gdif f
kd = . (2.5)
G
The relationship between kt and kd is approximated dierently by dierent decomposition
models, sometimes using additional parameters such as the variability of the clearness index,
∆kt , the solar elevation angle, αs and the dew point temperature, Tdp . Table 2.1 shows some
of the common decomposition models and the input variables they make use of.

2.2.3.3 Diuse Horizontal Irradiance (DHI)

Diuse Horizontal Irradiance (DHI) refers to the solar power received per unit area of a
horizontal surface from the irradiation that is scattered by particles in the atmosphere and
does not arrive on the surface on a direct path from the sun but from several other directions.

To maximise on energy production solar panels are often oriented to face the equator and
tilted at an angle, θT , usually set to equal the latitude, to try and maximise the normal
irradiance incident on the solar panel's surface. The orientation angle of the PV module is
 Stellenbosch University https://scholar.sun.ac.za

CHAPTER 2. THEORETICAL BACKGROUND 15

Table 2.1: Decomposition model

Model Input variables

Orgill and Holland (1977)[37] kt
Erbs (1982) [38] kt
DISC (1987) [39] kt , αs
Reindl 1 & 2 (1990) [40] kt , αs
Louche (1991) [41] kt
DIRINT (1992) [42] ∆kt , αs , Tdp

referred to as the collector/surface azimuth angle, φc , and is dened as the angle between
the horizon and the direction normal to the surface of the PV module. There are dierent
conventions for dening surface azimuth but the more commonly used one is the north-clockwise
convention. According to this convention north is the origin and the positive direction is
◦ ◦ ◦
clockwise therefore making east 90 , south 180 and west, 270 . The other convention is just
the reverse and takes south as the origin. To quantify the amount of irradiance incident on a
tilted module surface, irradiance data needs to be transposed into the plane of the array (POA).

2.2.3.4 Global Plane of Array (POA) Irradiance

Otherwise known as the Global Tilted Irradiance (GTI), it refers to the total solar power
received per unit area of a tilted surface. Ideally PV system designers aim for designs that
maximise the average, long-term Global POA Irradiance, GP OA , value and due to the fact that
the sun's position relative to the module position is ever-changing, a crucial step in estimating
PV plant performance is determining GP OA as a function of time, if reliable measured GP OA
data is not available. The GP OA depends on various factors such as:

ˆ the position of the sun

ˆ the mounting structure and orientation of the PV array

ˆ the direct beam and diuse irradiance components

ˆ the ground reectance (albedo)

ˆ the presence of obstructions that could cause shading (both near and far),

and is calculated mathematically as:

GP OA = Gbeam + Gground,ref + GSky,dif f (2.6)

Gbeam = Gdir ∗ cos(θAOI ) (2.7)

1 − cos(θT )
Gground,ref = G ∗ αλ (2.8)
2
where:
Gbeam is the direct beam component of irradiance in the plane of the array,
Gground,ref is the ground-reected component of irradiance in the plane of the array,
GSky,dif f is the sky diuse component of irradiance in the plane of the array,
θAOI is the angle of incidence of the irradiance on the tilted PV module surface,
αλ is the ground albedo and
θT is the tilt angle of the PV module.
 Stellenbosch University https://scholar.sun.ac.za

CHAPTER 2. THEORETICAL BACKGROUND 16

Calculation of incident irradiance on a tilted surface from horizontal irradiance data is

facilitated by transposition models, and each model diers from the others due to the assumptions
it is based on and the input variables it depends on. Table 2.2 shows some of the common
transposition models.

Table 2.2: Transposition models

Model Input variables

Isotropic [43] DHI, θT
Klucher [44] DHI, GHI, θT , θAZ , θZ , αs
Reindl [40] DHI, GHI, DNI, θT , θAZ ,
θZ , αs
Perez [45] DHI, DNI, Gextra , θT , θAOI ,
airmass

Hay-Davies [46] DHI, DNI, Gextra , θT , θAZ ,

θZ , αs

2.2.4 Solar radiation data sources

Solar resource data is usually obtained either by collecting actual ground measurements at
the location of interest or from satellite data. Ground measurements are obtained by using
equipment such as pyranometers and silicon sensors to collect hourly and sometimes sub-hourly
data onsite. To obtain long-term irradiance datasets the measured data is sometimes put
through the process of interpolation since extensive historical data from ground-based stations is
seldom available [47]. The other challenges presented by data obtained from ground measurements
include variable standards of calibration, maintenance and measurement periods and uncertainties
in the interpolated data.

Satellite data sources, on the other hand, usually have a wider geographical and historical
coverage. They are also not susceptible to most of the uncertainties and challenges that
ground-measured data sources are prone to. A list of some of the more globally applicable
sources of solar resource data is given in Table 2.3 [47], [48] and [49].
 Stellenbosch University https://scholar.sun.ac.za

CHAPTER 2. THEORETICAL BACKGROUND 17

Table 2.3: Sources of solar resource datasets

Data Type Coverage Access

Source
◦
SolarGIS satellite Regions between 60 North and paid
◦
data 50 South latitude angles; from
1994, 1999 or 2006 (depending on
region) to date

SoDa satellite Mainly Europe, Africa and the paid

Helioclim data Middle East; from February 2004
to date

Meteonorm ground global; from 1981- 2010 for most paid

data regions

3Tier satellite global; from 1997, 1998 or 1999, paid

data depending on region, to date

PVGIS depends on global; between 1981 and 2011, free

region and depending on region
time frame

NASA-SSE satellite global; from 1983 till 2005 free

data

2.2.5 Site selection

The process of selecting a site for a PV plant has an inuence on the PV project's nancial and
technical viability. Although there are no hard and fast rules when it comes to site selection,
especially considering the fact that viable projects have been developed in areas with steep
slopes and waste disposal sites [47], there are several factors worth considering. These factors
include:

ˆ Climate and solar resource - these have a direct bearing on the irradiation that the solar
PV modules receive and consequently, the plant energy output.

ˆ Topography and ground conditions - these determine how factors such as far-shading,
mounting system selection and the ground cover reectivity aect the PV module performances.

ˆ Land, especially the area available and policies for land-use in the region - this determines
how modules are laid out on the eld and in turn, their performance.

ˆ Availability and accessibility of a grid connection from the site - the grid needs to be
within reach such that energy losses are not incurred due to the use of longer cables
during transmission.

ˆ Environmental and social aspects - the eect of setting up a PV plant in the region needs
to be evaluated for its potential positive and negative impacts on the environment and
the area's inhabitants.

ˆ Costs and nancial incentives - this helps determine whether a PV project will be nancially
viable or not.
 Stellenbosch University https://scholar.sun.ac.za

CHAPTER 2. THEORETICAL BACKGROUND 18

2.3 PV System Components

PV systems are made up of dierent components and each component serves a specic purpose.
Some of the major components of Photovoltaic Grid-Connected Systems (PVGCS) are:

ˆ PV modules

ˆ Inverters

ˆ Mounting structures and other balance of system (BOS) components such as wiring,
overcurrent and over earth fault covering, surge protectors, instrumentation and control
systems (Scada or DCS) and disconnect devices.

2.3.1 PV modules
2.3.1.1 Photovoltaic eect
PV modules are responsible for the conversion of solar irradiation into electricity by means
of the photovoltaic eect. The photovoltaic eect is facilitated by semiconductor materials
such as silicon (Si), gallium arsenide (GaAs) and copper sulphate (Cu2S). Semiconductors are
specialised materials with conductivity that depends on the temporary availability of energy to
activate electrons in the crystal lattice. When photons are absorbed on the surface of a solar
cell, they raise the electrons in the lattice structure of the cell to a state of higher energy that
enables the the electrons move freely in the semiconductor. When electrons move from their
original positions, they create positive charges known as holes and this movement is facilitated
by two thin regions made by dopants. Dopants are specially added impurities that alter the
electrical properties of the semiconductors and help create regions of excess electrons, known
as type-n regions, and regions with an excess of positive holes, known as type-p regions. Where
these two regions meet they create what is known as a p-n junction which produces an internal
electric eld. When the photons create free electrons and holes close to the p-n junction, the
electric eld makes the electrons move towards the n-side and the holes towards the p-side
thereby creating an electromotive force (EMF) between the n and p regions. Wires are then
used to connect the n and p sides to supply electric current to an external circuit.

2.3.1.2 PV module technology

There are curently several types of PV modules available on the market and they all dier
in terms of their electrical and optical properties. These dierences cause certain types of
modules to perform better in certain environments or scenarios than others. PV modules can
be generally classied into crystalline silicone modules and thin lm modules with other minor
subdivisions as shown in Figure 2.4.

ˆ Monocrystalline silicon (Mono-c-Si) solar cells

Also referred to as single-crystalline solar cells, these solar cells are made from cylindrical
silicon ingots. Using the Czochralski process, four sides are cut out of the ingots to make
silicon wafers and this is what gives Mono-c-Si PV modules their characteristic look.
The solar cells of Mono-c-Si modules exhibit a dark uniform and evenly coloured look
as shown in Figure 2.5a because they are made from a single seed crystal of high-purity
silicon. Mono-c-Si solar panels have the highest eciency, typically ranging from 15% to
20% [50] and [51], due to their high-grade silicon content and tend to last the longest,with
a typical warranty of 25 years, compared to other types of modules. One of their major
downsides is the fact that they are the more expensive than polycrystalline and thin lm
modules.
 Stellenbosch University https://scholar.sun.ac.za

CHAPTER 2. THEORETICAL BACKGROUND 19

Figure 2.4: PV Cell Technologies

ˆ Multi-crystalline silicon (multi-c-Si) solar cells

This cell technology is also referred to as Polycrystalline silicon (p-Si). Multi-c-Si cells are
produced when raw silicon is melted and poured into a square mould and then cooled and
cut into rectangular wafers. This process creates variable shades in multi-c-Si cells due
to the visible edges and grain boundaries as shown in Figures 2.5c and 2.5d. Multi-c-Si
modules do not cost as much as Mono-c-Si modules, because their method of production
is cheaper than the Czochralski process, but the modules are just as durable so they
typically also carry a warranty of about 25 years as well. Their eciency is slightly lower
than Mono-c-Si module eciency, typically between 13% and 16%, but higher than that
of thin lm technologies [50] and [51].

ˆ Thin lm solar technology

Thin lm modules are produced by depositing thin layers of PV material on a substrate
like glass, metal or plastic. The types of thin lm photovoltaic (TFPV) solar cells
commonly used commercially include cadmium telluride (CdTe), copper indium gallium
selenide (CIGS) and amorphous thin-lm silicon (a-Si), which is silicon-based. TFPV
modules often have lower eciencies [50] and [51], compared to their crystalline counterparts
but are often easier to mass-produce and are generally less sensitive to high temperatures
and shading. This type of PV modules is, however,generally considered to be susceptible
to slightly faster degradation and has a shorter lifetime compared to the crystalline silicon
 Stellenbosch University https://scholar.sun.ac.za

CHAPTER 2. THEORETICAL BACKGROUND 20

modules [51]. That explains why thin lm modules usually come with shorter warranties.
Figure 2.5b shows a Thin Film solar cell.

(a) Monocrystalline solar cell (b) Thin Film solar cell

(c) Multicrystalline solar cell (d) Grain boundaries in multi-c-Si cells

Figure 2.5: Types of PV modules

With several module manufacturers operating in the solar PV industry today, it is essential
to know what else to look out for when it comes to PV module selection. One of the most
important criteria used for module selection is industry acceptance and towards this end,
PV system designers usually refer to the tier system. While tiers do not directly guarantee
module quality, they are considerably good at indicating it. Tier rankings may also vary in
certain criteria, depending on the entity that compiles the list but they are quite similar in
several respects. In general, most PV system designers and investors prefer Tier 1 module
manufactures, for bankability reasons. Tier rankings are, however, not ordinarily available to
the public but organisations such as Bloomberg New Energy Finance (BNEF) are working
towards bridging that gap. BNEF also has a solar insight service [52] that provides more
information about the tier system.
 Stellenbosch University https://scholar.sun.ac.za

CHAPTER 2. THEORETICAL BACKGROUND 21

2.3.2 Inverters
Solar power inverters are responsible for the conversion of the DC electricity produced by solar
PV modules into AC electricity. This is so that the electricity can be used by other appliances
on-site or fed into the grid and transmitted to other end-users. Inverters also go beyond just
that one application and are often typically useful in:

ˆ performing maximum power point tracking (MPPT) to maximise the energy harvest from
the PV system,

ˆ connecting the PV system to the grid in compliance with the grid-code and

ˆ providing basic monitoring and ancillary services.

There are several topologies of inverters and each one diers in architecture and capacity and
in turn exhibits strengths and weaknesses that make it more suitable for certain PV systems
than others are.

ˆ String inverters
The name of these inverters comes from the fact that they are connected to a string or
a couple of strings of PV modules. This type of setting ensures that there is a lower
risk to the entire system because if one inverter fails or has to be disconnected during
maintenance routines, the system can still supply power from the remaining functional
string inverters in the topology. The set up of a system with string inverters also means
that dierent string designs can be supported which is ideal for sites where multiple
orientations and variable numbers of modules per string are warranted. This type of
inverters is mostly used for residential or commercial PV plants because they often
have capacities ranging from 1 kW to 8 kW. String inverters have advantages such as
lower balance of system (BOS) and ongoing maintenance costs and simpler design due to
modularity. The downsides of this type of inverter include higher DC watt unit costs and
also an increased number of inverter connections and AC cabling.

ˆ Central inverters
Central inverters are similar to string inverters, but they are much bigger and therefore
capable of supporting more strings. Unlike in string inverter setups where strings run
directly to the inverter, in central inverter setups strings are rst connected together
in a common combiner box that then runs the DC power to the central inverter input
side. Central inverters generally require fewer component connections and lower DC
watt unit costs, but they tend to cost more to install and have a higher DC wiring and
combiner costs. They also tend to require larger spaces for inverter pads. Capacities of
such inverters usually range between 50 kW and 1000 kW and they are, therefore, mostly
suitable for commercial and utility scale PV plants.

ˆ Microinverters
Microinverters, as the name suggests, are a much smaller version of solar inverters that
convert DC to AC power right at the solar panel. Microinverters are part of the module
level power electronics (MLPE) family, together with power optimisers. They monitor
the performance of each PV module rather than a whole string or groups of strings. They
are increasingly becoming a popular choice of inverters for residential and commercial
PV systems, but they are usually regarded as less practical for utility scale PV systems
due to their higher costs per unit watt and the vast number of microinverters that will
be needed for such large PV plants. Microinverters are already being integrated into PV
 Stellenbosch University https://scholar.sun.ac.za

CHAPTER 2. THEORETICAL BACKGROUND 22

modules by some manufacturers to make what are known as AC modules. This makes
the modules relatively more expensive but ensures that the installation is cheaper and
relatively easier.

2.3.3 Mounting Systems

How the modules of a system are installed also has a bearing on the performance of the system as
well as its cost. Generally, ground-mounted systems can be classied into three main categories,
namely:

ˆ Fixed tilt - where the subarray is , mounted at xed tilt, θT , and azimuth, φc , angles.
As a rule of thumb, most designers use the latitude angle of the site as the optimal tilt
angle, and the azimuth as the direction facing directly towards the equator; north for
places in the southern hemisphere and south for places in the northern hemisphere.

ˆ Single-axis - where the subarray is mounted at a xed tilt angle and tracks the movement
of the sun from east in the morning to west in the evening about the north-south axis. How
far the subarray rotates towards the east and the west is determined by the tracker rotation
limit, θR . Some single-axis tracking systems employ a method known as backtracking
which adjusts the orientation of the rows of modules tracking the sun to prevent them
from shading each other.

ˆ Azimuth tracking - where the subarray is at a xed tilt angle but rotates in a horizontal
plane, tracking the daily movement of the sun and disregarding the azimuth angle.

ˆ Dual-axis - where the subarray rotates from east to west following the sun's daily
movement just like a single-axis tracking system and also tracks the sun's seasonal movement
throughout the year in the north to south direction.

Tracking systems use specialised algorithms that can pinpoint the sun's position with respect
to the site at any given time of the day and allow the subarrays to be oriented such that they
directly face in that particular direction to maximise solar irradiance incidence on the modules'
surfaces. Figure 2.6 shows the diagramatic representation of the above-mentioned mounting
methods.

2.4 PV System Performance Parameters

Several metrics used to evaluate a plant's operation to give an indication of a plant's nancial
and technical performance will be discussed in this section. These plant performance indicators
are used across the entire energy production chain in the PV system to identify opportunities
to improve its performance and reliability and consequently its yield and protability.
 Stellenbosch University https://scholar.sun.ac.za

CHAPTER 2. THEORETICAL BACKGROUND 23

(a) Fixed Tilt (b) Single-axis Tracking

(c) Azimuth Tracking (d) Dual-axis Tracking

Figure 2.6: Mounting methods as illustrated in the System Advisor Model documentation[5].

2.4.1 Energy production

The electrical energy output of a PV system is one of the most common ways by which the
performance of a plant can be measured. Energy output is usually set as a design goal and
the plant is regarded as performing really well if the production is greater or equal to the set
threshold or at least reasonably close to it. Factors that inuence energy production include the
amount of irradiance incident on the surface of the PV module, the module temperature, the
eciency of the PV modules and the inverters and the losses in the generation and transmission
of the energy. Theoretically, the energy production is calculated as the product of the total
system power output (usually given as kilowatts) and the time period of in question (usually
measured in hours):

T otal Energy (kW h) = T otal P ower(kW ) ∗ T ime(h). (2.9)

2.4.2 Specic Yield

Specic Yield is a way of measuring plant performance by stacking its energy production
against its peak installed capacity. It is often used to compare the operating results of PV
systems designed with dierent technologies or layouts on a standardised scale. Specic Yield
is calculated as:

T otal Energy (kW h)

Specif ic Y ield (kW h/kW p) = . (2.10)
P eak Capacity (kW p)

2.4.3 Performance Ratio

The performance ratio (PR) is also known as the quality factor (Q) because it is a measure of
the quality of a PV system. Another way of looking at the PR is as a measure of a plant's
energy eciency and reliability. PR illustrates the the relationship between the actual plant
yield and the theoretical energy output and it is usually expressed as a percentage as shown in
 Stellenbosch University https://scholar.sun.ac.za

CHAPTER 2. THEORETICAL BACKGROUND 24

Equation 2.11:

T otal P lant Output (kW h)

P erf ormance Ratio (%) = ∗ 100, (2.11)
Calculated N ominal Output (kW h)

where the nominal energy plant output is calculated using the total annual incident solar
irradiation at the generator surface of the PV plant, Annual POA Irrad., and the relative
eciency of the PV plant modules, ηM odule , as:

N ominal Output (kW h) = Annual P OA Irrad. (kW h) ∗ ηM odule . (2.12)

2.4.4 Capacity Factor

Capacity factor (CF) is a measure of the amount of energy produced by a PV plant compared to
its maximum possible output. It is calculated as a fraction or percentage, generally by dividing
the total energy the plant produces under normal operating conditions in a given period of
time by the total amount of energy the plant would have produced if it ran at its full capacity
during that period as given in Equation 2.13:

T otal Energy P roduced (kW h)

Capacity F actor = , (2.13)
M aximum P ossible Energy Output (kW h)

2.4.5 Levelised Cost Of Energy (LCOE)

The levelised cost of energy (LCOE) is the cost, per kilowatt-hour (kWh), of an operating
plant over an assumed nancial life. LCOE indicates a measure of the overall competitiveness
of energy generating technology as it takes into account all the costs incurred during the lifetime
of the generating technology and the units of energy produced during that same time frame.
It is a convenient way to also compare projects from dierent technologies, lifetimes and/or
capacities and this is particularly useful in assessing grid competitiveness and proximity to grid
parity (which occurs when an alternative source of energy can generate power at a LCOE that
is less than or equal to the price of purchasing power from the electricity grid). The simplied
formula for calculating LCOE is:

PN Cn
n=0
(1 + d)n
LCOE = P (2.14)
N Qn
n=1
(1 + d)n

where:
N is the analysis period,
Cn are the annual project costs in year n,
Qn is the amount of electricity generated by the system in year n, in kWh and
d is the discount rate which can be dnominal or dreal for nominal (including ination) or real
(without ination) discount rates respectively.

The annual cost summation starts from n = 0 to include the investment costs in the
calculations, while the total energy production summation only starts from n = 1, because
that is when the system starts producing energy.
 Stellenbosch University https://scholar.sun.ac.za

CHAPTER 2. THEORETICAL BACKGROUND 25

2.4.6 Losses
Loss analysis is an integral part of measuring the performance of a PV plant. Poor operating
conditions on a PV plant and the non-ideal nature of PV system components are usually the
chief causes of losses in a PV system. These losses are explained in more detail in the following
subsections.

2.4.6.1 Shading losses

Shading losses are usually a result of obstructions in the vicinity of a PV system casting shadows
on the solar PV modules or the modules in adjacent rows casting shadows on each other some
time during the day. In the case of most utility scale PV plants, trees, buildings and other
nearby obstructions can be avoided through a due diligence process when selecting the site. The
biggest shading challenge is, therefore, due to the inter-row shading of PV modules, especially
on space-constrained sites. To minimise these self-shading losses, PV system designers rely on
nding the optimum tilt angle and spacing of rows of PV modules. Equation 2.27 in subsection
2.6.1.4 deals with the calculation of the ideal row spacing such that the derate factor due to
shading does not go beyond 2.5% in a PV system, but there are several other methods and
algorithms used by system designers to quantify and mitigate shading losses [53] and [54].

2.4.6.2 Incidence angle modier (IAM) losses

The incidence angle modier losses are characterised by a decrease in the irradiance reaching the
PV module surface with respect to the irradiance under normal incidence due to reectance
caused by incidence angles that are not normal to the plane of the module surface. The
reectance phenomenon is consistent with the Fresnel's Laws of transmission and reections at
an interface of two transparent materials of dierent refraction indices, n1 and n2 . Although
in reality there are multiple and sometimes complex calculations that need to be taken into
account, in practice the irradiance eectively reaching the PV cell surface is often closely
estimated using the ASHRAE model which depends on only one PV module parameter, bo , as:

1
LIAM = 1 − bo ∗ ( − 1), (2.15)
cosθAOI
where θAOI is the incidence angle on the plane.

2.4.6.3 Soiling losses

When dirt accumulates on the surface of PV modules it can adversely aect the amount of
irradiance that reaches the PV cells for conversion into electrical energy. This accumulation of
dust, dirt and other contaminants is known as soiling and several studies are being currently
undertaken to nd better methods of quantifying the losses that arise from it. An alternative
way of estimating soiling losses without using expensive equipment or intricate methods is
by using precipitation levels. Monthly precipitation levels, in millimetres, can be used to
approximate soiling losses as shown in Table 2.4 derived from a study by ARUP which involved
module performance comparisons for locations in South Africa in [53].

2.4.6.4 PV losses due to irradiance levels

Losses due to irradiance are a result of the reduction of the eciency of a PV module in
conditions of low irradiance. This is because a PV module's eciency increases at high light
intensity and decreases at low light intensities relative to the PV modules' rated standard
2
irradiance conditions of 1000 W/m . Figure 2.7 shows the variation of module eciency with
irradiance at dierent temperature levels.
 Stellenbosch University https://scholar.sun.ac.za

CHAPTER 2. THEORETICAL BACKGROUND 26

Monthly precipitation (mm) Soiling loss

0 - 20 3%

20 - 50 2%

> 50 1%

Table 2.4: Soiling Loss Scale

PV module behaviour
16
Cell temp = 10 ⁰C Cell temp = 25 ⁰C Cell temp = 40 ⁰C Cell temp = 55 ⁰C Cell temp = 70 ⁰C

15

14

13
Efficiency at Pmax (%)

12

11

10

6
50 100 150 200 250 300 350 400 450 500 550 600 650 700 750 800 850 900 950 1000

Incident Global Irradiance (W/m2)

Figure 2.7: Variation of module eciency with irradiance at dierent cell temperature levels

2.4.6.5 Thermal losses

◦
Standard rated temperature conditions for PV modules are pegged at 25 C, but in reality, PV
modules operate mostly in temperature conditions higher than that. When modules operate
at higher temperatures their eciency tends to decrease and the losses associated with that
are referred to as thermal losses or PV losses due to temperature levels. Thermal losses are
calculated using an energy balance between the cell temperature readings from modules under
operating conditions and the prevailing ambient temperature. Figure 2.8 shows the variation
of module eciency with irradiance at dierent temperature levels.

U ∗ (Tcell − Tamb ) = α ∗ Ginc ∗ (1 − ηmodule ) (2.16)

where:
U is the thermal loss factor dependent on the mounting mode of the modules and wind velocity,
Tcell is the temperature of the PV cells,
Tamb is the ambient temperature,
α is the absorption coecient of solar irradiation,
GP OA is the global irradiance incident in the plane of the array, and
ηmodule is the eciency of the module.
 Stellenbosch University https://scholar.sun.ac.za

CHAPTER 2. THEORETICAL BACKGROUND 27

Module behaviour in response to temperature

16
Incident Irrad. = 1000 W/m2 Incident Irrad. = 800 W/m2 Incident Irrad. = 600 W/m2 Incident Irrad. = 400 W/m2 Incident Irrad. = 200 W/m2

15

14

13
Efficiency at Pmax (%)

12

11

10

8
2 4 6 8 10 12 14 16 18 20 22 24 26 28 30 32 34 36 38 40 42 44 46 48 50 52 54 56 58 60 62 64 66 68 70 72 74 76 78 80

Temperature (oC)

Figure 2.8: Variation of module eciency with temperature at dierent irradiance levels

2.4.6.6 Mismatch losses

Mismatch losses are mainly due to varying current drives in modules connected in a string.
Electrical properties such as the current-voltage (IV) characteristics of PV modules, even ones
of the same PV module model, are never exactly the same. In the case of a series connection,
the module with the lowest current drives the whole string and the corresponding power loss is
what is referred to as mismatch loss or circuit mismatch (CM). Mismatch losses are not that
easy to quantify so one of the ways that are used to estimate them is statistical sampling,
especially when being evaluated by simulation software [55]. A more common way, however, is
to derive the mismatch losses by calculating the dierence between the expected total power of
a string or array and the actual power realised by that particular string or array as follows:
PN
Pm,max − Pa,max
n=1
CM = PN (2.17)
n=1 Pm,max
where:
Pm,max is the maximum output power of each module,
Pa,max is the maximum output power of the entire array, and
n is the number of modules in an array.

2.4.6.7 DC Ohmic losses

DC Ohmic losses are mainly caused by the resistance in the cables that interconnect PV
modules, but they are also caused by transition resistance in fuses and terminal connections.
The DC ohmic power losses are, therefore, usually calculated as:

PDC,ohmic = I 2 R, (2.18)

where I is the average DC array current and R is the resistance in the cables.
Dierences in cable lenghts and sizes among parallel strings can lead to voltage drops, I ∗R ,
 Stellenbosch University https://scholar.sun.ac.za

CHAPTER 2. THEORETICAL BACKGROUND 28

thereby contributing to mismatch losses, so it is not unusual for DC ohmic losses and mismatch
losses to be combined and given as one loss value [56].

