i
PREDICTION OF PHOTOVOLTAIC (PV) OUTPUT VIA ARTIFICIAL NEURAL
NETWORK (ANN) BASED ON REAL CLIMATE CONDITION
FLORA ANAK CROCKER
A report submitted in
fulfillment of the requirement for the award of the
Degree of Master of Electrical Engineering
Faculty of Electrical and Electronics Engineering
Universiti Tun Hussein Onn Malaysia
JANUARY 2017
iv
In the name of Heavenly Father, YAHWEH, Omnipresent God, The Most Glory
Dedicated, in thankful appreciation for support, understanding, and encouragement
for my beloved father, mother, and sisters.
May God bless all of you
v
ACKNOWLEDGEMENTS
First and foremost, I would like to express my greatest gratitude to my
supervisor, Dr. Siti Amely Binti Jumaat, for the given guidance and enthusiasm
throughout on this project. Dr. Siti Amely has been an excellent mentor and has
provided unfailing support throughout my master project.
My appreciation also goes to my father, mother, and sisters who have been so
tolerant and support me all these years. I would like to extend my heartiest thanks to
them for their encouragement and patience.
It should be recognized that the success of this project was through the
support and advice of FKEE lecturers, Mr. Helmy Bin Abdul Wahab, Dr. Nor Aira
Binti Zambri, Dr. Goh Hui Hwang (as a panel) and Dr. Suriana Binti Salimin (as a
panel). My gratitude also goes to my co-supervisor Dr. Nur Hanis Binti Mohammad
Radzi and all others who have rendered support directly or indirectly to make this
project possible.
vi
ABSTRACT
Photovoltaic (PV) system most popular as harvesting energy and has major
challenged due to the difficult to control the output. The performance of PV panel
output is incompatible due to changing climate condition. It difficult to knowing the
power output of PV panel system for multiplicity amount of solar radiation and the
ambient environmental factors. Due to solar radiation, data is not always available in
remote areas; it would be assisted if the solar radiation from the system could be
predicted. This study will explore the potential of artificial neural network (ANN) to
be applying in the system of prediction of power output from photovoltaic (PV)
panel system. In order to test the efficiency and reliability of a proposed ANN model
experimental output compare with proposed mathematical equation. The objectives
of this project are to develop the ANN model that capable of predicting power
output, obtain power forecast model using ambient factor and identify the influence
of climate changing for electrical production. The proposal of this study to predict
the solar radiation, voltage, current, and power based on the real climate condition
along with PV panel system. The activation functions using for the hidden layer is
hyperbolic tangent. The training algorithm is used Levenberg-Marquardt
backpropagation. The meteorology data as input data was obtained from RET screen
database in the period from 1st January 2015 until 31st August 2016. There was five
selected location in Malaysia to be the subject test. From the result, average power
output was high level in January to March for all selected location except for
Kuching. While in low average solar radiation, the power output was at a low level.
This shows that the performance of power output is depending on the level of solar
radiation.
vii
ABSTRAK
Sistem fotovoltaik (PV) sangat terkenal sebagai sistem penuaian tenaga dan
menpunyai cabaran utama iaitu kesukaran mengawal keluarannya. Prestasi keluaran
bagi PV juga tidak selari oleh kerana perubahan keadaan iklim. Hal ini telah
menyebabkan kesukaran untuk mengetahui kuasa keluaran sistem panel PV
sekiranya terdapat kepelbagaian jumlah radiasi suria dan faktor persekitaran. Selain
itu, data radiasi suria adalah terhad bagi kawasan perdalaman, ianya amat menbantu
jika radiasi suria dapat diramal. Kajian ini akan meneroka potensi Artificial neural
network (ANN) untuk sistem ramalan kuasa keluaran dari sistem PV. Perbandingan
keluaran bagi eksperimen dengan persamaan matematik bagi menguji kecekapan dan
kebolehpercayaan model ANN yang dicadangkan. Objektif projek ini adalah untuk
membangunkan model ANN yang mampu meramalkan kuasa keluaran, mendapatkan
model ramalan bagi kuasa keluran dengan menggunakan faktor persekitaran dan
mengenal pasti pengaruh perubahan iklim dalam pengeluaran tenaga elektrik.
Cadangan kajian ini untuk meramalkan radiasi suria, voltan, arus, dan kuasa
berdasarkan keadaan iklim yang sebenar. Activation Function yang digunakan untuk
lapisan tersembunyi adalah hyperbolic tangent. Algoritma yang digunakan adalah
Levenberg-Marquardt backpropagation. Data meteorologi adalah sebagai data
masukan yang telah diperolehi daripada pangkalan data RET screen dalam tempoh
dari 1 Januari 2015 sehingga 31 Ogos 2016. Terdapat lima lokasi di Malaysia yang
dijadikan sebagai subjek kajian. Dari hasil eksperimen, purata kuasa keluaran adalah
tinggi pada bulan Januari hingga Mac untuk lokasi yang dipilih kecuali Kuching.
Sekiranya purata radiasi suria rendah, kuasa keluaran adalah pada rendah. Ini
menunjukkan bahawa prestasi kuasa keluran adalah bergantung kepada tahap radiasi
suria.
