Prediction of the Power Output of Solar Cells Using Neural Networks: Solar Cells Energy Sector in Palestine

Ibrahim Qasrawi & Mohammad Awad
International Journal of Computer Science and Security (IJCSS), Volume (9) : Issue (6) : 2015 280
Prediction of the Power Output of Solar Cells Using Neural
Networks: Solar Cells Energy Sector in Palestine
Ibrahim Qasrawi iqasrawi@outlook.com
Faculty of Engineering and Technology
Computer science
Arab American University-Jenin
Jenin, 240, Palestine
Mohammad Awad mohammed.awad@aauj.edu
Faculty of Engineering and Technology
Computer Engineering
Arab American University-Jenin
Jenin, 240, Palestine
Abstract
The prediction of the output power of solar cells in a given place has always been an important
factor in planning the installation of solar cell panels, and guiding electrical companies to control,
manage and distribute the energy into their electricity networks properly. The production of the
electricity sector in Palestine using solar cells is a promising sector; this paper proposes a model
which is used to predict future output power values of solar cells, which provides individuals and
companies with future information, so they can organize their activities. We aim to create a model
that able to connect time, place, and the relations between randomly distributed solar energy units.
The system analyzes collected data from units through solar cells distributed in different places in
Palestine. Multilayer Feed-Forward with Backpropagation Neural Networks (MFFNNBP) is used to
predict the power output of the solar cells in different places in Palestine. The model depends on
predicting the future produce of the power output of solar cell depending on the real power output
of the previous values. The data used in this paper depends on data collection of one day, month,
and year. Finally, this proposed model conduct a systematic process with the aim of determining
the most suitable places for an installation solar cell panel in different places in Palestine.
Keywords: Neural Networks, Solar Cell Energy, Prediction.
1. INTRODUCTION
Society starts searching for other energy resources with more awareness of their responsibility to
the environment. Renewable energy experiences a huge expansion, which raises many challenges
to the scientific and technical community [1]. Renewable energy is a clean and inexhaustible
energy, and it’s technology is developing rapidly [1]. In recent years, solar cells increasingly used
to produce the energy in Palestine, one of the problems faced are the unpredictability in their
production, since they depend on climatic variables of each moment [2]. Currently, distribution
companies are responsible for managing and selling the energy into their networks. Palestine has
seen an increase in the number of solar parks. The current Palestinian society aims to reduce
electrical energy dependence, a better use of resources and greater awareness of the environment.
There have been many initiatives in Palestine by the directives of support measures, renewable
energy, establishing growth targets for renewable technologies in order to get more energy
production comes from these resources [3]. Solar energy is the energy produced by thermonuclear
reactions continuously on the surface of the sun, this energy spread in the universe with different
energy forms such as light, heat, x-ray, and ultraviolet, these forms of energy forms the solar
spectrum. The solar energy is a vital factor of life on the earth, solar spectrum hits the earth with
half of its energy level because of the atmospheric shield and the reflection of the energy of the

