Sei sulla pagina 1di 19

Renewable and Sustainable Energy Reviews 56 (2016) 778796

Contents lists available at ScienceDirect

Renewable and Sustainable Energy Reviews


journal homepage: www.elsevier.com/locate/rser

Prediction and application of solar radiation with soft computing over


traditional and conventional approach A comprehensive review
Sthitapragyan Mohanty n, Prashanta Kumar Patra, Sudhansu Sekhar Sahoo
College of Engineering & Technology, Bhubaneswar 751003, India

art ic l e i nf o

a b s t r a c t

Article history:
Received 25 August 2015
Received in revised form
17 November 2015
Accepted 30 November 2015

Solar radiation data plays a crucial role in solar energy research and application. It provides the vital
information about the energy that strikes the earth and is highly useful for modeling and design of solar
thermal technologies and solar photovoltaic applications. As Conventional energy sources are depleting
day by day, it becomes necessary to use renewable energy sources like Solar, Wind, and Biomass etc.
Amongst all forms of renewable energy sources, solar energy is widely accepted as it is quite abundant
throughout the world. In many developing countries solar radiation data are not always available either
due to the unavailability of measuring instruments or due to the absence of meteorological stations. It is
also true that many countries fail to afford the costly measurement equipments and techniques involved
for measuring solar radiation. Thus, it is quite essential to develop models to measure accurate solar
radiation by using Various meteorological parameters such as(latitude, longitude and Altitude)/Climatological parameters (i.e. Sunshine duration, Humidity,clearness index, months, temperature, cloudiness,
wind velocity, atmospheric pressure, diffuse radiation, beam radiation, Global radiation extra terrestrial
radiation, evaporation) etc. The objective of this paper is to
i) To Study and review the model and techniques used for prediction of solar radiation.
ii) To identify the research gap and the best methods available in the literature review.
iii) To recommend appropriate techniques for solar energy predictions so that researchers nd it more
convenient and do their research implementations using these techniques for various applications.
& 2015 Elsevier Ltd. All rights reserved.

Contents
1.
2.
3.

Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Estimation/prediction of global solar radiation using Empirical model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Prediction of Global solar radiation using soft computing Approach . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
3.1.
Articial neural network . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
3.2.
ANFIS (Articial neuro-fuzzy inference system). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
3.3.
Radial Basis Function Network (RBFN) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4. Pros and cons of different models described in literature . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
5. Application of solar radiation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
5.1.
Application to Photovoltaic system. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
5.2.
Application to thermal . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
6. Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

778
779
784
784
788
790
792
792
792
793
794
794

1. Introduction

Corresponding author. Mobile: 919438180270.

http://dx.doi.org/10.1016/j.rser.2015.11.078
1364-0321/& 2015 Elsevier Ltd. All rights reserved.

The design and study of solar energy to gather information


about solar radiation with its components at a particular given

S. Mohanty et al. / Renewable and Sustainable Energy Reviews 56 (2016) 778796

Nomenclature
H
ANN
ANFIS
BP
MAPE
RMSE
MBE
MLR
LM
H0
T

Global solar radiation


Articial Neural Network
Articial Neuro Fuzzy Inference system
Back propagation
Mean Absolute Percentage Error
Root means square error
Mean bias error
Multiple linear regression
LevenbergMarquardt
Extra terrestrial radiation
Temperature

location is most important. Among all renewable energy sources,


solar energy is the most abundant and easily available energy
source. This energy caters not only the need of human being, but
also the plants and other organisms during photosynthesis. Solar
power is the conversion of sunlight into electricity, either directly
using photovoltaic (PV) or indirectly using concentrated solar
power (CSP).
Conventional energy sources are depleting day by day and to
overcome the dependency on conventional sources, many
researchers and organizations are advocating alternative fuels,
which are commercially viable, easy to use, less pollutant, and are
abundant in nature. Renewable energy sources like Solar Energy,
Wind Energy, Bio fuels, and tidal energy are preferred to conventional sources of energy as these are generated from natural
resources such as sunlight, wind, rain and various forms of biomass. These sources are not only renewable but also maintain
ecology with low environmental impact. Applications of solar
radiation data are found in solar heating, cooking, drying and
interior illumination of buildings etc. Several formulae and models
are developed by relating the global solar radiation to different
climatic and meteorological parameters such as latitude, longitude, altitude, maximum and minimum temperature, sunshine
duration and relative humidity. The main objective is to review the
global solar radiation either by traditional approach or by using
soft computing approaches. The measurement of solar radiation is
carried out hour basis, daily basis or monthly basis. The approaches used for prediction are:







2. Traditional approach
3. Soft computing approach
In traditional way, the approaches used for prediction are:
2.1 The dynamical approach
2.2 The empirical model

The dynamical approach is only useful for modeling large-scale


solar radiation prediction and may not predict short-term radiation. But for local scale & short term solar radiation prediction, the
soft computing approaches are used, which perform nonlinear
mapping between inputs and output.
Generally Empirical models are based entirely on data. These
models have uncertainty in terms of prediction. This section
describes prediction of solar radiation on the basis of
empirical model.

MLP
RBF
RNN
H
S
LONG
MSE
R2
S0
LAT
a,b
CC
R

779

Multilayer Perceptron
Radial Basis Function
Recurrent Neural Network
Solar Radiation
Sunshine duration
Longitude
Mean Squared Error
Correlation coefcient
Maximum daily sunshine
Latitude
Regression coefcients
Correlation Coefcient
Relative Humidity

2. Estimation/prediction of global solar radiation using


Empirical model
Angstrom [1] proposed the rst theoretical model for estimating global solar radiation based on sunshine duration. Prescott
[2,3] reconsidered the model to calculate monthly average daily
global radiation (MJ=m2 day) on a horizontal surface from monthly
average daily total insolation on an extraterrestrial horizontal
surface by using the following equation.
 