viii
TABLE OF CONTENT
TITLE i
DECLARATION iii
DEDICATION iv
ACKNOWLEDGEMENTS v
ABSTRACT vi
ABSTRAK vii
TABLE OF CONTENT viii
LIST OF FIGURES xii
LIST OF TABLES xv
LIST OF ABBREVIATIONS xvii
LIST OF SYMBOLS xviii
LIST OF APPENDICES xix
CHAPTER 1 INTRODUCTION
1.1 Project Background 1
1.2 Problem Statement 3
1.3 Objectives 4
1.4 Scopes 5
1.5 Chapter Outlines 6
CHAPTER 2 LITERATURE REVIEW
2.1 Introduction 7
2.2 Solar Radiation Prediction using Artificial
Intelligence (AI)
7
ix
2.3 Prediction of Photovoltaic Output Using
Artificial Neural Networks
10
2.4 Overview of Photovoltaic 11
2.4.1 PV Terminology 12
2.4.2 Photovoltaic Modules Technologies 13
2.4.3 Photovoltaic Electrical Characteristic 14
2.5 Neural Network (NN) 16
2.5.1 Basic Elements of ANN 16
2.5.1.1 Network Architecture 16
2.5.1.2 Weight and Biases 17
2.5.1.3 Activation Function 18
2.5.2 Multilayer Perceptron (MLP) 19
2.5.3 Multilayer Perceptron Training 20
CHAPTER 3 METHODOLOGY
3.1 Introduction 22
3.2 Project Review 22
3.3 Flow Chart for Development of Forecasting
Model
23
3.4 Data Collection 25
3.4.1 Meteorological Data: Input Data 25
3.4.2 Photovoltaic System Description 26
3.4.3 Proposed Equation 27
3.5 Designing an Artificial neural network (ANN)
model
29
3.6 Summary 33
x
CHAPTER 4 RESULT AND DISCUSSION
4.1 Introduction 34
4.2 Designation of Artificial Neural Network (ANN)
model for Forecasting
34
4.2.1 Selection of Number of Neuron for
Hidden Layer
35
4.2.2 Selection of Epochs 37
4.2.3 Selection of Learning Rate 38
4.3 Result of output ANN model 1: Solar Radiation 41
4.4 Result of modeling using ANN model 2 45
4.4.1 Result performance of output for ANN
Model 2
45
4.4.1.1 Batu Pahat 45
4.4.1.2 Melaka 47
4.4.1.3 Kuala Lumpur 48
4.4.1.4 Kuching 50
4.4.1.5 Kota Kinabalu 51
4.4.2 Result and Analysis of Current (Im) for
Selection Location
53
4.4.3 Result and Analysis of Voltage (Vm) for
Selection Location
58
4.4.4 Result and Analysis of Power Outputs for
Selection Location
63
4.4.5 Accuracy and Error 65
4.5 Summary 66
xi
CHAPTER 5 CONCLUSION
5.1 Conclusion 67
5.2 Recommendations 68
REFERENCES
APPENDICES
xii
LIST OF FIGURES
2.1 Configuration of PV Panel 12
2.2 I-V curve 14
2.3 P-V curve 15
2.4 Architecture of Network of Multilayer
Perceptron
17
2.5 Simple Neural Network 17
2.6 MLP network 19
3.1 Flow Chart for overall project 24
3.2 RET Screen front page of database 25
3.3 The flow chart for developing MLP network
using MATLAB
31
3.4 Project layout contain proposed ANN model 32
4.1 ANN Model 1 with selection value 40
4.2 ANN Model 2 with selection value 40
4.3 Actual vs. predict value of solar radiation
(Batu Pahat)
42
4.4 Actual vs. predict value of solar radiation
(Melaka)
42
4.5 Actual vs. predict value of solar radiation
(Kuala Lumpur)
42
4.6 Actual vs. predict value of solar radiation
(Kuching)
43
xiii
4.7 Actual vs. predict value of solar radiation
(Kota Kinabalu)
43
4.8 MSE performance of ANN model (Batu
Pahat)
46
4.9 Best Fit performance of ANN model (Batu
Pahat)
46
4.10 MSE performance of ANN model (Melaka) 47
4.11 Best Fit performance of ANN model
(Melaka)
48
4.12 MSE performance of ANN model (Kuala
Lumpur)
49
4.13 Best fit performance of ANN model (Kuala
Lumpur)
49
4.14 MSE performance of ANN model (Kuching) 50
4.15 Best fit performance of ANN model
(Kuching)
51
4.16 MSE performance of ANN model (Kota
Kinabalu)
52
4.17 Best fit performance of ANN model (Kota
Kinabalu)
52
4.18 Result for actual and predict data current for
Batu Pahat
55
4.19 Result for actual and predict data current for
Melaka
55
4.20 Result for actual and predict data current for
Kuala Lumpur
56
4.21 Result for actual and predict data current for
Kuching
56
xiv
4.22 Result for actual and predict data current for
Kota Kinabalu
56
4.23 Result for actual and predict voltage data for
Batu Pahat
60
4.24 Result for actual and predict data voltage for
Melaka
60
4.25 Result for actual and predict data voltage for
Kuala Lumpur
61
4.26 Result for actual and predict data voltage for
Kuching
61
4.27 Result for actual and predict data voltage for
Kota Kinabalu
61
4.28 Average value of power in watt for predict
data in all location regions
64
xv
LIST OF TABLES
2.1 Summary of Photovoltaic terminology 13
2.2 Transfer Function of Neural Network 18
3.1 Data Sheet of Kyocera KC175GH-2
polycrystalline panel
26
4.1 MSE for varies number of neurons 36
4.2 Best fit for varies number of neurons 36
4.3 MSE for varies number of Epochs 37
4.4 Best fit for varies number of Epochs 38
4.5 MSE for varies learning rate 39
4.6 Best fit for varies learning rate 39
4.7 Actual value of Solar Radiation in unit W/m2 44
4.8 Predict value of Solar Radiation in unit
W/m2
44
4.9 ANN model output performance (Batu
Pahat)
46
4.10 ANN model output performance (Melaka) 47
4.11 ANN model output performance (Kuala
Lumpur)
49
4.12 ANN model output performance (Kuching) 50
xvi
4.13 ANN model output performance (Kota
Kinabalu)
52
4.14 Average value current for actual data in unit
Amps
57
4.15 Average value current for predict data in unit
Amps
58
4.16 Average value voltage for actual data 62
4.17 Average value voltage for predict data 63
4.18 The error prediction for all selected location 65
xvii
LIST OF ABBREVIATIONS
ANN - Artificial Neural Network
PV - Photovoltaic
MSE - Mean Square Error
MAPE - Mean Absolute Percentage Error
RMSE - Root Mean Square Error
LM - Levenberg-Marquardt
MLP - Multilayer Perceptron
NN - Neural Network
xviii
LIST OF SYMBOLS
R - Correlation Coefficient
R2 - Determine coefficient
w - Weight
i - Neuron
j - Inputs from other neurons
Ta - Ambient Temperature (˚C)
Ws - Wind Speed (m/s)
Rh - Relative Humidity
G - Solar Radiation (W/m2)
Tmin - Minimum air temperature
Tmax - Maximum air temperature
Tc - Cell Temperature (˚C)
Voc - Open circuit voltage (V)
Isc - Short circuit current (A)
Vm - Voltage at maximum point
Im - Current at maximum point
P - Power generated by PV Panel
xix
LIST OF APPENDICES
APPENDIX TITLE PAGE
A MATLAB code for ANN Model 1 and 2 72
B Solar Panel Data Sheet 75
C Output for ANN Model 1: Solar Radiation 77
D Output for ANN Model 2 101
CHAPTER 1
INTRODUCTION
1.1 Project Background
The electrical energy consumption is growing exponentially as the world population
increase and increasing per capital demand. The production of the electricity is still
using natural resources such as fossil fuel and petroleum. These resources are used
by the world communities as their main resources and it kept increased day to day.