earth's surface, and this energy can be used in different fields like agriculture, heating, cooking,
and electricity production [4, 5]. One of the major solar energy usages is to produce electricity, the
process that can perform mainly in two methods, thermal solar energy which depends on heating
fluids in pipe network running turbines to produce electricity, and photovoltaic solar energy, which
uses photovoltaic phenomenon to produce electricity directly from the sunlight [6, 7].
In recent years, solar cells have a deployment unprecedented. The total energy supply from solar
cells in Palestine in the year 2007 accounted of 1402 ktoe. The indigenous production contributed
19% of total primary energy supply while the remaining quantity imported from Israel [3]. Solar cell
power in Palestine is a natural wealth, which must be explored; the electric distribution companies
receive the electricity in its electrical network from two resources; the direct electrical lines from
Israel and the solar cell panels installed in some areas in Palestine. There are no plans or
estimation process, which used by these companies to determine the demand and the production
of solar cells in such place, so the prediction of solar cell output power in such place, will help the
company to plan and manipulate the electricity in this area, which will produce stability in electrical
connection. For many years, they have tried to learn how to predict future events, so that, they can
take preventive action. weather conditions are one of the future events, which directly or indirectly
affect us every day, especially in electricity consumption. Therefore, they have developed methods
of climate prediction, supported by different models to predict solar cell output power using
mathematical prediction systems [8].
In general, a lot of researches have been developed for this purpose, but the most of them, use
statistical methods to get results and historical analysis for the collected energy, without using
prediction and classification models in order to create a solar energy map, that shows the best
regions and provide an advice for people who are concerned in the energy field. In [9] the author
used artificial neural network (ANN) control algorithms applied to the solar energy prediction, they
proposed an algorithm detects the optimal operation point for photovoltaic and thermal panels by
studying the (PV/T) model behavior considering irradiation and ambient temperature. Another
related study was made to determine the highest time horizon for generating solar energy prediction
by Ercan Izgi et al [10], they used small solar power system application to study and predict the
time horizon by dividing the study period of time into short term [5min] in medium term in April, and
[3 min] for short term, [40 min] for medium term in August, during April and August RMSE between
the measured value and testing value changed between 33-55 in April and 37-63 in August , ANN
algorithm used to predict the electricity generated in period 30-300 minutes. The external weather
conditions that affect the solar energy generation are studied by Esteban Velilla and others in [11],
they used two modules of solar cells [ mono-crystalline (55w) and organic solar module (12.4w),
the factors monitored by this study are the temperature, relative humidity, and irradiance, that are
used as inputs for an ANN algorithm, which developed by the team to train ,validate and test the
electric power generated, the result obtained of solar energy produced using organic solar module
was better than mono-crystalline module in the extreme conditions of (high temperature, high
humidity, and lower irradiance). Electricity sector in Palestine was studied by Ayman Abu AlKher in
[12] and others, the study has been divided in two parts, one part to study the current situation of
electricity production, consumption, and transmission, the second part is a comparison between
Palestine and other neighbors, to highlight the electricity consumption gap, and he used a
mathematical and economical model to predict the relation between electricity consumption and
economic growth. Solar radiation forecasting study made by Bader M. Alluhaidah and others [13]
explained the most effective variables that are used in the solar forecasting process as inputs for
ANN, the case study was made in Saudi Arabia. The simple structure offers better results in terms
of error between actual and predicted solar radiation values. A method for modelling and prediction
of PV-generated power [14] has been developed by Amin Mohammed Sabrian et al, this method
uses two kinds of ANN, general regression neural network (GRNN) and feed forward back
propagation (FFBP), in the modelling process, he used four inputs for both ANN (max temperature,
min temperature, mean temperature, and irradiance) with the power as an output. The data were
collected through 5 years from 2006 to 2010 period, it was split into two parts, first 3 years data
used for training and 2 years data used for testing, the result in both methods give good results,
where FFBP has better performance than GRNN.

One common method for data collection is the using of special data acquisition system based on
the wireless sensor network, this method of data collection uses the wireless sensor network
connected by Wi-Fi or Wi-Max technologies, to achieve a reliable connection between far sensor
nodes [15]. The other useful method is using the local electricity providers, where they have a very
useful projects for solar cells energy. They provided us with a very good amount of data that we
can use as a start point in our prediction and classification model, we get solar cells values from Al
Ojah village in the East, Al-Zababdeh village in the north, and Ramallah city in the middle of
Palestine. Another way of collecting previous data is the used of the data in the website "Solar
GIS", which is a website that integrated with satellites, google maps and other measurement tools,
to measure and estimates the atmospheric variables like temperature, humidity, and solar
irradiation, that are the main factors which affect the solar energy production all over the world.
The general objective of the proposed model is to develop a system based on Neural Networks
(NN), which can predict the short-term values of an output power of solar cells over days, months
and one year. Which aims to help companies in planning, managing and forecasting the suitable
time and place for the energy production from solar cells. The data used for the process of learning
in this model was taken for one year (one value in each 5 minutes) from different locations in
Palestine, which allow a possible approximating value of the total power acquired by solar cells. In
another hand, it analyzes the behavior of the time series that determine the value of solar cells in
one year for different cities of Palestine.
2. ARTIFICIAL NEURAL NETWORK
Supervised NNs depend on the provided input data, this data will be processed using special
activation functions, with the final goal of producing output. The current output of the NN will be
compared with the desired output for the purpose of training the NNs by updating the weights in
each epoch, which produce decreasing in approximation error. Artificial neurons are elements with
an internal state that changes depending on the signals it receives, such neurons also have a
transition function, which allows them to change the level of activation signals received from
neurons, whether connected or from the outside [16]. NNs have a basic structure consisting of a
series of entries that reach the neuron, and one or more output are connected to the input of another
neuron network, this structure is known as a multi-layered NN [17] as shown in Figure 1.
FIGURE 1: Basic Topology of Neural Network.
This type of structure has; an input layer which consists of input patterns. The connections in the
hidden layer which are weighted connections between the input layer and the hidden layer, a
processing units called neurons receives the input data from the input layer and processing it
depends on the activation function, finally the output layer that is used to calculate the output of
the NN. NNs are able to learn from input/output data to predict future value. Knowledge of this
value can be performed in a time step, which is obtained from samples available at time t, and it