2.4.6.8 Inverter losses

Losses in inverters are mainly due to ohmic and switching losses in semiconductors. Collectively,
all these losses are referred to as inverter losses. The eciency of an inverter, ηinv , is calculated
as the ratio of the the power available at the inverter output terminal, Pinv,out , to the power at
the inverter input terminal, Pinv,in . Consequently, the inverter losses, Pinv,loss , are given as:

Pinv,out
Pinv,loss = 100 ∗ (1 − ) = 100 ∗ (1 − ηinv ). (2.19)
Pinv,in

2.4.6.9 AC Ohmic and external transformer losses

The distribution and transfer of AC power from the inverter output terminals to the utility
meter usually lead to energy losses in the AC wiring and a reduction in the energy that enters the
transformers. Furthermore, hysteresis and eddy current losses as well as resistive and inductive
losses in the transformers also lead to a further reduction in the energy that gets injected into
the grid from the PV system. Ideally, eld data should be used to quantify the ohmic losses and
the transformer losses separately, but where measured eld data is not available, these losses,
PAC+transf , can be combined and calculated as:

Pgrid
P(AC+transf ) = 100 ∗ (1 − ), (2.20)
Pinverter,out

where Pgrid is the power supplied to the grid.

2.5 PV System Simulation Tools

Performance and nancial modelling are regular and essential processes in the design of PV
systems. Performance models are used to estimate the expected energy output of a given PV
system, while nancial models are used to approximate the costs involved in the process of the
energy production, including the construction and maintenance of the PV plants and in some
cases the proceeds expected after the energy is sold o to other users by the power producers.
This process is usually facilitated by the use of simulation software packages. Typical PV
simulations involve the prediction of plant performance and estimation of cost of energy based
on cost and design parameters specied for the site in question.

The procedure for modelling a PV system and predicting its energy yield, using hourly or
sub-hourly time-steps, in simulation software typically consists of the following steps:

ˆ Sourcing weather data, such as irradiance, wind speed and temperature from sources such
as the ones described in Section 2.2. Some simulation software comes with weather data
libraries, with an option for users to import data for sites the libraries do not cater for.

ˆ If DNI data is missing in the set, decomposition models can be used to approximate it
from the available irradiation datasets.

ˆ Using transposition models to calculate the irradiation incident on the plane of the tilted
array.
 Stellenbosch University https://scholar.sun.ac.za

CHAPTER 2. THEORETICAL BACKGROUND 29

ˆ Designing and sizing the PV system and modelling its performance and energy yield under
the site's operating conditions.

ˆ Applying the relevant losses or loss models.

ˆ Generating the plant performance report using the various system performance parameters.

There are several solar PV simulation software packages that one can choose from; some
available free of charge and others at a cost. The dierent software packages also come with
features that make them more suitable for certain tasks than others [57]. It is important for
PV system designers to select the right kind of design tool that serves their desired purpose, on
a platform suitable for them and at a cost that is within their budget. We came across several
design tools and they could basically be classied under ve categories:

ˆ Free - examples include SAM, PVLib, RETScreen, HOMER Legacy v.2.68, SKELION
and HYBRID2

ˆ Paid - examples include PVSyst, PVComplete, PVscout 2.0 Premium, SolarPro, Plan4Solar
PV, PV F-Chart PV Sol Premium, Homer, PolySun, EasySolar, HELIOS 3D Solarparkplanung
and Solarmapper

ˆ Online - examples include SISIFO, PV-GIS, PVWATTS, PV*SOL, DIAFEM and EASY-PV
for the free applications and Helioscope, SOLARPlus, Focus Solar, SolarDesignTool,
Aurorasolar, Solar Analytics, SolarGis PV Planner, PolySun, EasySolar and Solarmodel
for the paid applications.

ˆ Tools from inverter manufacturers - examples include ABB (Power One), Mastervolt,
Kaco, Fronius, Ingecon Sun Planner, SMA Sunny Design, Samil Power Design, SolarInfo
Design Software, Goodwe EZDesigner and Satcon Congurator.

Investigation of three of these software packages for the purposes of our design optimisation
will be discussed in the the following chapter.

2.6 Plant Design Fundamentals

2.6.1 Layout Design and Optimisation

This stage of the design process deals with the general layout of the PV plant so that the
PV system performance is at its best, operations and maintenance (O&M) procedures are not
inconvenienced and the design and full operation of the plant is all done at a reasonable cost.
Typically, the objectives for the layout design include:

ˆ selecting the tilt, orientation and conguration that optimises energy yield,

ˆ designing for a row-spacing (pitch) that minimises self-shading losses, while at the same
time maximising utilisation of the available land area,

ˆ creating access routes and leaving enough room between rows to allow for operations and
maintenance (O & M) procedures, and

ˆ laying out the PV modules to minimise cable lengths so that electrical losses are minimised.

Our focus on the layout design covers the tilt, azimuth, conguration and spacing of the rows
of PV modules to satisfy some of these objectives as given in the following sections.
 Stellenbosch University https://scholar.sun.ac.za

CHAPTER 2. THEORETICAL BACKGROUND 30

2.6.1.1 Conguration
The conguration of PV modules is mostly related to how the modules experience electrical
shading eects. There are typically two types of congurations of PV modules, portrait and
landscape, and they are as shown in Figure 2.9. Most PV modules currently on the market
come with bypass diodes to mitigate shading losses and, depending on how these diodes are
connected on the modules, their bypassing mechanism may or may not be fully utilised in
the one conguration compared to the other. Most bypass diodes are connected along the
length of the module so landscape conguration will mean smaller electrical shading losses in
design scenarios where inter-row spacing is limited so as to minimise the self-shading losses.
The portrait conguration, on the other hand, may be considered in cases where east and
west horizon shading losses are prevalent. Several rooftop installers also recommend portrait
congurations mainly to avoid running mounting rails parallel to the roof trusses. Dierent
PV module technologies have also been known to respond dierently to partial shading losses,
with some thin-lm module technologies emerging as the least aected, compared to mono
and polycrystalline technologies. All things considered, the decision of whether to mount
your modules using a landscape or portrait conguration is mostly determined by the module
manufacturing company's recommendation as well as site-specic constraints.

(a) Portrait (b) Landscape

Figure 2.9: Module congurations

2.6.1.2 Tilt
The highest energy output from PV arrays is achieved when the solar rays have a 90◦ incident
angle with the module surface [58]. To achieve this, in xed tilt systems, PV modules are
tilted at a certain angle from the horizon, which is typically equal to the latitude angle of
the site. This angle enables the PV modules to capture the highest amount of total annual
 Stellenbosch University https://scholar.sun.ac.za

CHAPTER 2. THEORETICAL BACKGROUND 31

irradiation. This optimal tilt angle based on the rule-of-thumb (ROT) might, however, be
subject to adjustments as a result of site-related constraints. These constraints that need to be
considered include:

ˆ Shading: higher tilt angles often lead to higher amounts of total irradiation captured,
but they can also lead to higher levels of inter-row shading.

ˆ Soiling and snowing: higher tilt angles tend to suer from dirt and snow accumulation
less, compared to lower ones, simply because the sharp incline of the panels makes snow
slide o them easily and rain and dew which wash the panels also gets to do so easily.

ˆ Wind loads: in regions that experience strong winds, lower tilt angles will be preferred
for their ability to withstand the wind speeds without needing too many reinforcements
on the mounting structures.

2.6.1.3 Azimuth
The ideal PV module surface azimuth angle, φc, is based on a rule-of-thumb (ROT) which
stipulates that the optimum orientation of a xed tilt PV system for the highest total energy
production is towards the equator. This means that for a site in the northern hemisphere,
the modules should face the true south direction and in the southern hemisphere, the modules
should face true north. It is not uncommon for designers to have to deviate from this ROT in
response to factors such as limitations in the design space or dierences in design objectives.
Time-of-Day (TOD) energy production preferences are one way in which design objectives can
have such an inuence on the selection of the azimuth angle. In such a design scenario, the
objective would, for instance, be to design a system that helps meet energy demands during a
specic time of the day and not necessarily the highest total energy output. Figure 2.10 shows
an example of how the power output prole of a system changes in response to the change in
the surface azimuth angle.

2.6.1.4 Row space (Pitch)

Pitch is essential to ensure that there is enough room for access to PV modules for O&M
purposes and that rows of PV modules do not shade each other. The distance left between
rows should also not be too large otherwise it leads to wasteful use of land. Two of the
most commonly used methods for calculating the pitch on a PV plant, using Figure 2.11 as a
complementary reference gure, are:

ˆ Using shadow analysis to predict and avoid shading problems between adjacent rows
during a certain window period [6] in the day. With this method, the winter solstice day
is selected as the worst case scenario to work from to calculate the ideal row-spacing, d:

d = Ls cos Φs (2.21)
γ
Ls = (2.22)
tan β
360
δ = 23.45[ ∗ (n − 81)] (2.23)
365
sin β = cos(L) cos(δ) cos(H) + sin(L) sin(δ) (2.24)

cos(δ) sin(H)
Φs = (2.25)
cos(β)
H = 15 ∗ Nh (2.26)
 Stellenbosch University https://scholar.sun.ac.za

CHAPTER 2. THEORETICAL BACKGROUND 32

80000
Azimuth : True North Azimuth: East Azimuth: West

70000

60000
Array DC Power Output (kW)

50000

40000

30000

20000

10000

Time of day

Figure 2.10: Eect of changing the azimuth on the power output prole of a PV system (in the
southern hemisphere)

where
β is the altitude angle,
γ is the height of the uppermost tip of the PV module measured vertically from the
ground,
Ls is the length of the shadow,
L is the latitude of the location,
δ solar declination angle,
H is the hour, in solar time,
Φs is the azimuth angle of the shadow,
Nh is the number of hours before solar noon and
n is the day number from the very rst day of the year.

The winter solstice is chosen because it is the day that experiences the longest north-south
shadows in a year due to the sun's position with respect to the site's location.

ˆ Using the ground cover ratio (GCR) to calculate the row space, d, so as to minimise
shading and improve the shading derate factor. The goal is to nd d such that the
resulting GCR leads to a desired derate factor using the equations:

CollectorArea, Ac w∗l
GCR = = , (2.27)
T otalGroundArea, Atot (w ∗ cosθ + d) ∗ l

where w and l are module dimensions.

 Stellenbosch University https://scholar.sun.ac.za

CHAPTER 2. THEORETICAL BACKGROUND 33

Figure 2.11: Reference gure for shading loss calculations

The approximate GCR can be selected using the corresponding mounting method and
desired shading derate factor according to the plot shown in Figure 2.12. PVWATTS
suggests that it is common industry practice to optimise land use on PV plants by using
a GCR value that corresponds to a shading derate factor of 0.975 which translates to
2.5% shading losses [6].

Shading derate factor for multiple rows PV arrays

1

0.98

0.96
Shading Derate Factor

dual axis tracking

0.94
single axis tracking
fixed tilt, 40 deg
0.92
fixed tilt, 30 deg
fixed tilt, 20 deg

0.9 fixed tilt, 10 deg

0.88

0.86
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8
Ground Cover Ratio (GCR)

Figure 2.12: Shading derate factors for dierent PV module congurations adapted from [6]

2.6.2 System sizing

Sizing the PV system refers to calculation of the number of PV modules and inverters that
will be able to meet a particular energy demand or performance objective. The whole process
goes beyond determining the number of PV modules needed to how many of them ought to be
connected in series and in parallel in order to match the inverter ratings under general operating
conditions. System sizing can be summarised in the following steps and can be executed by the
designer manually or using design tools. This process is usually initiated by rst selecting the
PV module and inverter models and then implementing the three fundamental steps:
 Stellenbosch University https://scholar.sun.ac.za

CHAPTER 2. THEORETICAL BACKGROUND 34

(a) Determine number of PV modules per string: The designer typically chooses an
initial number of PV modules that leads to a maximum power voltage closest to the
average of the inverter's minimum and maximum MPPT voltages. If the string open
circuit voltage exceeds the inverter's maximum DC input voltage, the number of modules
per string is decreased by one and this continues until the open circuit voltage is below
this limit.

(b) Calculate the number of parallel strings: The number of parallel strings of PV
modules to meet the desired array capacity is determined using the formula:

Parray
Nparallel = (2.28)
Nseries ∗ Pmodule
where:
Nparallel is the number of parallel strings of PV modules,
Parray is the total maximum DC power of the array,
Nseries is the number of series modules, and
Pmodule is the module maximum DC power.

(c) Calculate the number of inverters needed: The number of inverters of the chosen
model required to match the preferred DC to AC ratio as;

Nparallel ∗ Nseries ∗ Pmodule

Ninverters = (2.29)
RatioDC:AC ∗ Pmax,AC

where:
Nparallel ∗ Nseries ∗ Pmodule gives the total DC capacity of the system,
Ninverters is the number of inverters,
RatioDC:AC is the desired DC-to-AC ratio, and
Pmax,AC is the inverter maximum AC power.

Figure 2.13 shows the algorithm that summarises the array sizing calculation.

2.7 Chapter Summary

In this chapter, the fundamentals behind solar PV systems were presented. This covered
conditions favourable for solar PV projects, a breakdown of the PV system, system performance
parameters, the basics of PV system design and the role of simulation software in the design and
evaluation process. The purpose of this chapter was to highlight the key elements that make
up the PV system and its design procedure as a form of background for the design optimisation
procedures implemented in the thesis.
 Stellenbosch University https://scholar.sun.ac.za

CHAPTER 2. THEORETICAL BACKGROUND 35

Figure 2.13: System sizing algorithm

 Stellenbosch University https://scholar.sun.ac.za

Chapter 3
Preliminary Design Optimisation

3.1 Introduction
This chapter is the rst step towards formulating an optimisation framework and it involves
establishing the right kinds of design tools and optimisation strategies for the layout design.
The process of optimally designing PV system layouts can go a long way towards improving
system perfomance and lowering the corresponding system losses and in turn reducing system
and operation costs [27]. Layout optimisation, however, goes beyond just picking the ideal
module tilt and azimuth angles. There are several other design parameters that might also
need to be taken into account as well as practical and site-related constraints that inuence
the design decisions for PV systems [47]. Some of the common PV system design constraints
include:

ˆ area constraints

ˆ energy value

ˆ costs

ˆ PV technology and

ˆ incentives and policies.

In this chapter, a bit of background on work that has been done in the PV industry to
investigate plant layout optimisation is given in Section 3.2. This is followed by a brief discussion
of some of the optimisation criteria in Section 3.3 and then case study investigations of the
various strategies in Section 3.4.

3.2 Literature Review

In this section, several sources are highlighted for the purpose of shaping the optimisation
framework formulation process. The studies described in this section not only shed more light
on the work that has been done in this particular eld of study but also help develop a deeper
understanding of all the aspects involved in the design optimisation process.

K. Joyce, in 'Optimizing PV Plant Design to Achieve a Low Levelized Cost of Energy

(LCOE)', discusses some of the optimisation tools and methods that Black & Veatch [59]
implements to optimally design PV systems for a minimised LCOE by means of parametric
analyses [27]. Through an investigative approach to design parameter selection in constrained
design spaces, the study reveals the interdependence of design considerations that aect the

36
 Stellenbosch University https://scholar.sun.ac.za

CHAPTER 3. PRELIMINARY DESIGN OPTIMISATION 37

LCOE, and the importance of understanding their relationships and inuence on system costs
and performances to achieving a minimised LCOE. The signicance of constrained multivariable
optimisation is one of the main lessons that can be drawn from this article.

In the article, 'The 3 Best Ways to Optimize a Commercial PV System', authors P. Grana
and P. Gibbs review and suggest methods of optimisation for commercial PV system designers
[26]. The authors highlight how transformative and crucial some optimal design decisions are,
especially with all the technological advancements being experienced in the solar industry and
how that is aecting the prices of the PV components and in turn the focus of the PV system
design process. They go on to identify the three elements:

(a) module (row) spacing and tilt angle

(b) azimuth angle

(c) inverter and module level power electronics (MLPE) and shade tolerance,

as the best ways to optimise commercial PV systems. The connection they make between the
current trends in the solar industry and the eect the trends have on the focus of PV system
design optimisation is quite insightful and relevant to our work.

In 'A study into the optimisation and calculation of electrical losses in renewable energy
generation,' the author investigates calculation and optimisation of electrical losses in renewable
energy generation technologies [60]. The electrical losses are reviewed and the data collected
in the process is used as an empirical method to design a tool that can quickly and accurately
predict losses in future projects. This then forms a new way of obtaining bankable data to
assess the risks involved in future renewable energy projects. Although the work itself was
not directly linked to our particular focus in its entirety, there were certain elements in the
methodology that t in well with the focus of this chapter's focus that were drawn from it.

B. Goss, in 'Design process optimisation of solar photovoltaic systems,' is embedded in a

rooftop PV system project to analyse and identify opportunities for improvements in the design
procedures [10]. From the study, the author creates a design framework to accurately model
the performance of a PV system with special attention to three key areas, namely:

(a) safety assurance,

(b) design process integration and

(c) nancial optimisation.

The observations he makes particularly in the lack of design integration and nancial optimisation
in most commercially available design tools could be vital to our optimisation framework.

Authors, Louazene et al, Gong etal and Del Vecchio all have work that looked into the layout
optimisation of PV arrays [61], [62] and [63]. The inuence of the tilt angle and the various
tilting methods are investigated, along with some design constraints and design considerations.
These investigations are good starting points for the preliminary design optimisation formulation.
 Stellenbosch University https://scholar.sun.ac.za

CHAPTER 3. PRELIMINARY DESIGN OPTIMISATION 38

3.3 Optimisation Criteria

The criteria of our optimisation processes were broken down into two types, namely, design
parameters and design constraints. Design parameters are the variables that characterise a PV
system design such that when they are varied, the performance of the PV system also vary.
The goal is to nd and select the design variables that enable the system to exhibit its best
performance. Some of the design parameters that can be investigated, either individually or
simultaneously, are:

ˆ tilt angle, θT ,

ˆ azimuth angle, φC ,

ˆ mounting/racking system,

ˆ row spacing,

ˆ DC:AC ratio (also known as the inverter loading ratio, ILR),

ˆ the dierent PV module and inverter technologies.

Design constraints are limitations in the design space or the possible design parameters that
can be selected. The goal is to understand fully what these constraints are and design the
system optimally, while navigating the design constraints eciently. The design constraints
include, as mentioned earlier,:

ˆ area constraints - the amount of space available for the solar installation and its characteristics,
such as its terrain can heavily inuence the process of the layout optimisation process.

ˆ energy value and time of use - the desired time of use for the electrical energy generated
has a direct eect on the design parameter selection and optimisation.

ˆ costs - the cost of PV system components and the cost of energy production play a pivotal
role in determining a plant's nancial viability.

ˆ PV technology - the types of modules and inverters available on the market and their
attributes has an eect on the design parameters and the technical feasibility of a PV
project.

ˆ incentives and policies governing the solar industry in that particular country or region
can also put certain limits on some design parameters.

3.4 Optimisation tools and strategies: Case Studies

3.4.1 Test Case

A site in the Northern Cape Province of South Africa was selected as a benchmark for our
simulations. The Northern Cape is well-known for oering some of the best conditions in the
world for solar PV electrical energy generation [64]. Figure 3.1 shows the renewable energy
power plants in South Africa with solar PV projects mostly concentrated in the Nothern Cape
province. With about 17 solar PV plants now commercially operational in the province [7] and
[65], it was only tting that we use the region for our simulations.
 Stellenbosch University https://scholar.sun.ac.za

CHAPTER 3. PRELIMINARY DESIGN OPTIMISATION 39

Figure 3.1: Renewable energy power plants in South Africa courtesy of [7]

The test-case was an existing 75 MWp solar PV plant made up of 84 sub-systems, each
containing 3720 polycrystalline modules and an inverter. The conguration is such that there
are 24 series modules per string and 155 parallel strings feeding into each of the 84 inverters.
The entire system therefore consists of 312 480 modules over a total module area of 516 217
2
m . Table 3.1 summarises the conguration of the subsystems. To facilitate the PV system
performance simulations, ground measured data from weather stations on the site was used.
The dataset's integrity and reliability had already been checked and approved from previous
work done by T. Mahachi in [56].

Table 3.1: Sub-system conguration

Variable Description
Ground measurement data from site GHI, DHI, ambient temperature,wind speed

◦
Module orientation 0 , facing North

◦
Inclination 30

Type of installation ground mounted

PV module model BYD P6-30, 240 W

Number of modules per subsystem 3720 (24 in series in a total of 155 strings)

Inverter model SMA Sunny Central 800CP

Total number of modules and inverters on site 312 480 modules, 84 inverters

2
Total module area 516 217m

3.4.2 Comparison of the software packages

Dierent software packages have dierent features that make them more suitable for certain
tasks than others [57]. This investigation will focus on three of those packages, namely System
Advisor Model (SAM), Photovoltaic systems (PVSyst) and PVLib. In this subsection SAM,
 Stellenbosch University https://scholar.sun.ac.za

CHAPTER 3. PRELIMINARY DESIGN OPTIMISATION 40

PVSyst and PVLib are described in more detail. For each program, a brief summary of the
design tool will be given, closely followed by subsections in which some of their respective merits
and demerits are discussed. The purpose of the exercise was to determine which one of the
software packages would be the most ideal to use in the optimisation procedures to follow.

3.4.2.1 Summary of the Software Programs

Advisor Model (SAM)
SAM is a software package developed by the National Renewable Energy Laboratory (NREL)
in the United States of America. SAM was originally called the Solar Advisor Model, but
has since expanded its focus to also cover non-solar technologies such as geothermal power,
wind power and biomass power, hence the name change. It is categorised as a nancial and
performance modelling software because it helps assess the operation of a proposed or actual
renewable energy system and ties that in with an evaluation of the costs thereof. A typical
SAM simulation consists of the following basic steps:

1. Selecting, from the interface, the required renewable energy technology and nancing
option.

2. Selecting the appropriate set of simulation and nancing models. This is done by SAM
in the background, based on the options selected in Step 1.

3. Specifying the necessary input variables such as the weather data le, site location and
equipment. The supported weather les are typical meteorological year les (TMY2 or
TMY3) and energy plus weather les (EPW).

4. Running the simulation. For advanced analyses, the user can congure simulations for
sensitivity or optimisation before running the simulations. The user can then view the
simulation results on a variety of graphs and reports generated automatically by SAM
in the user interface's results section. There is also an option to export the results to a
third-party software package for presentation or further analysis.

SAM's features allow it to cater to the needs of a variety of users including project developers,
engineers, manufacturers, academic researchers, technology developers and policy makers and
analysts [5]. Figure 3.2 shows the interface on the SAM application.

Figure 3.2: A screenshot of the SAM application user interface

 Stellenbosch University https://scholar.sun.ac.za

CHAPTER 3. PRELIMINARY DESIGN OPTIMISATION 41

Photovoltaic Systems (PVSyst)

PVSyst is a PV-centric simulation tool that was originally developed at the University of
Geneva but is now a standalone company. The software package focuses on modelling, sizing,
simulating and analysing PV systems. PVSyst does have some sort of nancial modelling in
place, but it is primarily a performance modelling software [55].
A typical simulation in PVSyst consists of the following steps:

1. Dening the project. This is where the user creates the desired project on the user
interface, names it and selects the corresponding geographic location and meteo le to
be used. A number of sites and meteo les are already included in the PVSyst databases
but the user also has the option of importing his own. PVSyst supports several types of
weather les such as TMY2, TMY3 and EPW and les from sources such as Meteonorm,
Photovoltaic Geographical Information System (PVGIS), World Radiation Data Centre
(WRDC), Retscreen, Helioclim and SolarGIS.

2. Creating a system variant. This is where the user creates a calculation version of the
project created in step 1. On the interface, the user gets to dene various input parameters
such as module orientation, system conguration and loss parameters.

3. Running the simulation. The user runs the simulation and generates a variety of graphs
and reports for the analysis of the PV system. PVSyst allows the user to analyse the
results in the program, export them to a dierent program or save the variant for further
evaluation.

PVSyst oers the users extensive reports and breakdowns and valuable insights into the
engineering aspects of design and deployment. This allows PVSyst to cater for a wide range
of users including researchers and architects. Its interface is also multilingual and available
in other languages such as German, French, Spanish and Italian, in addition to English [55].
Figure 3.3 shows the interface on the PVSyst application.

Figure 3.3: A screenshot of the PVSyst application user interface

PVLib
PVLib is a product of the collaborative eorts of a group of PV professionals called the PV
Performance Modelling Collaborative (PVPMC), facilitated by Sandia National Laboratories
 Stellenbosch University https://scholar.sun.ac.za

CHAPTER 3. PRELIMINARY DESIGN OPTIMISATION 42

in the USA. The PVLib toolbox is a standard repository for PV system modelling and analysis
algorithms [66]. To date, it does not have any nancial models implemented so it is only a
performance modelling software package. The PVLib toolbox uses code-level modular programming
to perform the modelling of the system. The toolbox is available in MATLAB or python
programming languages. The PVLib toolboxes do not come with any graphical user interfaces
so the user is expected to call the functions, set the required parameters and execute the
simulation using the modules provided as building blocks. A typical simulation in PVLib
comprises the following steps:

1. Setting up the location and design of the PV system. This includes importing the
necessary python libraries and add-ons, collecting the weather data for the desired site
and setting the orientation of the PV modules or arrays. The weather data is obtained
from typical meteorological year les (TMY2 and TMY3).

2. Modelling the DC (direct current) IV (current-voltage) characteristics. This entails the

estimation of the chosen module's behaviour according to its IV characteristics given in
the module database, in response to the weather data given in step 1. Along with some
intermediate results, the user obtains the estimated DC power yield.

3. Converting the DC power to AC (alternating current). The user makes use of a PVLib
module that takes in the technical characteristics of a desired inverter as one of the
parameters and models the AC power output.

4. Plotting the graphs and tables. The user also has the option of exporting the simulation
results for presentation or further analysis in a third-party software package.

Access to PVLib is free and the software itself is easy to install. Its code is open-source
and therefore transparent and easy to edit or modify [67]. The modular approach in PVLib
allows users to draw better analyses on portions of the modelling chain separately. Because
of the programming literacy that PVLib requires, the program attracts users mostly from
specialised elds such as engineering and research. Figure 3.4 shows the interface on the of the
PVLib-python toolbox in use on, Jupyter Notebook [68], a notebook interface.