The dependency on these resources had lead to an increase in market price due to
depleting of sources. And else, increase the greenhouse gas (GHG) emission, which
has an effect on the climate.
2
There are three main important issues which are increasing oil prices, the
depleting of fossil fuel day to day and increasing the worst issues of environmental
problem. Then, by replacing fossil fuel that being depleted towards renewable energy
resources for instance winds power, solar power, hydropower, biomass and biofuels
(transportation). This alternative way gains popularity and gets attention from around
the world. Besides, it catches much attention from many foreign countries due to the
three main important challenging issues that are faced by worldwide power
generation [1]. There are a lot of study and research in order to improve the quality
performance of the electrical energy production using renewable energy as a resource
and one of that is solar energy [2].
Photovoltaic (PV) systems are one of solar energy and played an important
role to worldwide due to the fact of PV system is clean, pollution free, environment-
friendly, inexhaustible and secure energy source. The PV systems also provide the
direct method to convert solar energy into an electric energy. This PV system is
expandable power because it can be used in small and large power generation plant.
The PV system can be used either in the stand-alone installation and it also suitable
for residential, commercial and industrial applications. There are some problems in
term of initial cost, efficiency and reliability of power generation when solar power
had been dealing with PV system. Therefore, modeling and simulation are an
important role in the development and investigate the performances of PV itself as
well to design PV systems [3].
Malaysia is located in a tropical climate zone, experienced the extremely
rainfall and dry days (droughts) that take place regularly every year resulting from
the local tropical wet season. The ambient temperature is relatively uniform
temperatures throughout the year. The averages of ambient temperature are between
26°C to 32°C. The annual averages of solar radiations for Malaysia were from 4.21
kWh/m2 to 5.56 kWh/m2 [4]. Therefore, Malaysia has a potential location for
harvesting the solar power energy.
Hence, more research needs to be performed methodically in order to analyze
and increasing the performance of PV modules. Furthermore, it is expected in the
near future that the power generation of PV power system will increase and currently
PV system is popular as stand-alone power generation for the development of
industrial, business nation’s worldwide and also residential installation [5].
3
Nowadays, many methods used to increasing the solar PV performance. Most
of the method based on modeling of PV system to predict the output. In additional,
the input weather parameters such irradiation, temperature, relative humidity, and
wind speed are essential to use to evaluating PV outputs. Neural network (NN) a
popular method used to predict the Solar PV performance because it can run a
complex problem [6].
In additional, researchers work [7] use an artificial neural network (ANN) that
used to the prediction of photovoltaic module temperature. There is ambient factor
involve evaluating the PV performance. Therefore, from the results obtain are the
electrical efficiency and also power is calculated by depending on the prediction
modules temperatures. As the solar radiation increased, photovoltaic module
electrical efficiency decreased. Moreover, the power of photovoltaic module
increased as the solar radiation increased.
1.2 Problem Statement
Since photovoltaic (PV) system is gain its popularity in Malaysia that make the
number of PV system installation are drastically increasing. PV system has roles in
storage energy and solving environmental issues [8]. For instance, the PV system
standalone type mostly installed either in the building, housing, lighting street or
transportation. Moreover PV system can also consider as the best solution for
electrical production in remote areas for example in Malaysia rural areas. Generally,
using PV system harvesting energy could bring more challenged compared to
another conventional power plant for harvesting energy. Unlike another conventional
power plant, PV system output is difficult to control due to the nature resource from
the sun.
i. The facts of the climate condition also distress the performance of PV panel
output itself due to changing climate condition. This could make the PV panel
output is incompatible values with the rated value. Specifically, in Malaysia,
there are many factors that could influence the power output of PV panel
systems such as ambient temperature (Ta), wind speed (Ws), relative
4
humidity (Rh), dust, solar radiation or other factor relate to ambient factor
[9].
ii. This will cause difficult to knowing the power output of PV panel system due
to multiplicity amount of solar radiation and the ambient environmental
factors in certain location around Malaysia.
iii. Meanwhile, meteorology data is not always available in remote areas; as solar
radiation data is a need for research and development for solar energy
utilization. In particularly, found the cost and difficult to obtain solar
radiation measurement in Malaysia, due to meteorology station either in the
urban and rural area is not available to measure these data [10].
Therefore, an alternative way is important in order to generate the data such as
solar radiation and power output of PV panel system for a different location in
Malaysia. Thus, a prediction or forecasting tool is suitable to be used to generate
these data. One of example is a pyranometer to measured solar radiation data,
unfortunate this tool is costly and needlessness to predict output power for PV panel
system. Then, one of artificial intelligence technique have been used is an artificial
neural network (ANN). Recently, the Artificial Neural network (ANN) method has
the capability to predicting future system output values in order to afford a
comprehensive data for the solar energy potential in Malaysia. Besides, ANN can be
used to solving a complex problem in various fields of application such as prediction,
pattern recognition, identification, classification, etc. [11]. All the PV’s parameter
involves are investigated and trained using ANN method by using MATLAB
software.
1.3 Objectives
The objectives of this project are:-
i. To develop an ANN model as to predict the PV power output.
ii. To obtain power forecast model using ambient factor.
iii. To investigate the influence climate changing reliability in electrical
production.
5
1.4 Scopes
This project proposed to use Neural Network method to predict and forecast the PV
output. The scopes represent the limitation of the project in order to predict the PV
output by using ANN model.
i. Develop ANN model: This project will consist of two ANN models that
develop to predict PV output. The ANN model 1 has developed with five
inputs (Ta, Ws, Rh, Tmax, and Tmin) in order to predict one output
(solar radiation, G). The ANN model 2 consists of seven inputs (Ta, Ws,
Rh, Tc, Voc, Isc and G) for obtaining two outputs (voltage and current).