can generate a value for time t + 1. It is also possible to predict multiple time steps, which involves
taking other values known as the predicted values to generate new future value, NNs play good
rule in time series prediction.
The general main steps for the basis NN start by a collection of the data to be used in training,
validation, and testing. Preprocessing of the data which is an important stage that increases the
performance of the NN, and reduces the number of processors (neurons) required, this will lead to
reduce the number of neurons, and give the best curve fitting. Initialization of the NN starts by
assigning values for the input weights, this process that most likely done randomly, or using certain
algorithms according to the problem. In the NN training process, the activation function is applied
to find the relation between the inputs and the outputs, and then updating the weights using the
specified algorithms with a number of training data in order to get the best results. To check the
validation of the NN, this model uses another set of data called validation data. Finally, the
generalization process tests the NN for random and different dataset [17]. The Solar cells output
power prediction using silicon cells is affected by several factors like solar irradiation, climate,
temperature, relative humidity, dusty weather, cell direction, and the efficiency loss by the time [6].
In our proposed model we will use historical values of the solar cell output power recorded for a
period of time (one year), the output values are recorded in (5 min) time horizon, this output power
generated and recorded from the target output of the ANN [18].
In order to design the proposed NN, we must identify the problem domain and the factors that affect
or determine the problem, it’s known that the time series prediction is one of the most complex of
the real world application, and it’s also well known that the ANN has a good property of solving
such complex problems. The training process is the mapping process between the input/output
data of the NN when the input patterns provided to the NN with initial weights, the output of the NN
are given by the following expression:
1
( )
m
i ij j i
j
y f w x b

  1
Where Wij is the weights connection, and Xj are the value of the ith inputs for a simple of the NN, bi
is the NN bias, m is the number of neurons and f is the activation function. The general
approximation criterion which uses to determines the improvement of the prediction process, is the
error result, which comparing the actual output of the NN with the desired output in the learning
process, this error is basically calculated using the following expression:
id iaEr y y  2
Where yid is the desired value of the output for each ith element and yia is the actual value of the ith
element, normally this criterion is used as a termination condition to stop the prediction process. In
this paper, we use the root mean square error which presented by the following expression:
2
1
( )
n
id iai
y y
RMSE
n


 
 3
Where n is the number of the input data, and ϴ is the threshold value of the prediction process.
The training process continue to adjust the weights until the error criteria are satisfied, the weights
updated is performed by the equation 4:
1 . .i iw RMSE x  4
Where α is the learning rate.
One advantage of the multilayer perceptron neural networks (MLPNN), is that it can predict any
time series function if a configuration and an adequate number of neurons in the hidden layers are