Figure 3.4: A screenshot of the PVLib-python toolbox, running on Jupyter

 Stellenbosch University https://scholar.sun.ac.za

CHAPTER 3. PRELIMINARY DESIGN OPTIMISATION 43

SAM, PVSyst and PVLib were then compared to each other according to various evaluation
criteria:

(a) Cost and commercial availability

(b) Working platforms

(c) Version updates

(d) User-friendliness and ease of use

(e) Reporting and analysis options

(f ) Modelling exibility

(g) Performance and economic modelling capability

(h) Validation of simulation results with eld data

3.4.2.2 Cost and commercial availability

SAM and PVLib are free to download, but you need a registration key to activate your copy
of SAM. NREL provides SAM registration keys free of charge and uses your registration data
merely for statistical purposes [5].

As for PVSyst, after installing it users can use it for free in the evaluation mode for 30 days
during which it runs at full capacity. To enjoy the full capabilities thereafter, a user will have
to purchase a license, otherwise PVSyst will automatically switch to its demo mode in which
it runs with limited capabilities. The licensed mode is available in two avours, namely Pro30,
for PV systems up to 30 kW and Premium, the more expensive one which works with unlimited
sizes of PV systems. A full listing of the prices for the licenses can be accessed on the PVSyst
website [55]. PVSyst also supports academics with discounts on licenses bought for educational
purposes. The discounts are available for schools, universities and other educational institutes
on request directly from PVSyst.

3.4.2.3 Working platforms

PVSyst runs only on Windows client versions currently supported by Microsoft. The supported
versions are Vista, 7, 8 and 10 on both 32 bit and 64 bit systems. Other operating systems are
supported through the use of virtual machines running Windows such as VirtualBox [55].

SAM runs on Windows 7, 8 and 10 and is available in both 32 and 64 bit versions. SAM also
runs on OSX and Linux operating systems, but it is only available in 64-bit versions [5].

PVLIb comes in two dierent types of toolboxes, namely PVLib for MATLAB and PVLib-python
[67]. As the names suggest, the two toolboxes work on MATLAB and python environments
respectively. PVLib-python works on platforms that have python versions 2.7, 3.4 or 3.5 and
Pandas versions 0.13.1 or newer and these versions of python are available for Windows, Mac
OS X and Linux [69]. MATLAB is also supported on Windows, Mac OS X and Linux.

3.4.2.4 Version updates

PVSyst is updated regularly and the PVSyst website provides information with regards to
the improvements and xed errors associated with each update and the date each update was
released [70].
 Stellenbosch University https://scholar.sun.ac.za

CHAPTER 3. PRELIMINARY DESIGN OPTIMISATION 44

As is the case with PVSyst, SAM is update regularly and the NREL SAM website shows
release notes for each of the updates and revisions, highlighting the improvements made and
errors xed [71].

PVLib is also being continuously improved and expanded in its functionality. All the releases
are available on GitHub and are accompanied by a brief summary of the associated bug xes,
version documentation and acknowledgements of main contributors [72].

3.4.2.5 User-friendliness and ease of use

SAM comes with a graphical user interface (GUI). This makes it easy for users with no
experience in computer programming or developing computer models to build PV systems
and perform simulations. The GUI allows users to nd their way around the program with
relative ease. It provides access to input variables and simulation controls and it displays the
user's selections as well as graphs and tables after running the simulation.

PVSyst also has a graphical user interface (GUI) that makes it possible for users who are
not particularly experienced in computer modelling to build their own PV systems, perform
basic simulations and generally navigate their way through the software program with relative
ease. The interface in PVSyst allows the user to perform system designing and sizing and to
obtain results in graphical or tabular form easily. For most of the tasks, the user only has to
input certain required values in the appropriate elds and then execute actions by the touch of
a button.

PVLib does not oer a graphical user interface. It is essentially a collection of scripts and
functions that handles data, module and inverter parameters and takes in atmospheric and
irradiance models. The user runs these scripts and functions sequentially to perform PV system
simulations. The interface is therefore bound to present some challenges to users with no
programming skills or experience in computer modelling of PV systems.

3.4.2.6 Reporting and analysis options

SAM has a default graphing tool that not only enables the display of results as sets of default
graphs, but also allows the creation of custom graphs. Alternatively, the results can be presented
in the form of tables or exported to other software packages for presentation or further analysis.
SAM also provides a number of options for simulation results analysis, namely [73]:

(a) Parametric analysis - This enables the user to assign multiple values to input variables in
order to create graphs showing the relationship between one or more input variables to a
metric result. This operation is ideal for the optimisation of input variables.

(b) Sensitivity analysis - This option enables a user to investigate how sensitive an output
metric is to variations in certain input variables.

(c) Statistical analysis -This analysis option allows the user to investigate uncertainties in
one or more input variables on output metrics.

(d) Multi-level subsystems - This allows users to model power systems as a combination of
subsystems. This could be ideal, for instance, in a situation where each subsystem is
meant to be oriented in a dierent direction.
 Stellenbosch University https://scholar.sun.ac.za

CHAPTER 3. PRELIMINARY DESIGN OPTIMISATION 45

(e) Excel Exchange - This option allows the user to link an input variable in SAM to a cell
or range of cells in a Microsoft Excel Workbook.

(f ) P50/P90 analysis - This is a method of calculating the probability that a system's total
annual output will exceed a certain value, for a site of which the weather data is available
for many years.

(g) Exchange Variables - This option allows the user to input his own variables for use with
a custom Transient System Simulation (TRNSYS) deck or Excel Exchange.

PVSyst, like SAM, documents simulation results into reports, summarising the simulation
parameters, main results and system quality. The results can also be viewed as tables or
exported to a third party software package like Microsoft Excel for further analysis or presentation.
The only analysis option that PVSyst oers is the P50/P90 [55].

PVLib allows the user to view simulation results by plotting the respective metric's graph
or exporting the results to a third-party software package such as Microsoft Excel for further
analysis. The original PVLib toolbox does not oer any analysis options.

3.4.2.7 Modelling exibility

PVSyst comes as a complete desktop application and does not have an allowance for scripting.
That means that the user can only model a PV system the design of which adheres to certain
predened standards in PVSyst. It is therefore not possible for the user to introduce or modify
any of the thermal, electrical or optical models in PVSyst [74]. Users can however dene devices
not found in the PVSyst component databases using the manufacturers' datasheets, thereby
creating new devices in the database.

SAM also comes as a complete desktop application but it has a built-in scripting language,
SAM User Language (SamUL). This allows the user to automate tasks and to perform more
complex analyses from within SAM [5]. With SAM, a user also cannot introduce new thermal,
electrical or optical models but can link and import variables from other other programs such as
Microsoft Excel and TRNSYS. The user is also allowed to dene devices that cannot be found
in the component database using the component's datasheet. In August 2017 SAM actually
became open source and one of the main reasons for that was so that users can change the
code, build their own versions of SAM and add their own features and capabilities, in addition
to providing more transparency in their algorithms and promoting collaborations [75].

PVLib is built on the very basis of modelling exibility. The availability and free access to
the PVLib source code makes it possible and easy for users to modify models, databases and
input methods for a range of research and design goals [66] and [67].

3.4.2.8 Performance and economic modelling capability

SAM has a lot of performance models for photovoltaic systems, concentrating solar power, solar
water heating, wind power, geothermal and biomass power. The photovoltaic systems in SAM
are further subdivided into three categories:

(a) Flat Plate PV, which models the system using sub-models for the components (modules
and Inverters) of the system. This is the model we will use for our grid-connected PV
system performance model.
 Stellenbosch University https://scholar.sun.ac.za

CHAPTER 3. PRELIMINARY DESIGN OPTIMISATION 46

(b) High-X Concentrating PV, which models concentrating photovoltaic (CPV) systems.

(c) PVWatts, which models crystalline silicon-based systems and is an implementation of

NREL's PVWatts model which is web-based.

SAM oers a signicant number of nancial models. The models calculate the cash-ows
and nancial metrics for dierent nancing options chosen by the user. The nancial models
given in SAM are for residential projects, commercial (excluding Commercial PPA) projects,
commercial projects with power purchase agreements (PPA) and utility projects [5].

PVSyst only deals with photovoltaics and it splits its performance models into four main
categories:

(a) Grid-connected - This is a type of system that is made up of components constituting the
PV array i.e. modules/ strings of them, inverters and everything up to the connection to
the grid.

(b) Stand-alone - A system which should constitute of modules, a battery and a regulator.

(c) A water pumping system.

(d) DC Grid - A system made up of a PV-array and a load prole meant for public transport
network.

PVSyst also oers economic evaluation to model project costs and investments. With the
nancial model, the user can project running costs and deduce the long-term protability,
especially for grid-connected systems [55].

The PVLib original toolbox has a performance model that currently only supports grid-connected
systems and does not oer any nancial models. There is, however, room for improvement in
PVLib and for more experienced users to expand on PVLib's capabilities and add elements
that will help them model other systems [66], [67] and [76].

3.4.2.9 Comparisons of Simulation results

The 75 MWp solar plant was modelled in the three programs to simulate the total energy yield
for the whole year using the Hay-Davies [46] transposition model. Weather data measured on
the site for the year 2014 was used in the simulations and the simulated total energy yield of
each software package was compared to the total yield recorded at the test site for the year
2014. Table 3.2 summarises the losses calculated for the 75 MWp system from the plant's
measured data.

Table 3.3 shows a summary of the total yield produced by the simulations in each of the
software programs and their percentage dierences from the actual plant yield. The percentage
dierence was calculated as:

Y ieldactual − Y ieldsimulated
Dif f (%) = · 100% (3.1)
Y ieldactual
In Figure 3.5, the three simulated annual yields are compared to the plant's actual yield for
the year 2014 and in Figure 3.6, a similar comparison is drawn based on the energy proles on
randomly selected clear sky and cloudy days. The results show consistency and that PVSyst
simulated the yield more accurately, with a calculated percentage dierence of 3.4%, followed
 Stellenbosch University https://scholar.sun.ac.za

CHAPTER 3. PRELIMINARY DESIGN OPTIMISATION 47

Table 3.2: Summary of plant losses

Loss Value (%)

Shading -2.5

Reection -2.5

PV loss due to irradiance level -0.45

Module mismatch and DC ohmic -1.77

Soiling loss -1.75

PV loss due to temperature -6.3

Inverter loss -2.6

Table 3.3: 2014 Total Annual Yield Comparison

Source Total Yield Dierence

(MWh) (%)
Actual yield 149868.2 -

PVSyst 144813.6 3.37

SAM 144046.4 3.86

PVLib 142276.8 5.07

Figure 3.5: Total energy yield for the year 2014

by SAM with 3.9% and then lastly PVLib with approximately 5.1%. The results also show
that the three software programs were generally conservative in their yield predictions and this
is probably because of the overestimation of losses in some of the programs' default losses. The
default losses were used in some of the system modelling stages where the actual values for the
losses could not be calculated from the available plant data or the design tool did not provide
an option for actual values to be input for particular system losses but rather estimated them
 Stellenbosch University https://scholar.sun.ac.za

CHAPTER 3. PRELIMINARY DESIGN OPTIMISATION 48

Figure 3.6: Clear sky and cloudy day energy proles

using inbuilt system loss models.

3.4.2.10 Conclusions
Analysis of the software packages revealed that each software package has its own merits that
makes it more suitable than the others for certain simulations, and demerits that serve as
disadvantages in certain instances. In terms of accuracy and robustness, PVSyst seemed to
outperform the rest, closely followed by SAM. In terms of modelling exibility, PVLib and
SAM did well, while PVSyst was held back by its rigid form. In terms of other general
characteristics and functionalities, it seemed SAM would be the most suitable, of the three
software packages evaluated in this paper, for the next section of work involving optimisation
strategy formulation and implementation. SAM proved to be a fairly accurate, complete, easy
to install and user-friendly solar PV simulation tool that also comes with the perks of modelling
exibility and a wide range of reporting and analysis options which could potentially come in
handy.

3.4.3 Iterative optimisation approach

One way of determining design parameters for optimally designing a PV system is by iterating
through all possible values and then choosing the one value that best satises the design
objective(s). When there are more alternative system congurations to choose from, it might
also be worthwhile to compare the performances of the options with each other simultaneously.
Parametric analysis enables the system designer to quantitatively and qualitatively compare
dierent PV systems with each other simultaneously. In this example, we used an iterative
approach to determine the optimal tilt angle that would yield the highest amount of energy per
year, month and season. The results were then compared to those of the rule-of-thumb, ROT,
recommendation (tilt = latitude) as well as the yield from single and dual axis tracking systems
using parametric analysis. NREL SAM's scripting ability and parametric analysis features were
implemented towards this end.

3.4.3.1 SAM scripting and parametric analysis

SAM has an in-built scripting language known as LK. It is a powerful, yet simple, scripting
language that is designed to be compact, fast and easy to embed in other applications. It is
 Stellenbosch University https://scholar.sun.ac.za

CHAPTER 3. PRELIMINARY DESIGN OPTIMISATION 49

quite similar to Visual Basic for Applications (VBA) in Microsoft Excel and works on Windows
and OSX platforms alike [77]. In SAM, particularly, it helps extend the inbuilt functionality of
the program and enables users to automate redundant tasks and large numbers of simulations.
Some of the features of the LK scripting language include:

ˆ ability to use user-written functions,

ˆ conditional statements (if, else, etc...) and loops (while, for, etc...) just like any other
programming language,

ˆ ability to implement data structures such as hash tables, arrays (single and multidimensional),
and

ˆ a built-in library of functions for simple input and output tasks as well as more complex
subroutines such as regressions, sampling, plotting and accessing web services.

SAM's parametric analysis enables PV system designers to assign more than one value to
input variables and assess their inuence on the output results and it can even take the place of
the iterative approach in some cases. Parametric analyses in SAM produce tables and graphs
that can also be exported to third-party platforms for further investigation and presentation.

3.4.3.2 Procedure
The following procedure was used in this investigation to optimise energy production for our
test site by calculating optimal tilt angles for dierent periods in the year and comparing the
dierent system setups with each other:

1. Use SAM script to iterate through possible tilt angles, θT , to nd the optimum tilt angle,
θT,opt , that gives the highest total annual energy yield.

2. Use iterative SAM script to nd the optimum tilt angles, θT m,opt , that maximise the total
energy yield for each month.

3. Derive the seasons' optimum tilt angles, θT s,opt , from the averages of the tilt angles of
the three months that constitute that particular season according to the South African
seasons given in Table 3.4 obtained from [78].

4. Derive the biannual optimum tilt angles, θT b,opt , by averaging the optimum tilt angles in
the 6 months between the two equinoxes, in March and September. For the biannual tilt
method, summer tilt, therefore, is an average of the October to March period, while the
winter tilt is the average for the period starting in April and ending in September. This
method of calculating the biannual tilt led to projection of a higher energy yield than the
rule-of-thumb approximation for the biannual tilt which stipulates that the optimum tilt
◦ ◦
angle should be 15 more than the latitude for the six winter months and 15 less than
the latitude for the six summer months [58].

5. Compare all the above-mentioned methods' monthly yields with that of the θT,opt =
Latitude and the tracking systems. The dierent systems were simulated and their total
monthly and annual energy yields were compared all at once using SAM's parametric
analysis.

Figure 3.7 shows the algorithm for the iterative process that was implemented in SAM LK to
loop through the possible tilt angles, 0 to 90 degrees, to optimise the systems for maximum
energy yield.
 Stellenbosch University https://scholar.sun.ac.za

CHAPTER 3. PRELIMINARY DESIGN OPTIMISATION 50

Figure 3.7: Algorithm for iterative tilt angle optimisation

Table 3.5 summarises the monthly, seasonal and biannual tilt angles as calculated for our
test site. In all the comparisons, the same model and number of PV modules and inverters
were used. Only the tilt methods and mounting methods were varied, accordingly.
 Stellenbosch University https://scholar.sun.ac.za

CHAPTER 3. PRELIMINARY DESIGN OPTIMISATION 51

Table 3.4: Table of season calendar dates

S. Hemisphere Dates N. Hemisphere

Autumn 1 March - 31 May Spring

Winter 1 June - 31 August Summer

Spring 1 September - 30 November Autumn

Summer 1 December - 28/29 February Winter

Table 3.5: Table with the calculated monthly, seasonal and biannual tilt angles

Tilt angles

Month Optimal tilt angle Seasonal Tilt Biannual Annual

January 0 3 8 32

February 8 3 8 32

March 26 44 8 32

April 52 44 52 32

May 54 44 52 32

June 63 58 52 32

July 56 58 52 32

August 52 58 52 32

September 34 16 52 32

October 15 16 8 32

November 0 16 8 32

December 0 3 8 32

3.4.3.3 Results
Figures 3.8 and 3.9 show the results of the parametric analysis of various tilting and mounting
methods. The plot shown in Figure 3.8 shows the monthly variation of system yield for various
tilting and mounting methods. Energy production seems fairly consistent for most of the
methods save for the two tracking systems which show signicant drops in energy in winter; the
single-axis tracker dropping more signicantly than the dual-axis tracking system. Figure 3.9
shows the comparison of annual energy yield for various tilting and mounting methods. The
percentage dierence between each mounting or tilting method's yield with that of an optimally
tilted xed system is shown on top of each bar. The results show that adjusting the tilt angle
of the xed tilt systems periodically can boost energy production. This is not only valuable in
optimising annual energy production, but also optimising energy production for specic months
or seasons. The rule of thumb was also proven to be fairly accurate as its energy production
◦
was within 0.1% of the actual optimal energy yield produced at at tilt angle of 32 .
 Stellenbosch University https://scholar.sun.ac.za

CHAPTER 3. PRELIMINARY DESIGN OPTIMISATION 52

Figure 3.8: Monthly Energy Yield comparison of various tilting and mounting methods

Figure 3.9: Total Energy Yield Comparison for the various tilt and mounting methods

3.4.3.4 Conclusion
The parametric analysis helped us quantify the dierences in the yield of the various mounting
methods. Tracking systems generally performed signicantly better than any of the other
 Stellenbosch University https://scholar.sun.ac.za

CHAPTER 3. PRELIMINARY DESIGN OPTIMISATION 53

systems, but the adjustable tilt systems oer a less expensive way to drive system performance
so they might be worth considering depending on how practical that might be in the particular
site's case. The seasonal, monthly and biannual tilt methods showed great potential in boosting
energy production and would be interesting compromises between xed tilt and single-axis
tracking systems. The iterative approach was very useful in determining the optimal tilt angles
and automation of the process by means of scripting helped circumvent the monotony that is
usually part of iterative processes. The main challenge with this method is the fact that it is
extremely tedious to go through each and every one of the possible design variables, especially
if there are other parameters that also need to be designed for. Evolutionary programming
methods and their ingenious iterative approaches, as will be discussed in the Optimisation
Strategy Chapter, would prove useful in this regard.

3.4.4 Multi-objective optimisation

It is not uncommon for the PV system design process to involve multiple design parameters
and multiple design objectives. When this happens, there might be need to nd a reasonable
compromise so that the nal PV system does not optimise one objective or parameter over the
other. This is usually where multi-objective optimisation comes in handy. One such design
scenario occurs when PV system designers are designing for optimal tilt and row spacing.
These two design variables tend to be antagonistic because optimally designing one tends to
compromise the other, especially in a constrained design space, as explored in the following
design scenario investigation.

In the case of xed tilt systems on space-constrained sites, PV system designers would
typically want to design the system for maximum power density, so as to t in as many solar
panels as they can in the available area. In designing for maximum power density, rows of
modules are placed closer together and the tilt is lowered to reduce inter-row shading. This
results in a higher peak capacity but the modules will be less productive individually.

In cost-constrained projects, however, system designers would probably rather design for
maximum panel energy yield to ensure that the system gives the best value for money in
module performance since module costs alone make up a substantial portion of the system
costs compared to any other component in the PV system [79]. To get the maximum energy
yield from each module, the designer will have to design for an optimum tilt and space the
modules far apart to avoid shading. While this is good for energy production as it maximises
the amount of irradiation reaching each panel's surface and reduces shading, this design decision
reduces the number of modules that can t into a unit area.

A multi-objective optimisation for tilt and row spacing, to nd a compromise between the
two design considerations, was performed in Helioscope using shading analysis. Helioscope was
not part of the initial PV system tool investigation as given in Section 3.4.2, but it was found
suitable for this particular investigation.

3.4.4.1 About Helioscope

Named after the instrument that is used to observe the sun and sunspots, Helioscope is a
PV system design tool from Folsom Labs that incorporates system layout and performance
modelling to simplify the solar PV system design process [31]. It is a cloud-based software
program designed to be easier and cheaper to use than other commercially available design tools
such as PVsyst. Helioscope uses the same underlying algorithms as PVsyst, also comes with
 Stellenbosch University https://scholar.sun.ac.za

CHAPTER 3. PRELIMINARY DESIGN OPTIMISATION 54

an extensive component library and has global weather coverage and enables users to design
PV systems with relative ease, without requiring them to know too many engineering-level
intricacies [80]. Simulation results from Helioscope were evaluated by BEW Engineering [81]
and they proved to be fairly consistent with PVsyst results [80].

Helioscope, however, can only handle PV systems of up to 5 MW in capacity so it was

necessary for us to scale down the size of our test-case to an appropriate level. Scaling down
in no way compromises our investigations because the main goal was not to calculate shading
losses precisely but rather identify their trends within the context of our test-case's design
parameters and constraints. Helioscope's robust shade and sketch-up integration made it ideal
for the pitch-tilt multi-objective optimisation investigation even though SAM and PVsyst could
have also been implemented with a bit of extra eort. Compared to PVSyst, dening a shading
scene in Helioscope is much easier and straightforward. With SAM, the optimisation would
tend to lean more towards the optimisation of the ground cover ratio, since SAM uses GCR
in its shading computations rather than directly using the pitch [54], instead of the intended
pitch-tilt multi-objective optimisation even though the ultimate eect would be the same.

3.4.4.2 Procedure
2
A multi-objective optimisation was performed over an area 174661.7m in size with a baseline
◦
system design set as 30 tilt, north-facing azimuth and an inter-row spacing of 4.14m based on
a scaled down version of the test-case design specication. The simulation was repeated for
◦ ◦
20 and 40 tilt angles and the inter-row spacing was varied from 1.14m to 7.14m in one-metre
increments. The assumptions in this case are:

ˆ that the conguration of the PV modules is set according to the recommendation of the
module manufacturer in the portrait conguration,

ˆ ◦
the azimuth is at its optimum at 0 , facing the true north direction, and

ˆ no other obstacles are in the vicinity of the solar array to cause shading so all shading
losses in this case can be attributed to inter-row shading.

Specic Yield, dened by the equation:

AnnualEnergytotal (kW h)
Y ieldspecif ic = , (3.2)
InstalledCapacitytotal (kW p)
is essential in this analysis because as we vary the row-spacing and tilt angles, the number of
PV modules that can t into the space provided also changes and this, in turn, aects the
nameplate capacity of each modelled system. Specic Yield, therefore, allows us to compare
the overall performance of each of the 21 systems on a standardised scale.

3.4.4.3 Results
Figures 3.10 and 3.11 show the results of the shading losses and the specic yield as the tilt
angles and inter-row spacing were varied. From the gures, it can be deduced that the lower tilt
◦
angle, 20 , showed fairly consistent performance for the most part, regardless of the inter-row
spacing. The sensitivity of the system's performance also seemed to increase as the tilt angle
◦ ◦ ◦
was increased and this is mostly evident in the gradient of the 20 , 30 and 40 line graphs in
the sub-'2.14m inter-row spacing' region. It was also noted that specic yield increases as the
inter-row spacing increases until a certain distance beyond which increasing the row spacing
will not do much to improve the system performance. This distance is just over 4.14 m. There
 Stellenbosch University https://scholar.sun.ac.za

CHAPTER 3. PRELIMINARY DESIGN OPTIMISATION 55

is also evidence of a minimum row spacing below which the system performance will start
decreasing drastically and this is around 2.14 m for the three tilt angles. A row distance below
2.14 m would generally prove more taxing where module performance is concerned, while a row
distance above 5.14 m would be wasteful in land utilisation so for system designers, the 2.14 m
to 5.14 m row distance region would be an area of key interest.

Figure 3.10: Graph of variation of system performance with inter-row spacing

Figure 3.11 also shows a similar trend and conrms that the closer module rows are placed
to each other, the more they make shade on each other and the less productive the system
performance will be.

3.4.4.4 Conclusion
Multi-objective optimisation could be useful to determine reasonable compromises in the PV
system design process. The approach, however, could suer if the number of variables or
objectives increase as it is already a rigorous process with two design variables and objectives,
judging by this design example. Optimisation methods that take into account multiple design
objectives and multiple variables can be quite useful in this regard. This is noted and implemented
in the optimisation strategy chapter.
 Stellenbosch University https://scholar.sun.ac.za

CHAPTER 3. PRELIMINARY DESIGN OPTIMISATION 56

Figure 3.11: Graph showing variation of shading losses with inter-row spacing

3.4.5 Other non-conventional design alternatives

As mentioned in [26], recent trends and changes in the solar PV industry have warranted a
revisit to design methods to determine what changes need to be made for adaptation purposes.
Methods that do not adhere to the traditional design conventions, such as the rule-of-thumb, are
becoming a popular way of optimising system performance in various aspects and also helping
designers bypass certain design constraints. One such example is the east-west orientation
method.

The east-west orientation is a racking style where consecutive PV module rows are oriented
facing the east and west directions in an alternating fashion as shown in Figure 3.12. The

Figure 3.12: A pictoral representation of the east-west orientation.

east-west orientation is based on an understanding of the apparent motion of the sun due to
 Stellenbosch University https://scholar.sun.ac.za

CHAPTER 3. PRELIMINARY DESIGN OPTIMISATION 57

the earth's rotation as it rises from the east and sets in the west as shown in Figure 3.13. As
such, in this orientation the east-facing modules produce more energy in the morning while
the west-facing modules cater for the afternoon. Even though the east-west orientation usually
results in a lower peak, the set up enables modules to capture peak power production for longer
durations and during the earlier and later hours of the day. It also allows modules to be tted
closer to each other since it results in minimal shading. The system will consequently have a
higher power density since more PV modules will be able to t in per unit area.

Figure 3.13: Depiction of the apparent path of the sun as the day and year passes.

In addition to improved power density and reduced shading, east-west orientation also has a
variety of other advantages [82]. These advantages include:

ˆ improved aerodynamics since the panels will be back to back at a low tilt. This results
in less drag and less air-scoop unlike in North or South facing systems where the setups
are vulnerable to southerly or northerly winds.

ˆ more ecient use of inverter capacity because of the ability to use over-paneling as a
consequence of a lower peak output.

ˆ lower costs due to less maintenance and more ecient use of components such as mounting
material.

ˆ more consistent energy supply during the day and potential use in matching facilities'
demand curves especially in regions where Feed-In Taris (FITs) are in place.