Meanwhile, the value of solar radiation (G) as input in ANN model 2 is
obtained from the output of ANN model 1. Multilayer perceptron (MLP)
network and Levenberg Marquardt (LM) algorithm is proposed for
developing this ANN model.
ii. Prediction of the PV output: This studies set out to predict or forecast
the PV output in term of power by using ambient factors. In this project
is involving meteorology data from five locations in Malaysia: Batu
Pahat (Johor), Melaka, Kuala Lumpur, Kuching (Sarawak), and Kota
Kinabalu (Sabah). The meteorology data are used from January 2015
until August 2016 and it obtains from RET Screen database.
Mathematical equations are used to be express in ANN model as the
relationship of output and target.
iv. Performance analysis: This involves the pattern of PV output for
selected locations. Then, three error statistics are used to evaluate the
performance of ANN model as inspect the influence climate changing
effect the output. There are consisting of the mean squared error (MSE),
the mean absolute percentage error (MAPE), and root means squared
error (RMSE). The correlation coefficient (R) represent the relationship
between the actual and predict value of output.
6
1.5 Chapter Outlines
This report consists of five chapters. The current chapter mainly presents the problem
statement, objectives, and importance of this study. It also provides a common
development method used in the creating ANN model as the prediction model.
Chapter 2 consists of previous studies and researches on artificial intelligence
(AI) and artificial neural network (ANN) which is relevant in developing and creates
the ANN model in order to predict solar radiation, power, current, and voltage. In
this chapter also discusses the theoretical of photovoltaic (PV) and neural network.
Chapter 3 discusses the methodology used for this study. It describes in detail
the process for the short listing of alternatives to develop this project that illustrated
using a flow chart. Besides, the detail of meteorology data as input data has been
taken in RET Screen database. The PV panel is used a polycrystalline panel and
detail brief in this chapter. Then, the mathematical equations are a manifestation in
order to obtain PV output. This chapter approach provides an illustration of project
layout that consists of ANN Model 1 and 2.
Chapter 4 details the findings of outputs for ANN Model 1 (solar radiation)
and ANN Model 2 (voltage and current). Moreover, also discuss the selection of the
number of hidden layers, the number of epochs and learning rate for developing
ANN model. These chapter present results of ANN Model 1 and 2 by illustrating in
figure and table for each location. Also in it discuss the performance of ANN model
in order to predict the solar radiation, voltage, current and power output. The
evaluation accuracy and error of ANN model to predict PV output also discussed in
this chapter.
Chapter 5 concluded the discussion of findings of the experiment, and the
further recommendation in order to improve techniques, input parameter, research
areas, and variables involve in this project.
CHAPTER 2
LITERATURE REVIEW
2.1 Introduction
The main purpose of this section is to provide the essential background of PV power
generation. Besides, also reviews the previous study and research done on the PV
module performance via modeling and simulation using artificial neural network
(ANN) method. Hence, theoretical of ANN also discuss in this section.
2.2 Solar Radiation Prediction using Artificial Intelligence (AI)
Solar power is a crucial alternative energy for electrical energy production.
Photovoltaic (PV) is one of the solar power harvesting systems that use to convert
solar energy into electrical energy. Then, the performance of PV solar electricity is a
vital matter that should be taken into account by the producers of electric power. If it
is ignored, conversely it will result in efficiency for solar PV output will decline and
its impact will be involved in the outage.
8
Then, the solar radiation data play an important role in the production of
output for PV system under the diversity of ambient environmental condition. Solar
radiation is known as energy from the sun which transfers to the earth’s surface in
the form of radiant. This solar radiation data can measure by relevant equipment such
as a pyranometer; it is installed in the weather stations [12]. According to
A.Babatunde et al [12], the solar radiation data have to forecast from one location for
application in other location. This is due to the high cost and difficult to install this
type of equipment in various different locations [13]. However, by using this meter
there is the lack of accuracy for the data and it could be a missing data because of
surrounding effect.
Besides, another issue for solar radiation equipment is difficult to the
maintenance of the measuring instrument [14]. Thus, solar radiation data is not
always available at every station for instance in Sarawak, Malaysia [15] due to the
high cost to install measurement equipment where the solar radiation data only
available in the Kuching station. There is a lacking of information for meteorological
data in centers stations especially in rural areas [16]. Therefore, an alternative
method is by prediction method to obtain the data. When the data are not available,
here important use to estimate and predict it from theory or empirical models that
have been developed based on the measured values [17].
There are more study and researchers on the prediction method which
suitable to obtain data, for instance, the solar radiation data based on influence from
ambient environment condition like ambient temperature, relative humidity, and
wind speed [12]. From the studies, ANN is choosing as the method to forecast the
solar radiation data. ANN is adapting the human brain that processing of information
by parallel distributed structure known as a neural network.
According to S.Shanmuga Priya and M.H Iqbal [13], they were used ANN
method to estimate and predict global solar radiation (GSR). Their research found
that prediction using ANN more accurate compared with the conventional model.
The subject locations for their research are in Bangalore, Thiruvananthapuram,
Chennai and Hyderabad. They were used ANN fitting tool (nftool) which consists of
a standard two layers feedforward neural network (FFNN) trained with Levenberg-
Marquardt (LM) algorithm. They were developed ANN model with four inputs
parameter; average temperature, maximum temperature, minimum temperature, and
altitude.
9
The performances of model evaluated according to regression (R value) and mean
absolute percentage error (MAPE). The ANN model achieves the regression (R
value) is 88.62% for the whole model and MAPE are Bangalore (5.13%),
Thiruvananthapuram (6.29%), Chennai (7.39%) and Hyderabad (8.09%).
In addition, A.K Yadav and S.S. Chandel [14] were used four inputs; latitude,
longitude, sunshine hours and height above sea level in order to predict the solar
radiation in India station. ANN model creates using neural network fitting toolbox
(nftool) and the Levenberg-Marquard (LM) algorithm is used in this analysis. The
statistical analysis is R= 91.96 % for the total response and found that RMSE varies
from 0.0486 to 3.5262 for Indian region.
Likewise, M. Benghamen et al [17] studies four combinations inputs; global
irradiation (HG), diffuse radiation (HD), air temperature (Ta), and relative humidity
((Hµ) in the case for estimating and modeling of the daily global solar radiation. The
choice location data is in Al-Madinah, Saudi Arabia. They were choosing feed-
forward neural network (FFNN) with backpropagation training algorithms. From the
result obtained, the R-value = 97.65% with MPE = 2.2903 and RMSE = 0.044251.