H
S
ab
1
H0
S0
where H is the monthly average global radiation on horizontal
surface,S is the monthly average daily bright sunshine hours,S0 is
the maximum possible monthly average daily sunshine hours or
the day length,H 0 is the monthly average daily extraterrestrial
radiation,a and b are the regression coefcients.
The solar radiation can also be estimated by using higher order
correlations. Benson et al. [4] used a quadratic form of relationship
between daily global/extraterrestrial radiation and actual/maximum possible hours of sunshine duration.
   2
H
S
S
ab
2
c
H0
S0
S0
Falayi et al. [5] Used multi linear regression equations to predict the relationship between global solar radiations with one or
more combinations of meteorological parameters: clearness index,
mean of daily temperature, the ratio of maximum and minimum
daily temperature, relative humidity and relative sunshine duration for Iseyin Nigeria from (1995 to 1999),and is shown in Table 1.
Multiple linear regressions (MLR) is an approach where the relationship occurs between a dependent variable and several independent variables.The value of correlation coefcient (r), Root
Mean Square Error (RMSE),Mean Bias Error (MBE) and Mean Percentage Error (MPE) were determined for each equation. The best
accuracy can be obtained by calculating equation with the highest
value of r and least value of RMSE, MPE and MBE.
Augustine and Nnabuchi [6] compare the measured global solar
radiation with the value calculated by using an AngstromPrescott
correlation equation having regression coefcient 0.29 and 0.49
(Shown in Fig. 1).
H
0:29 0:42n=N
H0

Medugu et al. [7] Used angstrom model for Estimating mean


monthly global solar radiation in YolaNigeria from 2004 to 2007

780

S. Mohanty et al. / Renewable and Sustainable Energy Reviews 56 (2016) 778796

Table 1
MBE, RMSE and MPE of Different equation and its corresponding r% [5].
Equation

R2

R


H=H 0 0:20765 0:7475 S=Smax


H=H 0 0:97877 0:05722T
H=H 0 1:1973  0:00829RH
H=H0 1:7217  1:691RH


H=H 0 0:5475 0:5987 S=Smax  :0035RH


H=H 0 0:8758 0:5168 S=Smax 0:0194RH


H=H 0  0:02144 0:541 S=Smax 0:0194T


H=H 0 1:1203 0:4690 S=Smax  1:5956 0:0041RH


H=H 0 0:856 0:676 S=Smax  0:010T  0:004RH


H=H 0 1:309 0:601 S=Smax  0:9990  0:01287T
H=H 0 0:7162 0:0106RH  2:684 0:0324T


H=H 0 1:3467 0:5305 S=Smax  1:567 0:0033RH 0:00806T

MBE

RMSE

MPE

0.9350

0.8746

 0.00018

0.03765

 0.0321

0.8828
0.7529
0.8629
0.9702

0.7794
0.5669
0.7447
0.9414

 0.00461
 0.00019
 0.00020
 0.00021

0.04961
0.06996
0.05373
0.02574

0.06931
0.12284
0.13299
 0.0915

0.9822

0.9646

 0.00026

0.02001

 0.0521

0.9473

0.8984

 0.00019

0.03389

 0.0208

0.9864

0.9728

 0.00020

0.01752

0.9718

0.9445

 0.00020

0.02516

0.02956
 0.0996

0.9849

0.9701

 0.00021

0.01863

 0.0823

0.9464
0.9870

0.8957
0.9748

 0.00019
 0.00019

0.03411
0.01705

 0.0278
0.02125

Table 2
Different model with its corresponding RMSE, MBE and MPE value [8].
Model

RMSE

MBE

MPE

Original Hargreaves
Hargreaves models with (linear Regression)
Hargreaves Models with ( Power Regression)

4.55
1.59
1.62

4.30
0.82
0.85

 34.25
 7.36
 7.63

Fig. 1. Comparison between Measured and Predicted Solar Radiation [6].

Fig. 2. Mean Monthly Global Solar Radiation for 2004, 2005, 2006 and 2007 against Months [7].

by using sunshine duration, and is shown in Fig. 2. The comparison


between the measurement and estimation was carried out
according to the t-statistic. The smaller the value of t, the better is
the models performance.
The graph shown in Fig. 2 shows the maximum values of global
solar radiation appears in March, April and May with 24.38
Kt day  1,24.92 MJ=m2 day  1 and 24.54 MJ=m2 day  1 respectively,
during dry season while minimum values of solar radiation,
20.31MJ=m2 day  1 and 20.77 MJ=m2 day  1 have been observed in
August and September, respectively.
Sansui Yekinni K [8] used three empirical models to estimate
solar radiation at Ibadan, Nigeria. The input parameters used in
this model are maximum and minimum temperature for a period
of six years, which is clearly shown in Table 2. The model performance of three empirical models was compared based on the
MBE, RMSE and MPE.Based on the RMSE, Hargreaves model with
linear regression produces the best coefcient of determination,
while the Hargreaves-Samani model gives the worst with larger
values of RMSE.For MBE, the result shows that the Hargreaves
model with linear regression is the best while the HargreavesSamani is the worst. With respect to MPE Hargreaves model linear

regression offers best correlation, while the Hargreaves-Samani


model gives the worst.
Sa et al. [9] used two procedures for modeling daily global solar
radiation based upon higher order statistics. The rst one uses the
(lost component) and the second one uses the (clearness index). The
result shows that the transformation yields the lost component is
better than the classical one using clearness index.The minimum
values of the statistical indicators (NMBE-2.28%, NRMSE-0.16% and
t1.75) shows using lost component method is efcient for
representing a non-Gaussian processes characterized by fast uctuations such as daily solar radiation.Taha Ahmed Tawk Hussein
[10] developed a computer mathematical model to estimate the total
amount of hourly solar radiation that reaches the earth's surface in a
day by using the available meteorological data, such as sunshine
duration, cloud cover, maximum and minimum temperature covering the year 2009 (shown in Table 3).Based on the Mean bias error
and correlation coefcient, Comparison between measured and
predicted data has been carried out of selected regions of Egypt.
GholamrezaJanbaz Ghobadi et al. [11] estimate global solar
radiation by using meteorological parameter temperature at sari
station from 20002010, and it is shown in Fig. 3.Two solar

S. Mohanty et al. / Renewable and Sustainable Energy Reviews 56 (2016) 778796

Table 3
Correlation coefcient of predicted and measured hourly solar radiation for
(a) Shebin Elkom,(b) Belbees and (c) El-Mansoura [10].
Regions

Correlation coefcient

No of observation

Shebin Elkom
Belbees
El-Mansoura

0.9851
0.9945
0.9883

408
306
612

781

mean global solar radiation on a horizontal surface by using single


input parameter sunshine duration for Jaipur having latitude 26.92
N and longitude 75.87 E,and is shown in Table 5. Predicted
values of monthly mean global solar radiation were compared
with observed values using statistical parameters coefcient of
determination R2 ,mean bias error MBE and root mean Square error
RMSE.Among them Cubic correlation shows best result in comparison to logarithmic correlation.