3.4.5.1 Procedure
A comparison of North-facing and East-West orientation was conducted to simulate the potential
◦ ◦ ◦
benets of the east-west orientation using tilt angles of 20 , 30 and 40 for a space-constrained
2
piece of land measuring 174661.7m on the test-case site. The pitch in each scenario was
determined such that it was the shortest possible distance that resulted in 0% inter-row shading
losses and the goal was to t as many PV modules as possible in the given area. Using the
same tilt angle at each iteration, energy yield, plant performance and installed capacities for
the two orientations were simulated in Helioscope and compared.
 Stellenbosch University https://scholar.sun.ac.za

CHAPTER 3. PRELIMINARY DESIGN OPTIMISATION 58

3.4.5.2 Results
Figures 3.14, 3.15 and 3.16 show the summary of results. Figure 3.14 shows how the east-west
orientation can be used to improve installed capacity on a space-constrained piece of land
since it results in very little inter-row shading. Figure 3.15 shows how the poor individual
module performance can be compensated for by the increased installed capacity to boost the
total annual energy yield of a system in an east-west orientation in a space constrained design
space. In projects with high xed costs, designers will also be able to boost the system's energy
production capacity per unit area without too much of a negative impact on overall project
costs.

The low specic yield for the east-west orientation shown in Figure 3.16 is due to the fact
that module layout will not be optimally designed to produce the highest amount of energy at
solar noon, when GHI is supposed to be at its highest, and at any particular time before and
after solar noon, only half of the modules will be optimally oriented for maximum energy yield.
This results in a relative under-performance of the east-west systems, compared to systems
oriented facing towards the equator.

Figure 3.14: Comparison of installation capacity showing the increase in the number of PV
modules that can t into the site due to the east-west orientation.

3.4.5.3 Conclusion
The east-west orientation method was particularly interesting as it goes against convention, but
proved to be potentially useful in overcoming issues that come with many space-constrained
locations and projects with high xed costs. For a small sacricial drop in system performance,
system designers will be able to t more PV modules per unit area using the east-west orientation
and get more energy yield, compared to the north-facing orientation. In addition to that, the
 Stellenbosch University https://scholar.sun.ac.za

CHAPTER 3. PRELIMINARY DESIGN OPTIMISATION 59

Figure 3.15: Comparison of energy output of the PV systems indicating the increases in the
energy production due to the east-west orientation.

Figure 3.16: Comparison of the system performances showing a reduction the percentage drop
in production due to the east-west orientation.
 Stellenbosch University https://scholar.sun.ac.za

CHAPTER 3. PRELIMINARY DESIGN OPTIMISATION 60

east-west orientation method harnesses intrinsic value in several other plant design aspects
and ties in together with what was pointed out earlier in Section 3.2, that the focus of PV
system design is starting to take more interest in the balance of system aspects as well as
other PV system performance indicators to optimise PV system designs. We concluded that
it would be worthwhile to devise a plant layout strategy that would take into account several
design objectives simultaneously. Multi-objective optimisation methods that are discussed in
the Optimisation Strategies Chapter will enable us to achieve this goal.

3.5 Summary
In this chapter, various elements that go into plant design optimisation were reviewed. The
process involved investigating the tools and strategies currently being implemented by some PV
system designers to nd their purposes and strengths and to identify areas that could use some
improvement. Some of the key insights from this preliminary plant design and optimisation
process are:

ˆ Every design tool is made with a particular goal in mind and this sets PV system design
tools apart from each other. It is important for PV system designers to knowledgeably
select design tools that will enable them to perform their modelling and analysis thoroughly
and accurately.

ˆ Iterative optimisation approaches, as investigated in this study, are very helpful in the
optimisation process but with increased dimensionality in design objectives and variables,
it might get more cumbersome to implement them. This could be solved by using
evolutionary programming methods which make use of algorithms that can navigate
complicated iterations.

ˆ In cases involving several, sometimes antagonistic, variables and objectives, multi-objective

optimisation could be implemented to circumvent the challenges of having to prioritise one
objective over the other(s). Optimisation methods that perform operations on multiple
variables and/or multiple objectives could be helpful in this regard.

ˆ Nonconventional design optimisation processes also play an important role in overcoming

certain design constraints and challenges. They also are better adapted to the current
trends in the PV industry. The east-west orientation method used in the design example
brought multi-objective optimisation into the picture as a way to make the design process
more holistic to intentionally develop systems that excel in satisfying several design
objectives simultaneously.

All these conclusions were then used to formulate an optimisation framework for utility scale
PV systems that combines the accurate PV system design modelling in reputable, bankable
design tools and a more comprehensive approach to PV system layout design.
 Stellenbosch University https://scholar.sun.ac.za

Chapter 4
Optimisation Strategies

4.1 Introduction
The main objective of this project is to investigate strategies to optimise PV plant design
layout. In PV system layout design, optimisation is the process of improving the PV system's
orientation and setup to better achieve the desired design goal. For design layout optimisation,
this process involves selecting design parameters such as module tilt and azimuth angle and the
ground cover ratio (GCR) to optimise energy yield, plant performance and/or cost of energy
production for the desired project site. The main elements needed for the successful design
layout optimisation are:

ˆ PV system modelling tools that are capable of accurately estimating the response of the
plant design objectives to variations in the design parameters,

ˆ Clearly dened site conditions and constraints for the plant being designed, and

ˆ An eective method to go through the possible solutions to select the best option.

In this chapter, these elements are explored and made use of in the development of an optimisation
framework for PV system layout using evolutionary programming techniques.

4.2 Evolutionary Algorithms (EAs) in the design of

grid-connected PV systems
Evolutionary Algorithms (EAs) are a part of evolutionary computation which is a metaheuristic
(higher-level) optimisation procedure. EAs use techniques adopted from nature and biological
evolution to nd solutions for complex problems and have been used widely in several science
and engineering elds to date. These methods include Ant Colony, Particle Swarm Optimization
(PSO) and Genetic Algorithms (GAs) [83] and [84]. In the renewable energy industry, and more
specically in solar PV, this new generation approach to system design is being used in several
ways to replace and sometimes consolidate the traditional design approach which often follows
rigorous procedures such as iteration and linear programming. Some of the areas in which EAs
have been used extensively include PV-wind hybrid system designs and methods for Maximum
Power Point Tracking (MPPT). Sinha and Chandel did a substantial review of recent trends
in optimisation techniques implemented in solar PV-wind hybrid systems in [85]. The major
outcomes from this review indicate that EA methods are not only improving over time but are
also proving to be useful, especially towards tackling the involute problems often encountered
in the renewable energy research sector.

61
 Stellenbosch University https://scholar.sun.ac.za

CHAPTER 4. OPTIMISATION STRATEGIES 62

Several sources also show the implementation of optimisation algorithms in Maximum Power
Point Tracking (MPPT). Sreedhar and Jagadeesh, briey describe and review several of these
optimisation techniques and nd that the biologically inspired algorithms generally perform
better than the other methods they were compared to such as the Constant Voltage Tracking
(CVT) method, the Perturbation and Observation (P&O) method and the curve-tting method
in [86].

Authors such as Kornelakis et al [87], [88] and [89], Weinstock and Appelbaum [90] and
Gallardo [91] have some of the best work in the optimisation of layout design and sizing of
grid-connected PV systems using EA techniques which is more relevant to the main focus
of this thesis. Weinstock and Appelbaum present a single objective-multivariable solar eld
design of xed tilt solar panels. Kornelakis et al present methodology to optimise system
design and evaluate project economic viability for grid-connected PV systems on an island in
Greece. The PVGCS modelling uses mathematical equations to facilitate the design process
by optimising various device-type combinations and then selecting the one which is the most
economically protable. Gallardo builds up on work related to that done by Weinstock and
Appelbaum to successfully design a general methodology for designing PV systems based on
ecodesign principles. The work by Gallardo is very elaborate and the only drawbacks to it are,
like the author states, the lack of an approach that allows the integration of the optimisation
algorithm with commercially available PV system simulation tools and, to a reasonable extent,
the computational expense as evidenced by the time the simulations took to run using Gallardo's
methodology. The focus of Gallardo's research also is more towards optimisation for more
eco-friendly PV system designs, as opposed to the focus of this thesis which is optimisation of
the PV plant designs for better performance and energy production.

4.3 Optimisation Methodology

4.3.1 Simulation Software

Three software packages were evaluated for the purpose of optimisation and SAM was selected as
the nal option. From the initial evaluation, SAM's capabilities presented three big opportunities
for its use in optimisation processes, namely:

(a) The parametric analysis functionality which could enable us to perform iterative optimisation
processes in SAM with relative ease

(b) The ability of SAM to interface with other software programs which meant that optimisation
algorithms could be implemented on a dierent platform to provide optimal parameters
that could be integrated with SAM simulations

(c) SAM's scripting ability which means that code could be written to automate some
processes or to formulate algorithms that could be used for optimisation in SAM.

For the purposes of this optimisation strategy, the second option was chosen to take advantage
of the Graphical User Interface (GUI) in SAM to facilitate the initial plant design set up and
then exporting the setup to MATLAB for integration with the optimisation toolbox. Using
SAM's GUI, the respective weather le of the desired location is loaded and the inverter and
module models are selected from the components databases in the SAM software libraries. The
system sizing is performed by specifying the desired array size and DC to AC ratio, followed
by the conguration and orientation of the modules and inverters. Through SAM's sizing
algorithm, the system conguration at reference conditions is automatically generated, based
 Stellenbosch University https://scholar.sun.ac.za

CHAPTER 4. OPTIMISATION STRATEGIES 63

on the module and inverter models' electrical properties. Shading, irradiance, DC, AC and
transformer losses to be taken into account in the simulations as well as the parameters for the
nancial modelling are also set up. Figure 4.1 shows the owchart that summarises the SAM
simulation setup process. After completion of the setup, the corresponding code is generated in
MATLAB format in preparation for the next stage of the optimisation which will be facilitated
by the MATLAB Optimisation Toolbox.

Figure 4.1: SAM simulation process ow

4.3.2 Module Technology

The approach used in comparing the module technologies was to keep the system design
parameters constant as far as possible in a similar way to the current method. The datasheets
of the modules are given in Appendices A, B and C.

4.3.3 Optimisation Toolbox

The MATLAB Optimisation Toolbox provides various routines that can be used to determine
parameters that minimise or maximise objectives or to solve equations. This is facilitated by
solvers (subroutines) that can be used to perform constrained or unconstrained optimisation
on a wide variety of problems including, discrete, continuous and large-scale problems. The
optimisation problem is set up using three main constituents namely:

ˆ objective function

ˆ constraints

ˆ function

4.3.3.1 Objective Function

An objective function is a method that shows the relationship between a design variable and
the outcome that is meant to be optimised. The user creates an objective function, denoted
symbolically by f(x), which is a measure of the performance of a system as the variable 'x' is
varied. 'x' can, therefore, be viewed as a column vector of design variables which can aect the
performance of the system. Conventionally, optimisation in the toolbox entails minimization of
the objective function, therefore, in the event that maximisation is preferred, the optimisation
procedure is simply the minimisation of the negative of the objective function, -f(x). In our
optimisation problem cases the objective function is built from the MATLAB code generated
from the SAM simulation setup and the plant layout parameters make up the design variable
column vector.
 Stellenbosch University https://scholar.sun.ac.za

CHAPTER 4. OPTIMISATION STRATEGIES 64

4.3.3.2 Constraints
Constraints are limitations in the design space which in PV system design procedures are
hugely inuenced by the design goals themselves, the simulation software restriction or from
site-specic limitations. Where applicable, the user denes a set of constraints in the problem
setup to restrict the solution space according to these limitations. Constraints are generally
classied as one of the following types:

ˆ Linear inequalities of the form A*x ≤ b, where A is a matrix and b is a vector

ˆ Linear equalities of the form Aeq *x = beq , specied by matrix Aeq and the vector beq

ˆ Lower and upper bounds on the variables, specied as vectors

ˆ Nonlinear constraint functions that dene the nonlinear constraints

4.3.3.3 Function
The user also needs to choose an appropriate solver for the particular optimisation problem.
There are four main subdivisions for the optimisation algorithms in the toolbox depending on
the type of problems that need to be solved:

ˆ Linear or nonlinear problems

ˆ Constrained or unconstrained problems,

ˆ Single or multi-variable problems

ˆ Single or multi-objective problems

The PV system layout optimisation problem using SAM software falls under nonlinear constrained
optimisation and our test cases looked into multivariable problems of both single and multi-objective
natures.

4.4 Single Objective Optimisation

The optimisation toolbox provides three main single-objective optimisation functions. These
are:

ˆ fminbnd - to nd a minimum of a single-variable function over a xed interval

ˆ fmincon - to nd a minimum of a constrained nonlinear multivariable function

ˆ fseminf - to nd a minimum of a semi-innitely constrained nonlinear multivariable

function

Our optimisation problem cases do not involve semi-innite constraints so the only options were
the fminbnd and fmincon functions. The fminbnd is only limited to cases where a single variable
is being evaluated and this restricts its usefulness in our optimisation problems. The fmincon
function was then selected because it oers more functionality with its ability to evaluate
multiple variables simultaneously.
 Stellenbosch University https://scholar.sun.ac.za

CHAPTER 4. OPTIMISATION STRATEGIES 65

4.4.1 Algorithm
fmincon has ve algorithm options namely:

ˆ Interior point,

ˆ Trust region reective,

ˆ Sequential quadratic programming (SQP) and

ˆ Active-set.

According to the MATLAB optimisation toolbox documentation, it is recommended that the

user attempts to implement the interior point algorithm rst before turning to the other options.
The interior point algorithm is a large scale algorithm, capable of handling not only large, sparse
problems but small dense problems as well.

The original problem is denoted by

min f (x), (4.1)
x

subject to h(x)= 0 and g(x)≤ 0,

where h(x) represents equality constraints and g(x) represents inequality constraints. To solve
the original problem, it is approximated using the barrier function as
X
min fµ (x, s) = min f (x, s) − µ ln(si ), (4.2)
x,s x,s
i

subject to h(x)= 0 and g(x)+ s = 0 for each µ> 0 .

The approximate problem represents the original problem as a sequence of equality constrained
problems which are easier to solve than the original inequality constrained problem. The step
attempts to solve he resulting Karush-Kuhn-Tucker (KKT) equations from the approximate
problem using a linear approximation. The KKT conditions are similar to the condition that
the gradient must be zero at a minimum, only that they are modied to hold for constrained
problems. To solve the approximate problem the algorithm implements Newton's method for
minimization. If the Newton (direct) step fails, the algorithm tries to implement a conjugate
gradient step which uses a trust region. A trust region is a neighbourhood N around the point
x where the behaviour of the function f(x) can be reasonably reected by a simpler function q.
The trust-region subproblem is denoted by:

min {q(s), s ∈ N )} (4.3)

s

To run the optimisation, the user supplies the objective function, the constraints and a
preferred starting point for the solver. In addition to that, it is imperative that the approximation
of derivatives by the solver is done using central dierences and not forward dierences. This
way the algorithm takes twice as many function evaluations but yields more accurate results.

4.5 Multi-objective Optimisation

In many real-life situations and processes, optimisation problems are often multi-objective in
nature, and the design process for PV plant layouts is not exempt from that. In several
situations it may not be adequate to represent plant design optimisation as a single-objective
optimisation so multi-objective optimisation will have to be implemented. There are often
instances when several factors such as techno-economic and sometimes environmental design
 Stellenbosch University https://scholar.sun.ac.za

CHAPTER 4. OPTIMISATION STRATEGIES 66

objectives need to be taken into account simultaneously during the layout design stage. This
increases the problem's dimensionality and as a result, it is very rare to nd a single optimal
solution for that particular problem, especially if two or more of the design objectives are
antagonistic. Multi-objective Optimisation (MOO) methods can therefore be implemented in
such scenarios to nd some reasonable compromises in the achievement of design objectives.
MOO procedures enable the optimisation of multiple design variables for several, often antagonistic,
design objectives simultaneously in a constrained design space using mathematical programming
techniques. The generic mathematical formulation of a multi-objective optimisation problem
is to nd a design vector:
→
−
X = [x(1), x(2), . . . , x(n)]T , (4.4)

subject to m inequality constraints:

→
−
gi ( X ) ≤ 0 i = 1, 2, . . . , m

and p equality constraints:

→
−
hi ( X ) = 0 i = 1, 2, . . . , p

to optimise the vector function:

→
− →
− →
− →
−
f ( X ) = [f1 ( X ), f2 ( X ), . . . , fq ( X )]T .
−→∗
As stated before, it is very rare in such problem cases to nd a single X such that

−→ →
− →
−
fi (X ∗ ) ≤ fi ( X ) ∀ X ∈ F, i = 1, 2, . . . , q ,

where F is the feasible region that satises the constraints.

The reasonable trade-os among dierent design objectives that MOO procedures nd in
such scenarios are a type of optimum solutions known as Pareto optimum solutions. Named
−→∗
after Vilfredo Pareto who formulated it, a Pareto optimal is a point X ∈ F that for every
→
−
X ∈F is either

−→ →
−
fi (X ∗ ) = fi ( X ) i = 1, 2, . . . , q

or at least

−→ →
−
fi (X ∗ ) < fi ( X ).
−→∗
In layman's terms, a solution vector X
is a Pareto optimal if there are no other solutions
−→∗
better than it with respect to all design objectives. X needs to be the best solution in
at least one objective to be accepted according to the Pareto optimality criterion. For that
−→∗
reason, X is regarded as a non-dominated solution, and a set of these non-dominated (Pareto
optimal) solutions form a surface known as a Pareto front [88] and [92]. Figure 4.2 shows
an illustration of this in a design space, F (the shaded area) for a bi-objective optimisation
problem. The boundary, in red, between P1 and P2 shows the Pareto front and points x1 and
x3 are non-dominated and hence Pareto optimal solutions, but x2 is not.

The Pareto front illustrates the nature of the trade-os that need to be made in order
to nd good, non-dominated solutions. Multi-Criteria Decision Making methods can then
be implemented to select the best solution from the Pareto front, based on the designer's
preferences or weighting on the various design objectives.
 Stellenbosch University https://scholar.sun.ac.za

CHAPTER 4. OPTIMISATION STRATEGIES 67

Figure 4.2: Illustration of a Pareto optimal solution set from a 2 dimensional performance
space

In the following sections, a multivariable-multiobjective optimisation approach to plant design

layout is presented. Subsection 4.5.1 is dedicated to the algorithm implemented in the MOO
while subsection 4.5.2 pays particular attention to the multi-criteria decision making process
and subsection 5.3.3 addresses the formulation of the MOO problem.

4.5.1 Algorithm
As can be established from section 4.2 there are various evolutionary or stochastic methods
that can be implemented to solve constrained and unconstrained optimisation problems of
a non-linear nature. These methods, e.g Genetic algorithms, Neural Network Particle Swarm
Optimisation and Dierential Evolution, are especially suitable where the mathematical properties
of the problem such as derivability and continuity are dicult to establish. The method selected
for our multi-objective optimisation cases is based on genetic algorithms (GA). GAs are based
on the Darwinian theory of evolution and the principle of the survival of the ttest. This
process is what is used to rene a population of possible solutions in search and optimisation
problems until a working solution or set of solutions is found.

4.5.1.1 Biological Background

The theory of evolution states that in any population, there are distinguishing traits among
individuals that can make certain individuals better adapted to survive and reproduce than
others. These traits, if heritable, can be passed on to ospring and new generations of
better-suited individuals survive while the weaker ones become extinct. The traits are contained
in material called deoxyribonucleic acid (DNA) and each cell of a living organism contains
strings of DNA known as chromosomes. Each chromosome contains a set of genes or blocks of
DNA which, during the process of reproduction, are recombined to form new chromosomes and
in turn new individuals.
Similarly, in genetic algorithms, there is a set of possible solutions in the design space known
as a population. Each population is made up of individuals, which are the GA equivalence of
chromosomes. Fit individuals from the current population are then selected, recombined and
sometimes mutated to form new individuals. This process is repeated until the best individual
or set of individuals is found.

4.5.1.2 Simple Genetic Algorithm

A simplied version of a genetic algorithm goes through the following steps:

1. Generate an initial population,

 Stellenbosch University https://scholar.sun.ac.za

CHAPTER 4. OPTIMISATION STRATEGIES 68

2. Assign tness value (a measure of how close an individual is to the desired solution) to
all individuals,

3. Select individuals from the current generation for reproduction,

4. Produce new ospring by means of mutation and/or recombination,

5. Compute new tness for all individuals and get rid of all unt individuals,

6. Use stopping criteria to check if the best solution or solution set has been found. if not,
repeat steps 3 to 6.

Figure 4.3 is a ow chart of the process as described above.

Start

Generate initial population N

Evaluate fitness value of

every individual

Stopping criteria
satisfied?

No

Apply selection operator

Incorporate new individual into
population

Apply crossover operator

Yes

Apply mutation operator

No
Maximum number of
generations reached?

Yes

Stop

Figure 4.3: Genetic algorithm ow chart

4.5.1.3 Genetic Algorithm Operators

There are three basic operators that are implemented on current generations of GA populations
to produce new populations. These are selection, crossover and mutation.

Selection By this operation, a number of individuals is randomly chosen out of the current
generation and then the ones with the highest tness values are selected out of them.
The better an individual's tness is, the more likely that individual is to be chosen. GAs
usually have several methods to perform this selection operation. Some of the commonly
used ones are:

ˆ Roulette wheel selection

ˆ Rank selection

ˆ Tournament selection

ˆ Steady state selection.

 Stellenbosch University https://scholar.sun.ac.za

CHAPTER 4. OPTIMISATION STRATEGIES 69

The multi-objective genetic algorithm in the MATLAB optimisation toolbox uses tournament
selection in which two random individuals are chosen from the current generation and the
ttest of the two is picked to be a parent in the next population generation. Figure 4.4
shows an example of a tournament selection in which n(=2) chromosomes are chosen from
the current generation and the chromosome with the highest tness value out of the 2 is
selected to be a parent in the process of creating the next generation.

Figure 4.4: Illustration of the tournament selection operation

Crossover The crossover genetic operator is used to vary the programming of chromosomes
from one generation to another. In this process, a new individual is created by the
combination of dierent elements from the parents selected from the previous generation.
Figure 4.5 shows examples of the crossover operation using two methods but there are
several other crossover techniques such as uniform and half-uniform crossover and three
parent crossover, that one can also make use of.

(a) Single-point crossover (b) Two-point crossover

Figure 4.5: An illustration of the crossover operation

Mutation This genetic operator is mainly used to preserve and maintain diversity in the
population, from one generation to another. If there is no diversity in the population, the
evolution of the individuals will slow down and eventually stop at local minima whilst
the algorithm is meant to nd global minima. Mutation is often realised by altering one
or more gene values in a chromosome to dierentiate it from its initial state, often aided
by a predened mutation probability. The key is to keep the probability low such that
the search does not end up being a primitive random search. An easy way to illustrate
mutation is by using a bit string:
10100011 ⇒ 10100010
In this example, the child, on the right hand side of the arrow, is produced by the mutation
of the last bit of the parent, on the left hand side of the arrow, and the bit that is
mutated is known as the mutant gene. Bit string mutation is but one example of several
mutation techniques that can be applied to GAs. Other methods include ip bit, uniform,
non-uniform, gaussian and shrink mutation.
 Stellenbosch University https://scholar.sun.ac.za

CHAPTER 4. OPTIMISATION STRATEGIES 70

4.5.2 Multiple Criteria Decision Making (MCDM)

MCDM is a part of operations research that deals with multiple, conicting criteria in decision
making. This is almost inevitable when it comes to multi-objective optimisation as there might
be solutions which are the best in at least one dimension but not a great solution in other
dimensions. A simple illustration of such a scenario is the performance space shown in Figure
4.6. Solution A is better with respect to objective F1 but worse with respect to F2 , while
solution B is better with regards to objective F2 but worse with respect to F1 . If the Pareto
criterion were to be applied to this performance space, the Pareto front would be made up
of solutions A and B and it would not be easy to select the best solution of the biobjective
problem out of the two. With increased dimensionality and complexity of relationships between
the objectives, there is also likely to be an increase in the number of non-dominated solutions
in the Pareto Set. Deciding on the best option out of all the provided solutions would therefore
denitely take a dierent turn. Methods such as prioritisation and weighting of some of the
objectives over others would therefore be helpful, and that is what MCDM is about.

2-dimensional Performance Space

F2

F1

Figure 4.6: 2-dimensional performance space from a bi-objective optimisation problem

There are several approaches to MCDM. The list includes the Simple Additive Weighting
(SAW), the TOPSIS method, Analytical Hierarchical Processes (AHP) and Goal Programming.
One of the most popular of these methods is TOPSIS which stands for Technique for Order
Preference by Similarity to Ideal Solution. The TOPSIS method is ideal for identifying the
solution that is the most attractive over all dimensions and it is applicable even if the dimensions
have incommensurable units [93]. Some of the common implementations of TOPSIS in multiobjective
optimisation include the multi-objective ecodesign of PVGCS in [94], chemical engineering
applications in [95] and the multi-objective optimisation for multicontaminant industrial water
network design in [96].

4.5.2.1 TOPSIS
The TOPSIS method selects the solution based on proximity to two articial solutions that it
hypothesises:
 Stellenbosch University https://scholar.sun.ac.za

CHAPTER 4. OPTIMISATION STRATEGIES 71

ˆ Positive Ideal Solution (PIS) - which constitutes the best attributes in all the dimensions,
and

ˆ Negative Ideal Solution (NIS)- which has the worst level of all the attributes across all
the dimensions.

The following steps show the stages of the TOPSIS method and how they are implemented:
Step 1: Construction of the decision matrix.
In this step, a matrix of m alternatives by n criteria is created.

Criteria

C1 C2 ... Cn

A1 x11 x12 ... x1n

A2 x21 x22 ... x2n
Alternatives

.
.
.

Am xm1 xm2 ... xmn

Table 4.1: Decision matrix

Step 2: Creation of normalised decision matrix N.

Each objective or criteria might come in its own unit of measurement which is dierent from
the next one and this makes comparisons across criteria very dicult. These are what are
referred to as incommensurable units. The purpose of this step is therefore to transform every
value in the decision matrix into a normalised, non-dimensional value between 0 and 1 so that
the dierent criteria can be compared. There are several ways to perform the normalisation
procedure and Milani et al [97] and Celen [98], amongst other authors, have done studies into
them. Celen's comparative studies to analyse normalisation procedures conclude that the vector
normalisation procedure in Equation 4.5 generates the most consistent results. [98]. This is the
method that was selected as it is one of the most commonly used. The normalisation equation
is given as:

x0ij
N = [zij ]m∗n , zij = qP , j = 1, . . . , n and i = 1, . . . , m. (4.5)
m 0 2
(x
i=1 ij )

x0ij = xij for all benet attributes and

1
xij
x0ij =
for all cost attributes.