The ANN model only uses one hidden layer and numbers of a neuron are in between
three to five neurons.
Moreover, James Mubiru [18] used ANN model for predicting monthly
typical daily direct solar radiation for case studies locations in Uganda. This paper
present six input; monthly average daily sunshine hours, maximum temperature,
longitude, latitude and altitude and one output are solar radiation. The output from
this paper researcher gets a good agreement between the expected and measured
values of direct solar irradiation. A correlation coefficient of 0.998 is obtained with
mean bias error of 0.005 MJ/m2 and root means square error of 0.197 MJ/m2. The
comparison between the ANN and empirical model emphasized the recommend use
ANN model for prediction. The application of the proposed ANN model can be
extended to other locations with similar climate and terrain.
Moreover, there is two type of neural network general use in prediction and
forecasting multi-layer perceptron (MLP) and radial bias function (RBF). Represent
by M.A. Behrang et al [19] which considered the combination input parameters;
daily mean air temperature, relativity humidity, sunshine hours, evaporation and
wind speed. Since this combination training and testing for two general neural
networks, there were show different result between both networks. Consequently,
10
MLP network was showed the good performance as MAPE=5.21% is lower and
R2=99.75% higher while comparing with RBF network (MAPE is 6.53% and R2 of
99.45%).
2.3 Prediction of Photovoltaic Output Using Artificial Neural Networks
Nowadays, PV system is high demand among power generation. Indeed the
performance, reliability and also quality will become a major issue for the rising
photovoltaic market worldwide. Since solar energy obtains from solar PV module is
not stable, so there become ineffective use solar energy due to the unpredictable of
the output power for PV modules [20]. Moreover, the efficiency of PV could be
decreased when there is a disturbance if many connections of PV as in PV generation
to the grid. For this reason clearly prediction tools needed to optimize the
performance of PV system. Also, evaluate the performance efficiency with a
combination of the weather conditions.
Many methods from studies that suitable to be used as the prediction tools,
like an adaptive system [21]. Artificial Neural Network (ANN) is one of the adaptive
systems could be used for modeling PV modules in two ways; either to predict the
equivalent circuit parameter or to generate the I-V curves of PV modules under
different weather conditions by using meteorology data. There are several advantages
use ANN; no need a prior knowledge of the parameter, less computational effort and
can solve the multivariable problem [16].
According to Mabel U. Olanipekun et al [20] used 3 layers ANN model that
to predict the power output generated by a CIGS thin film PV module. The inputs of
ANN network are the non-uniform; such solar radiation, module temperature, open
circuit voltage (Voc) and the short circuit current (Isc) in order to predict the PV
power output. From the experimental show coefficient, R2 = 0.8882, correlation
coefficient, cc = 0.9424, relative mean square error, RMSE = 22.9034 and mean
absolute error, MAE = 19.0016 that figured the potential of ANN model use to
forecast the output power from PV module.
Whereas, A. Saberian et al [21] utilized two model of neural network
structures there was a general regression neural network (GRNN) and feed forward
back propagation (FFBP). This model has been used to model a photovoltaic panel
output power and estimated the generated power. Both models have training and
11
testing using the same inputs are maximum temperatures, minimum temperature,
mean temperature, and irradiance; the output is the power. There was involving some
mathematical equation used to estimate the generated power using both of ANN
models. The meteorology data are used for these studies for training and testing both
of ANN models. The result was FFBP has shown a better performance compared
with GRNN with low error and R-value is 0.999 (99.9%) model accurate prediction.
Meanwhile, in Valerio Lo Brano et al [22] paper have been used three
different types of ANN model. There was Multilayer perceptron (MLP), a recursive
neural network (RNN), and a gamma memory (GM) trained with the back
propagation. All the data use for this model obtained from weather monitoring
system together with data acquisition system. The overall result this model of ANN
achieve 95% accurate of the output with the lower error. Therefore, the ANN is the
most accurate neural network compared to other AI method, this because ANN
simple and less computational method.
2.4 Overview of Photovoltaic
Solar power is rising as a major power source, gradually becoming more affordable
and proving to be more reliable for stand-alone compare with the utilities. During
1839, Edmund Becquerel discovered the process that using sunlight to produces
electricity in a solid material. However, in the century the scientist enhanced method
by learning that certain material can be used to convert the sunlight energy into the
electrical energy due to the photovoltaic effect. The energy from the sun will transfer
in form of radiant by two ways type; light and heat. This PV effect is the basic
principle process of PV cell to convert the sun energy into electricity energy. The
process is when there are the light and heat shine on the PV cell; it may receive the
reflected light and heat absorbed. The absorbed of heat and light generates electricity
energy.
The early 1950s, there is technology such as a spin-off of transistor develop
as a medium to convert light energy to electricity energy. There are also thin layers
of pure silicon for PV cell involve in the conversion of energy when it exposed to
sunlight energy the small amount of electricity will be produced. However, this
technology is expensive due to material and location.
12
Figure 2.1 and Table 2.1 shown the configuration of PV and the function of every
configuration; there are a cell, modules, panels, and arrays [23].
Figure 2.1: Configuration of PV Panel [23]
2.4.1 PV Terminology
The common of PV terminology are including PV cells, modules, panels, and arrays.
Solar cell component is used for converting sunlight to electricity, where sunlight
frees electron arresting in the silicon material. Thus, a free electron cannot return to
the positively charged and generating current. It is interconnected in parallel and
series to produce the voltage and current.
A PV module is combining of connected with the solar cells that are
compressed between a glass cover and weatherproof backing. The modules are
normally framed with aluminum frames suitable for mounting. Besides, the PV
module is rated at total power output or peak watts. The peak watt value is an amount
of PV module’s power output that produce under the standard test condition (STC) in
where the module operating temperature of 25°C in full noontime sunshine
(irradiance) of 1000W/m2.
13
Additional, solar panel is the gathered of PV modules in order to increase the
power output. Meanwhile, PV arrays are the interconnected of PV modules either in
series or parallel to form an array of modules. This will increase the total power
output to the needed voltage and current of the solar panel [24]. The summary of the
common PV terminology is listed in Table 2.1.