available. The MLPNN is an excellent time series prediction, although it is impossible to find a
single configuration for each application. The learning process of the MLPNN with backpropagation
algorithm is not fixed to any application; a successful method is to try different settings until you get
the desired response. The choice of training patterns is performed depending on the explicit needs
of the prediction, which will show on the output and the quality of information available. Any
changes in the patterns of training require different training parameters of the NN, but the training
process remains the same.
3. METHODOLOGY
Developing a methodology to establish prediction is a relationship as accurately and precisely as
possible. The values of future solar cell energy output require the knowing of the previous solar cell
energy output, these values are used as input to the model, xt =F(xt,…,xt-tw) + εt , where xt is the
forecasting forward steps with respect to time t, F is the modeling function between the previous
and future values, εt is the modeling error. Predicting of solar energy output can be performed
using different techniques as; prediction by numerical models, prediction by statistical methods,
time series prediction based on the application of statistical techniques linear and nonlinear, and
prediction coefficient cloud cover from satellite images. Obtaining a final prediction model of solar
energy output based on the time series prediction using NNs. Predicting the energy output of solar
cell using only numerical models has a high bias and a high mean square error, which depends
on the distribution function of the radiance data for the station (at positions predominantly clear sky
conditions with errors are smaller). For this reason, we used Multilayer feed- forward with
backpropagation neural networks (MFFNNBP) and data of different months, each month presents
one season of the year.
Multilayer feed-forward with backpropagation neural networks (MFFNNBP) is an MLPNN that
passes the inputs and the weights from one layer to the next one through the feed forward process
and then it performs the weights update to be back-propagated to the previous layers in order to
recalculate the weights. Our proposed ANN architecture has three parts, one for producing solar
energy prediction depending on previous real output measured along one year "2014" from three
main solar panels located in Ain-Mousbah – Ramallah, Al-Oja – Jerico, and Al-Zababdeh- Jenin.
These three energy collection points provide the power output within 5,10,30,60 minute’s horizon
of one month or one-year time horizon, then the data was processed according to the following
process shown in figure 2.
FIGURE 2: Proposed Solar Energy Prediction ANN.
The sigmoid activation function, f1 is used [14]. In MLPNN, the output of a layer will be an input
for the next layer passing from the input layer to the output layer; the equations used for this
procedure are illustrated as follows:
Input: time series /
1 hour time horizon
2014
MLPANN:
1, to 10 Neurons
Predicted solar
energy product

2
1
1
( . )
n
jk
j
output f out w

  5
Where the output of the first hidden layer out1, which calculated using the following expression:
1
1
1
( . )
n
i ij
j
out f in w

  6
Where f1 and f2 are the activation functions for output layer and hidden layer, which calculated as
in the following expressions:
1 1
1 x
f
e


7
Where, x = input vector. Depending on equations above, the weights are updated use as the
following expression:
( )n
jkn
jk
jk
dE w
w
d w
   9
Where µ is the learning rate (normally between 0 and 1). The final output depends on all earlier
layer's output, weights, and the algorithm of learning used [14]. Using data collected from previously
mentioned solar panels, this data was preprocessed to reduce the noise from the input signal, this
process increases the performance of the ANN and reduces the prediction error, using Matlab
function called smooth function, then takes the signal and filters the noise by using average filter
smooths data, then replacing each data point with the average of the neighboring data points
defined within the span. This process is equivalent to low-pass filtering with the response of the
smoothing given by the next difference equation:
        12 1 1 ...ys i N y i N y i N y i N         10
Where ys(i) is the smoothed value for the ith data point, N is the number of neighboring data points
on either side of ys(i), and 2N+1 is the span.
The backpropagation process calculates the gradient decent error between the desired and the
predicted output considering the new weights each time, this gradient is almost always used in a
simple stochastic gradient descent algorithm to find the weights that minimize the error. Different
algorithms are used for training the feed forward with backpropagation neural networks, which train
the NN and reduce the error values by adjusting and updating the weights and the biases of the
connections that form the neural network, two kinds of training algorithms are available to slow
convergence according to steepest descent methods with better generalization, and fast
convergence according to newton's method, but these methods are complex because of the
complex matrix calculations [19]. In our paper, we use one of the fast convergence algorithms,
which is the Levenberg Marquardt Algorithm (LM) training algorithms [20], implemented by Matlab
7.1, and we use it in two steps; one is the training of time series using the time as input and the
power generated by solar energy points as output, and the second step is to train the data produced
by the factors that affect the energy production along the time.
2
f x 8