Fig. 3. Correlation of measured and estimated radiation by using Angstrom model (a) Calibration (b) Validation [11].Table 4.

Fig. 4. Comparison between the measured and predicted Global Solar Radiation [12].

radiation models are calibrated developing a new model. The


accuracy of the models was compared on the basis of the statistical
error tests such as mean bias error (MBE), Root mean square error
(RMSE), correlation coefcient (r) and the t-test. Angstrom and
Prescott model showed better estimation of the monthly average
daily global solar radiation on a horizontal surface for a Sari station
in comparison to other models.
The statistical tests of MBE, RMSE, r and t-test for the period
20002010 were determined as
Ituen, Eno.E et al. [12] developed model with regression
equations to predict the monthly global solar radiation based on
measured air temperature, relative humidity and sunshine hour
values between from 1991 to 2007 for Uyo,in Niger Delta Region,
Nigeria (shown in Fig. 4) by Using the Angstrom model.After
considering statistical indicators that are MBE,RMSE and MPE the
equation with the highest value of correlation coefcient (r) and
the least values of RMSE,MBE and MPE are chosen as the
best model.
Marwal [13] used six empirical correlations AngstromPrescott
linear correlation and modied functions such as quadratic, cubic,
exponential, logarithmic and power function to predict monthly

Table 4
Statistical test of different models [11] is shown in Table 4.
Model

R2

RMSE

MBE

Calibration
Validation

0.86
0.86

2.464
5.149

 0.136
 4.628

2.06
66.1

Table 5
Correlations with their computed regression coefcients and statistical parameters
[13].
Correlation

R2

MBE

RMSE

Linear
Quadratic
Cubic
Exponential
Logarithmic
Power

0.8050
0.8423
0.8551
0.8006
0.8005
0.8517

 0.0753
 0.0396
 0.0363
 0.1520
 0.8368
 0.1086

1.3073
1.1997
1.1425
1.4223
1.6393
1.2390

782

S. Mohanty et al. / Renewable and Sustainable Energy Reviews 56 (2016) 778796

Tolabi et al. [14] used imperialist competitive algorithm to


estimate monthly average daily global solar radiation on a horizontal surface for four different climate cities of Iran. Results show
that the imperialist competitive algorithm is a suitable method to
nd the best experimental coefcients based on Angstrom model
and its predicted coefcients have more accuracy than coefcients
estimated by statistical regression techniques.
Khorasanizadeh et al. [15] present a statistical comparative
study to demonstrate the merit of day of the year-based models
for estimation of horizontal global solar radiation of Birjand,
located in the sunny belt region of Iran. 12 models have been
selected from the literature. By utilizing the long-term measured
data and via statistical regression techniques the models have
been established and their performances evaluated through several statistical indicators. To identify the suitability of the DYB
models for monthly mean daily estimation, new regression constants have been developed for all of the nominated models and
their performances. Owing to accurate estimation, simplicity and
more or less similar performance as SDB models, DYB models
seem qualied as proper alternatives of SDB models. So in this
study the best DYB models used for estimation of daily and
monthly mean daily horizontal global solar radiation, have been
recommended for utilization in Birjand city. Because of similar
climate conditions, the results are also applicable for the whole
South Khorasan province and its neighboring regions.
Al-Rawahi [16] predicts hourly terrestrial solar radiation on a
horizontal surface and inclined surface: direct beam, diffuse and
global from measured daily averaged global solar radiation,and it
is shown in Fig. 5. The predicted hourly solar radiation incident on
a horizontal surface was compared with hourly data measured
locally at the Seeb Meteorological Centre of Oman. The gure
shows the average received solar radiation, where the tilt angle is
kept same throughout the year. When the tilt angle of a solar
collector is xed, a tilt angle of about 25degree will receive the
maximum solar radiation.
Almorox et al. [17] estimates and compares ve models to
predict global solar radiation of Canada de Luque, Crdoba,
Argentina by taking temperature as the input parameter. The
performance of the models is measured and compared on the
basis of statistical indicators such as R2 ,RMSE, MBE, MAPE
and MPE.
Kaplanis [18] describes two new reliable approaches to estimate hourly global solar radiation on a horizontal surface. The
predicted global solar hourly radiation values are compared with
the estimation from two existing packages and the recorded solar
radiation for the two biggest cities of Greece.
Kacem Gairaa [19] developed seven models for predicting the
global solar radiation on a horizontal plane for estimating the
global solar radiation from sunshine duration and from two
meteorological parameters (air temperature and relative humid-

ity) and is shown in Table 6.The root mean square error (RMSE),
Mean bias error (MBE), correlation coefcient (CC), and percentage
error (e) have been computed to test the accuracy of the proposed
models. Comparison between the measured and the calculated
values have been made. The result shows the linear and quadratic
models are the most suitable for estimating the global solar
radiation.Abdalla and Ojosus models give the best performances
with a CC of 0.898 and 0.892.
After comparison between the estimated and measured annual
average values of the global solar radiation, the annual percentage
error is calculated, which lies between  4.047% and 0.639%.That
means the linear quadratic models and Abdalla and Ojosu are the
suitable models to estimate the annual global solar radiation on a
horizontal surface in Gharda a region.
Kaplanis, Kaplani [20] described the stochastic prediction of the
hourly intensity of the global solar radiation, I (h; nj) for any day nj
at a site, as shown in Fig. 6. The predicted results of the hourly
global solar radiation for winter, autumn and spring seasons were
also compared to the results provided by the METEONORM
package.
J.K. Yohanna et al. [21] used an empirical model for determining the monthly average daily global solar radiation on a horizontal surface of Makurdi, Nigeria (Latitude 7_70N and Longitude
8_60E).The model was developed by using Angstrom-Prescott
equation. After prediction the measured solar radiation is compared with the solar radiation predicted by the model having HH0
0:17 0:68n=N with an MBE of 0:17% and RMSE of 1:22%.This
shows good performance in determining the monthly average
daily global solar radiation for Makurdi, Nigeria.
Mejdoul [22] proposes a statistical comparison between measured data of mean hourly global radiation at two different climate
regions located in Morocco and three predicting models based
upon statistical test error as root mean square error (RMSE),Mean
bias error (MBE) and correlation coefcient (R).A comparative
study has been done between measured data and the three correlations (WLJ,CPR and CPRG) in terms of statistical indicators such
as the root mean square error (RMSE),the Mean bias error (MBE)
and the correlation coefcient (R).
Table 6
Estimated and Annual percentage error of different models [19].
Model

Estimated Value

Error %

Linear
Quadratic
Logarithmic
Exponential
Abdalla
Ojosu
Hargreaves

5852.45
5857.43
6108.62
5847.95
5846.03
5833.50
5887.97

0.316
0.231
 4.047
0.393
0.425
0.639
 0.289

Fig. 5. Average daily incident solar radiation energy in Seeb/Muscat area for different tilt angles [16].