The term benet attributes refers to the criteria where the higher the value, the better is
whereas cost attributes are criteria where less is better.

Step 3: Creation of weighted matrix V.

Weights of importance are assigned to each criteria, relative to others, and a weighted matrix,V,
is created by multiplying elements in the normalised matrix N by the corresponding criterion
weight:
V = [vij ]m∗n = [wj ] ∗ [zij ]m∗n j = 1, . . . , n and i = 1, . . . , m. (4.6)
Pn
where wj is the weight of the j -th criterion, j=1 wj .
 Stellenbosch University https://scholar.sun.ac.za

CHAPTER 4. OPTIMISATION STRATEGIES 72

Step 4: Determination of the positive and negative ideal solutions (PIS and NIS).

A+ = {v1+ , . . . , vn+ } where vj+ = {(max(vij ) if j ∈ J)}, (min(vij ) if j ∈ J 0 )} (4.7)

i i

A− = {v1− , . . . , vn− } where vj− = {(min(vij ) if j ∈ J)}, (max(vij ) if j ∈ J 0 )} (4.8)

i i

where J is the index set of benet attributes and J' is the index set of cost attributes.
Step 5: Calculation of the Euclidean distance.
The measure of the separation of each alternative from the PIS and NIS is determined using
the euclidean distance. The respective distances from the PIS and NIS are calculated as:

v
u n
uX
di = t (vij − vj+ )2 ,
+
i = 1, . . . , m. (4.9)
j=1

v
u n
uX
di = t (vij − vj− )2 ,
−
i = 1, . . . , m. (4.10)
j=1

Step 6: Determination of the relative closeness (closeness coecient) to the ideal solution.
The relative closeness of each alternative to the ideal solution is calculated as:

d−
i
c+
i = , i = 1, . . . , m. (4.11)
d+
i + d−
i

where 0 ≤ c+
i ≤ 1.
The higher the value the value of c+
i the higher the rank of that alternative and the more
favourable it is. This is essentially one way of saying that the ideal solution is the one solution
that is suciently close to the PIS, while suciently distant from the NIS.

4.6 Conclusion
In this chapter, the optimisation strategy was broken down and each vital step was explained
in more detail to shed some light as to its importance in the optimisation process and what
each step entails. It was noted that evolutionary computing could be quite useful in the
optimisation of the PV system design process given the number of design objectives that need
to be evaluated simultaneously. The optimisation procedure also ideally ought to take into
account the complex relationships of various design objectives in response to changes in the
design variables, while ensuring that the process adheres to constraints in the design space,
something that several functions in the MATLAB optimisation toolbox are capable of. Based on
the problem cases under investigation, the appropriate optimisation algorithms in the MATLAB
Toolbox were selected and implemented in conjunction with a reliable PV system simulation
software program, SAM, to perform multivariable-single- and multi-objective optimisation of
the system design while concurrently reducing the redundancy in some of the modelling steps
and improving the speed, accuracy and precision of the design variable selection. An extra
step was introduced in the multiobjective optimisation case to assist with the selection of the
best solution from the non-dominated solution set returned by the genetic algorithm. The step
uses the TOPSIS method to perform the multi-criteria decision making (MCDM) process. This
MCDM procedure was preferred as it is fairly easy to implement and successfully navigates its
way around the challenges of incommensurable units and increased dimensionality presented
by the multiple design objectives in the problem cases.
 Stellenbosch University https://scholar.sun.ac.za

Chapter 5
Results and Analyses

5.1 Introduction
In this chapter, the proposed methodology for multivariable optimisation is evaluated by means
of some design examples. First, two single-objective problems are investigated to nd the
optimal layouts of the PV modules, characterised by a combination of the tilt angle, azimuth
angle and ground cover ratio and sometimes the rotational limit, that result in maximum annual
energy output (Qout ), maximum performance ratio of the plant and minimum system losses. A
detailed outline of each problem's formulation as well as its optimisation results are given and
three evaluations are given for each problem, to cater for the three PV module technologies in
our investigation.

In a similar fashion, two multi-objective optimisation cases are also investigated. The
optimisation is achieved by means of a multi-objective genetic algorithm and each genetic
algorithm run returns a set of layouts that are best at optimising one or more objectives in
one way or another. The dierent layouts are then ranked from the best to the worst using a
multicriteria decision making (MCDM) method known as TOPSIS, which stands for Technique
for Order of Preference by Similarity to Ideal Solution. Using the TOPSIS method, the layouts
by each module technology are compared to each other and ranked, and the same is done for
the best layout of each module technology to nd out which technology yields the best results.

Another design example is formulated to illustrate how multiple objectives can be taken
into consideration simultaneously, but with dierent weights to resemble precedence of certain
design objectives over others. This is because multi-objective design problems rarely have
equally important design goals. Some objectives are usually more important than others, and
weighting can help take that into account.

The structure of the chapter is as follows: single-objective optimisation is covered in Section

5.2 and multi-objective optimisation problems are covered in Section 5.3. Section 5.4 will then
summarise the ndings of the optimisation framework investigated in this chapter.

5.2 Single objective optimisation

5.2.1 Solver Options

The execution of the fmincon function required additional options to be selected. The purpose
of these options is to facilitate the convergence of simulation runs towards optimal solutions

73
 Stellenbosch University https://scholar.sun.ac.za

CHAPTER 5. RESULTS AND ANALYSES 74

and can be summarised as follows:

ˆ The initial starting point of the iterations was set to be: θT = 0◦ , φC = 0◦ and a
GCRinv = 1.1111

ˆ The 'approximated derivatives' option was set such that the function would search for
central dierences in successive iterations as opposed to forward dierences for better
accuracy. The penalty for the central dierences option was that it would be relatively
slower than the forward dierences option, but with the central dierences option the
average simulation run was seven seconds which is fast enough and, in fact, signicantly
faster than other normal iterative methods.

ˆ The stopping criteria were set as summarised in Table 5.1

Table 5.1: fmincon stopping criteria

Criteria Value

Maximum number of iterations 1000

Maximum function evaluations 3000

X tolerance 10−10
Unbounded threshold 10−20
Function tolerance 10−4
Constraint tolerance 10−3

5.2.2 Case 1: Fixed Tilt System

5.2.2.1 Problem formulation
→
−
In this particular case, the objective was to nd the optimal design layout vector X to maximise
annual energy fed into the grid, maximise the performance ratio and minimise system losses of
a xed tilt PV system on a site in the Northern Cape Province of South Africa. The objective
functions, therefore, were:

→
−
1. Maximise total annual energy yield, Qout,max : f1 ( X )

→
−
2. Maximise performance ratio, P Rmax : f2 ( X )
→
−
3. Minimise system losses (DC + AC losses), System Lossesmin : f3 ( X )

and they were executed separately using the fmincon solver in MATLAB.

The design variables, module tilt angle, module azimuth angle and ground cover ratio
(GCR), were passed to the optimisation tool in the form of a vector. As a general rule, as
described in the fundamental background chapter, the optimal tilt and azimuth angles ought
to be selected to facilitate maximum energy capture by the PV modules while the optimal GCR
 Stellenbosch University https://scholar.sun.ac.za

CHAPTER 5. RESULTS AND ANALYSES 75

needs to be selected to prevent reductions in energy capture due to shading or wastage of land
area available for the PV plant set-up. The design vector of the xed tilt problem case was:
   


 θT 




 x(1)


→
−    
X = φC = x(2) , (5.1)

 
 
 


GCR 
 x(3)
inv

where,
θT = module surface tilt angle
φC = module surface azimuth angle
GCRinv = the inverse of the GCR.
x(3) is set to be the inverse of the GCR value and not the GCR so that it ts perfectly with the
inequality formulation which is of the form A ∗ x ≤ b as described in the Optimisation Strategy
Chapter.

Constraints
The constraints of the problem were:

ˆ Linear inequalities
This constraint arose from the restriction in the land area available for the PV system
layout design. The test case is an actual plant which occupies 105 hectares so the
maximum land area available for the design layout was set to equal that. In SAM,
land area is calculated as:
T MA
LandArea = ∗ 0.0002471
GCR
(5.2)
= T M A ∗ GCRinv ∗ 0.0002471

= T M A ∗ 0.0002471 ∗ GCRinv
where:
TMA is the total module area and the constant, 0.0002471, converts square meters to
acres in equation 5.2 since computation of system costs in SAM is done by using the cost
of land in dollars per acre. Consequently, the maximum available land area constraint
also needed to be converted and set to 260 acres. The A matrix in the formulation of the
land area inequality has three elements:

 A[0] = 0, for the tilt angle,

 A[1] = 0, for surface azimuth and

 A[2] = T M A*0.0002471.
The T MA varies from one PV module model to another so the A[2] values are derived
as follows:

ˆ Lower and upper bounds

The upper and lower bounds of the design variables were set to put some practicality
◦ ◦
limits to the design space. According to SAM the θT must be a value between 0 and 90
◦ ◦
and the φC must be an angle between 0 and 360 so the constraints were set as shown
in Table 5.4. The GCR must be a value between 0.01 and 0.99 according to SAM but we
decided on a range a range starting from 0.1 to 0.9 since most practical setups actually
estimate the ideal range of GCR to be from 0.3 to 0.7 in most xed tilt case [99]. The
values of these side constraints were all the same for the dierent PV module technologies
and are as shown in Table 5.4.
 Stellenbosch University https://scholar.sun.ac.za

CHAPTER 5. RESULTS AND ANALYSES 76

Table 5.2: Inequality constraint parameters

2 2
Module Technology Module Area(m ) Number of Modules TMA (m ) A[2]

MonoSi 1.852 245988 455569.8 112.6

PolySi 2.126 237932 505843.4 124.995

Thin Film 0.72 667610 480679.2 118.6

Table 5.3: Inequality constraint parameters

A
Module Technology b
A[0] A[1] A[2]

MonoSi 0 0 112.6 260

PolySi 0 0 124.995 260

Thin Film 0 0 118.6 260

Table 5.4: Lower and upper bounds of solar PV system design variables

Bounds θT φC GCRinv
Lower 0 0 1.111

Upper 89 359 10

5.2.2.2 Results
The fmincon constrained minimisation function was applied as described in Subsubsection
5.2.2.1 for each of the three PV module technologies. The entire design space was evaluated
in about seven seconds, on average, in each of the individual single objective optimisation runs
and the three optimal layouts, one for each module technology, were compiled and tabulated
as shown in Table 5.5. The results show the layouts that lead to the maximisation of the total
annual energy production in each of the three PV module technologies' cases. Analysis of the
performance of the dierent technologies showed that Thin Film modules performed the best,
followed by the mono-crystalline silicon modules and then polycrystalline modules in the given
design example.

Table 5.5: Fixed-tilt total annual energy output, Qout , optimisation results

Design Parameters
Module Technology Qout (kW h)
θT φC GCRinv
MonoSi 28.693 1.26 2.309 1.55 ·108
PolySi 27.97 0.298 2.08 1.51 ·108
Thin Film 30.523 0.099 2.192 1.58 ·108
 Stellenbosch University https://scholar.sun.ac.za

CHAPTER 5. RESULTS AND ANALYSES 77

The same process of constrained minimisation was performed to nd orientations that would
lead to maximised performance ratios (PR) for the three PV module technologies being investigated.
Each simulation run was also seven seconds long, on average, and the results of each PV module
technology's constrained minimisation were compiled and documented as shown in Table 5.6.
The results show the design parameters that would result in a module layout optimised for
maximum PR in each module technology's case and compares the nal maximised PR values
of these technologies against each other. From the results, it can be concluded that the Thin
Film modules led to the highest PR, followed by monocrystalline and then lastly polycrystalline
modules. In all cases, the orientations resulting in maximised PR were characterised by low
tilt and azimuth angles and relatively high ground cover ratios of approximately 50% of the
installation area.

Table 5.6: Optimisation results for maximum PR in MonoSi, PolySi and Thin Film PV module
technologies

Design Parameters
Module Technology PR
θT φC GCRinv
MonoSi 0.99 0.99 2.101 0.839

PolySi 1.081 1.081 2.071 0.83

Thin Film 0.99 0.99 2.101 0.8457

The constrained minimisation procedure was performed to nd orientations that would lead
to the minimisation of system losses in the same xed tilt PV system and again, simulation
runs were performed for the three PV module technologies being investigated. The average
simulation run for each technology was approximately seven seconds and the results were
compiled and documented as shown in Table 5.7. The results show the design parameters
that would result in a module layout optimised for minimum system losses in each module
technology's case and compares the nal minimised system loss values of these technologies
against each other. From the results, it can be concluded that the Thin Film modules led
to the lowest total system losses, followed by monocrystalline and then lastly polycrystalline
modules. In all cases the orientations resulting in minimised system losses were characterised
by low tilt and azimuth angles and relatively high ground cover ratios of approximately 50% of
the installation area.

Table 5.7: Optimisation results for minimum system losses in MonoSi, PolySi and Thin Film
PV module technologies

Design Parameters
Module Technology System Losses (%)
θT φC GCRinv
MonoSi 1.872 6.848 2.101 13.31

PolySi 0.303 1.168 1.841 15.465

Thin Film 0.99 0.99 2.101 12.57

 Stellenbosch University https://scholar.sun.ac.za

CHAPTER 5. RESULTS AND ANALYSES 78

The optimal layouts for total energy output, performance ratio and system losses were
compared in terms their satisfaction of each of the objectives investigated in Case 1. The results
were also compared to simulation results generated by mounting polySi modules according to
the test site's tilt angle, azimuth angle and GCR. These comparisons were drawn only using
polySi modules because that is the technology currently in use at the test site. The idea was to
illustrate how close or far o the existing plant layout, as well as the layouts optimised for the
other two objectives were, from the actual optimised values and graphs of these comparisons
are given in Figure 5.1(a-d). As can be deduced from the graphs, there is denitely room
for improvement in the existing plant layout's case to get to the optimal values of the three
design objectives. Furthermore, the results show that optimising the design layout for one of
the objectives does not necessarily optimise the layout to satisfy the other two objectives. To
simultaneously satisfy the dierent objectives, multi-objective optimisation techniques would
need to be implemented and this is discussed in Section 5.3.

Pe r fo r m a n ce Rat i o
Total Annual Yield
1.512

0.83
0.827
Total Annual Energy Yield (*108 kWh)

1.48

Performance Ratio
1.47

0.808

0.801
1.44

Current plant layout Q_max layout PR_max layout Loss_min layout Current plant layout Q_max layout PR_max layout Loss_min layout

(a) Total Annual Energy Output comparison (b) Performance Ratio comparison

System Losses
Comparison of the Layouts
18.34
18.05

27.97
30
30
System Losses (%)

25

20
16.48

0.303 1.168 0.543

15 Loss_min layout
1.081 1.081 0.482 PR_max layout
10
15.465

0.298 0.48 Q_max layout

5
0 0.4086
Current plant layout
0

Current plant layout Q_max layout PR_max layout Loss_min layout

(c) System Losses comparison (d) Layouts under investigation

Figure 5.1: Comparison of the performance of PolySi module technologies with respect to each
objective for the xed tilt case, Case 1

5.2.3 Case 2: Single Axis Tracking System

5.2.3.1 Problem formulation
→
−
In this particular case, the objective was to nd the optimal design layout vector X to maximise
annual energy fed into the grid, maximise PR and minimise system losses for a single-axis
 Stellenbosch University https://scholar.sun.ac.za

CHAPTER 5. RESULTS AND ANALYSES 79

tracking PV system on a site in the Northern Cape Province of South Africa. The objective
functions for single axis tracking systems were the same as in the xed tilt case: The objective
functions, therefore, were:

→
−
1. Maximise total annual energy yield, Qout,max : f1 ( X )

→
−
2. Maximise performance ratio, P Rmax : f2 ( X )
→
−
3. Minimise system losses (DC + AC losses), System Lossesmin : f3 ( X )

and they were executed separately using the fmincon solver.

The design parameters vector, however, was made up of the module tilt angle, the module
azimuth angle, the tracker rotation limit and ground cover ratio (GCR). The tracker rotation
limit gives the maximum or minimum allowable rotation angle for the one-axis tracker systems
in SAM. A value of θRot degrees is such that the tracker is allowed to rotate θRot degrees about
the center line in both directions from the horizontal [5]. The design vector of the single-axis
tracking problem case was:

   


 θT 




 x(1) 



 
 
 

  x(2)

→
−  φC  
X = = , (5.3)


 θRot 




 x(3)



 
 
 


GCR   
inv
  x(4)
where,
θT = module surface tilt angle
φC = module surface azimuth angle
θRot = tracker rotation limit
GCRinv = the inverse of the GCR.

Constraints
The constraints of the problem were set up as follows:

ˆ Linear inequalities
The test case was adopted from the size of an actual 75 MWp single-axis tracker plant
which occupies 225 hectares so the maximum land area available for the design layout
was set to 556 acres.

Table 5.8: Inequality constraint parameters

A
Module Technology b
A[0] A[1] A[2] A[3]

MonoSi 0 0 0 112.6 556

PolySi 0 0 0 124.995 556

Thin Film 0 0 0 118.6 556

 Stellenbosch University https://scholar.sun.ac.za

CHAPTER 5. RESULTS AND ANALYSES 80

ˆ Lower and upper bounds

The upper and lower bounds of the design variables were set up the same way as in
the xed tilt case (see Table 5.4). The upper and lower bounds of the tracker rotation
◦ ◦
limit were set to 5 and 85 as they are the minimum and maximum allowable values,
respectively. Table 5.9 shows a summary of the upper and lower bounds for the one-axis
tracker problem.

Table 5.9: Lower and upper bounds on solar PV system design variables

Bounds θT φC θRot GCRinv

Lower 0 0 5 1.111

Upper 89 359 85 10

5.2.3.2 Results
Constrained minimisation was performed on the tracking systems to determine the technology
and module layout that would lead to a maximised energy production on the 75 MWp test-plant.
Table 5.10 shows the simulation results, with monocrystalline technology leads to the highest
energy output, followed by Thin Film technology and lastly polycrystalline modules, in the
same design space.

Table 5.10: Single-axis tracking total annual energy output, Qout , optimisation results

Design Parameters
Module Technology Qout (kW h)
θT φC θRot GCRinv
MonoSi 36.04 355.5 40.1 4.675 2.03 ·108
PolySi 26.5 10.66 43.06 3.706 1.99 ·108
Thin Film 33.7 2.48 63.4 4.68 2.09 ·108

Results of design parameters for a maximised PR for each module technology were compiled
and tabulated as given in Table 5.11. The results also show that for systems with the same
nameplate capacities, monocrystalline technology had the highest PR, followed by Thin Film
technology and then polycrystalline modules would have the worst.

Table 5.11: Single-axis tracking performance ratio, PR, optimisation results

Design Parameters
Module Technology PR
θT φC θRot GCRinv
MonoSi 88.33 87.5 113.45 1.69 0.877

PolySi 88.7 88.9 16.8 2.01 0.845

Thin Film 87.3 4.05 61.03 4.92 0.86

 Stellenbosch University https://scholar.sun.ac.za

CHAPTER 5. RESULTS AND ANALYSES 81

Constrained optimisation to determine orientations which would result in minimum system

losses was performed. The results generated from the optimisation were compiled and documented
as shown in Table 5.12. From the results, it can be deduced that the Thin Film technology would
result in the lowest system losses, followed by monocrystalline and lastly the polycrystalline
technology.

Table 5.12: Single-axis tracking performance ratio, PR, optimisation results

Design Parameters
Module Technology System Losses (%)
θT φC θRot GCRinv
MonoSi 1.081 1.081 6.176 2.116 15.88

PolySi 2.12 6.491 12.826 3.54 16.79

Thin Film 2.443 2.183 7.488 1.363 14.128

5.3 Multi-objective optimisation

→
−
The purpose of this optimisation was to nd the optimal design layout vector, X, to optimise
three plant performance indicators simultaneously for both a xed tilt PV system and a single
axis tracking system using a site in the Northern Cape Province of South Africa as a test site. In
the formulation of multi-objective optimisation problems, certain aspects, namely the objective
function and the solver options of the algorithm, remained unchanged for the dierent cases
investigated and were given as in Subsection 5.3.1. The rest of the features of the problems,
which diered from one case to the other, are then given within the corresponding case's problem
formulation in the subsequent subsections.

5.3.1 Solver Options

The execution of the multi-objective optimisation genetic algorithm needed additional options
to be selected. The purpose of these options is to facilitate the convergence of simulation runs
towards optimal solutions and can be summarised as:

ˆ The initial population was created using the 'Feasible population' function which creates
a random population which satises the bounds and the linear constraints provided by
the user.

ˆ The selection operator used the 'Tournament' function. This function selects two random
individuals from the current generation and chooses the best individual out of the two to
be a parent.

ˆ The reproduction operator used a crossover fraction of 0.8 to create children at each new
generation. As such, 80% of each successive generation was created from crossover.

ˆ Mutation was responsible for the creation of the remaining 20% of the children at each
new generation and was facilitated by the 'Adaptive feasible' function. This function
randomly generates adaptive directions for progression based on the last successful or
unsuccessful generation.
 Stellenbosch University https://scholar.sun.ac.za

CHAPTER 5. RESULTS AND ANALYSES 82

ˆ The Pareto front population fraction is kept at 0.35 to keep the most t population down
and maintain diversity in the population and is facilitated by the in-built 'distancecrowding'
function. Population diversity, as described in earlier sections, is crucial to the fruitful
exploration of a GA to ensure that the population converges but does not do so prematurely.

ˆ The stopping criteria were set as summarised in Table 5.13

Table 5.13: MOGA stopping criteria

Criteria Value

Maximum number of generations 1000

Time limit ∞
Fitness limit -∞

Stall time limit ∞

Stall generations 100

Function tolerance 10−4

Constraint tolerance 10−3

5.3.2 Choice of design objectives

The multi-objective framework was formulated in such a way that it could be used for the
simultaneous optimisation of several, and sometimes antagonistic, plant performance indicators
with the hopes of expanding the focus of the design procedure to look at more than one
objective at a time. Unfortunately, at the time the work was done there was not enough for the
levelised cost of energy (LCOE) to be included as one of the design objectives as cost-breakdown
information for a practical nancial model that exemplies current industry trends and practices
was not available. This was mainly because the solar market in Africa is still fairly small so
condentiality is extremely important in such matters, as projects can be easily identiable. As
a result data on cost breakdowns are hardly readily availed by solar developers in the country
[100]. In the end, a design example featuring land use, performance ratio and energy output as
design objectives was used to present the framework.

5.3.3 Case 1 - Fixed Tilt, Equal Weights

5.3.3.1 Objective Function Formulation
The xed tilt multi-objective problem was characterised by three objective functions which
were:

→
−
1. Maximise total annual energy yield, Qout : f1 ( X )
→
−
2. Minimise land area usage, Land: f2 ( X )
→
−
3. Maximise performance ratio, P R : f3 ( X )
 Stellenbosch University https://scholar.sun.ac.za

CHAPTER 5. RESULTS AND ANALYSES 83

The resulting design objective vector was:

→
− 
   
Qout
  f ( X ) 
 1

 
  

→
− 
→
−
 
F ( X ) = Landarea = f2 ( X ) , (5.4)
 →
f (−

 
  

 PR 
   
3 X)


5.3.3.2 Problem Formulation

Design parameters
The design parameters vector was the same as the one for the single-objective optimisation
problem case, given in Subsubsection 5.2.2.1 for xed tilt systems. The design vector is
characterised by the three design parameters as:
   


 θT 




 x(1)


→
−    
X = φC = x(2) , (5.5)

 
 
 


GCR 
 x(3)
 
inv

where,
θT = module surface tilt angle
φC = module surface azimuth angle
GCRinv = the inverse of the GCR.

TOPSIS Parameters
The application of the TOPSIS method requires that each design objective be assigned the
correct criteria sign and a weight, λ. The criteria sign range is a positive unit (+1), if the design
objective is to be maximised, and a negative unit (-1) if the design objective is to be minimised.
The initial investigation assumed equal weighting of a unit for each objective. Table 5.14
summarises the criteria and weights for this particular investigation's TOPSIS implementation.

Table 5.14: Criteria and weights for the TOPSIS method

Qout Land area usage PR

Criteria sign +1 -1 +1

Weight 1 1 1

Constraints
The corresponding bounded constraints were formulated the same way as described in Subsubsection
5.2.2.1 and are summarised in Table ??. The linear inequality constraint as applied to the
single-objective problems was not used in the multi-objective cases because land usage was to
be used as a design objective instead.
 Stellenbosch University https://scholar.sun.ac.za

CHAPTER 5. RESULTS AND ANALYSES 84

Table 5.15: Lower and upper bounds of solar PV system design variables

Bounds θT φC GCRinv
Lower 0 0 1.111

Upper 89 359 10

5.3.3.3 Results
As described in the optimisation strategy chapter, the multi-objective optimisation for xed
tilt systems was executed by means of a multi-objective genetic algorithm (MOGA). The
antagonism in the various objectives being designed for meant that no single layout could lead
to the optimisation of all the objectives at once so a Pareto front was generated. The Pareto
front represented several alternative ways of laying out the arrays in each module technology's
case in a way that would be benecial to the achievement of one or more of the three design
objectives. Each layout in the solution set was dened by the tilt angle, azimuth angle and
the ground cover ratio, GCR, (given as an inverse, GCRinv for the purpose of the simulations,
refer to Subsubection 5.2.2.1) and shows the corresponding inuence the parameters have on
the three design objectives.

The TOPSIS method was then used to nd an acceptable compromise in the layouts such
that the selection of the nal layout would be based on each alternative's proximity to the ideal
solution and distance from the non-ideal solution. The solutions were ranked from the best using
the relative closeness coecient as described in the TOPSIS nethod outline in the Optimisation
Strategy Chapter. It is worth mentioning again that in this evaluation, as indicated in the
problem formulation, the importance of all objectives was assumed to be the same so all the
objectives were weighted equally. Tables 5.16, 5.17 and 5.18 show the Pareto fronts of the
monocrystalline, polycrystalline and thin lm module technologies, respectively.

An additional step was then taken to select the best performing layouts from each module
technology's Pareto front and pit them against each other to determine the best performing
module technology in this design scenario. Table 5.19 shows the resulting ranking of the
three dierent technologies after they were put through the TOPSIS method. The Thin Film
technology performed the best out of the three, followed by the monocrystalline technology and
then lastly the polycrystalline technology.
 Table 5.16: Fixed Tilt MonoSi Pareto front
CHAPTER 5.

