Available online at www.sciencedirect.

com
Available online at www.sciencedirect.com
Energy
Procedia
Energy Procedia
Energy 00 (2011)
Procedia 000–000
14 (2012) 2066 – 2074
www.elsevier.com/locate/procedia

2011 2nd International Conference on Advances in Energy Engineering (ICEAA 2011)

An Assessment of Diffuse Solar Energy Models in Terms of

Estimation Accuracy
Tamer Khatiba, Azah Mohameda, Marwan Mahmoudb and K. Sopianc
a
Department of Electrical, Electronic & System Engineering, Faculty of Engineering & Built Environment, Universiti
Kebangsaan Malaysia, Bangi 43600, Selangor, Malaysia.
b
Department of Electrical Engineering, An-Najah National University, Nablus, Palestine
c
Solar Energy Research Institute, Universiti Kebangsaan Malaysia, Bangi 43600, Selangor, Malaysia

Abstract

This paper presents models for diffuse solar energy on a horizontal surface for main five sites in Malaysia. The
diffuse solar energy is modeled using linear, non-linear and ANN models. Three statistical values are used to evaluate
the developed solar energy models, namely, the mean absolute percentage error, MAPE, root mean square error,
RMSE and mean bias error, MBE. The results showed that the ANN models are superior compared to the other
models in which the MAPE of the ANN model is 1.53%, while the MAPE of the linear and nonlinear models are 4.35%
and 3.74%, respectively. The accurate ANN models can therefore be used to predict solar energy in Malaysia and
nearby regions.
© 2011 Published by Elsevier Ltd. Selection and/or peer-review under responsibility of the organizing committee
of
© 2nd International
2011 Published byConference
Elsevier on Advances
Ltd. Selectioninand/or
Energypeer-review
Engineeringunder
(ICAEE).
responsibility
Open accessofunder
[ICAEE 2011] license.
CC BY-NC-ND
Kewqords: Solar energy modeling, diffuse solar energy, ANN, Malaysia

1. Introduction

Solar energy is part of the sun’s energy which falls at the earth’s surface. This energy is available for
many applications, such as increasing water’s temperature or moving electrons in a photovoltaic cell.
Moreover, it supplies energy to natural processes like photosynthesis. Solar energy is secure, clean and
available on the earth throughout the year. Such clean energy is very important to the world, especially at
the time of high fossil fuel costs and the critical situation of the atmosphere resulting from fossil fuel
applications. The solar energy data provides information on how much is the sun potential at a location on
the earth’s during a specific time period. These data are very important for designing and sizing solar
energy systems. Due to the high cost and installation difficulties in solar measurement, the solar energy
data are not always available. Therefore, there is a demand to develop alternative ways of predicting the
solar energy data [1]. In Malaysia, there are cities/regions that do not have measured solar energy data
and to address this problem, a predication model has to be developed to estimate the potential of solar
energy based on location coordinates.
Many solar energy models have been developed which include the linear, nonlinear and ANN models.
Linear models for global solar energy [2] have been developed for calculating the solar energy linear

1876-6102 © 2011 Published by Elsevier Ltd. Selection and/or peer-review under responsibility of the organizing committee of 2nd International
Conference on Advances in Energy Engineering (ICAEE). Open access under CC BY-NC-ND license.
doi:10.1016/j.egypro.2011.12.1209
2 Tamer Khatib
Tameret al.\ / Energy
Khatib/ Procedia
Energy 14 (2012)
Procedia 2066
00 (2011) – 2074
000–000 2067

model’s parameters. Non-linear models [3] including fuzzy logic models for global solar energy have
been in use in many countries such as USA, Canada, Russia, UK, India, Turkey and Pakistan. As for the
diffuse solar energy models, linear and nonlinear models for specific countries have been presented in [4-
10]. However, in recent years, ANN is used in solar energy modeling for locations with different latitudes
and climates, such as Saudi Arabia, Oman, Spain, Turkey, China, Egypt, Cyprus, Greece, India, Algeria
and the UK [11]. As for Malaysia, little work has been done on solar energy modeling. The only
significant solar energy modeling works can be found in [12, 13] in which solar energy data for three
locations has been considered in [12]. In [13], a solar energy model has been developed for monthly solar
energy based on the least square linear regression analysis using eight data locations. Consequently, an
accurate model for solar energy should be developed to provide a comprehensive database for the solar
energy potential in Malaysia. The main objective of this research is to develop accurate models for diffuse
solar energy in Malaysia using ANN for providing diffuse solar energy predictions for many different
locations in Malaysia. This work has been done based on long term data for solar irradiations (1975-2004)
taken from the five sites in Malaysia. These data were provided by the Solar Energy Research Institute of
Universiti Kebangsaan Malaysia (UKM).