S. Mohanty et al. / Renewable and Sustainable Energy Reviews 56 (2016) 778796

783

Fig. 6. Predicted hourly global solar radiation Im, pr (h; 17), and the measured I mes (h; 17) [20].

Tu rk Tog rul et al. [23] used different regression method to


estimate monthly mean solar radiation in turkey by using different
meteorological parameters. After calculating from the equations,
the monthly mean global solar radiation was developed and
compared by measuring values for six cities in Turkey. Two statistical tests, root mean square error (RMSE) and mean bias error
(MBE), and t-statistic were used to evaluate the accuracy of the
correlations.
Li et al. [24] used a new empirical model for estimating daily
global solar radiation in china on a horizontal surface by the day of
the year. The performance of the model is evaluated by comparing
with three trigonometric correlations at nine representative stations of China using statistical error tests such as the mean absolute percentage error (MAPE),Mean absolute bias error (MABE),
Root mean square error (RMSE) and correlation coefcients (r).The
results show that the new model provides better estimation and
has good adaptability to highly variable weather conditions.
Empirical modeling is most important and economical tool for
estimating solar radiation. A trigonometric model in conjunction
with a sine and cosine wave for estimating daily global solar
radiation is proposed in this work.
Yingni Jiang [25] used several empirical equations to estimate
monthly mean daily diffuse solar radiation for eight typical
meteorological stations in China. Estimated values are compared
with measured values in terms of statistical error tests such as
mean percentage error (MPE), Mean bias error (MBE), Root mean
square error (RMSE).Here the author used quadratic model. H d =

2


H g 0:945 0:675H g =H o  0:166 H g =H o  0:173 S=So  0:079

2
S=So and compared it with other empirical equations. According


to MPE, MBE and RMSE, the model H d =H g 0:945  0:675 H g =H o

2



2
 0:166 H g =H o  0:173 S=So  0:079 S=So has the best performance based on the measured data of eight stations in China with
MPE of 1.75%, MBE of  0:03MJ=m2 and RMSE of0:78MJ=m2 .

Table 7
Regression coefcient of Different models after prediction [26].
Models
H m =H 0 a bS=S0
H m =H 0 aT 0:5


H m =H 0 a b RH=100
H m =H 0 a bT avg
H m =H 0 a bTR


H m =H 0 a b RH=100  TR
H m =H 0 a bp

b
0.2493
0.1495

0.5659

0.8454

0.4603

 1.113
1.4192
0.7711

0.0641
 1.197
 0.465

0.5904

 0.0218

Adaramola [26] estimates monthly average global solar radiation (Table 7) in Akure, Nigeria by using meteorological data such
as sunshine duration, temperature and humidity. The Angstrom
Page correlation predicted the monthly average daily global solar
radiation, which is better than the other correlations developed. In
the absence of the sunshine hour data, it was found that the
temperature based correlations can be used to predict the global
solar radiation within a reasonable level of accuracy in Akure.
Bulut and Bu yu [27] uses a simple model for estimating the
daily global radiation in Turkey. The model is based on a trigonometric function, which has only one independent parameter, i.e.
the day of the year. The model is tested for 68 locations in Turkey
using the data measured during 10 years duration. The statistical
indicators of the model such as mean absolute error, root-meansquare error and correlation coefcient are found to be at acceptable levels. It was found that the model can be used for estimating
monthly values of global solar-radiation with a high accuracy.
Musa et al. [28] estimates monthly mean Global Solar radiation
of Maiduguri, Nigeria by using Angstrom model for ve years from
2006 to 2010 based on daily sunshine duration as shown in Fig. 7.

Fig. 7. Monthly mean sunshine hours from 2006 to 2010 [28].

784

S. Mohanty et al. / Renewable and Sustainable Energy Reviews 56 (2016) 778796

After observation the February, March and April months give the
peak amount of solar radiation where as the June, July and August
months give the least amount of solar radiation.

3. Prediction of Global solar radiation using soft computing


Approach
The global solar radiation can also be estimated /predicted by
using soft computing approaches such as Multilayer perceptron
Neural Network(MLP),Radial basis function Network (RBFN),
Recurrent Neural Network (RNN),Support Vector Machine (SVM),
Genetic Algorithm (GA),Articial Neuro-fuzzy inference system
(ANFIS) and Hybrid Network to predict solar radiation of a
particular place.So a comparative study can be done between a
conventional approach,i.e.which is Multiple Linear Regression
(MLR) with the soft computing approach.
Several Application of Articial neural networks are found in
various elds such as character recognition, image compression,
aerospace, defense, mathematics, engineering, medicine, electronic nose, economics, meteorology, psychology, neurology and
many others. They have been also used for prediction and
regression analysis in weather, solar and Market trend forecasting.

index used to predict the Global solar irradiation. The clearness


index of measured and predicted outputs are compared and the
errors are calculated. Here the author considered ve main sites of
Malaysia for testing. The average MAPE, MBE and RMSE for the
predicted global solar irradiation are 5.92%, 1.46% and 7.96.
Jiang [25] used feed-forward back propagation neural network
for estimating mean monthly, daily diffuse solar radiation for eight
cities (Haerbin, Lanzhou, Beijing, Wuhan, Kunming, Guangzhou,
Wulumuqi and Lasa) of China. The input parameters are monthly
mean daily clearness index, sunshine percentage and mean
monthly daily diffuse fraction is the output. The Comparison result
shows that the RMSE values of ANN model are more accurate than
empirical model.
Lubna. B. Mohammed et al. [32] used Nonlinear Autoregressive
Exogenous (NARX) model to predict hourly solar radiation in
Amman, Jordan. Meteorological data for the years from 2004 to
2007 were used for training while the data of the year 2008 were
used for testing, as depicted in Table 11. The performance of NARX
model was examined and compared with different training algorithms The comparative analysis of different training algorithms is
evaluated on the basic of statistics (coefcient of determination
Table 8
Architecture, MSE and MAE for the developed ANN Model [29].