θT φC GCRinv Qout,year (kWh) Land area used PR Closenesss Coe Rank

0,537667 1,833885 1,1111 1,37E+08 125,078 0,828987 0,996688 1

0,11285 1,833885 1,1111 1,37E+08 125,078 0,829139 0,996595 2

5,836793 0,33445 1,173611 1,41E+08 132,1149 0,815404 0,992478 3

39,03782 4,211961 2,175078 1,52E+08 244,8513 0,818444 0,88021 4

2,409527 87,42346 3,286091 1,37E+08 369,9195 0,829873 0,755103 5

21,96655 25,43673 4,042909 1,52E+08 455,1155 0,830035 0,669924 6

RESULTS AND ANALYSES

27,80457 341,3956 4,391459 1,54E+08 494,3522 0,826861 0,630687 7

7,930363 100,9917 4,419519 1,35E+08 497,511 0,831378 0,627498 8

53,8891 36,7796 6,278051 1,4E+08 706,7283 0,836357 0,418276 9

44,47812 25,1907 6,897904 1,49E+08 776,506 0,833371 0,348525 10

58,30252 53,38639 7,530551 1,3E+08 847,7239 0,836904 0,277252 11

Stellenbosch University https://scholar.sun.ac.za

77,46206 43,17247 7,829544 1,13E+08 881,3819 0,844139 0,243555 12

64,1818 17,19639 7,947074 1,34E+08 894,6124 0,837465 0,230379 13

63,17443 60,14977 8,248634 1,23E+08 928,5594 0,838721 0,196395 14

85

71,35178 53,80541 8,98052 1,18E+08 1010,949 0,842703 0,113989 15

82,01798 47,62673 9,945921 1,07E+08 1119,625 0,847102 0,005277 16

81,90861 47,63308 9,992796 1,07E+08 1124,902 0,847163 3,2E-05 17

 Table 5.17: Fixed Tilt PolySi Pareto front

θT φC GCRinv Qout,year (kWh) Land area used PR Closenesss Coe Rank

CHAPTER 5.

0,537741 1,833397 1,1111 1,36E+08 138,8807 0,818096 0,992065 1

28,29722 3,860474 1,244104 1,36E+08 155,5054 0,725212 0,983087 2

24,75675 5,836019 1,516114 1,46E+08 189,5051 0,784854 0,954327 3

28,30241 2,231479 2,197376 1,52E+08 274,6587 0,813254 0,87783 4

39,20124 38,6273 4,006285 1,45E+08 500,7612 0,81696 0,674376 5

64,68456 5,674133 4,46337 1,32E+08 557,894 0,820813 0,62286 6

23,42938 53,17622 4,744507 1,44E+08 593,0344 0,820625 0,59136 7

RESULTS AND ANALYSES

56,02497 0,932095 5,393802 1,42E+08 674,1924 0,821583 0,518347 8

71,89996 21,94978 5,908869 1,22E+08 738,5726 0,825497 0,460228 9

83,05503 47,52825 6,949164 1,03E+08 868,6031 0,826391 0,34309 10

78,54437 49,37105 7,37889 1,08E+08 922,3163 0,82862 0,294813 11

48,30448 45,19499 7,546371 1,39E+08 943,2504 0,824498 0,27644 12

Stellenbosch University https://scholar.sun.ac.za

80,34592 55,92698 8,137169 1,05E+08 1017,096 0,830328 0,20952 13

74,8226 45,1918 8,792787 1,15E+08 1099,045 0,831131 0,13594 14

31,86059 0,047765 9,982456 1,54E+08 1247,746 0,821109 0,023654 15

86

84,76675 58,83206 9,938267 98740368 1242,223 0,832901 0,00695 16

79,96354 57,64601 9,999997 1,05E+08 1249,939 0,833491 0,002761 17

79,96354 57,64606 9,999993 1,05E+08 1249,938 0,833491 0,002761 18

 Table 5.18: Fixed Tilt Thin Film Pareto Front

θT φC GCRinv Qout,year (kWh) Land area used PR Closenesss Coe Rank

CHAPTER 5.

27,325 0,084147 1,111208 1,53E+08 121,2701 0,815541 0,996733 1

0,537667 1,833885 1,1111 1,4E+08 121,2583 0,845734 0,989613 2

20,21536 44,58465 1,620504 1,48E+08 176,8513 0,834797 0,942366 3

2,780583 42,33594 1,616985 1,42E+08 176,4673 0,846024 0,942297 4

39,36899 6,912212 2,097899 1,56E+08 228,9512 0,842332 0,888934 5

28,89885 1,69588 3,153415 1,58E+08 344,1434 0,843911 0,770195 6

80,70094 38,51896 3,707639 1,11E+08 404,6277 0,855992 0,706633 7

RESULTS AND ANALYSES

54,37466 42,24433 4,011111 1,4E+08 437,7468 0,849752 0,673438 8

87,21361 72,61783 4,601405 93928168 502,1676 0,857782 0,605728 9

83,2639 44,81732 5,70721 1,08E+08 622,8481 0,864582 0,481912 10

41,3404 45,04646 6,33199 1,47E+08 691,0324 0,848151 0,412783 11

65,59724 32,29402 7,053186 1,33E+08 769,7391 0,859228 0,331358 12

Stellenbosch University https://scholar.sun.ac.za

31,39818 0,024528 7,874206 1,58E+08 859,34 0,844238 0,241043 13

66,38543 58,35785 8,472173 1,24E+08 924,5983 0,862137 0,172015 14

78,81813 73,67347 8,922294 1,04E+08 973,7215 0,867816 0,120456 15

87

88,59699 92,18474 9,238952 83788152 1008,279 0,871105 0,084248 16

88,82833 72,52951 9,987533 93686824 1089,975 0,872834 0,005719 17

88,95333 72,52951 9,987533 93541896 1089,975 0,872853 0,005636 18

 CHAPTER 5.

Table 5.19: Fixed Tilt Ranking of the best congurations from the 3 dierent technologies
RESULTS AND ANALYSES

Technology θT φC GCRinv Qout,year (kWh) Land area PR Closenesss Coe Rank

MonoSi 0,537667 1,833885 1,1111 1,37E+08 125,078 0,828987 0,53475 2

PolySi 0,537741 1,833397 1,1111 1,36E+08 138,8807 0,818096 0,000269 3

Thin Film 27,325 0,084147 1,111208 1,53E+08 121,2701 0,815541 0,998583 1

Stellenbosch University https://scholar.sun.ac.za

88
 Stellenbosch University https://scholar.sun.ac.za

CHAPTER 5. RESULTS AND ANALYSES 89

5.3.4 Case 2 - Single Axis Tracking, Equal Weights

5.3.4.1 Problem Formulation
5.3.4.2 Objective Function Formulation
The single axis tracking multi-objective problem was characterised by three objective functions
which were:
→
−
1. Maximise total annual energy yield, Qout : f1 ( X )
→
−
2. Minimise land area usage , Land area: f2 ( X )

→
−
3. Maximise performance ratio, P R: f3 ( X )

This is because the tracking systems make use of backtracking and this deals with the PV
system's self-shading loss problems.

The resulting design objective vector was:

→
− 
   


 Qout


 f 1 (


 X )

→
−   
→
−

F ( X ) = Landarea = f2 ( X ) , (5.6)
 →
f (−

 
  

 PR 
   
X ) 3

Design parameters
The design parameters vector was the same as that of the tracking case given in Subsubsection
5.2.3.1. The design parameters vector was, therefore:
   
θT


 x(1)

 





 
 
 

   
→
− φ

C
 x(2)
 
X = = , (5.7)


 θRot  

x(3)





  
  


GCR    x(4) 
inv

where,
θT = module surface tilt angle
φC = module surface azimuth angle
θRot = tracker rotation limit
GCRinv = the inverse of the GCR.

TOPSIS parameters
Table 5.20 summarises the criteria and weights for this particular problem's TOPSIS implementation:

Table 5.20: Criteria and weights for the TOPSIS method

Qout Land area PR

Criteria sign +1 -1 +1

Weight 1 1 1
 Stellenbosch University https://scholar.sun.ac.za

CHAPTER 5. RESULTS AND ANALYSES 90

Table 5.21: Lower and upper bounds of solar PV system design variables

Bounds θT φC θRot GCRinv

Lower 0 0 5 1.111

Upper 89 359 85 10

Constraints
The constraints of the problem were set up as shown in Table 5.21:

5.3.4.3 Results
Pareto fronts generated from the multi-objective optimisation of single axis tracking systems
are shown in Tables 5.22, ?? and 5.24. The bold values in each column represent the best case
scenario for each design objective. After the TOPSIS method was applied, the various layouts
were ranked according to their closeness to the ideal solution, as shown in the 'Rank' column.
The best and the worst layouts in the Pareto front were highlighted green and red, respectively.

An additional step was also taken to select the best performing layouts from each module
technology's Pareto front and pit them against each other to determine the best performing
module technology in this design scenario. Table 5.25 shows the resulting ranking of the
three dierent technologies after they were put through the TOPSIS method. The Thin Film
technology was by far the best out of the three, followed by the monocrystalline technology
and then lastly the polycrystalline technology.
 Table 5.22: MonoSi Pareto Front - Tracking

θT φC θRot GCRinv Qout,year (kW h) Land area used PR Closeness Coe Rank
CHAPTER 5.

1,100167 2,583885 5 1,1111 1,44E+08 125,078 0,829645 0,965658 1

76,44965 80,87462 16,84505 1,33134 1,09E+08 149,8707 0,851926 0,942216 2

79,25495 116,2148 19,12457 1,195229 79550760 134,5485 0,842703 0,93326 3

28,21867 16,61694 52,01599 1,775623 1,87E+08 199,8842 0,815622 0,924355 4

85,03451 65,35196 46,85173 1,742779 1,22E+08 196,1869 0,849249 0,915677 5

78,9039 86,68032 27,5212 2,339926 1,1E+08 263,4086 0,851245 0,853183 6

23,29035 5,411698 27,67661 3,070497 1,91E+08 345,6498 0,82684 0,779856 7

RESULTS AND ANALYSES

69,19155 58,57939 53,54066 3,409069 1,54E+08 383,7633 0,844432 0,740398 8

75,91149 43,82272 58,70844 3,924963 1,62E+08 441,8382 0,850968 0,683138 9

84,63264 49,4144 52,94807 4,232635 1,44E+08 476,4732 0,854779 0,647951 10

31,92984 15,76828 39,62676 4,917822 1,98E+08 553,6055 0,823243 0,573172 11

68,66479 35,03547 53,1935 5,6165 1,74E+08 632,2567 0,846011 0,494249 12

Stellenbosch University https://scholar.sun.ac.za

81,99318 49,83343 75,61133 6,397774 1,64E+08 720,2057 0,857117 0,406504 13

27,13342 8,136346 68,6656 7,178502 2,1E+08 808,0933 0,823059 0,322523 14

34,71267 18,85969 53,59603 9,061377 2,04E+08 1020,051 0,82365 0,122316 15

91

30,85514 8,161073 69,24016 9,997416 2,11E+08 1125,422 0,823229 0,065612 16

30,93809 8,160811 69,305 9,999602 2,11E+08 1125,668 0,823233 0,065595 17

88,44413 24,04567 80,99968 9,896312 1,73E+08 1114,041 0,862403 0,049112 18

 Table 5.23: Evaluation of PolySi modules in a xed tilt with dierent weights

θT φC GCRinv Qout,year (kWh) Land area used PR CC1 Rank 1 CC2 Rank 2
CHAPTER 5.

0,537741 1,833397 1,1111 1,36E+08 138,8807 0,818096 0,992065 1 0,984262 1

28,29722 3,860474 1,244104 1,36E+08 155,5054 0,725212 0,983087 2 0,978318 2

24,75675 5,836019 1,516114 1,46E+08 189,5051 0,784854 0,954327 3 0,954003 3

28,30241 2,231479 2,197376 1,52E+08 274,6587 0,813254 0,87783 4 0,877938 4

39,20124 38,6273 4,006285 1,45E+08 500,7612 0,81696 0,674376 5 0,674627 5

64,68456 5,674133 4,46337 1,32E+08 557,894 0,820813 0,62286 6 0,62283 6

23,42938 53,17622 4,744507 1,44E+08 593,0344 0,820625 0,59136 7 0,591711 7

RESULTS AND ANALYSES

56,02497 0,932095 5,393802 1,42E+08 674,1924 0,821583 0,518347 8 0,518798 8

71,89996 21,94978 5,908869 1,22E+08 738,5726 0,825497 0,460228 9 0,460157 9

83,05503 47,52825 6,949164 1,03E+08 868,6031 0,826391 0,34309 10 0,342707 10

78,54437 49,37105 7,37889 1,08E+08 922,3163 0,82862 0,294813 11 0,294631 11

48,30448 45,19499 7,546371 1,39E+08 943,2504 0,824498 0,27644 12 0,277653 12

Stellenbosch University https://scholar.sun.ac.za

80,34592 55,92698 8,137169 1,05E+08 1017,096 0,830328 0,20952 13 0,209374 13

74,8226 45,1918 8,792787 1,15E+08 1099,045 0,831131 0,13594 14 0,136323 14

31,86059 0,047765 9,982456 1,54E+08 1247,746 0,821109 0,023654 15 0,046102 15

92

84,76675 58,83206 9,938267 98740368 1242,223 0,832901 0,00695 16 0,006944 16

79,96354 57,64601 9,999997 1,05E+08 1249,939 0,833491 0,002761 17 0,005475 17

79,96354 57,64606 9,999993 1,05E+08 1249,938 0,833491 0,002761 18 0,005475 18

 Table 5.24: Thin Film Pareto Front - Tracking

θT φC θRot GCRinv Qout,year (kW h) Land area used PR Closeness Coe Rank
CHAPTER 5.

0,537667 5,021385 5,000244 1,1111 1,46E+08 121,2583 0,845125 0,958096 1

74,74597 34,015 11,92926 1,245089 1,32E+08 135,881 0,863666 0,94689 2

27,561 13,805 50,78984 1,747066 1,89E+08 190,6635 0,82315 0,927085 3

86,37965 116,8384 24,08482 1,142483 74407160 124,6833 0,863696 0,916301 4

86,27028 116,6958 23,94029 1,11221 73763328 121,3794 0,863213 0,916266 5

76,80504 48,04264 39,97448 1,888153 1,44E+08 206,0608 0,862856 0,902674 6

82,92486 91,30166 32,3949 2,104921 1,05E+08 229,7175 0,868975 0,87016 7

RESULTS AND ANALYSES

86,80354 78,94392 45,61889 2,342562 1,15E+08 255,6521 0,86932 0,849455 8

84,02272 95,73228 12,76826 2,323053 94695968 253,523 0,872356 0,846258 9

88,08156 76,71907 56,06746 2,659097 1,22E+08 290,1966 0,869536 0,817827 10

2,77192 10,48428 7,444234 3,933046 1,57E+08 429,2273 0,848502 0,681767 11

19,58151 18,14759 19,35673 4,431561 1,84E+08 483,6319 0,840679 0,627664 12

Stellenbosch University https://scholar.sun.ac.za

4,899029 5,558087 55,75197 5,438474 2E+08 593,52 0,839135 0,516313 13

58,50853 16,70521 43,65259 6,036098 1,92E+08 658,7407 0,850272 0,449517 14

29,43773 8,679419 76,01778 9,999433 2,13E+08 1091,274 0,830289 0,083727 15

93

31,15916 10,51302 75,44469 9,999677 2,13E+08 1091,3 0,830501 0,083485 16

5,499363 9,631891 78,33934 9,8723 2,05E+08 1077,399 0,83897 0,081053 17

88,72978 28,26615 82,06904 9,978879 1,75E+08 1089,03 0,878846 0,062304 18

 CHAPTER 5.

Table 5.25: Ranking of the best congurations from the 3 dierent technologies under investigation after applying TOPSIS.
RESULTS AND ANALYSES

Technology θT φC θRot GCRinv Qout,year (kW h) Land area PR Closeness Coe Rank

MonoSi 1,44E+08 125,078 0,829645 1,100167 2,583885 5 1,1111 0,168518 3

PolySi 1,74E+08 166,2364 0,80861 28,57031 13,67791 39,9932 1,329956 0,674982 2

Thin Film 1,46E+08 121,2583 0,845125 0,537667 5,021385 5,000244 1,1111 0,872967 1
Stellenbosch University https://scholar.sun.ac.za

94
 Stellenbosch University https://scholar.sun.ac.za

CHAPTER 5. RESULTS AND ANALYSES 95

5.3.5 Case 3 - Fixed Tilt, Dierent Weights

5.3.5.1 Problem formulation
In many design scenarios design objectives hardly have the same level of importance. This
can be factored in by changing the 'weight' of each objective in the application of the TOPSIS
method according to the particular PV project's focus. The greater the weight of the objective
is, relative to the others, the more important it is in that particular design scenario. To illustrate
this, a xed tilt problem was conceptualised where:

ˆ the main goal was to design a 75 MWp plant for a maximised annual energy yield so
Qout,year was assigned a unit weight,

ˆ second to this goal was achieving a maximised performance ratio and minimising Land
area use so PR and Land area objectives were assigned weights of 0.5 each.

While the rest of the problem formulation remained the same as the one given in Case 1 of the
multi-objective optimisation problems (Subsubsection 5.3.3.2), the weights were dierent and
were as summarised in Table 5.26.

Objective Qout Land area PR

Criteria sign +1 -1 +1

Weight 1 0.5 0.5

Table 5.26: Criteria and weights for the TOPSIS method

5.3.5.2 Results
The closeness coecient (CC) of the layouts in the Pareto front and ranks of each alternative
layout were compared between the initial case, where equal weights were used, and the new
case where the dierent weights were implemented. This was done for all three PV module
technologies and the results were recorded and tabulated as shown in Tables 5.27, 5.28 and
5.29.

As can be observed from the results, the weights have a bearing on how close an alternative
layout can be regarded to be to the ideal solution. The results in this example now favoured
layouts leading to higher energy production, more than those leading to minimised land use or
maximised PR which is what was expected. While the ranks in all three technologies remained
the same, the closeness coecient of each of their alternative layouts was evidently aected.
 Table 5.27: Evaluation of MonoSi modules in a xed tilt system with dierent weights.
CHAPTER 5.

θT φC GCRinv Qout,year (kWh) Land area used PR CC1 Rank 1 CC2 Rank 2

0,537667 1,833885 1,1111 1,37E+08 125,078 0,828987 0,996688 1 0,993397 1

0,11285 1,833885 1,1111 1,37E+08 125,078 0,829139 0,996595 2 0,993212 2

5,836793 0,33445 1,173611 1,41E+08 132,1149 0,815404 0,992478 3 0,991187 3

39,03782 4,211961 2,175078 1,52E+08 244,8513 0,818444 0,88021 4 0,880225 4

2,409527 87,42346 3,286091 1,37E+08 369,9195 0,829873 0,755103 5 0,755067 5

21,96655 25,43673 4,042909 1,52E+08 455,1155 0,830035 0,669924 6 0,669983 6

RESULTS AND ANALYSES

27,80457 341,3956 4,391459 1,54E+08 494,3522 0,826861 0,630687 7 0,630765 7

7,930363 100,9917 4,419519 1,35E+08 497,511 0,831378 0,627498 8 0,627488 8

53,8891 36,7796 6,278051 1,4E+08 706,7283 0,836357 0,418276 9 0,418361 9

44,47812 25,1907 6,897904 1,49E+08 776,506 0,833371 0,348525 10 0,348728 10

58,30252 53,38639 7,530551 1,3E+08 847,7239 0,836904 0,277252 11 0,277328 11

Stellenbosch University https://scholar.sun.ac.za

77,46206 43,17247 7,829544 1,13E+08 881,3819 0,844139 0,243555 12 0,243532 12

64,1818 17,19639 7,947074 1,34E+08 894,6124 0,837465 0,230379 13 0,230525 13

63,17443 60,14977 8,248634 1,23E+08 928,5594 0,838721 0,196395 14 0,19645 14

96

71,35178 53,80541 8,98052 1,18E+08 1010,949 0,842703 0,113989 15 0,114037 15

82,01798 47,62673 9,945921 1,07E+08 1119,625 0,847102 0,005277 16 0,005277 16

81,90861 47,63308 9,992796 1,07E+08 1124,902 0,847163 3,2E-05 17 6,21E-05 17

 Table 5.28: Evaluation of PolySi modules in a xed tilt with dierent weights

θT φC GCRinv Qout,year (kWh) Land area used PR Closenesss Coe Rank

CHAPTER 5.

0,537741 1,833397 1,1111 1,36E+08 138,8807 0,818096 0,992065 1 0,984262 1

28,29722 3,860474 1,244104 1,36E+08 155,5054 0,725212 0,983087 2 0,978318 2

24,75675 5,836019 1,516114 1,46E+08 189,5051 0,784854 0,954327 3 0,954003 3

28,30241 2,231479 2,197376 1,52E+08 274,6587 0,813254 0,87783 4 0,877938 4

39,20124 38,6273 4,006285 1,45E+08 500,7612 0,81696 0,674376 5 0,674627 5

64,68456 5,674133 4,46337 1,32E+08 557,894 0,820813 0,62286 6 0,62283 6

23,42938 53,17622 4,744507 1,44E+08 593,0344 0,820625 0,59136 7 0,591711 7

RESULTS AND ANALYSES

56,02497 0,932095 5,393802 1,42E+08 674,1924 0,821583 0,518347 8 0,518798 8

71,89996 21,94978 5,908869 1,22E+08 738,5726 0,825497 0,460228 9 0,460157 9

83,05503 47,52825 6,949164 1,03E+08 868,6031 0,826391 0,34309 10 0,342707 10

78,54437 49,37105 7,37889 1,08E+08 922,3163 0,82862 0,294813 11 0,294631 11

48,30448 45,19499 7,546371 1,39E+08 943,2504 0,824498 0,27644 12 0,277653 12

Stellenbosch University https://scholar.sun.ac.za

80,34592 55,92698 8,137169 1,05E+08 1017,096 0,830328 0,20952 13 0,209374 13

74,8226 45,1918 8,792787 1,15E+08 1099,045 0,831131 0,13594 14 0,136323 14

31,86059 0,047765 9,982456 1,54E+08 1247,746 0,821109 0,023654 15 0,046102 15

97

84,76675 58,83206 9,938267 98740368 1242,223 0,832901 0,00695 16 0,006944 16

79,96354 57,64601 9,999997 1,05E+08 1249,939 0,833491 0,002761 17 0,005475 17

79,96354 57,64606 9,999993 1,05E+08 1249,938 0,833491 0,002761 18 0,005475 18

 Table 5.29: Evaluation of Thin Film modules in a xed tilt system with dierent weights

θT φC GCRinv Qout,year (kWh) Land area used PR CC1 Rank 1 CC2 Rank 2
CHAPTER 5.

27,325 0,084147 1,111208 1,53E+08 121,2701 0,815541 0,996733 1 0,993503 1

0,537667 1,833885 1,1111 1,4E+08 121,2583 0,845734 0,989613 2 0,979472 2

20,21536 44,58465 1,620504 1,48E+08 176,8513 0,834797 0,942366 3 0,941639 3

2,780583 42,33594 1,616985 1,42E+08 176,4673 0,846024 0,942297 4 0,940234 4

39,36899 6,912212 2,097899 1,56E+08 228,9512 0,842332 0,888934 5 0,889245 5

28,89885 1,69588 3,153415 1,58E+08 344,1434 0,843911 0,770195 6 0,771021 6

80,70094 38,51896 3,707639 1,11E+08 404,6277 0,855992 0,706633 7 0,704131 7

RESULTS AND ANALYSES

54,37466 42,24433 4,011111 1,4E+08 437,7468 0,849752 0,673438 8 0,673878 8

87,21361 72,61783 4,601405 93928168 502,1676 0,857782 0,605728 9 0,602593 9

83,2639 44,81732 5,70721 1,08E+08 622,8481 0,864582 0,481912 10 0,48102 10

41,3404 45,04646 6,33199 1,47E+08 691,0324 0,848151 0,412783 11 0,415614 11

65,59724 32,29402 7,053186 1,33E+08 769,7391 0,859228 0,331358 12 0,33367 12

Stellenbosch University https://scholar.sun.ac.za

31,39818 0,024528 7,874206 1,58E+08 859,34 0,844238 0,241043 13 0,249519 13

66,38543 58,35785 8,472173 1,24E+08 924,5983 0,862137 0,172015 14 0,175808 14

78,81813 73,67347 8,922294 1,04E+08 973,7215 0,867816 0,120456 15 0,121779 15

98

88,59699 92,18474 9,238952 83788152 1008,279 0,871105 0,084248 16 0,083991 16

88,82833 72,52951 9,987533 93686824 1089,975 0,872834 0,005719 17 0,011349 17

88,95333 72,52951 9,987533 93541896 1089,975 0,872853 0,005636 18 0,011185 18

 Stellenbosch University https://scholar.sun.ac.za

CHAPTER 5. RESULTS AND ANALYSES 99

5.3.6 Case 4 - Single Axis Tracking, Dierent Weights

5.3.6.1 Problem Formulation
The single axis tracking multi-objective problem was also subjected to prioritised weights as in
Case 3 following the same conceptual design in which:

ˆ the main goal was to design a 75 MWp plant for a maximised annual energy yield so
Qout,year was assigned a unit weight,

ˆ second to this goal was achieving a maximised performance ratio and minimising Land
area use so PR and Land area objectives were assigned weights of 0.5 each.

The corresponding weights were, therefore, as summarised in Table 5.30.

Objective Qout Land area PR

Criteria sign +1 -1 +1

Weight 1 0.5 0.5

Table 5.30: Criteria and weights for the TOPSIS method

5.3.6.2 Results
The TOPSIS method was implemented and the resulting rankings and closeness coeecient
(CC) were tabulated and compared to those from the equally-weighted single axis tracking
multi-objective problem given in Subsubsection 5.2.3.1. The results for the monocrystalline,
polycrystalline and thin lm technology modules were as shown in Tables 5.31, 5.32 and 5.33.

5.4 Conclusions
The optimisation framework presented in this chapter was successfully integrated with a reputable
PV system design tool. This enabled the framework to utilise functionalities in both the
optimisation toolbox and the design tool to quickly, easily and accurately model PV system
performance. The toolbox provided several ways of optimising the design process, of which
two were fully explored in this chapter to tackle the design of a PV module layout in a
design space dened by multiple variables which ought to be taken into account simultaneously.
SAM, with its inbuilt subroutines and predened variables enabled the evaluation of several
plant performance parameters with relative ease and this can be extended to touch on other
parameters, system layouts such as can be dened by splitting the PV system into up to four
dierent subarrays, and use of other time periods, such as specic hours of the day or particular
months or seasons of the year, not covered in this chapter. The framework is therefore adaptable
to several other design objectives and design variables to best suit the design goals within the
connes of SAM's capabilities.

The fmincon solver's interior-point algorithm and the multi-objective genetic algorithm (MOGA),
implemented in the optimisation process, enabled multiple design parameters to be taken
into account simultaneously, within the connes of predened constraints in the design space.
This proved to be less cumbersome and faster than the original trial method that the design
engineers currently use and the iterative approach that was investigated in the Preliminary
Table 5.31: Evaluation of layouts of MonoSi modules in a single-axis tracking system with objectives with dierent weights.