2. Solar radiation component

Solar radiation is classified into two main parts, extraterrestrial solar radiation (������ ) and global solar
irradiation (�� ) [1]. The variable ������ stands for the solar radiation above the atmosphere while �� is
the solar under the atmosphere. The extraterrestrial solar radiation, ������ is given b
���
������ � �� �� � ���� ��� � �� (1)
���

where �� is the solar constant, 1,367 ���� , and N is the number of days.
The components of a global solar radiation (�� ) on a horizontal surface are shown in Figure 1. In general,
�� consists of two parts and is given by,
�� � �� � �� (2)
����� �� ��� �� are the beam or direct and diffused solar radiations, respectively.
The direct solar radiation is the beam which falls directly from the sun, while the diffuse solar radiation is
the radiation which is being scattered by clouds and other particles in the sky.

Fig. 1 Solar irradiation components on a horizontal surface

2068 TamerTamer
Khatib et al.\ Energy
Khatib/ / EnergyProcedia
Procedia00
14(2011)
(2012)000–000
2066 – 2074 3

The angle between the fallen beam and the ground is called the altitude angel (α). While the incidence
from the zenith is represented by (ξ) which is called the incidence angle. In the case of horizontal surface,
the incidence angle is equal to the zenith angle (θ). The incidence angle can be calculated by using the
following equations
� � � � �� ���� (3)
� � � � �� ���� (4)
in which the altitude angel (α) is given by,
��� � � ��� � ��� δ � ��� � ��� δ ��� ��
(5)
where L is the latitude, δ is the angle of declination and �� is the hour angle.
The declination angle is the angle between the earth–sun vector and the equatorial plane, and it is given
by
��� �������
� � ����� ��� � � (6)
���
Meanwhile, the hour angle is defined as the angular displacement of the sun from the local point and it
has a positive value at AM and a negative value at PM. It can be calculated using,
�� � �� ��� � �� �����
(7)
����� �� is the solar time, which is the time measured with respect to the sun. �� is given by,
�
�� � ��� � ��� � ���� � ���� (8)
�
where LMT is the time at the moment of calculation , EOT is the equation of time and Lzt is the local
standard meridian.
The equation of time (EOT) is the difference between the apparent solar time and the mean solar time,
both taken at a given place or at another place with the same geographical longitude, at the same real
instant of time. EOT is given by
��� � ���� ������� � �������� � ������� (9)
where B can be calculated by using,
���
�� ��� � ��� (10)
���
The local standard meridian (Lzt) is given as ,

��� � ���� ���� �� ��� � ��

(11)

2.1 Solar energy calculation

The global solar energy on a horizontal surface is the average of global solar radiation multiplied by the
length of the solar day which is from sun shine to sun set. The solar day length (So) is calculated by
using,
�
�� � ��� �� �� ��� � ��� �� (12)
��
 Tamer Khatib et al.\ / Energy Procedia 14 (2012) 2066 – 2074 2069
4 Tamer Khatib/ Energy Procedia 00 (2011) 000–000

Based on this, the extraterrestrial ሺ‫ܧ‬௘௫௧௥௔ ሻ, global (‫) ்ܧ‬, beam (‫ܧ‬஻ ) and direct solar energy ሺ‫ܧ‬஽ ) can be
expressed as follows,
‫ܧ‬௘௫௧௥௔ ൌ ‫ܩ‬௘௫௧௥௔ ܵ௢ (13)
‫ ்ܧ‬ൌ ‫ܵ ்ܩ‬௢ (14)
‫ܧ‬஻ ൌ ‫ܩ‬஻ ܵ௢ (15)
‫ܧ‬஽ ൌ ‫ܩ‬஽ ܵ௢ (16)
Following equation (2), we get,
‫ ்ܧ‬ൌ ‫ܧ‬஻ ൅  ‫ܧ‬஽ (17)
The global solar energy strongly depends on a factor called the sun shine ratio. High sun shines ratio
means high solar energy and vice versa. The sun shine ratio is given by,

ͲǤͲ ൏ ܵൗܵ ൏ ͳǤͲ

ௌ ሺଵ଴ିଵǤଶହ஼ሻ
ൌ (18)
ௌ೚ ଵ଴ ௢

where ܵ௖ is the number of shining hours in solar day, from sun shine to sun set), and C is the measured
daily mean cloud cover with values from 0 for clear sky to 10 for overcast.
Diffuse irradiance is measured by placing a shadow band over a pyranometer. As a result, it is very
important to appreciate the order of measurements prior to any modeling study both for solar radiation
and sunshine duration or daylight. However, the measured solar energy records can be used for
developing solar energy models. These models calculate solar energy components using historical
metrological data. These models are used to calculate solar energy at sites where there is no solar energy
measuring device. In this research, the historical data consisting of diffuse solar energy and sun shine
hours for Malaysia are used. These data were taken for the five main sites in Malaysia.