3.1. Articial neural network


Neural networks approaches have been widely used for prediction and estimation of solar radiation. The most common form
of neural network is the multilayer perceptron. The structure of
ANN is characterized by its input layer, one or more hidden layer
and output layer, and is shown in Fig. 8. Two parameters weight
and bias are connected between layers.
This section shows a number of solar energy prediction/estimation and its applications using articial neural network.
Premalatha et al. [29] used articial neural network to estimate
global solar radiation of India as presented in Table 8. The input
parameters used in this model are maximum ambient temperature, minimum ambient temperature and minimum relative
humidity. Depending upon the number of input parameters, various models are developed and tested in order to get better results.
Emad et al. [30] predicts monthly average Global Solar Radiation by Using Articial Neural Network in Qena, Upper Egypt, and
is shown in Table 9. The author compares the ANN model with
Empirical model and shows the model using ANN gives better
result in comparison to Empirical model. A correlation coefcient
of 0.977 was obtained having mean bias error (MBE) of the 48 Wh
=m2 and root mean square error (RMSE) of 115Wh=m2 .
Tamer et al. [31] uses Multilayer perceptron articial neural
network to estimate Global solar energy for Malaysia based on
input parameters latitude, longitude, day number and sunshine
ratio, as shown in Table 10. The output parameter is the clearness

Model

Input parameters

Architecture

MSE

MAE

1
2
3
4

f t; T max
f t; T min
f t; T max ; RH min
f t; T min ; RHmin

2-24-1
2-32-1
3-36-1
3-36-36-1

0.011
0.008
0.048
0.029

8.39
6.65
18.03
12.34

Table 9
Results of correlation and error analysis of two models [30].
Model

MBE

RMSE

Empirical
ANN

0.960
0.977

 335
48

540
115

Table 10
MBE and RMSE values of different sites of Malaysia [31].
Different Sites

MBE

RMSE

KualaLumpur
Alor Setar
Johor Bharu
Kuching
Ipoh

 0.0087
0.161
0.043
0.036
0.105

0.348
0.419
0.342
0.353
0.380

Training
Development
of ANN Model

Input
data

Selection of
Parameters using
ANN Model

Testing

Error Calculation
using (RMSE,MSE,MAPE)

Selection of
Prediction model with
minimum error
Fig. 8. Methodology used for prediction of solar radiation.

S. Mohanty et al. / Renewable and Sustainable Energy Reviews 56 (2016) 778796

785

Table 11
Performance of different training algorithms based on statistical criteria [32].
Algorithm

RMSE

MBE

Training

Validation

Training

Validation

Training

Validation

Training

Trainlm
Trainrp
Trainscg
Traincgb
Traincgf
Traincgp
Trainoss

42.8367
49.2078
53.2732
47.2268
49.0563
48.1758
49.1726

48.3991
50.2298
52.6080
49.0884
49.8144
49.2361
49.8859

25.5612
28.9444
31.3656
28.0998
29.5055
28.3929
28.7343

28.5317
30.6432
32.5375
29.7275
30.9015
29.9944
30.1949

0.99157
0.98884
0.98692
0.98974
0.98891
0.98931
0.98886

0.98916
0.98832
0.98718
0.98884
0.98852
0.98878
0.98848

(R2 ), root mean squared error (RMSE) and mean bias error (MBE)).
By using different training algorithm, The MarquardtLevenberg
learning algorithm with a minimum root mean squared error
(RMSE) and maximum coefcient of determination (R) was found
as the best period when applied in NARX model.
Szen et al. [33,34] applied ANN model for estimation of solar
radiation in Turkey by using meteorological and geographical data
(mean sunshine duration, mean temperature and month take as
input parameters) and solar radiation as output parameter. The
learning algorithm used in this network is scaled conjugate gradient, Pola-Ribiere conjugate gradient, Levenberg Marquardt and
a logistic sigmoid transfer function. The MAPE value of the MLP
network after prediction is found to be 6.73%.
Rajesh et al. [35] developed a New Regression Model to Estimate Global Solar Radiation Using Articial Neural Network by
using sunshine duration as input. The monthly average global solar
radiation data of four different locations in North India were
analyzed by using a neural network tting tool. The network
shows the data was best tted when the regression coefcient is
0.99558 and validation performance 0.85906. The values of a and
b with its MPE and MBE values computed for four stations of
North India have been presented as in tabular form.
Voyant et al. [36] studied the effect of exogenous meteorological variables during the prediction of daily solar radiation.
After prediction the root mean square error (RMSE) is found to be
0.5%, 1% of Corsica Island, France. But the combination of both
endogenous and exogenous variables decreases the RMSE value by
1% improving prediction accuracy.
Laidi Maamar et al. [37] used an articial neural network (ANN)
for the estimation of daily global solar radiation (DGSR) on a
horizontal surface by using parameters from the meteorological
station located inside the University. Six input parameters elevation, longitude, latitude, air temperature, relative humidity, and
wind speed were used to predicate the measured data of 2011 for
training and testing the neural networks. The optimized network
with the lowest error during the training was obtained with one
with six neurons in the input layer, six neurons in the hidden, and
one neuron in the output layer.
AI-Alawi and AI-Hinai [38] used Multilayer feed forward network with a back propagation training algorithm to predict global
solar radiation for Seeb locations Based on input location, monthly
mean pressure, mean temperature, mean vapor pressure, mean
relative humidity, mean wind speed and mean sunshine hours.
The prediction gives an MAPE range from 5.43 to 7.30.
Hasni et al. [39] estimated global solar radiation by using input
parameters air temperature, relative humidity in south-western
region of Algeria. The training is done using LM feed-forward back
propagation algorithm. The hyperbolic tangent sigmoid and
purelin transfer function used in hidden and output layers. The
MAPE, R2 are 2.9971%, 99.99%.
Lu et al. [40] used ANN model for estimating daily global solar
radiation over China using Multi-functional Transport Satellite