θT φC θRot GCRinv Qout,year (kWh) Land area PR CC1 Rank 1 CC2 Rank 2
CHAPTER 5.

1,1001671 2,583885 5 1,1111 1,44E+08 125,078 0,829645 0,965658 1 0,933707 1

76,449654 80,87462 16,84505 1,33134 1,09E+08 149,8707 0,851926 0,942216 2 0,897307 3

79,254946 116,2148 19,12457 1,195229 79550760 134,5485 0,842703 0,93326 3 0,875607 5

28,218669 16,61694 52,01599 1,775623 1,87E+08 199,8842 0,815622 0,924355 4 0,92182 2

85,034514 65,35196 46,85173 1,742779 1,22E+08 196,1869 0,849249 0,915677 5 0,886648 4

78,903904 86,68032 27,5212 2,339926 1,1E+08 263,4086 0,851245 0,853183 6 0,831237 6

23,290351 5,411698 27,67661 3,070497 1,91E+08 345,6498 0,82684 0,779856 7 0,780735 7

RESULTS AND ANALYSES

69,191546 58,57939 53,54066 3,409069 1,54E+08 383,7633 0,844432 0,740398 8 0,737278 8

75,911485 43,82272 58,70844 3,924963 1,62E+08 441,8382 0,850968 0,683138 9 0,682291 9

84,632635 49,4144 52,94807 4,232635 1,44E+08 476,4732 0,854779 0,647951 10 0,645429 10

31,929841 15,76828 39,62676 4,917822 1,98E+08 553,6055 0,823243 0,573172 11 0,5774 11

68,664793 35,03547 53,1935 5,6165 1,74E+08 632,2567 0,846011 0,494249 12 0,497555 12

Stellenbosch University https://scholar.sun.ac.za

81,993183 49,83343 75,61133 6,397774 1,64E+08 720,2057 0,857117 0,406504 13 0,410247 13

27,133417 8,136346 68,6656 7,178502 2,1E+08 808,0933 0,823059 0,322523 14 0,336836 14

34,71267 18,85969 53,59603 9,061377 2,04E+08 1020,051 0,82365 0,122316 15 0,159303 15

100

30,855144 8,161073 69,24016 9,997416 2,11E+08 1125,422 0,823229 0,065612 16 0,123143 16

30,938086 8,160811 69,305 9,999602 2,11E+08 1125,668 0,823233 0,065595 17 0,123114 17

88,444133 24,04567 80,99968 9,896312 1,73E+08 1114,041 0,862403 0,049112 18 0,091905 18

 Table 5.32: Evaluation of PolySi modules in a single-axis tracking system with dierent weights.

θT φC θRot GCRinv Qout,year (kWh) Land area PR CC1 Rank 1 CC2 Rank 2
CHAPTER 5.

28,570315 13,67791 39,9932 1,329956 1,74E+08 166,2364 0,80861 0,965564 1 0,946927 1

3,2896077 1,783148 5,000185 1,1111 1,45E+08 138,8807 0,819371 0,956801 2 0,917419 2

44,987454 33,35379 13,947 1,619432 1,62E+08 202,4191 0,821706 0,934466 3 0,915226 3

79,102403 62,10481 14,54495 1,798239 1,15E+08 224,7689 0,84172 0,900802 4 0,85797 6

84,442914 108,9202 10,00559 1,638795 79648472 204,8394 0,840845 0,896294 5 0,83059 7

84,480742 109,003 10,009 1,705936 79666936 213,2316 0,840649 0,891873 6 0,827653 8

75,139569 54,54231 36,10246 2,006985 1,36E+08 250,8608 0,835562 0,888359 7 0,862516 5

RESULTS AND ANALYSES

6,8373735 11,05188 47,65152 2,310485 1,84E+08 288,7966 0,816101 0,864462 8 0,862721 4

83,714473 102,1834 29,12561 3,739012 96562984 467,3538 0,842458 0,697101 9 0,677483 9

83,859626 98,90498 23,54674 4,177602 95736920 522,1748 0,842989 0,649122 10 0,63288 10

9,0977445 2,464374 12,61781 4,525765 1,67E+08 565,6931 0,824323 0,616312 11 0,61767 11

73,922883 45,96376 5,810664 5,519751 1,23E+08 689,9352 0,839829 0,502603 12 0,498508 12

Stellenbosch University https://scholar.sun.ac.za

61,47455 44,09715 61,74716 8,135417 1,79E+08 1016,878 0,832304 0,218638 13 0,24186 13

60,966014 22,36667 62,96374 9,200731 1,93E+08 1150,035 0,832924 0,116878 14 0,168001 14

42,128701 28,718 63,16714 9,503582 2E+08 1187,89 0,819638 0,096911 15 0,15867 15

101

27,624343 10,68017 69,11961 9,999757 2,09E+08 1249,909 0,816451 0,083595 16 0,154291 16

64,797783 23,1929 66,01007 9,999854 1,91E+08 1249,921 0,836621 0,072844 17 0,135769 17

86,659078 24,84075 80,46224 9,957074 1,73E+08 1244,573 0,850025 0,061835 18 0,116156 18

 Table 5.33: Evaluation of Thin Film modules in a single axis tracking with dierent weights.

θT φC θRot GCRinv Qout,year (kWh) Land area PR CC1 Rank 1 CC2 Rank 2
CHAPTER 5.

0,5376671 5,021385 5,000244 1,1111 1,46E+08 121,2583 0,845125 0,958096 1 0,919812 2

74,745975 34,015 11,92926 1,245089 1,32E+08 135,881 0,863666 0,94689 2 0,901908 3

27,561 13,805 50,78984 1,747066 1,89E+08 190,6635 0,82315 0,927085 3 0,923264 1

86,379654 116,8384 24,08482 1,142483 74407160 124,6833 0,863696 0,916301 4 0,845605 5

86,270279 116,6958 23,94029 1,11221 73763328 121,3794 0,863213 0,916266 5 0,845471 6

76,805043 48,04264 39,97448 1,888153 1,44E+08 206,0608 0,862856 0,902674 6 0,879149 4

82,924864 91,30166 32,3949 2,104921 1,05E+08 229,7175 0,868975 0,87016 7 0,830795 7

RESULTS AND ANALYSES

86,803536 78,94392 45,61889 2,342562 1,15E+08 255,6521 0,86932 0,849455 8 0,820321 8

84,022722 95,73228 12,76826 2,323053 94695968 253,523 0,872356 0,846258 9 0,806966 9

88,081556 76,71907 56,06746 2,659097 1,22E+08 290,1966 0,869536 0,817827 10 0,796881 10

2,7719204 10,48428 7,444234 3,933046 1,57E+08 429,2273 0,848502 0,681767 11 0,67959 11

19,581509 18,14759 19,35673 4,431561 1,84E+08 483,6319 0,840679 0,627664 12 0,631228 12

Stellenbosch University https://scholar.sun.ac.za

4,8990292 5,558087 55,75197 5,438474 2E+08 593,52 0,839135 0,516313 13 0,525322 13

58,508529 16,70521 43,65259 6,036098 1,92E+08 658,7407 0,850272 0,449517 14 0,459727 14

29,437727 8,679419 76,01778 9,999433 2,13E+08 1091,274 0,830289 0,083727 15 0,154516 15

102

31,159163 10,51302 75,44469 9,999677 2,13E+08 1091,3 0,830501 0,083485 16 0,154105 16

5,4993627 9,631891 78,33934 9,8723 2,05E+08 1077,399 0,83897 0,081053 17 0,148638 17

88,729784 28,26615 82,06904 9,978879 1,75E+08 1089,03 0,878846 0,062304 18 0,117153 18

 Stellenbosch University https://scholar.sun.ac.za

CHAPTER 5. RESULTS AND ANALYSES 103

Design Chapter. The MOGA was also able to design for multiple objectives at the same time,
which was a better way of designing PV systems to simultaneously optimally satisfy various,
sometimes antagonistic, design objectives as it takes into account more plant performance
indicators.

The MOGA, when handling antagonistic design objectives, tends to yield multiple alternative
PV system layouts as options to optimally design for one objective or the other. To select the
best option out of these alternatives, a multi-criteria decision making (MCDM) tool known
as TOPSIS is implemented. TOPSIS, which stands for Technique for Order of Preference by
Similarity to Ideal Solution, enables all alternative options to be evaluated and ranked according
to how closely they resemble the ideal solution. This method was successfully implemented to
choose the best layout and the best module technology for each design scenario. The weights
can be used to remodel the ideal solution and its makeup such that more emphasis is placed
on design objectives that are deemed more important than others in that particular design
scenario. This is a better design approximation, compared to designing a system with a single
design objective in mind, which results in the oversimplication of the design process especially
in cases where more than one design objective needs to be achieved and the objectives are
potentially antagonistic. This is also better than performing a multi-objective design with all
objectives weighted equally because that is seldom the case in reality.
 Stellenbosch University https://scholar.sun.ac.za

Chapter 6
Conclusions

In this chapter, conclusions drawn from the other chapters of this thesis are summarised. This
is then followed by a section containing recommendations in relation to the work presented in
this project.

6.1 Conclusions

6.1.1 Introduction
In this opening chapter, a general context around the need for renewable energy sources was
given. Emphasis was, however, placed on solar energy and the state of the solar industry was
evaluated in the worldwide as well as the South African context. Thereafter, the research
problem of the lack of an ecient, integrated and hollistic approach to PV system design
optimisation was noted and investigated. The project's objectives were then laid out such that
frameworks for design optimisation could be formulated to:

ˆ replace or complement the current, sometimes arduous, plant design procedures,

ˆ be easily adaptable to various design scenarios since not all projects are ever the same,
especially in terms of constraints and design objectives,

ˆ provide a faster and more well-rounded approach to the design process by simultaneously
designing for multiple parameters and/or multiple objectives,

ˆ yield accurate and bankable PV system designs by integrating powerful optimisation

algorithms with reliable PV system design tools.

6.1.2 Theoretical Background

In this chapter, the fundamentals behind solar PV systems were presented. This covered
conditions favourable for solar PV projects, a breakdown of the PV system, system performance
parameters, the basics of PV system design and the role of simulation software in the design and
evaluation process. The purpose of this chapter was to highlight the key elements that make
up the PV system and its design procedure as a form of background for the design optimisation
procedures implemented in the thesis.

6.1.3 Preliminary Design

In this chapter, various elements that go into plant design optimisation were reviewed. The
process involved investigating the tools and strategies currently being implemented by some

104
 Stellenbosch University https://scholar.sun.ac.za

CHAPTER 6. CONCLUSIONS 105

PV system designers to nd out their purposes and strengths and to identify areas that could
do with some improvement. Some of the key insights from this preliminary plant design and
optimisation process were:

ˆ The vital role system design tools play in the design process and the importance of
selecting tools that enable developers to perform system modeling thoroughly and accurately.

ˆ The need for optimisation frameworks capable of handling multiple iterations and the
simultaneous optimisation of several design variables and objectives.

ˆ How some nonconventional design optimisation processes also play an important role in
overcoming certain design constraints and challenges and shed more insight with regards
to the complex relationship of several design variables and objectives.

These key deductions were then used to formulate optimisation frameworks for the design of
utility scale PV systems.

6.1.4 Optimisation Strategies

In this chapter, the optimisation strategy was broken down and each vital step was explained in
more detail to shed some light as to its importance in the optimisation process and what each
step entails. It was noted that evolutionary computing could be quite useful in the optimisation
of the PV system design process given the number of design objectives that needed to be
evaluated simultaneously. The optimisation procedure was modelled to take into account the
complex relationships of various design objectives in response to changes in the design variables
while ensuring that the process adhered to constraints in the design space. Based on the problem
cases under investigation, the appropriate optimisation algorithms in the MATLAB Toolbox
were selected and implemented in conjunction with a reliable PV system simulation software
program, SAM, to perform single- and multi-objective optimisation of the system design. A
multi-criteria decision making (MCDM) procedure was implemented to assist with the selection
of optimal layouts according to their satisfaction of the design objectives in the multi-objective
optimisation cases.

6.1.5 Results and Analysis

The optimisation frameworks presented in this chapter were successfully integrated with a
reputable PV system design tool. This enabled the frameworks to utilise functionalities in
both the optimisation toolbox and the design tool to quickly, easily and accurately model PV
system performance. The toolbox provided several ways of optimising the design process, of
which two were fully explored in this chapter to tackle the design of a PV module layout in a
design space dened by multiple variables which ought to be taken into account simultaneously.
SAM, with its inbuilt subroutines and predened variables enables the evaluation of several
plant performance parameters with relative ease and this can be extended to the design of
several other performance parameters, system layouts and timeframes, making the frameworks
discussed in this chapter adaptable.

The fmincon solver's interior-point algorithm and the multiobjective genetic algorithm (MOGA),
implemented in the optimisation process, enabled multiple design parameters to be taken into
account simultaneously, within the connes of predened constraints in the design space. This
proved to be less cumbersome and much faster than the method that the design engineers
currently use as mentioned in the problem statement. The MOGA was also able to design
 Stellenbosch University https://scholar.sun.ac.za

CHAPTER 6. CONCLUSIONS 106

for multiple objectives at the same time, which was a better way of designing PV systems to
simultaneously optimally satisfy various, sometimes antagonistic, design objectives as it takes
into account several plant performance indicators at the same time.

The MOGA, when handling antagonistic design objectives, tends to yield multiple alternative
PV system layouts in its Pareto optimal solution set. To select the best option out of these
alternatives, a multi-criteria decision making (MCDM) tool known as TOPSIS is implemented.
TOPSIS, which stands for Technique for Order of Preference by Similarity to Ideal Solution,
enables all alternative options to be evaluated and ranked according to how closely they resemble
the ideal solution. This method was successfully implemented to choose the best layout and the
best module technology for each design scenario. The TOPSIS weights were used to remodel
the ideal solution and its makeup such that more emphasis was placed on design objectives
that are deemed more important than others in that particular design scenario.

6.2 Recommendations
Strong evidence has recently emerged pointing to the possibility of signicant amounts of savings
in the PV system design if certain practices in the balance of system (BoS) design process are
also improved. One such example is the cost-saving wiring method known as the leapfrog
method as shown in Figure 6.1. According to the author of 'Cost-saving PV source-circuit
wiring method' [8], the leapfrog method leads to material cost savings of up to $20 000 on a 5
MW PV system with 72-cell modules. BoS optimisation could, therefore, be worth investigating
and adding to the overall plant design optimisation.

(a) Daisy chain wiring

(b) Leapfrog wiring

Figure 6.1: Illustration of source circuits connected using the daisy chain and leapfrog wiring
methods. Figures adopted from [8]

A drawback that is common in the application of multiobjective genetic algorithms (MOGA)

is that an increase in the dimensions or objectives of the problem leads to a larger computational
burden and, in turn, aects how quickly the optimisation process can be executed. Multivariate
reduction techniques such as principal component analysis (PCA) can then be used in such
cases to reduce some of the redundant objectives and consequently the computational burden
of multi-objective optimisation procedures.
 Stellenbosch University https://scholar.sun.ac.za

CHAPTER 6. CONCLUSIONS 107

Alternative implementations of the optimisation frameworks can be achieved by means of

other design tools, optimisation toolboxes and evolutionary programming techniques. For
instance, a SAM simulation setup can be exported to python form and integrated with a
python-based optimisation toolbox such as the distributed evolutionary algorithms in python
(DEAP) to execute the frameworks presented in this thesis. This is but one of many other ways
in which evolutionary programming techniques can be harnessed in order to improve the PV
system design process. The given alternative toolboxes and software, and many others, can be
taken into consideration for future work, depending on designers' preferences, prociency and
various other factors.
Stellenbosch University https://scholar.sun.ac.za

Appendices

108
 Stellenbosch University https://scholar.sun.ac.za

Appendix A
Canadian Solar PV Module Datasheet

109
 Stellenbosch University https://scholar.sun.ac.za

MAXPOWER
CS6X-310|315|320|325 P
The high quality and reliability of Canadian Solar’s
modules is ensured by 15 years of experience in
module manufacturing, well-engineered module
design, stringent BOM quality testing, an
automated manufacturing process and 100% EL
testing. linear power output warranty

KEY FEATURES product warranty on materials

and workmanship
Excellent module efficiency of
up to 16.94 %
Management System Certificates*
ISO 9001:2008 / Quality management system
Outstanding low irradiance ISO/TS 16949:2009 / The automotive industry quality management system
performance: 96.0 % ISO 14001:2004 / Standards for environmental management system
OHSAS 18001:2007 / International standards for occupational health & safety

Product CERTIFICATEs*
High PTC rating of up to 91.97%
IEC 61215 / IEC 61730: TÜV-Rheinland / VDE / KEMCO / MCS / CE / CEC AU / INMETRO
UL 1703 / IEC 61215 performance: CEC listed (US)
UL 1703: CSA / IEC 61701 ED2: VDE / IEC 62716: VDE / IEC 60068-2-68: SGS
IP67 junction box for long-term Take-e-way / UNI 9177 Reaction to Fire: Class 1
weather endurance

* As there are different certification requirements in different markets, please contact

Heavy snow load up to 5400 Pa,
your local Canadian Solar sales representative for the specific certificates applicable to
wind load up to 2400 Pa the products in the region in which the products are to be used.

CANADIAN SOLAR INC. is committed to providing high quality

solar products, solar system solutions and services to customers
around the world. As a leading PV project developer and
manufacturer of solar modules with over 15 GW deployed around
the world since 2001, Canadian Solar Inc. (NASDAQ: CSIQ) is one of
the most bankable solar companies worldwide.

Canadian Solar Inc.

545 Speedvale Avenue West, Guelph, Ontario N1K 1E6, Canada, www.canadiansolar.com, support@canadiansolar.com
 Stellenbosch University https://scholar.sun.ac.za

ENGINEERING DRAWING (mm) CS6X-320P / I-V CURVES

A A
Rear View Frame Cross Section A-A 10 10

9 9
35 8 8

7 7

6 6

5 5

40
Grounding
hole 4 4
Middle
bar 2–Ø5
3 3
1954
1400
1155

2 2
11
G

1 1
- +
V V
0 0
8-11 ˣ 7 A A Mounting Hole 5 10 15 20 25 30 35 40 45 5 10 15 20 25 30 35 40 45 50
Mounting
7
hole
R 1000 W/m2 5°C
800 W/m2 25°C
40 932 600 W/m2 45°C
982
11

400 W/m2 65°C

Electrical Data | STC* mechanical Data

CS6X 310P 315P 320P 325P Specification Data
Nominal Max. Power (Pmax) 310 W 315 W 320 W 325 W Cell Type Poly-crystalline, 6 inch
Opt. Operating Voltage (Vmp) 36.4 V 36.6 V 36.8 V 37.0 V Cell Arrangement 72 (6 ˣ 12)
Opt. Operating Current (Imp) 8.52 A 8.61 A 8.69 A 8.78 A Dimensions 1954 ˣ 982 ˣ 40 mm
Open Circuit Voltage (Voc) 44.9 V 45.1 V 45.3 V 45.5 V (76.9 ˣ 38.7ˣ 1.57 in)
Short Circuit Current (Isc) 9.08 A 9.18 A 9.26 A 9.34 A Weight 22 kg (48.5 lbs)
Module Efficiency 16.16% 16.42% 16.68% 16.94% Front Cover 3.2 mm tempered glass
Operating Temperature -40°C ~ +85°C Frame Material Anodized aluminium alloy
Max. System Voltage 1000 V (IEC) or 1000 V (UL) J-Box IP67, 3 diodes
Module Fire Performance TYPE 1 (UL 1703) or Cable 4 mm2 (IEC) or 4 mm2 & 12 AWG
CLASS C (IEC 61730) 1000V (UL), 1150 mm
Max. Series Fuse Rating 15 A Connector T4-1000V or PV2 series
Application Classification Class A Per Pallet 26 pieces, 620 kg (1366.9 lbs)
Power Tolerance 0~+5W Per Container (40‘ HQ) 624 pieces
* Under Standard Test Conditions (STC) of irradiance of 1000 W/m2, spectrum AM
1.5 and cell temperature of 25°C.

Electrical Data | Noct* temperature characteristics

CS6X 310P 315P 320P 325P Specification Data
Nominal Max. Power (Pmax) 225 W 228 W 232 W 236 W Temp. Coefficient (Pmax) -0.41 % / °C
Opt. Operating Voltage (Vmp) 33.2 V 33.4 V 33.6 V 33.7 V Temp. Coefficient (Voc) -0.31 % / °C
Opt. Operating Current (Imp) 6.77 A 6.84 A 6.91 A 6.98 A Temp. Coefficient (Isc) 0.053 % / °C
Open Circuit Voltage (Voc) 41.3 V 41.5 V 41.6 V 41.8 V Nominal Operating Cell Temperature 45±2 °C
Short Circuit Current (Isc) 7.36 A 7.44 A 7.50 A 7.57 A
* Under Nominal Operating Cell Temperature (NOCT), irradiance of 800 W/m2,
spectrum AM 1.5, ambient temperature 20°C, wind speed 1 m/s.

PERFORMANCE AT LOW IRRADIANCE Partner section

Outstanding performance at low irradiance, average relative
efficiency of 96.0 % from an irradiance of 1000 W/m2 to 200
W/m2 (AM 1.5, 25°C).

The specification and key features described in this datasheet may deviate slightly and
are not guaranteed. Due to on-going innovation, research and product enhancement,
Canadian Solar Inc. reserves the right to make any adjustment to the information
described herein at any time without notice. Please always obtain the most recent
version of the datasheet which shall be duly incorporated into the binding contract made
by the parties governing all transactions related to the purchase and sale of the products
described herein.

Caution: For professional use only. The installation and handling of PV modules requires Scan this QR-code to discover solar
professional skills and should only be performed by qualified professionals. Please read the
safety and installation instructions before using the modules. projects built with this module

Canadian Solar Inc. July 2016. All rights reserved, PV Module Product Datasheet V5.51_EN
 Stellenbosch University https://scholar.sun.ac.za

Appendix B
Kyocera PV Module Datasheet

112
 Stellenbosch University https://scholar.sun.ac.za

HIGH EFFICIENCY MULTICRYSTAL PHOTOVOLTAIC MODULE

KD 300-80 P Series
KD315GX-LPB KD320GX-LPB

Cutting Edge Technology

As a pioneer with over 35 years in the solar energy
industry, Kyocera demonstrates leadership in the
development of solar energy products. Kyocera’s
Kaizen Philosophy, commitment to continuous
improvement, is shown by repeatedly achieving
world record cell efficiencies.

Quality Built In
• UV stabilized, aesthetically pleasing black
anodized frame
• Supported by major mounting structure
manufacturers

• Easily accessible grounding points on

all four corners for fast installation
• Proven junction box technology with 12 AWG
PV wire to work with transformerless inverters
• Quality locking MC4 plug-in connectors to
provide safe and quick connections

Reliable
• Proven superior field performance
• Tight power tolerance
• Only module manufacturer to pass rigorous
long-term testing performed by TÜV Rheinland

Qualifications and Certifications

UL Listing
QIGU.E173074

NEC 2008 Compliant, UL 1703, and ISO 14001

UL1703 Certified and Registered, UL Fire Safety Class C, CEC, FSEC
 Stellenbosch University https://scholar.sun.ac.za

KD 300-80 P Series

electrical SPECIFICATIONS Module Characteristics

Standard Test Conditions (STC) Dimensions: 65.43in/51.97in/1.8in
STC = 1000 W/M2 irradiance, 25oC module temperature, AM 1.5 spectrum* length/width/height (1662mm/1320mm/46mm)

Weight: 60.6lbs (27.5kg)

KD315GX-LPB KD320GX-LPB

Pmp 315 320 W

PACKAGING SPECIFICATIONS
Vmp 39.8 40.1 V
Modules per pallet: 20
Imp 7.92 7.99 A Pallets per 53’ container: 22

Voc 49.2 49.5 V Pallet box dimensions: 66in/53in/47in

length/width/height (1675mm/1330mm/1190mm)
Isc 8.50 8.60 A Pallet box weight: 1323 lbs (600kg)
KD 315-80
Ptolerance +5/-3 +5/-3 % 5.24in 1.8in
133mm 46mm

Nominal Operating Cell Temperature Conditions (NOCT)

NOCT = 800 W/M2 irradiance, 20oC ambient temperature, AM 1.5 spectrum*

136mm
5.36in
11.06in
281mm

(IP65)
TNOCT 45 45 o
C

1040mm

1290mm
40.9in

50.8in
Pmax 226 230 W

Vmp 35.8 36.1 V

STABILIZER BAR
21.65in
550mm

Imp 6.34 6.40 A

Voc 45.0 45.3 V

1662mm
65.43in

Isc 6.88 6.96 A

PTC 276.4 280.9 W

21.65in
550mm

Temperature Coefficients STABILIZER BAR

Pmax -0.46 -0.45 %/oC

11.06in
281mm

Vmp -0.52 -0.51 %/oC

Imp 0.0064 0.0065 %/oC

51.97in
1320mm 0.87in
22mm
Voc -0.36 -0.36 %/oC Weight: 60.6lbs (28.0kg)
.47in
12mm
Isc 0.061 0.060 %/oC
46mm
1.8in

Operating
Temp -40 to +90 -40 to +90 %/oC
1.18in Short Side
30mm

System Design
11.06in
281mm

.41in
10.3mm

Series Fuse Rating 15 A

46mm
1.8in

Maximum DC System Voltage (UL) 600 V

Long Side 1.18in
30mm
Hailstone Impact 1in (25mm) @ 51mph (23m/s)
Expanded View of Grounding Holes Frame Cross Section Diagrams
* Subject to simulator measurement uncertainty of +/- 3%.
KYOCERA reserves the right to modify these specifications without notice.
WARNING: Read the instruction
NEC 2008 Compliant manual in its entirety prior to
UL 1703 Listed handling, installing & operat- MOUNTING HOLES DRAINAGE HOLES GROUND SYMBOL
ing Kyocera Solar modules. .35in (9mm) .35in (9mm)

040513 Our Valued Partner

KYOCERA Solar, Inc. 800-223-9580 800-523-2329 fax www.kyocerasolar.com

 Stellenbosch University https://scholar.sun.ac.za

Appendix C
First Solar PV Module Datasheet

115
 Stellenbosch University https://scholar.sun.ac.za

First Solar Series 4™

PV Module
ADVANCED THIN FILM SOLAR TECHNOLOGY

INDUSTRY BENCHMARK SOLAR MODULES

As a global leader in PV energy, First Solar’s advanced thin film solar modules
have set the industry benchmark with over 10 gigawatts (GW) installed worldwide
and a proven performance advantage over conventional crystalline silicon solar
modules. Generating more energy than competing modules with the same power
rating, First Solar’s Series 4™ and Series 4A™ PV Modules deliver superior
performance and reliability to our customers.