Table 2.1: Summary of Photovoltaic terminology [23]
Cells The semiconductor device is adapted sunlight into direct
current (DC) electricity.
Modules
PV modules consist of PV cell circuits sealed in an
environmentally protective seal and are the fundamental
building block of PV systems.
Panels PV panels include one or more PV modules gathered as a
pre-wired, field-installable unit.
Array PV array is the comprehensive power-generating unit,
involving of any number of PV modules and panels.
2.4.2 Photovoltaic Modules Technologies
There are currently four commercial production technologies for PV Modules which
used for some application around world [23, 24]
i. Single Crystalline
This is the oldest and most expensive production PV module material, but it's
also the most efficient to convert sunlight energy technology. However, PV
Module efficiency averages are about 10% to 12% [23].
ii. Polycrystalline or Multi-crystalline
This PV material technology has a little lower conversion efficiency of
sunlight energy interconnected to single crystalline. Meanwhile, the
manufacturing costs are lower and module efficiency averages about 10% to
11% almost same with single crystalline [23].
14
iii. String Ribbon
This is a modification or improvement of polycrystalline technology. It has
less module efficiency averages up to 7% until 8%, and this cost it much
lower the polycrystalline [23].
iv. Amorphous or Thin Film
The production of the thin film is by using vaporized the silicon material and
deposited on glass or stainless steel. The cost is lower than any other
technologies and the PV module efficiency averages are up to 5% (until 7%)
[23].
2.4.3 Photovoltaic Electrical Characteristic
The PV characteristic is representing by I-V curve (current-voltage relationship) as
shown in Figure 2.2 and 2.3. The I-V curve is plotted current versus voltage from
short circuit current, Isc through loading to open circuit voltage, Voc [24].
Figure 2.2: I-V curve[24]
15
Figure 2.3: P-V curve [24]
This curve is used to represents the performance of PV cells, modules, and
arrays. In order to achieve the high output, the PV module or cell will be exposed to
a constant level of irradiance while maintaining cell temperature, loads resistance,
and current produced [24].
There are two endpoints involve; open circuit voltage, Voc and short circuit
current, Isc. The open circuit voltage, Voc is the voltage that across the positive and
negative terminal under opens circuit voltage with no current due to the infinite load
resistance. Besides, Isc is the current produced by the positive and negative terminal
of cell or module. The voltage between the terminals is zero because there is zero
load resistance [24].
Moreover, the PV cell or module is operating at several of voltage and
current. According to the curve, there is difficult to determine the real efficiency of
the PV cell or module. Instead, the selected a point at the maximum power point
(Pm). Power is the product of the voltage-time current, hence Pm is the product of
maximum voltage (Vm) time maximum current (Im). Since there is no power produce
at short circuit (no voltage) and at open circuit (no current), the maximum power
value is determined at knee point of the curve. Although it representing the
maximum efficiency of the performance of PV cell or module [24, 25].
16
2.5 Neural Network (NN)
According to writer Simon Haykin, [26] a neural network is a massively parallel
distributed processor which is the unit with a simple processing, has the ability to
sorting experiential knowledge and making it available to use. It is two aspects
neural network look like a human brain. First, the knowledge or input is learned by
the network through a learning process and secondly, the inter-neuron connection
strengths is used as storage the knowledge, which known as synaptic weights.
Artificial Neural Network (ANN) is a field of study for human brain system,
which is expected to be applying in real life situation. ANN is able to act like a
human brain which to solve the problem by learning method compared to
conventional computing which is pre-programmed. ANN has a processing unit to
process information and response to the input given before giving the output desired.
Hence, it can be recognized that ANN is used as a method to replacement human
brain as a computing device which is far more powerful than conventional logic
algebra computation.
2.5.1 Basic Elements of ANN
Neural Network concept and theory involve of large networks and many elements.
To study the relationship, important things to known are learn about relevant
elements that are essential in the assembly neural network models.
2.5.1.1 Network Architecture
The organization of neuron into layers and the pattern of connection within and in
between layer are called as the architecture of the network. The neurons within a
layer are found to be fully interconnected or not interconnected. The number of
layers in the network can be defined to be a number of layers of the weighted
interconnected link between the particular the neurons [27]. If there are two layers of
interconnected weights are present, then it is found to have hidden layers as depicted
in Figure 2.4.
17
Figure 2.4: Architecture of Network of Multilayer Perceptron
2.5.1.2 Weight and Biases
The neurons are connected to each other by using the link called weights in the
network. Weight is known as the information used by the neural network to solve a
problem. Figure 2.5 show simple neural network with weight and bias [27].
The weight carries information denoted by W1 and W2, where X1 and X2 are
input and Y is output. The value can be fixed or take randomly. Weight value can
also set to zero or calculated using some method. When initialization the weight, it
also influences the performance network. A bias is a weight on a connection from a
unit whose activation is always 1. When increase the bias will increase the network
input. Bias can be either 1 or specified value [27].
Figure 2.5: Simple Neural Network
18
Net = ∑ XiWi + b (2.1)
Where, Net = Network
Xi = Input
Wi = Weight if neuron I to the output neuron
b = bias
2.5.1.3 Activation Function
An activation function also is known as a transfer function and it can be a linear and
nonlinear function. The selecting of the transfer function is to satisfy some
specification of the problem that the neuron is attempting to solve [27]. There are
three of common transfer function uses for ANN model as listed in Table 2.2;
Table 2.2: Transfer Function of Neural Network [27]
Transfer Function Description
Linear transfer function
- Function: f(x) = x
- a = purelin (wp + b)
Log-Sigmoid
- Function: 𝑦 =1
1+𝑒−𝑥
- Range is from 0 to 1
- a = logsig (wp + b)
Hyperbolic Tangent
- Function: 𝑦 =1−𝑒−2𝑥
1+𝑒2𝑥
- Range is from -1 to 1
- a = tangsig (wp + b)
19
2.5.2 Multilayer Perceptron (MLP)
MLPs are the most common type of feed-forward networks us for the neural
network. The MLP which consist of three types of the layers: an input layer, a hidden
layer and an output layer as shown in Figure 2.6 [28].