(MTSAT) data. The model takes daytime mean air mass, surface
altitude as different input combinations and daily clearness index
as output. The results show that the ANN model by using daytime
mean air mass, surface altitude inputs give better correlation value
to model than the model which uses only surface altitude as input.
Yildiz et al. [41] used two models (ANN-1, ANN-2) for the
estimation of solar radiation in Turkey. The ANN-1model uses
inputs as latitude, longitude, and altitude month and meteorological and surface temperature where as ANN-2model uses
latitude, longitude, altitude, month and satellite and surface
temperature as inputs. The regression values for model ANN-1 and
ANN-2 are 80.41%, 82.37% respectively.
Ouammi et al. [42] applied ANN model for estimating monthly
solar irradiation of 41 Moroccan sites for the period 1998 to 2010
by taking inputs longitude, latitude and elevation. The predicted
solar irradiation varies from 5030 to 6230Wh=m2 =day.
Sivamadhavi [43] used multilayer feed forward (MLFF) neural
network based on back propagation algorithm to predict monthly
mean daily global radiation in Tamil Nadu, India. Various geographical and meteorological parameters of three different locations were used as input parameters. Out of 565 available data,
530 data were used for training and the rest were used for testing
the articial neural network. A 3-layer and a 4-layer MLFF networks were developed and the performance of the developed
models was evaluated based on mean bias error, mean absolute
percentage error, root mean squared error and Students t-test.
Linares-Rodriguez et al. [44] used articial neural network to
generate synthetic daily global solar radiation by using data total
cloud cover, skin temperature, total column water vapor and total
column ozone at Andalusia (Spain), and is presented in Table 12.
The model used measured data for nine years from 83 ground
stations. The accuracy of the model is evaluated by using following
statistical errors (mean bias error, root mean square error, correlation coefcient(R).
A. Mellit et al. [45] embedded articial intelligent technique
such as,a Field Programmable Gate Array for predicting global
solar radiation at Al-Madinah (Saudi Arabia) from 1998 to 2002,
that is represented in Table 13.The parameters used in this model
are temperature, humidity, sunshine duration, day of the year. In
this paper six different models are developed by varying the
number of input data.

G f t; T; S; RH ; G f t; T; S; G f t; T; RH ; G f t; S; RH ;

G f t; T ; G f t; S
The correlation coefcient lies between 89% and 97%, and the
MBE varied between 4% and 6%.The model concludes with the
sunshine duration that provides much better results, which will
increases the performance of the predictor.
Kadirgama et al. [46] used Articial Networks: for estimating
solar radiation of East Coast Malaysia. The input parameters are
temperature, time, wind chill, pressure and Humidity. The maximum mean absolute percentage error was found to be less than

786

S. Mohanty et al. / Renewable and Sustainable Energy Reviews 56 (2016) 778796

7.74% and R-squared (R2 ) values were found to be about 98.9% of


the testing stations. Similarly for training stations the MAPE and Rsquared is about 5.398% and 97.9%.
Elminir et al. [47] used the ANN model to predict the diffuse
fraction (K D ) in hourly and daily basis. The meteorological input
parameters, used here are long-wave atmospheric emitted, air
temperature; relative humidity and atmospheric pressure. The
result shows that ANN based model used for diffuse fraction is
more suitable for predicting the diffuse fraction in hourly basis
than the regression model.

Table 12
Forecasting capability of the model. Error values of the ANN model, for data from
2009/01/01 to 2009/09/30 [44].
Stations

MBE

RMSE

65 training station (17745 values)


18 testing stations
All stations (22659 values)

0.14
0.49
0.22

2.83
3.016
2.87

0.94
0.93
0.94

Table 13
Correlation coefcient of models and architecture [45].
Conguration

Accuracy

Architecture

G f t; T; S; RH
G f t; T; S
G f t; T; RH
G f t; S; RH
G f t; T
G f t; S

0.9720
0.9749
0.8978
0.9730
0.8927
0.9724

4-4-1
3-4-1
3-4-1
3-4-1
2-4-1
2-4-1

Table 14
Statistical error estimation of different combination model [50].
Combined model

MBE

RMSE

MPE

R2

MPL-1
MPL-2
MPL-3
MPL-4

0.167
0.103
0.856
0.248

0.295
0.288
2.117
0.901

 7.72
 4.17
 39.5
 11.9

0.95
0.96
0.54
0.93

Table 15
Regression plot and Error value analysis of ANN during training [51].
Station

MSE

MAPE

SSE

V2

Ahmadabad
Mangalore
Mumbai
Kolkata

0.027
0.053
0.002
0.033

1.280
2.146
0.278
1.989

0.33
0.63
0.02
0.40

99.5
99
99.9
99.3

Angela et al. [48] used ve years of global solar radiation data in


Uganda to estimate the monthly average of daily global solar
irradiation on a horizontal surface based on a single parameter,
sunshine hours, using the articial neural network technique. A
correlation coefcient of 0.963 was obtained with a mean bias
error of 0.055 MJ=m2 and a root mean square error of 0.521MJ=m2 .
Sanusi et al. [49] used Articial Neural Networks (ANNs) to
predict daily solar radiation in Sokoto having latitude and longitude (13 03N, 5 14E) based on parameters such as sunshine
hours, air temperature, relative humidity along with day number
and month number. After Comparison the model shows the following results in tabular form.
Lazzs et al. [50] embedded ANN model (shown in Table 14) to
estimate hourly global solar radiation for La Serena in Chile. The
input parameters used in this model are wind speed, relative
humidity, air temperature and soil temperature. The regression
coefcient (R 96%), shows the strong correlation between hourly
global solar radiation and meteorological data.
Amit Kumar Yadav [51] used Articial Neural Network tting
tool for Predicting Solar Radiation for 12 Indian Stations with
different climatic parameters such as latitude, longitude, sunshine
hours and height above sea level, and is shown in Table 15. The
geographical and sunshine hour data for cities Ahmadabad, Mangalore, Mumbai, Kolkata, Chandigarh, Dehradun, Jodhpur, Luck
now, are used for training and the geographical and sunshine hour
data for cities Nagpur, New Delhi, Shillong, Vishakhapatnam, is
used for testing. The results of ANN model are compared with the
measured data on the basis of root mean square error (RMSE) and
mean bias error (MBE).RMSE with the ANN model varies 0.04863.562 for the Indian region. The Study indicates that the selected
ANN model has lower RMSE.
Yacef et al. [52] prepares a comparative study between Bayesian Neural Network (BNN), classical Neural Network (NN) and
empirical models (shown in Table 16) for estimating the daily
global solar irradiation (DGSR) of Al-Madinah (Saudi Arabia) from
1998-2002.A comparative study also has been made between
Bayesian network with classical neural network and empirical
model developed by AngstromPrescott equation. The performance of different models is measured by calculating RMSE, MBE
and MAE for training and testing of different data shown in
Table 16.
Seyed Fazel Ziaei Asl et al. [53] used multilayer perceptron
(MLP) neural network to predict daily global solar radiation based
on meteorological variable daily mean air temperature, relative
humidity, sunshine hours, evaporation, wind speed, and soil
temperature values from 20022006 for Dezful city in Iran having
latitude 32 16' N and longitude 48 25 E, as shown in Fig. 9. After
prediction the model will produce the result mean absolute percentage error (MAPE) 6.08% and absolute fraction of variance (R2 )
99.03% (on testing data) and mean square error (MSE) 0.0042 and
sum of square error (SSE) 5.9278 (on training data).
Ibeh et al. [54] used angstrom and MLP ANN models to estimate
mean monthly global solar radiation on horizontal surface based
on meteorological parameters such as maximum temperature,
relative humidity, cloudiness and sunshine duration for Warri-