PROVEN ENERGY YIELD ADVANTAGE

• Generates more energy than conventional crystalline silicon solar
modules with the same power
• Superior temperature coefficient resulting in greater energy yield in typical
field operating temperatures
• Superior spectral response resulting in a proven energy yield advantage in
humid environments
• Anti-reflective coated glass (Series 4A™) enhances energy production

117.5 WATT MODULE ADVANCED PERFORMANCE & RELIABILITY

EFFICIENCY OF 16.3%
• Long-term power-output warranted for 25 years
• Compatible with advanced 1500V plant architectures
• Highly predictable energy in all climates and applications
• Independently certified for reliable performance in high temperature, high
humidity, extreme desert and coastal environments based on accelerated
life and stress tests

CERTIFICATIONS & TESTS

• PID-Free, Thresher Test, Long-Term Sequential Test, and ATLAS 25+1
• IEC 61646 1500V, IEC 61730 1500V, CE
• IEC 61701 Salt Mist Corrosion, IEC 60068-2-68 Dust and Sand Resistance
• ISO 9001:2008 and ISO 14001:2004
• UL 1703 and ULC 1703 Listed Class B Fire Rating (Class A Spread of Flame)
• CSI Eligible (CA-USA), FSEC (FL-USA), MCS (UK), CEC Listed (Australia),
JET (Japan)2, SII (Israel), InMetro (Brazil)2

3
MODULE WARRANTY
100% • 25-Year Linear
97%
95%
Performance
Power Output

Warranty4
90%
• 10-Year Limited
85%
Product Warranty
80%
0 5 10 15 20 25
Years

PD-5-401-04-2 | December 2015

First Solar, Inc. | firstsolar.com | info@firstsolar.com
 Stellenbosch University https://scholar.sun.ac.za

FIRST SOLAR SERIES 4™

PV MODULE

MECHANICAL DESCRIPTION MODULE NUMBERS AND RATINGS AT STC 5,6

Length 1200mm FS-4105-2 FS-4107-2 FS-4110-2 FS-4112-2 FS-4115-2 FS-4117-2
NOMINAL VALUES
Width 600mm FS-4105A-2 FS-4107A-2 FS-4110A-2 FS-4112A-2 FS-4115A-2 FS-4117A-2

Weight 12kg Nominal Power (± 5%) PMPP (W) 105.0 107.5 110.0 112.5 115.0 117.5

Thickness 6.8mm Voltage at PMAX VMPP (V) 67.8 68.6 69.4 70.2 70.5 71.2

Area 0.72m2 Current at PMAX IMPP (A) 1.55 1.57 1.59 1.60 1.63 1.65

Leadwire 2.5mm2, 610mm Open Circuit Voltage VOC (V) 86.0 86.6 87.2 87.7 87.8 88.2

Connectors MC49 Short Circuit Current ISC (A) 1.74 1.75 1.75 1.75 1.78 1.79

Bypass Diode None Module Efficiency % 14.6 14.9 15.3 15.6 16.0 16.3

Cell Type Thin-film CdTe Maximum System Voltage VSYS (V) 1500 7
semiconductor, Limiting Reverse Current IR (A) 4.0
up to 216 cells
Maximum Series Fuse I CF (A) 4.0
Frame Material None
MODULE NUMBERS AND RATINGS AT 800W/m2, NOCT8 45°C, AM 1.5 6
Front Glass 3.2mm heat
strengthened Nominal Power (± 5%) PMPP (W) 78.3 80.1 82.0 83.9 85.8 87.6

Series 4ATM Voltage at PMAX VMPP (V) 62.6 63.1 64.1 65.0 65.5 65.9
includes anti- Current at PMAX IMPP (A) 1.25 1.27 1.28 1.29 1.31 1.33
reflective coating
Open Circuit Voltage VOC (V) 81.0 81.6 82.1 82.6 82.7 83.1
Back Glass 3.2mm
tempered Short Circuit Current ISC (A) 1.40 1.41 1.41 1.41 1.44 1.44
Encapsulation Laminate TEMPERATURE CHARACTERISTICS
material with
edge seal Module Operating (°C) -40 to +85
Temperature Range
Load Rating 2400Pa10
Temperature Coefficient
TK (PMPP) -0.34%/°C
of PMPP
Temperature Coefficient
TK (VOC) -0.29%/°C
of VOC
Temperature Coefficient
TK (ISC) +0.04%/°C
of ISC

SUPERIOR TEMPERATURE COEFFICIENT MECHANICAL DRAWING

120
STC Typical Field
DC Power Output Relative to Power at STC (%)

Operating
Temperatures
110

100

More Energy for Same

Nameplate Watts
90

Linear (First Solar)

80
Linear (Multicrystalline Si)

70
1 Device package meets ATLAS 25+.
0 10 20 30 40 50 60 70
2 Testing Certifications/Listings pending.
Module Temperature (°C) 3 Limited power output and product warranties subject to warranty terms and conditions.
4 Ensures 97% rated power in first year, -0.7%/year through year 25.
5 Standard Test Conditions (STC) 1000W/m2, AM 1.5, 25°C
END-OF-LIFE RECYCLING 6 All ratings ±10%, unless specified otherwise. Specifications are subject to change.
7 Application Class A for 1000V (class II), Application Class B for 1500V (class 0)
• Recycling services available through First Solar’s 8 Nominal Operating Cell Temperature: Module operation temperature at 800W/m2

industry-leading recycling program or customer- irradiance, 20°C air temperature, 1m/s wind speed.
9 Multi-Contact MC4 (PV-KST4/PV-KBT4)
selected third party. 10 Higher load ratings can be met with additional clips or wider clips, subject to testing

Disclaimer
The information included in this Module Datasheet is subject to change without notice and is provided for informational purposes only. No contractual rights are established or should be
inferred because of user’s reliance on the information contained in this Module Datasheet. Please refer to the appropriate Module User Guide and Module Product Specification document for
more detailed technical information regarding module performance, installation and use.
The First Solar logo, First Solar™, and all products denoted with ® are registered trademarks, and those denoted with a ™ are trademarks of First Solar, Inc.

firstsolar.com | info@firstsolar.com
 Stellenbosch University https://scholar.sun.ac.za

List of References

[1] W. Moomaw, P. Burgherr, G. Heath, M. Lenzen, J. Nyboer, and A. Verbruggen,

 Annex II: Methodology,IPCC Special Report on Renewable Energy Sources and
Climate Change Mitigation , vol. 16, no. 10, p. NP, 2014. [Online]. Available:
http://www.ncbi.nlm.nih.gov/pubmed/25180255

[2] N. M. Haegel, R. Margolis, T. Buonassisi, D. Feldman, A. Froitzheim, R. Garabedian,

M. Green, S. Glunz, H.-M. Henning, B. Holder, I. Kaizuka, B. Kroposki, K. Matsubara,
S. Niki, K. Sakurai, R. A. Schindler, W. Tumas, E. R. Weber, G. Wilson,
M. Woodhouse, and S. Kurtz,  Terawatt-scale photovoltaics: Trajectories and
challenges, Science , vol. 356, no. 6334, pp. 141143, 2017. [Online]. Available:
http://www.sciencemag.org/lookup/doi/10.1126/science.aal1288

[3] Solargis,  Free download of global and regional solar resource maps / Overview | Solargis.
[Online]. Available: http://solargis.com/products/maps-and-gis-data/free/overview/

[4] Earthworks Magazine,  South Africa's renewable energy projects

- Earthworks magazine South Africa. [Online]. Available:
http://earthworksmagazine.co.za/south-africas-renewable-energy-projects/

[5] NREL SAM U.S. Department of Energy Oce of Energy Eciency and Renewable
Energy,  System Advisor Model (SAM). [Online]. Available: https://sam.nrel.gov/

[6] G. M. Masters, Wind Power Systems, in Renewable and Ecient Electric Power Systems ,
2nd ed. Wiley, John & Sons, 2004.

[7] Earthworks,  South Africa's renewable energy projects

- Earthworks magazine South Africa. [Online]. Available:
http://earthworksmagazine.co.za/south-africas-renewable-energy-projects/

[8] C. Colp,  Cost-saving PV source-circuit wiring method, 7.3 , 2014.

[9] J. Skye,  Why Is Renewable Energy Needed? [Online]. Available:

http://greenliving.lovetoknow.com/Why_Do_We_Need_Renewable_Energy

[10] B. Goss,  Design process optimisation of solar photovoltaic systems, Ph.D. dissertation.

[11] BBC,  BBC Bitesize - Higher Geography - Reasons for increase

in demand for energy - Revision 1. [Online]. Available:
http://www.bbc.co.uk/education/guides/zpmmmp3/revision/1

[12] C. Philibert,  Three Reasons Why Renewable Energy Is So Important

To The Power Industry | Power Conversion. [Online]. Available:
https://www.gepowerconversion.com/inspire/three-reasons-why-renewable-energy-so-important-po

118
 Stellenbosch University https://scholar.sun.ac.za

LIST OF REFERENCES 119

[13] UCSUSA,  Benets of Renewable Energy Use |

Union of Concerned Scientists. [Online]. Available:
http://www.ucsusa.org/clean-energy/renewable-energy/public-benets-of-renewable-power#.WWy

[14] INTERNATIONAL ENERGY AGENCY,  Snapshot of Global Photovoltaic Markets,

pp. 116, 2016.

[15] Energy Trend,  PV Price Quotes. [Online]. Available:

http://pv.energytrend.com/pricequotes.html

[16] S. Ringbeck and J. Sutterlueti,  BoS costs: status and optimization to reach
industrialgrid parity, Prog. Photovolt: Res. Appl. , vol. 27, no. February 2013, pp.
659676, 2012. [Online]. Available: http://dx.doi.org/10.1002/pip.1160

[17] Department of Energy, State of Renewable Energy in

Souh Africa , 2015, vol. 14, no. 3. [Online]. Available:
http://www.tandfonline.com/doi/abs/10.1080/21550085.2011.605855

[18] L.-A. Steenkamp,  Powering up: A look at section 12B allowance for renewable
energy machinery - The SA Institute of Tax Professionals. [Online]. Available:
http://www.thesait.org.za/news/269950/Powering-up-A-look-at-section-12B-allowance-for-renewab

[19] Green Energy Solutions,  Updated SARS Tax

Incentive for Solar Power! [Online]. Available:
http://www.greenenergysolutions.co.za/updated-sars-tax-incentive-for-solar-power/

[20] City Power Johannesburg,  CITY OF JOHANNESBURG TARIFFS. [Online].

Available: https://www.citypower.co.za/customers/Pages/Tari-Info.aspx

[21] Electricity Services Department and Environmental Resource Management

Department of the City of Cape Town,  Safe and Legal Installations
of Rooftop Photovoltaic Systems, pp. 23, 2016. [Online]. Available:
http://resource.capetown.gov.za/documentcentre/Documents/Procedures, guidelines
and regulations/Safe and Legal PV installations March 2016.pdf

[22] eThikwini Municipality,  eThikwini MunicipalityTaris. [Online]. Available:

http://www.durban.gov.za/City_Services/electricity/Taris/Pages/default.aspx

[23] Green Business Guide,  Reverse metering in the Nelson Mandela Bay
Metro - The Green Business GuideThe Green Business Guide. [Online]. Available:
http://www.greenbusinessguide.co.za/reverse-metering-in-the-nelson-mandela-bay-metro/

[24] SAURAN,  SAURAN network - Southern African Universities Radiometric Network.

[Online]. Available: http://www.sauran.net/Page/About

[25] A. Nakumuryango and R. Inglesi-Lotz,  South Africa's performance on renewable energy

and its relative position against the OECD countries and the rest of Africa, Renewable
and Sustainable Energy Reviews , vol. 56, pp. 9991007, 2016. [Online]. Available:
http://dx.doi.org/10.1016/j.rser.2015.12.013

[26] P. Grana,Paul; Gibbs,  The 3 Best Ways to

Optimize a Commercial PV System. [Online]. Available:
https://www.greentechmedia.com/articles/read/The-3-Best-Ways-to-Optimize-a-Commercial-PV-S
 Stellenbosch University https://scholar.sun.ac.za

LIST OF REFERENCES 120

[27] K. Joyce,  Optimizing PV Plant Design to Achieve a Low

Levelized Cost of Energy by, pp. 18. [Online]. Available:
http://bv.com/docs/reports-studies/optimizing-pv-plant-design-to-achieve-a-low-levelized-cost-of-e

[28] Merriam-Webster,  Dictionary and Thesaurus | Merriam-Webster, 2015. [Online].

Available: http://www.merriam-webster.com/

[29] R. Terlinden, T. Fricke, A. Linke, and C. Rost,  Redening Bankability

in Utility-Scale PV | Sunny. The SMA Corporate Blog. [Online]. Available:
https://en.sma-sunny.com/en/redening-bankability-in-utility-scale-pv/

[30] G. Algermissen,  Bankability: What it Means and Why it

is Important for Your Solar Project, 2015. [Online]. Available:
https://blog.pickmysolar.com/solar-bankability-what-it-means-and-why-it-is-important

[31] Folsom Labs,  Optimizing a Solar Array's Tilt and Module Spacing | AltEnergyMag.
[Online]. Available: http://www.altenergymag.com/content.php?post_type=2372

[32] I. Woofenden,  ENERGY BASICS: PV System

Types | Home Power Magazine. [Online]. Available:
https://www.homepower.com/articles/solar-electricity/equipment-products/energy-basics-pv-syste

[33] Solar Las Vegas,  Solar Power - The 7 Main Types of

Photovoltaic (PV) Systems - Universal Solar Direct. [Online]. Available:
https://www.universalsolardirect.com/solar-power-the-7-main-types-of-photovoltaic-pv-systems/

[34] Aladdin Solar LLC.,  PV System Types: Grid-Tied, Battery Backup, O-Grid, Direct
PV. [Online]. Available: http://www.aladdinsolar.com/pvsystems.html

[35] National Grid UK,  Grid code. [Online]. Available:

http://www2.nationalgrid.com/UK/Industry-information/Electricity-codes/Grid-Code/

[36] Florida Solar Energy Center,  How a PV System Works. [Online]. Available:
http://www.fsec.ucf.edu/en/consumer/solar_electricity/basics/how_pv_system_works.htm

[37] J. F. Orgill and K. G. T. Hollands,  Correlation equation for hourly diuse radiation on
a horizontal surface, Solar Energy , vol. 19, no. 4, pp. 357359, 1977.

[38] D. G. Erbs, S. A. Klein, and J. A. Due,  Estimation of the diuse radiation fraction
for hourly, daily and monthly-average global radiation, Solar Energy , vol. 28, no. 4, pp.
293302, 1982.

[39] E. L. Maxwell, A quasi-physical model for converting hourly global

to direct normal insolation, pp. 3546, 1987. [Online]. Available:
http://rredc.nrel.gov/solar/pubs/PDFs/TR-215-3087.pdf

[40] P. G. Loutzenhiser, H. Manz, C. Felsmann, P. A. Strachan, T. Frank, and G. M. Maxwell,

 Empirical validation of models to compute solar irradiance on inclined surfaces for
building energy simulation, Solar Energy , vol. 81, no. 2, pp. 254267, 2007.

[41] A. Louche, G. Notton, P. Poggi, and G. Simonnot,  Correlations for direct normal and
global horizontal irradiation on a French Mediterranean site, Solar Energy , vol. 46, no. 4,
pp. 261266, 1991.

[42] R. Perez,  Article Dynamic global-to-direct irradiance conversion models, vol. 98, no. 1,
pp. 354369, 2015.
 Stellenbosch University https://scholar.sun.ac.za

LIST OF REFERENCES 121

[43] B. Y. Liu and R. C. Jordan,  The interrelationship and characteristic distribution of

direct, diuse and total solar radiation, Solar Energy , vol. 4, no. 3, pp. 119, 1960.
[Online]. Available: http://linkinghub.elsevier.com/retrieve/pii/0038092X60900621

[44] T. M. Klucher,  Evaluation of models to predict insolation on tilted surfaces, Solar

Energy , vol. 23, no. 2, pp. 111114, 1979.

[45] R. Perez, P. Ineichen, R. Seals, J. Michalsky, and R. Stewart,  Modeling daylight

availability and irradiance components from direct and global irradiance, Solar Energy ,
vol. 44, no. 5, pp. 271289, 1990.

[46] J. E. Hay,  Calculating solar radiation for inclined surfaces: Practical approaches,
Renewable Energy , vol. 3, no. 4-5, pp. 373380, 1993.

[47] International Finance Corporation,  Utility-Scale Solar Photovoltaic Power Plants.

[48] C. Budig, J. Orozaliev, and K. Vajen,  Comparison of Dierent Sources of Meteorological

Data for Central Asia and Russia, EuroSun 2010 , p. 8, 2010.

[49] T. Mahachi and A. J. Rix,  COMPARISON OF VARIOUS LONG TERM AVERAGED

METEOROLOGICAL DATA SOURCES Conference Paper · January 2017, no.
February, 2017.

[50] Solar Green Australia,  The Three Types of Solar Cells. [Online]. Available:
http://www.solargreen.net.au/the-three-types-of-solar-cells.html

[51] Energy Informative,  Which Solar Panel Type is Best?

Mono-, Polycrystalline or Thin Film? [Online]. Available:
http://energyinformative.org/best-solar-panel-monocrystalline-polycrystalline-thin-lm/

[52] J. Chase,  Bloomberg New Energy Finance PV Module Maker Tiering System, no. July,
pp. 14, 2014.

[53] Arup,  First Solar Energy Yield Simulations Module Performance Comparison for Four
Solar PV Module Technologies, no. 1, 2015.

[54] C. Deline, A. Dobos, S. Janzou, J. Meydbray, and M. Donovan,  A simplied model of

uniform shading in large photovoltaic arrays, Solar Energy , vol. 96, pp. 274282, 2013.
[Online]. Available: http://dx.doi.org/10.1016/j.solener.2013.07.008

[55] PVSyst,  PVSyst Software. [Online]. Available: http://www.pvsyst.com/en/software

[56] T. Mahachi,  Energy yield analysis and evaluation of solar irradiance models for a utility
scale solar PV plant in South Africa, no. December, 2016.

[57] M. Lalwani, M; Kothari, D.P; Singh,  Investigation of Solar Photovoltaic Simulation

Softwares, vol. 1, no. 3, pp. 585601, 2010.

[58]  Optimum Tilt of Solar Panels. [Online]. Available: http://www.solarpaneltilt.com/

[59]  About Us | Black & Veatch. [Online]. Available: https://www.bv.com/about-us

[60] A. Engineering,  A study into the optimisation and calculation of electrical losses in
renewable energy generation, 2014.
 Stellenbosch University https://scholar.sun.ac.za

LIST OF REFERENCES 122

[61] M. L. Louazene, D. Korichi, and B. Azoui,  Optimization of global solar radiation of

tilt angle for solar panels, location: Ouargla, algeria, Journal of Electrical Engineering ,
vol. 13, no. 1, pp. 106111, 2013.

[62] X. Gong and M. Kulkarni,  Design optimization of a large scale rooftop photovoltaic
system, Solar Energy , vol. 78, no. 3, pp. 362374, 2005.

[63] D. Del Vecchio,  Optimizing a PV Array with Orientation

& Tilt | Home Power Magazine. [Online]. Available:
https://www.homepower.com/articles/solar-electricity/design-installation/optimizing-pv-array-orie

[64] G. Warner,  Solar energy in South Africa

Challenges and opportunities. [Online]. Available:
http://www.solarcentury.com/za/media-centre/solar-energy-south-africa-challenges-opportunities/

[65] D. Oliveira,  Northern Cape renewable Energy Mecca. [Online]. Available:

http://www.engineeringnews.co.za/article/northern-cape-renewable-energy-mecca-joemat-pettersso

[66] W. F. Holmgren, R. W. Andrews, A. T. Lorenzo, and J. S. Stein,  PVLIB Python 2015,

42nd IEEE PVSC , pp. 15, 2015.

[67] Sandia National Labs and R. Andrews,  PVLIB _ Python Documentation, 2016.

[68]  Project Jupyter | About. [Online]. Available: http://jupyter.org/about.html

[69] Python Software Foundation,  About Python. [Online]. Available:

https://www.python.org/about/

[70] PVSyst,  PVSyst Software Development. [Online]. Available:

http://www.pvsyst.com/en/software/software-development

[71] NREL SAM U.S. Department of Energy Oce of Energy Eciency and Renewable
Energy,  System Advisor Model (SAM) Release Notes. [Online]. Available:
https://sam.nrel.gov/sites/default/les/content/updates/releasenotes.html

[72] W. F. Holmgren,  PVLib-python release notes, 2016. [Online]. Available:

https://github.com/pvlib/pvlib-python/releases

[73] SAM Analysis Options,  SAM Help - Analysis Options. [Online]. Available:
https://www.nrel.gov/analysis/sam/help/html-php/index.html?samul_write_a_simple_script.htm

[74] Z. Ren, S. Jacques, S. Bissey, N. Batut, A. Schellmanns, and A. Caldeira,  PVLab:

an innovative and exible simulation tool to better size photovoltaic units, Renewable
Energies and Power Quality Journal , vol. 1, no. 12, 2014. [Online]. Available:
http://www.icrepq.com/icrepq%2714/241.14-Ren.pdf

[75]  SAM Open Source | System Advisor Model (SAM). [Online]. Available:
https://sam.nrel.gov/opensource

[76] PVLib-python,  PVLib-python Modules. [Online]. Available:

http://pvlib-python.readthedocs.io/en/latest/modules.html

[77] A. P. Dobos,  LK Scripting Language Reference, pp. 146, 2015.

[78] Timeanddate,  Seasons: Meteorological and Astronomical. [Online]. Available:

https://www.timeanddate.com/calendar/aboutseasons.html
 Stellenbosch University https://scholar.sun.ac.za

LIST OF REFERENCES 123

[79] D. Chung, C. Davidson, R. Fu, K. Ardani, and R. Margolis,  U.S. Photovoltaic Prices
and Cost Breakdowns : Q1 2015 Benchmarks for Residential , Commercial , and
Utility-Scale Systems, National Renewable Energy Laboratory , no. September, p. 42,
2015. [Online]. Available: http://www.nrel.gov/docs/fy15osti/64746.pdf

[80] K. Trabish, Herman,  Better, More Bankable Solar PV Installation

Designs From Any Computer? | Greentech Media. [Online]. Available:
https://www.greentechmedia.com/articles/read/better-faster-bankable-solar-pv-installation-designs

[81] BEW,  Bew Engineering. [Online]. Available: http://bewengineering.co.uk/

[82] SunPeak,  SunPeak's solar installation technology; East

West panel orientation | SunPeak. [Online]. Available:
https://www.sunpeakpower.com/sunpeak-technology-details

[83] A. Mellit, S. A. Kalogirou, L. Hontoria, and S. Shaari,  Articial intelligence techniques

for sizing photovoltaic systems: A review, Renewable and Sustainable Energy Reviews ,
vol. 13, no. 2, pp. 406419, 2009.

[84] D. Gómez-Lorente, I. Triguero, C. Gil, and A. E. Estrella,  Evolutionary

algorithms for the design of grid-connected PV-systems, EXPERT SYSTEMS WITH
APPLICATIONS , 2012.

[85] S. Sinha and S. Chandel,  Review of recent trends in optimization techniques

for solar photovoltaic-wind based hybrid energy systems, Renewable and
Sustainable Energy Reviews , vol. 50, pp. 755769, 2015. [Online]. Available:
http://linkinghub.elsevier.com/retrieve/pii/S1364032115005006

[86] S. Sreedhar and D. Jagadeesh,  A Review on Optimization Algorithms for MPPT in Solar
PV System under Partially Shaded Conditions, pp. 2332, 2016.

[87] a. Kornelakis and E. Koutroulis,  Methodology for the design optimisation and the
economic analysis of grid-connected photovoltaic systems, IET Renewable Power
Generation , vol. 3, no. July 2008, p. 476, 2009.

[88] A. Kornelakis,  Multiobjective Particle Swarm Optimization for the optimal design of
photovoltaic grid-connected systems, Solar Energy , vol. 84, pp. 20222033, 2010.

[89] A. Kornelakis and Y. Marinakis,  Contribution for optimal sizing of grid-connected

PV-systems using PSO, Renewable Energy , vol. 35, no. 6, pp. 13331341, 2010. [Online].
Available: http://dx.doi.org/10.1016/j.renene.2009.10.014

[90] D. Weinstock and J. Appelbaum,  Optimal Solar Field Design of Stationary Collectors,
Journal of Solar Energy Engineering , vol. 126, no. 3, p. 898, 2004. [Online]. Available:
http://solarenergyengineering.asmedigitalcollection.asme.org/article.aspx?articleid=1457085

[91] D. Szubert,  Lire la première p a rtie de la thèse, 2015. [Online]. Available:

http://ethesis.inp-toulouse.fr/archive/00002416/02/perez_gallardo_partie_2_sur_2.pdf

[92] D. A. Coley,  An Introduction to Genetic Algorithms for Scientists and Engineers.

[93] K. Yoon and C.-L. Hwang, Multiple attribute decision

making: an introduction , 1995. [Online]. Available:
http://books.google.com/books?hl=en&lr=&id=Fo47SWBuEyMC&oi=fnd&pg=PP8&dq=Multip

[94] C. Doctorale,  0Rïsentïe Et Soutenue Par, Sciences-New York , pp. 0180, 2011.
 Stellenbosch University https://scholar.sun.ac.za

LIST OF REFERENCES 124

[95] C. Azzaro-pantel, E. Nationale, G. Hernandez-rodriguez, F. Morales-mendoza,

L. Pibouleau, C. Azzaro-pantel, S. Domenech, and A. Ouattara,  Applications
of Metaheuristics in Process Engineering, no. April, 2014. [Online]. Available:
http://link.springer.com/10.1007/978-3-319-06508-3

[96] M. Boix, L. Montastruc, L. Pibouleau, C. Azzaro-Pantel, and S. Domenech, A

multiobjective optimization framework for multicontaminant industrial water network
design, Journal of Environmental Management , vol. 92, no. 7, pp. 18021808, 2011.

[97] A. S. Milani, A. Shanian, R. Madoliat, and J. A. Nemes,  The eect of normalization

norms in multiple attribute decision making models: A case study in gear material
selection, Structural and Multidisciplinary Optimization , vol. 29, no. 4, pp. 312318,
2005.

[98] A. Çelen,  Comparative analysis of normalization procedures in TOPSIS method: with an

application to Turkish deposit banking market, Informatica , vol. 25, no. 2, pp. 185208,
2014.

[99] Folsom Labs,  Mechanical Design. [Online]. Available:

https://www.folsomlabs.com/modeling/design/mechanical

[100] International Renewable Energy Agency, Solar Pv in Africa: Costs and Markets , 2016,
no. September.