FIGURE 3: Flowchart of MFFNNBP Model.
4. RESULT AND DISCUSSION
Experiments have been performed to test the proposed prediction model. The model is simulated
in MATLAB 7.1 under Windows 7 with processor i5. In this section, different examples are given
to verify the procedure in the proposed model. Three different results are presented; One day, one
month and one year. The results of the validity of the model in prediction samples of I/O data,
compared with real results of the solar cell energy product of the last year in different areas in
Palestine. The results are obtained in 5 executions; {# of neurons} the set of neurons used in each
MFFNNBP. # of Epochs is the number of the execution cycle of the MFFNNBP. RMSEtest is the
Root mean squared error of the training.
4.1 One Day Solar Energy Prediction
For one day prediction, we use data for each hour, which present the mean of all the read values
in each 5 minutes. The day (1- Aug-2014) was selected with good climate conditions, like clear sky,
long daytime, high solar irradiance, and medium humidity level, these conditions are highly effected
the solar energy production. Applying the proposed model for one day of solar prediction produces
the following results:
Calculate the prediction output
Start
Set initial weights for the NN
Calculate the error between the
predicted and the actual output
Read the actual output
Calculate ∆𝒘 for all weights from the
hidden layer to the output layer
Calculate ∆𝒘 for all weights from the
input layer to the hidden layer
Update the weights of the network
Error ≤ ϴ
c
Stop
Increment neurons by 1
Forward
Phase
Backward
Phase
Backward
Phase

Solar prediction NN (# of neurons) Epochs #
Train
Data
Test
Data
Validation
Data
RMSEtest
2 11 70% 15% 15% 7.6X10-5
4 29 70% 15% 15% 3.16X10-5
6 19 70% 15% 15% 1.64X10-5
8 22 70% 15% 15% 3.3 X10-6
10 19 70% 15% 15% 1.8 X10-6
TABLE 1: Results of Applying Proposed MFFNNBP in One-Day.
As seen from table 1, MFFNNBP plays a good role in solar prediction for one day, the real output
for one day is a curve with some noise that are removed by using a smooth function as shown in
figure 4. With one hidden layer of a small number of neurons, the proposed model predicts the
future energy produced by the solar cell value in one day. This evaluated to find the RMSE of the
MFFNNBP. Table 1 represents the test RMSE. In addition, figure 5 shows that the production
process produces a fitting curve, with small values of RMSE.
FIGURE 4: Power Generated for One Day.
4.2 One Month Prediction
To prove the process of predictions of the proposed models, we Applying the proposed MFFNNBP
to predict energy that can produce in one month as the second part of the study, where July period
of time was selected. Data was recorded for ALFA’A solar panel along the year 2014, and we built
the training set for the July month record. The process was accomplished by taking the mean value
of the generated power every 10 days with 10 readings per a day that can be illustrated in figure 5.
Table 2 shows the results of applying proposed MFFNNBP for the training set which is formed of
30 examples, the results show that the best prediction result was achieved when hidden layer of
10 neurons was trained to give the following figures. According to the figure 6, the best prediction
is achieved with 10 neurons in the hidden layer and 18 epoch. MFFNNBP produce small RMSE in
monthly measured.

FIGURE 5: Mean Values Solar Energy Produced in July 2014.
Solar prediction NN (# of neurons) Train Test Validate Epochs # RMSEtest
2 70% 15% 15% 8 0.091638
4 70% 15% 15% 9 0.072802
6 70% 15% 15% 11 0.000436
8 70% 15% 15% 18 0.000413
10 70% 15% 15% 18 0.000236
TABLE 2: Model Training Results for Energy Prediction at July (One-Month).
FIGURE 6: The Best Prediction Result of the Training Process for One-Month.

4.3 One Year Prediction
In order to get energy production along one year we need to deal with complex time series, figure
7 below shows the time series signal for one year [1-Jan-2014 to 1-Jan-2015], so we divided the
year into four parts representing the four seasons of the year, dealing with each part as the mean
value of the produced solar energy. As we see the high complexity of this signal, so we need to
process it before we use it as target function to be predicted using the MFFNNBP model. This
process started by; dividing the year into four parts, remove the night time where the energy
produced goes to zero, reduce the large amount of data as Mean of one month = Mean [mean (1st
,2nd , 3rd days) + mean (14 ,15 , 16 days) + mean (28 ,29 , 30 days), reduced the noise using
smoothing function, normalization of the signal data, and applying the proposed model using
several probabilities to get the best prediction results.
FIGURE 7: Energy Produced on One-Year.
FIGURE 8: The Best Prediction Result of the Training Process for One-Year.