Table 16
Performance of the different models during Training and Testing [52].
Model

Classical NN (3 inputs-20 hidden units)


Bayesian NN (3 inputs-20 hidden units)
Bayesian NN (2 inputs-20 hidden units)
Bayesian NN (2 inputs-2 hidden units)

Training Set

Test Set

RMSE

MBE

MAE

RMSE

MBE

MAE

6.0024
8.2493
7.5815
15.0593

0.1144
0.0747
0.0509
0.3704

4.1155
4.8432
4.6670
5.0525

17.0582
12.7968
9.3184
8.4173

4.6977
3.8352
3.3139
3.0658

7.0969
6.3513
5.9386
5.9092

S. Mohanty et al. / Renewable and Sustainable Energy Reviews 56 (2016) 778796

787

Fig. 9. Comparison between measured and estimated daily GSR (testing data) [53].

Fig. 10. Comparison between Measured, MLP Predicted and Empirical Model predicted of solar radiation [54].

Nigeria from 1991 to 2007, and is shown in Fig. 10.To compare the
performance of ANN models and Angstrom-Prescott model, statistical analysis using Mean bias error (MBE), Root mean square
error (RMSE) and Mean percent error (MPE)) have been taken. The
result shows that ANN model provides better performance in
comparison to AngstromPrescott empirical model.
Azeez [55] used feed forward back propagation neural network
to estimate monthly average global solar irradiation on a horizontal surface for Gusau, Nigeria based on the input parameters
sunshine duration, maximum ambient temperature and relative
humidity and solar irradiation as output parameter. After Statistical analysis the results (R99.96, MPE0.8512, and
RMSE0.0028) show the best correlation between the estimated
and measured values of global solar irradiation.
Rahimikhoob [56] estimated global solar radiation of Iran from
(1994 to 2001) for training and from (2002 to 2003) for testing by
using Articial neural network with maximum and minimum air
temperature as input, and it is shown in Fig. 11 and Table 17.The
empirical Hargreaves and Samani equation (HS) is used for the
comparison. The comparison result shows the ANN model was
superior to the calibrated Hargreaves and Samani equation.

Koca et al. [57] used articial neural network (ANN) model to


estimate the solar radiation with different parameters (latitude,
longitude, and altitude, month of the year and mean cloudiness)
for the seven cities of the Mediterranean region of Anatolia in
Turkey,which is presented in Table 18. The output of the model has
been compared with the output by changing the number of inputs
of different cities.
Krishnaiah et al. [58] used Neural Network approach for estimating hourly global solar radiation (HGSR) in India. Here the
author takes solar radiation data from seven Indian stations for
training the ANN and data from two Indian stations for testing.
Multi layer feed forward neural network with back propagation
Table 17
Model and the calibrated Hargreaves and Samani equation [55].
Model

R2

RMSE

RE%

ANN
Calibrated HS

0.89
0.84

2.53
3.64

13.83
19.84

0.97
0.89

Fig. 11. Comparative results of the measured GSR with estimated by Using ANN) [55].

788

S. Mohanty et al. / Renewable and Sustainable Energy Reviews 56 (2016) 778796

Table 18
R2 values of the ANN method with different input parameters [57].
Station No. of parameters

Isparta

Lat,long,Altitude
Month

Lat,long,Altitude Month,Avg
cloudness

Lat,Long,Altitude,Month,
Avg temp

Lat,Long,Altitude,Month,Avg, Lat,Long,Alt,Month,Avg,Wind Lat,Long,Alt,


humidity
Velocity
Month,AvgCloudiness,
Sunshine
duration

0.9971

0.9974

0.9959

0.9978

0.9934
0.9920

K. Maras

0.9916

0.9931

0.9534

0.9821

0.9916
0.9898

Mersin

0.9960

0.9906

0.9373

0.9763

0.9839
0.9879

Adana

0.9936

0.9945

0.9446

0.9883

0.9920
0.9810

Antakya

0.9943

0.9944

0.9062

0.9879

0.9872
0.9868

Ave.R(%)

0.9945

0.994

0.9474

0.9864

0.9896
0.9875

learning is used for the modeling and testing.The performance of


the model can be evaluated on the basis of root mean square error
(RMSE), Mean bias error (MBE) and absolute fraction of variance
(r2).The results show that the neural network approaches are more
suitable to predict the solar radiation as compared to traditional
regression models.
Rehman and Mohandas [59] used RBF network approach for
modeling of diffuse and direct normal solar radiation for sites in
Saudi Arabia based on input data: day, global solar radiation,
ambient temperature and relative humidity. The result indicates
that RBF (50 hidden neurons, 0.1 spread constant) predicts direct
normal solar radiation with MAPE of 0.016 and 0.41 for diffuse
solar radiation.
Senkal [60] uses articial neural networks (ANNs) for the
estimation of solar radiation in Turkey from August 1997 to
December 1997 for 12 cities (Antalya, Artvin, Edirne, Kayseri,
Kutahya, Van, Adana, Ankara, Istanbul, Samsun _Izmir, Diyarbakir.
The Meteorological and geographical parameters used in this
model are (latitude, longitude, altitude, month, mean diffuse
radiation and mean beam radiation).Correlation values indicate a
relatively good agreement between the observed ANN values and
the predicted satellite values.The maximum correlation coefcient
was obtained as 99.93% for Van station by using physical method,
while the minimum correlation coefcient was obtained as 84.51%
for Kayseri station.
enkal and Kuleli [61] applied ANN and physical model to
estimate solar radiation for 12 cities in Turkey. The input parameters used in this model are latitude, longitude, altitude, and
month, mean diffuse radiation and mean beam radiation. Out of 12
cities, data of 9 cities are used for training a neural network and
remaining 3 cities are used for testing. The RMSE value after
training, using the MLP and the physical model is 54 W=m2 and 64
W=m2 and after testing the values becomes 91 W=m2 and
125W=m2 .
enkal [62] used generalized regression neural network
(GRNN) for estimating solar radiation in Turkey by taking input
latitude, longitude, altitude, surface emissivity, land surface temperature and solar radiation as output. The statistical analysis gives
RMSE with R2 values are 0.1630MJ=m2 , 95.34% after training and
0.3200MJ=m2 , 93.41% after testing respectively.
Vassiliki et al. [63] predicts cloud amount that affects solar
radiation, using neural network soft computing approach based on
following parameters, i.e. air temperature, dew point, air humidity,
sea level pressure, visibility, wind speed, wind direction and
amount of clouds. A total of sixteen years of data of