Figure 2.6: MLP network
The schematic of MLP network that consists of two layers perceptron, for
input layer, x(t) contain three input parameter, hidden layer consist three hidden units
and output, y(t) consist three output as shown in Figure 2.6. The bias weight is not
shown in the figure. As known that, bias weight is represented the extra weight for
each unit which fixed to 1 and it can be changing by adjusted same value with weight
in the network. Besides, a layer of perceptron noticed by counting the layer of weight
not layer unit [28].
For the hidden layer to output layer weights:
∆𝑤𝑗𝑘 = −𝜂𝜕𝐸
𝜕𝑊𝑗𝑘= −𝜂𝛿𝑘𝑦𝑗 (2.2)
where, 𝛿𝑘 =𝜕𝐸
𝜕𝑎𝑘= (𝑦𝑘 − 𝑡𝑘)𝑦𝑘(1 − 𝑦𝑘) (2.3)
20
For the input layer to hidden layer weights:
∆𝑤𝑖𝑗 = −𝜂𝜕𝐸
𝜕𝑊𝑖𝑗= −𝜂𝛿𝑗𝑦𝑗 (2.4)
where, 𝛿𝑗 =𝜕𝐸
𝜕𝑎𝑗= ∑(𝛿𝑘𝑊𝑗𝑘 𝑦𝑗)(1 − 𝑦𝑘) (2.5)
Equations (2.2) and (2.4) only differences between in definition of the 𝛿𝑠 in
Equations (2.3) and (2.5). The hidden layer unit, 𝛿𝑗 depend on the 𝛿𝑘𝑠 for the entire
output unit that connected using weight, 𝑤𝑗𝑘 . Moreover, the convergence is faster if
there is momentum term is added to Equations (2.3) and (2.5). Hence,
∆𝑤𝑗𝑘 = 𝑤𝑗𝑘(𝜏 + 1) − 𝑤𝑗𝑘 (𝜏) = −𝜂𝛿𝑘𝑦𝑗 + 𝛼(𝑤𝑗𝑘(𝜏) − 𝑤𝑗𝑘(𝜏 − 1)) (2.6)
∆𝑤𝑖𝑗 = 𝑤𝑖𝑗(𝜏 + 1) − 𝑤𝑖𝑗(𝜏) = −𝜂𝛿𝑗𝑦𝑖 + 𝛼(𝑤𝑖𝑗(𝜏) − 𝑤𝑖𝑗(𝜏 − 1)) (2.7)
where learning rate is 0 < 𝛼 < 1.0 and 𝜏 is the iteration number [28].
2.5.3 Multilayer Perceptron Training
The important to getting best training performance for MLP is by choose architecture
for given input parameter. Most an architecture example used is MLP with two-layer
perceptron, which with sigmoid transfer function (non-linear). This model can
approximate arbitrary accuracy the problem solving with choosing one hidden layer
units. Model selection is depending on the complexity of the problem that MLP
model needed to be solved. However, if there are too many hidden units will cause
the model overfitting and effect the interpolation of the test set becomes poor quality.
During training, the number j of hidden unit is determined with the weight for
the first layer, 𝑤𝑖𝑗 and 𝑤𝑗𝑘 weight of second layer. Thus, at least ten network
different with different hidden layer should be trained. The initial of weight is in
small random of value near zero and can be both positive and negative. Instead that,
the learning rate is considerable to generalization performance of network, then
reasonable value is in range 0.05 to 1.0 as default.
21
Before training process, the data should be partitioned into three types; train,
test and validate set. Input data is usually needed to normalized according to desired
range (0 to 1) or (-1 to 1). The pattern from training set will present the network
pattern and one epoch of training is set. It can be a hundred epochs occur during the
learning process.
Nevertheless, this learning process has to stop criteria when validation stop
decreases. The validation set is used to decide to stop training. Hence stop criterion
should reach the minimum point of error on the validation set in order to get good
performance [28].
CHAPTER 3
METHODOLOGY
3.1 Introduction
This chapter will discuss all the methods that use to develop and implement in this
project. The main purpose of this section is to provide the chronology of the project
in form of flow chart. Besides, also discuss the step by step to develop the ANN
Model use for achieving the object of this project using MLP network with three
layers.
3.2 Project Review
The project has investigated the performance of PV panel influence by weather
condition: temperature, relative humidity, wind speed and solar radiation using ANN
model. Conversely, this project will use weather data such air temperature (Ta),
temperature minimum, (Tmin), temperature maximum, (Tmax), relative humidity,
(Rh), wind speed, (Ws), and solar radiation, (G) collected from RET screen software
database. This project divides into two parts; first part it ANN Model 1 to predicts
the solar radiation data of locations are Batu Pahat in Johor, Melaka, Kuala Lumpur,
Kuching in Sarawak and Kota Kinabalu in Sabah.
23
23
Second, ANN Model 2 uses to predict the output from PV panel in term of
voltage and current. Although meteorology data used as input data of locations are
Batu Pahat in Johor, Melaka, Kuala Lumpur, Kuching in Sarawak and Kota Kinabalu
in Sabah for the second model. Then, these data will be training and testing using
ANN model that develop in MATLAB software. The target data for output will
calculate using the proposed equations that briefly more in Section 3.4. The network
topology selected for this project is MLP because of it is simple topology for
forecasting and prediction. Further, there are two ways could be used to create and
develop the ANN model either using Neural Network Toolbox or MATLAB coding.
Thus, MATLAB coding is chosen for creating and develops the ANN model in this
project as in Appendix A.
3.3 Flow Chart for Development of Forecasting Model
This project achievement method will be dividing into several stages in order to
make the project flow to become more systematic, manageable and easier to
troubleshoot. Figure 3.1 show the overall implementation process of the project that
will make the project run successfully.
24
24
Figure 3.1: Flow Chart for overall project
69
69
REFERENCES
[1] M. Bouzguenda, T. Salmi, A. Gastli, and A. Masmoudi, “Evaluating solar
photovoltaic system performance using MATLAB,” 2012 First Int. Conf.
Renew. Energies Veh. Technol., pp. 55–59, 2012.
[2] N. A. Hidayatullah, “Impacts of Distributed Generation on Smart Grid,”
Master, no. February 2011.
[3] F. B. S. Mekhtoub, “PV Cell Temperature / PV Power Output Relationships
Homer Methodology Calculation,” Ipco, vol. 2, no. 1, pp. 1–12, 2014.