Many models have been presented in the literature for modeling diffuse solar energy on a horizontal
surface. In general, there are three kinds of global solar energy models, namely, linear, non-linear, and
artificial neural network models. Here, the results of these models will be compared with measured
values so as to select the most accurate model.

3. Modelling of diffuse solar energy

The relationship between the average daily diffuse and global solar radiations incident on a horizontal
surface and the sky clearness index is described in this section. This relationship can be found from direct
meteorological observations or through an empirical relationship as studied in [24-26].

3.1 Linear models

Many linear models described the relation betweenࡱࢊ Ȁࡱࢀ and the clearness index, ࡷࢀ which equals to
ࡱࢀ Ȁࡱࢋ࢚࢞࢘ࢇ .The general equation of a linear model that calculates the diffuse solar energy can be
expressed as follows,
ࡱࢊ
ൌ ࢇ ൅ ࢈ࡷࢀ (19)
ࡱࢀ

where a and b are the coefficients of the model.

The following linear expression [6] is used for data obtained from ten meteorological stations located
between 40°N and 40°S latitudes,
2070 Tamer Khatib et al.\ / Energy Procedia 14 (2012) 2066 – 2074
Tamer Khatib/ Energy Procedia 00 (2011) 000–000 5

��
� �� � � �� ���� (20)
��

Tuller [7] considered data from four actinometrical stations of Canada and suggested the following
expression:
��
� �� �� � �� ���� (21)
��

For latitudes less than 50°, the Tuller equation can be recommend whereas for latitudes more than 50°,
the Iqbal equation [4] is more suitable. Based on the data from southern Canadian stations, the same
linear equation with restrictions on KT is used as follows:
��
� �� �� � �� ���� ��� � � �� � �� �� (22)
��

However, based on the 12 meteorological stations from India, Modi et al [8] developed a regression
equation:
��
� �� ���� � �� ������ ��� �� � �� � �� ��� (23)
��

In this research work, the linear model described in (19) is used to calculate diffuse solar energy for
Malaysia. The calculation of the coefficients, a and b is done by using the MATLAB fitting tool. Table 1
shows the developed model coefficients.

Table 1 Diffue solar energy model's parameter

Station a b
Alor Setar 1.2602 -1.3265
Ipoh 1.2867 -1.3771
Johor Bharu 1.2407 -1.2885
Kuala Lumpur 1.2841 -1.3765
Kuching 1.2398 -1.2848
Average 1.2623 -1.3307

Using the average coefficients vales in Table 2, a liner model for diffuse solar energy in Malaysia is
derived as follows:

��
� ������ � �������� (24)
��

3.2 Non linear models

The same relationship between �� ��� and the clearness index �� can be described by the following non-
linear model.
��
� � � ��� � ��� � � ��� � (25)
��

Kenisarin et al [36] used 34 sets of USSR meteorological station data for establishing such a relationship
by using the 12 daily values of �� ��� and �� . A non-linear of third-order polynomial model is taken into
consideration for the determination of the correlation type. The cubic polynomial model with the least
square error is given as follows:
��
� ����� � ������� � ������� � � ������� � ����� � �� ���� (26)
��
 Tamer Khatib et al.\ / Energy Procedia 14 (2012) 2066 – 2074 2071
6 Tamer Khatib/ Energy Procedia 00 (2011) 000–000

Lui and Jordan [9] developed a graphical relationship between �� ��� and KT , which is expressed later by
Klein [5] as:
��
� �.39 � 4.027�� � 5.53��� � � 3.�0��� � (27)
��

Additionally, with five sets of actinometrical station data from the USA, Collares et al [10] suggested an
equation whose coefficient varies with the seasons:
�� � �
� .775 � 0.347 ���� � 90� � � 0.505 � 0.2�� ���� � 90� ���� �2��� � 0.9�� (28)
�� ��� ���

However, based on the data from the four US stations, Collares et al [10] recommended the use of the
following equations:
��
� �.39� � 3.50��� � 4.��9�� � � 2.�37�� � ���� � �0� ��� 0.3 � �� � 0.�� (29)
��

and
��
� �.3�� � 3.022�� � 3.427�� � � �.�2��� � ���� � �0� ��� 0.3 � �� � 0.�� (30)
��

In this work, the non-linear model expressed in (25) is used to estimate the diffuse solar energy for
Malaysia. The MATLAB fitting tool is also used in calculating the developed model parameters.
However, The model coefficients are shown in Table 2.
Table 2 Diffuse solar energy nonlinear model's parameters
Station a b c d
Alor Setar 0.9540 0.8834 -4.7559 3.1623
Ipoh 0.9503 0.9314 -4.8726 3.2134
Johor Bharu 0.9522 0.8799 -4.6988 3.0503
Kuala Lumpur 0.9434 0.9895 -5.0309 3.3477
Kuching 0.9486 0.9507 -5.0524 3.4974
Average 0.9497 0.9270 -4.8821 3.2542

Thus, a recommended nonlinear model for the diffuse solar energy in Malaysia is derived by using the
average coefficient values of Table 6 as follows,

��
� 0.9497 � 0.9270�� � 4.��2��� � � 3.2542�� � (31)
��

3.3 ANN model for the diffuse solar energy

Artificial neural networks (ANNs) are information processing systems that are non-algorithmic, intensely
parallel and learn the relationship between the input and output variables by studying the previously
recorded data. In ANN, the neurons are connected by a large number of weighted links, over which
signals or information can pass. A neuron receives inputs over its incoming connections, combines the
inputs, generally performs a non-linear operation and outputs the final results. MATLAB was used to
train and develop the ANNs for the diffuse solar energy modeling. The ANN model adopted is the feed
forward, multilayer perceptron (FFMLP) network because it is the most commonly used ANN that learns
from examples. A schematic diagram of the basic ANN architecture is shown in Figure 2. The network
has three layers: the input, hidden and output layers. Each layer is interconnected by connection strengths,
called weights [14].
2072 Tamer Khatib
Tamer et al.\Energy
Khatib/ / Energy Procedia
Procedia 0014 (2012)
(2011) 2066 – 2074
000–000 7

Fig. 2 ANN model for diffuse solar energy perdition

The network has four inputs comprising of latitude, longitude, clearness index and day number and one
output variable which is the diffuse solar energy. There is a single output node to represent the
estimated daily clearness index prediction as the output.

4. Test Results

All the developed models are evaluated based on evaluation criteria with the aim of determining the most
accurate model for diffuse solar energy. The testing of the models is based on a long term solar energy
data (1975 to 2005) for the chosen sites. The data for the period 1975-2004 are used to model and train
the developed models, while the data for the year 2005 is used for testing the developed models. Thus, the
developed models calculate the daily diffuse solar energy for Malaysia in the year 2005. To evaluate the
developed models, three error statistics are used, namely, the mean absolute percentage error (MAPE),
mean bias error (MBE) and root mean square error (RMSE). The MAPE is an indicator of accuracy in
which it usually expresses accuracy as a percentage and is given by:
� I�I�
���� � ∑���� (34)
� I
where I is the real value and Ip is the predicted value. The calculated MAPE is summed for every fitted or
forecasted point in time and divided again by the number of fitted points, n. Most ANN models are also
evaluated using MBE and RMSE. MBE is an indicator for the average deviation of the predicted values
from the measured data. A positive MBE value indicates the amount of overestimation in the predicted
global solar energy and vice versa. On the other hand, RMSE provides information on the short-term
performance of the models and is a measure of the variation of the predicted values around the measured
data. It indicates the scattering of data around linear lines. RMSE also shows the efficiency of the
developed ANN in predicting future individual values. A large positive RMSE implies a big deviation in
the predicted value from the measured value. MBE and RMSE are given as follows:
�
��� � ∑������� � � �� �
�
(35)
 Tamer Khatib et al.\ / Energy Procedia 14 (2012) 2066 – 2074 2073
8 Tamer Khatib/ Energy Procedia 00 (2011) 000–000

� �
���� � � ∑��� ��� � � �� ��
�
(36)
Figures 3 shows the calculated values of the developed diffuse solar energy models for Kuala Lumpur,
the ANN model gives the most accurate results compared to the other three models with the linear and
nonlinear models giving almost similar performance. Tables 3 shows the evaluation results of the
developed diffuse solar energy models.