Figure 8 shows the best result for one-year prediction which was achieved when we applied the
proposed model with 80 neurons in the hidden layer, and also shows the complexity of one-year
time series solar energy prediction. As we see from figure 8 the error was at its minimum value with
90 neurons hidden layer is used [0.00564], which is the best error value, while when using more
than 90 neurons the system will produce an overfitting problem. In all tables and approximation
curves, we can observe the RMSE of the prediction, which clearly shows that the prediction is
closest to the real value, regardless of the loss in the efficiency of the solar cells in energy
production decreases by %5 in each year, all with a small number of epochs. The prediction
process using the proposed model MFFNNBP was achieved using 80 neurons in the hidden layer,
as shown in table 2 below:
Solar prediction NN(# of neurons) Train Test Validate Epochs # RMSEtest
20 70% 15% 15% 7 0.070022
30 70% 15% 15% 8 0.067588
40 70% 15% 15% 9 0.064546
50 70% 15% 15% 22 0.039261
60 70% 15% 15% 11 0.035956
70 70% 15% 15% 11 0.01505
80 70% 15% 15% 10 0.008242
90 70% 15% 15% 12 0.00564
100 70% 15% 15% 12 0.008293
TABLE 3: Model Training Results for Energy Prediction for One-Year.
As we can see in the tables and figures, it is possible to observe the correlation values between
real solar cell output power and power output obtained from the prediction using the proposed
model MFFNNBP, that are measured in the same place. It is clear that we can consider the
possibility of including other information as input to the model like, temperature, radiation, and
humidity, and the time duration of the sunlight. This may perform better prediction of solar cell
output power. Hence, the idea to predict more deeply the correlation between climatic parameters
using NNs. The use of the MFFNNBP model for prediction of climatic variables becomes a viable
solution for the knowledge of these values. The implementing of this model to solve a prediction
problem is necessary to determine the optimum areas, where the solar cell panels can produce a
more efficient power output. The obtained result, which use the MFFNNBP model for solar energy
prediction, different attributes like training algorithm, training data, a region of study will change the
values of the prediction result. This result obtained using real output data (previous data) as a
measure of the prediction improvement, or prediction of the actual output using some input
parameters. In the future work we will study the prediction process using the input data parameters
that affect the power output of the solar cells like; temperature, radiation, and humidity, and the
time duration of the sunlight. Furthermore, we will apply the model in two different types of solar
cells; mono-crystalline and organic solar module, in the aim to determine the best type for our
region.
5. CONCLUSION
In this paper, we used the energy produced data from solar panel located in different Places in
Palestine for the year 2014, to train and test prediction technique that uses multilayer feed forward
with backpropagation neural networks (MFFNNBP) trained using Levenberg-Marquardt algorithm,
the model predicted the solar cells energy production for one day, one month and finally for the
whole year with very high accuracy and low relatively number of processing units (neurons) to
accomplish this task. The use of MFFNNBP for the prediction of solar cell energy output becomes
a viable solution for the knowledge of these future values. In fact, the introducing of another
meteorological system such as radiation, temperature, and sunshine hours and humidity may