[4] Ayu Wazira Azhari, Kamaruzzaman Sopian, Azami Zaharim, “A New
Approach For Predicting Solar Radiation In Tropical Environment Using
Satellite Images – Case Study Of Malaysia,” Environ. Dev., vol. 4, no. 4, pp.
373–378, 2008.
[5] C. B. Ahmed, M. Kassas, and S. Essamuddin, “PV-standalone monitoring
system performance using LabVIEW,” Int. J. Smart Grid Clean Energy, vol.
3, no. 1, pp. 44–50, 2014.
[6] B. Bora, O. S. Sastry, and R. Dahiya, “Artificial neural network based
modeling of PV module to predict the output,” no. December 2013.
[7] I. Ceylan, O. Erkaymaz, E. Gedik, and A. E. Gurel, “The prediction of
photovoltaic module temperature with artificial neural networks,” Case Stud.
Therm. Eng., vol. 3, pp. 11–20, 2014.
[8] J. Wong, Y. S. Lim, J. H. Tang, and E. Morris, “Grid-connected photovoltaic
system in Malaysia: A review on voltage issues,” Renewable and Sustainable
Energy Reviews. 2014.
[9] A. M. Ameen and J. Pasupuleti, “Sensitivity Of Artificial Neural Network
Based Model For Photovoltaic System Actual Performance,” pp. 241–244,
2014.
[10] T. Khatib, A. Mohamed, K. Sopian, and M. Mahmoud, “Solar Energy
Prediction for Malaysia Using Artificial Neural Networks,” vol. 2012, pp. 82–
87, 2012.
[11] K. Karabacak and N. Cetin, “Artificial neural networks for controlling wind–
PV power systems: A review,” Renew. Sustain. Energy Rev., vol. 29, pp. 804–
70
70
827, 2014.
[12] A. Babatunde and O. Olusola, “Regional Estimation of Solar Radiation using
Routinal Meteorological Variables,” J. Environ. Earth Sci., vol. 5, no. 17,
2015.
[13] S. S. Priya and Mohammad Hashif Iqbal, “Solar radiation prediction using
Artificial Neural Network,” Int. J. Comput. Appl., vol. 116, no. 16, pp. 28–31,
2015.
[14] A. K. Yadav and S. . Chandel, “Artificial Neural Network based Prediction of
Solar Radiation for Indian Stations,” Int. J. Comput. Appl., vol. 50, no. 9, pp.
1–4, 2012.
[15] A. . Jakhrani, A. . Othman, A. R. . Rigit, and S. . Samo, “Model for Estimation
of Global Solar Radiation in Sarawak, Malaysia,” World Appl. Sci., vol. 14,
pp. 83–90, 2011.
[16] H. G. Chua, B. C. Kok, and H. H. Goh, “Solar Insolation Forecast Using
Artificial Neural Network For Malaysian Weather,” no. April, pp. 4–9, 2012.
[17] M. Benghanem, A. Mellit, and S. N. Alamri, “ANN-based modelling and
estimation of daily global solar radiation data: A case study,” Energy Convers.
Manag., vol. 50, no. 7, pp. 1644–1655, 2009.
[18] J. Mubiru, “Using Artificial Neural Networks to Predict Direct Solar
Irradiation,” Adv. Artif. Neural Syst., vol. 2011, pp. 1–6, 2011.
[19] M. A. Behrang, E. Assareh, A. Ghanbarzadeh, and A. R. Noghrehabadi, “The
potential of different artificial neural network (ANN) techniques in daily
global solar radiation modeling based on meteorological data,” Sol. Energy,
2010.
[20] M. U. Olanipekun, J. L. Munda, G. Chen, and A. S. Kumar, “Estimation of
maximum power generated from CIGS photovoltaic modules under non -
uniform conditions.”
[21] A. M. Saberian, H. Hizam, M.A.M Radzi, M.Z.A Ab Kadir, and Mirzaei,
“Modelling and Prediction of Photovoltaic Power Output Using Artificial
Neural Networks,” Int. J. Photoenergy, vol. 2014, 2014.
71
71
[22] V. Lo Brano, G. Ciulla, and M. Di Falco, “Artificial Neural Networks to
Predict the Power Output of a PV Panel,” Int. J. Photoenergy, vol. 2014, p. 12,
2014.
[23] Solar direct, “Solar Direct,” Solar direct. [Online]. Available:
http://www.solardirect.com/pv/pvlist/pvlist.htm. [Accessed: 21-Mar-2016].
[24] R. Foster, “Solar Energy,” in Renewable Energy And The Environment,
United State of America: CRC Press Taylor & Francis Group, 2010, pp. 135–
155.
[25] E. Durán, J. M. Andújar, J. M. Enrique, and J. M. Pérez-Oria, “Determination
of PV generator I-V/P-V characteristic curves using a DC-DC converter
controlled by a virtual instrument,” Int. J. Photoenergy, vol. 2012, 2012.
[26] S. Haykin, Neural Network: A Comprehensive Foundation, Second. India:
Person Prentice Hall, 2005.
[27] S. Sumathi, S. N. Sivanandam, and S. N Deepa, Neural Network Using
Matlab 6.0, Second. New Dehli: Tata McGraw-Hill Education, 2006.
[28] L.Tarassenko, A Guide Neural Computing Application. New York, 1998.
[29] W. Zhou, H. Yang, and Z. Fang, “A novel model for photovoltaic array
performance prediction,” Appl. Energy, 2007.
[30] G. Tamizhmani, L. Ji, Y. Tang, L. Petacci, and C. Osterwald, “Photovoltaic
Module Thermal/Wind Performance: Long -Term Monitoring and Model
Development For Energy Rating,” 2003.
[31] S. S. B. Jamal, “An Application of Artificial Neural Network for DC servo
Motor,” Universiti Teknologi Malaysia, 2009.
[32] K. A. Baharin, H. Abdul Rahman, M. Y. Hassan, and C. K. Gan, “Short-term
forecasting of solar photovoltaic output power for tropical climate using
ground-based measurement data,” J. Renew. Sustain. Energy, 2016.
[33] A. Ghazali M. and A. M. Abdul Rahman, “The Performance of Three
Different Solar Panels for Solar Electricity Applying Solar Tracking Device
under the Malaysian Climate Condition,” Energy Environ. Res., vol. 2, no. 1,
pp. 235–243, 2012.