Fig 3 Global solar energy for Kuala Lumpur using the developed models

Table 3 Evaluation on the developed diffuse solar energy models

MAPE RMSE RMSE (%) MBE MBE (%)

Kuala Lumpur 4.21 0.138 4.89 0.009 0.315
Johor Bharu 4.20 0.136 4.94 0.003 0.091
Linear model Ipoh 3.61 0.124 4.33 -0.006 -0.194
Kuching 4.31 0.146 5.29 0.010 0.358
Alor Setar 5.42 0.186 6.65 -0.003 -0.101
Average 4.35 0.146 5.22 0.003 0.094
MAPE RMSE RMSE (%) MBE MBE (%)
Kuala Lumpur 3.79 0.130 4.61 0.005 0.168
Johor Bharu 3.11 0.114 4.13 0.003 0.096
Non-linear model Ipoh 3.33 0.121 4.21 -0.00004 0.002
Kuching 3.49 0.129 4.68 0.009 0.332
Alor Setar 4.97 0.185 6.62 -0.004 -0.146
Average 3.74 0.136 4.85 0.003 0.090
MAPE RMSE RMSE (%) MBE MBE (%)
Kuala Lumpur 1.51 0.055 1.97 0.002 0.058
Johor Bharu 1.41 0.045 1.63 0.0007 0.025
ANN model Ipoh 1.18 0.048 1.68 -0.0041 -0.143
Kuching 1.50 0.053 1.91 0.0043 0.016
Alor Setar 2.07 0.079 2.85 -0.0074 -0.265
Average 1.53 0.056 2.01 -0.0009 -0.062
2074 Tamer
Tamer Khatib
Khatib/etEnergy
al.\ / Energy Procedia
Procedia 14 (2012)
00 (2011) 2066 – 2074
000–000 9

From the MAPE, RMSE and MBE results shown in Tables 3, it is shown that the ANN model for diffuse
solar energy calculation performs better than the linear and nonlinear models. Comparing the accuracy of
the non-linear and linear models based on the MAPE, RMSE and MBE values, it is noted that the
nonlinear model is more accurate compared to the linear model.

5. Conclusion

Linear, non-linear and ANN models for calculating diffuse solar energy have been developed for the
five sites in Malaysia. The developed models have been evaluated based on three statistical values,
namely, mean absolute percentage error, MAPE, root mean square error, RMSE and mean bias error.
Results showed that ANN based model is the most accurate model in calculating the diffuse solar energy
in Malaysia. The MAPE values for ANN based diffuse solar energy models is 1.53%. While the MAPE
values for the models for the diffuse solar energy are 4.35% and 3.74%, respectively..

References

[1] Kreider J F, Kreith F .1981. Solar energy handbook. McGraw-Hill, New York.
[2] Angström A (1956) On the computation of global radiation from records of sunshine. Arkiv Geof 2:471–479.
[3] Abdalla Y A G, Baghdady M K (1985) Global and diffuse solar radiation in Doha (Qatar). Solar and Wind Technology
2:209.
[4] Iqbal M (1979) A study of Canadian diffuse and total solar radiation data 1. Monthly average daily horizontal radiation.
Solar Energy 22:81–86.
[5] Klein S A (1977) Calculation of monthly average insolation on tilted surface. Solar Energy 19:325– 329.
[6] Page J K (1961) The estimation of monthly ea values of daily total short wave radiation on vertical and inclined surfaces
from sunshine records for latitudes 40°N–40°S, Proceedings of the UN conference on new sources of energy, paper no. 598, 4, pp
378–390.
[7] Tuller S E (1976) The relationship between diffuse, total and extraterrestrial solar radiation. Solar Energy 18:259–263.
[8] Modi V, Sukhatme S P (1979) Estimation of daily total and diffuse insolation in India from weather data. Solar Energy
22:407–411.
[9] Lui B Y H, Jordan R C (1960) The interrelationship and characteristic distribution of direct, diffuse and total solar
radiation. Solar Energy 4:1–19.
[10] Collares–Pereira M, Rabl A (1979) The average distribution of solar radiation: correlations between diffuse and
hemispherical and between daily and hourly insolation values. Solar Energy 22:155–164.
[11] Adel Mellit, Soteris A. Kalogirou. Artificial intelligence techniques for photovoltaic applications: A review. Progress in
Energy and Combustion Science 34 (2008) 574–632.
[12] Donald G. S. Chuah and S. L. Lee. Solar radiation in peninsular Malaysia: Statistical presentations. Energy Convers
Manage 1982;22(1):71–84.
[13] Kamaruzzaman Sopian and Mohd.Yusof Hj.Othman. Estimates of monthly average daily global solar radiation in
Malaysia.Renew Energy 1992;2(3):319-325.
[14] Caudill M, Butler C. Understanding neural networks: computer explorations – volume 1 basic networks. Massachusetts:
The MIT Press; 1993.