perform a significant improvement in prediction. From these results, we can predict the solar cell
energy output for the next year, and we notice that August was the best month of the solar cell
energy production, that is because of the clear sky, intermediate temperature, and long daytime
which gives a long time of solar irradiation to produce energy. By looking at the figures obtained by
the model you can perceive that this model predicts the future solar cell energy output in an
accurate form.
6. REFERENCES
[1] N. L. Panwar, S. C. Kaushik, and S. Kothari, “Role of renewable energy sources in
environmental protection: a review,” Renewable and Sustainable Energy Reviews, vol. 15, no.
3, pp. 1513–1524, 2011.
[2] Yaseen, Basel T. Q. “Renewable Energy Applications in Palestine”, Palestinian Energy and
Environment Research Center (PEC)- Energy Authority, 2nd International Conference on the
Palestinian Environment 2007.
[3] Yaseen, Basel T. Q. “Renewable Energy Applications in Palestine”, Palestinian Energy and
Environment Research Center (PEC)- Energy Authority, 2nd International Conference on the
Palestinian Environment 2007.
[4] L. Mart´ın, L. F. Zarzalejo, J. Polo, A. Navarro, R.Marchante, and M. Cony, “Prediction of global
solar irradiance based on time series analysis: application to solar thermal power plants energy
production planning,” Solar Energy, vol. 84, no. 10, pp. 1772– 1781, 2010.
[5] Duffie.J.A, and Beckman.W.A.(2006), Solar Engineering of Thermal Processes. John Wiley
Sons, 3rd edition.
[6] R.Ramaprabha and Dr.B.L.Mathur,”A Technique to extract maximum Power from Photovoltaic
Panels”,Proc. of IEEE Int. Conf. on Recent Advancements and Applications of Computer in
Electrical Engineering, pp. 447 –449, Bikaner, Rajastan, India, Mar. 24-25, 2007.
[7] Engin Karatepe, Mutlu Boztepe and Metin Colak, “Development of suitable model for
characterizing photovoltaic arrays with shaded solar cells”, Solar Energy, 2007, pp 329-340.
[8] A. Mellit and A. M. Pavan, “A 24-h forecast of solar irradiance using artificial neural network:
application for performance prediction of a grid-connected PV plant at Trieste, Italy,” Solar
Energy, vol. 84, no. 5, pp. 807–821, 2010.
[9] Errachdi Ayachi, Saad Ihsen, Benrejeb Mohamed,"A Comparative Study of Nonlinear Time-
Varying Process Modeling Techniques: Application to Chemical Reactor.” Journal of Intelligent
Learning Systems and Applications Vol.4 No.1 (2012).
[10] Ercan _Izgi a, Ahmet O¨ ztopal b, Bihter Yerli b, Mustafa Kemal Kaymak b,Ahmet Duran Sahin
b,” Short–mid-term solar power prediction by using artificial Neural networks”, December 2011,
Elsevier, SciVerse ScienceDirect, solar energy 86(2012) 725-733.
[11] Esteban Velilla a, Jaime Valencia a, Franklin Jaramillo. Performance evaluation of two solar
photovoltaic technologies under atmospheric exposure using artificial neural network models,
Elsevier, solar energy [2014].
[12] Ayman Abualkhair,” Electricity sector in the Palestinian territories: Which priorities for
Development and peace?” Elsevier, Energy Policy 35 (2007) 2209–2230.
[13] Bader M. Alluhaidah, “MOST INFLUENTIAL VARIABLES FOR SOLAR RADIATION
FORECASTING USING ARTIFICIAL NEURAL NETWORKS”, Dalhousie University Halifax,
Nova Scotia [June 2014].

[14] Aminmohammad Saberian, H. Hizam, M. A. M. Radzi,M. Z. A. Ab Kadir, andMaryam Mirzaei,
“Modelling and Prediction of Photovoltaic Power Output Using Artificial Neural Networks”,
Hindawi Publishing Corporation International Journal of PhotoenergyVolume 2014, Article ID
469701, 10 pages, [April 2014].
[15] N. S. Shrirao ,Mr D. H. Bodkhey, Mr. Sapan Kumar Singh. “A Review of Sensor Networks:
Challenges and solutions “, International Journal of ICT and Management, February 2013
Vol- I Issue –I.
[16] K.Hornik, M. Stinchcombe, and H.White, “Multilayer feedforwardnetworks are universal
approximators,” Neural Networks, vol. 2, no. 5, pp. 359–366, 1989.
[17] M. T. Hagan, H. B. Demuth, and M. Beale, Neural Network Design, PWS Publishing Company,
Boston,Mass, USA, 1995.
[18] Simon S. Haykin, (2008), Neural Networks-A Comprehensive Foundation, 2nd Edition. ISBN-
13: 978-0132733502.
[19] K. Madsen, H.B. Nielsen, and O. Tingleff. Methods for Non-Linear Least Squares Problems. T
echnical University of Denmark, 2004. Lecture notes.
[20] levmar- Manolis I. A. Lourakis, [February 11, 2005],” A Brief Description of the Levenberg-
Marquardt Algorithm Implemented”, Institute of Computer Science-Foundation for Research
and Technology - Hellas (FOR TH)V assilika V outon, P .O. Box 1385, GR 711 10 Heraklion,
Crete, GREECE.

Prediction of the Power Output of Solar Cells Using Neural Networks: Solar Cells Energy Sector in Palestine

More Related Content

What's hot (19)

Similar to Prediction of the Power Output of Solar Cells Using Neural Networks: Solar Cells Energy Sector in Palestine (20)

Recently uploaded (20)

Prediction of the Power Output of Solar Cells Using Neural Networks: Solar Cells Energy Sector in Palestine