Alexandroupolis, Greece has been divided into two parts.i.e.the


data of fteen years are used for training and remaining 1 year of
data used for testing.
Elminir et al. [64] Predicted hourly and daily diffuse radiation of
Egypt by using neural network and compared it with two linear
regression models. The performances of the models were assessed
on the basis of the mean bias error (MBE), RMSE and correlation
coefcient (r) between predicted and measured data. The result
shows that the ANN model is more suitable to predict diffuse
radiation in hourly and daily scales than the regression models.
Fei Wang et al. [65] used Articial Neural Network (ANN) for
modeling short-term solar irradiance forecasting based on Statistical Feature Parameters.The comparison of measured data with
the forecasted values shows the proposed model is reliably and
more effective.
Ozgur et al. [66] used Articial Neural Networks to predict
hourly solar radiation in Turkey on the basis of six parameters
latitude, longitude, altitude, day of the year, hour of the day, and
mean hourly atmospheric air temperature. Two different models
have been analyzed for training and testing. The results obtained
from both models were compared by calculating Mean squared
error (MSE), coefcient of determination (R2 ), and Mean absolute
error (MAE) as shown in Fig. 12.
Santamouris et al. [67] used one atmospheric deterministic
model and two intelligent data-driven techniques for estimating
Global solar radiation on the earths surface. The following parameters such as the air temperature, the relative humidity, and the
sunshine duration are used to predict solar radiation hourly values
of the year 1995.The comparison of the three methods shows the
proposed intelligent technique gives better performance of global
solar radiation during the warm period of the year, while during
the cold period the atmospheric deterministic model gives better
performance.
3.2. ANFIS (Articial neuro-fuzzy inference system)
The ANFIS is a multilayer feed-forward network, which uses
neural network learning algorithms and fuzzy reasoning to map
inputs into an output. ANFIS derives its name from adaptive
neuro-fuzzy inference system which uses a combination of least
squares and back propagation algorithm for estimation of activation function.ANFIS is based on conventional mathematical tools
that combine the properties of fuzzy logic and neural networks to
form a hybrid intelligent system. It enhances the ability to learn

S. Mohanty et al. / Renewable and Sustainable Energy Reviews 56 (2016) 778796

789

Fig. 12. Relative error of the articial neural network models for prediction of global solar radiation in Turkey [66].

and adapt automatically. This section describes the prediction and


estimation of solar radiation by using ANFIS technique.
Rahoma et al. [68] uses Articial neuro-fuzzy inference system
to generate daily solar radiation data recorded on a horizontal
surface in National Research Institute of Astronomy and Geophysics, Helwan, Egypt (NARIG) for ten years (19912000), where
solar radiation measurements are not available. The paper uses
ANFIS as a combination of fuzzy logic and neural network techniques to gain more efciency. The prediction shows TS fuzzy
model gives a better accuracy of approximately 96% and a root
mean square error lower than 6%.The results show that the identied TS fuzzy model provides better performances.
Mohammad et al. [69] applied potential of adaptive neurofuzzy system for prediction of daily global solar radiation by day of
the year. Estimation of the horizontal global solar radiation by day
of the year nday is particularly appealing as there is no need of
using any specic meteorological input data or even precalculation analysis. So an intelligent optimization technique
based upon adaptive Neuro-fuzzy inference system (ANFIS) was
applied to develop a model for the estimation of daily horizontal
global solar radiation using nday as the input. Long-term measured
data for Iranian city of Tabass was used to train and test the ANFIS
model. The statistical result shows the ANFIS model provides
accurate and reliable predictions. After Statistical analysis, the
mean absolute percentage error, Mean absolute bias error, Root
mean square error and correlation coefcient were found to be
3.9569%, 0.6911MJ=m2 , 0.8917 MJ=m2 and 0.9908MJ=m2 respectively. The daily bias errors between the ANFIS predictions and
measured data fell in the range of 3 to 3MJ=m2 .
Iqdour et al. [70] used fuzzy systems for modeling the daily
solar radiation data recorded on a horizontal surface in Dakhla in
Morocco. In Table 19, the performances of the fuzzy model have
been compared with a linear model using the SOS techniques. The
prediction results of the TS fuzzy model are compared with the
linear model.
Mohanty [71] used ANFIS for prediction and comparison of
monthly average solar radiation data of Bhubaneswar location.
Comparison also has been done on the basis of mean absolute
percentage error.
T.R.Sumithira et al. [72] used an adaptive neuro-fuzzy inference
system (ANFIS) to predict the monthly global solar radiation
(MGSR) in Tamilnadu of 31 districts (in Fig. 13). Comparison of the
predicted and measured value of monthly global solar radiation
(MGSR) on a horizontal surface was evaluated by calculating root
mean square error (RMSE), Mean bias error (MBE), and coefcient
of determination (R2 ) for testing locations.

Table 19
Comparison between measured and predicted data using two models [70].
Statistical indicators

RMSE

TS Fuzzy model
Linear model

0.505
0.612

96%
89%

Mellit et al. [73] used an adaptive Neuro-fuzzy inference system


(ANFIS) model for estimating sequence of monthly mean clearness
index (K t ) and daily solar radiation data in isolated areas of
Algerian location with some geographical coordinates (latitude,
longitude and altitude) and meteorological parameters such as
temperature, humidity and wind speed and is shown in Fig. 14. A
comparison also has been made between ANFIS and ANN by
evaluating the root mean square error (RMSE) and mean absolute
percentage error (MAPE).
The result shows ANFIS gives better performance in Comparison to ANN architecture such as (RBFN, MLP and RNN).The main
advantage of this model is it can estimate K t from only the geographical coordinates of the site.
Mohanty et al. [74] Used soft computing approaches (MLP, RBF,
ANFIS) for comparison and prediction solar radiation data of
eastern India from 1984-1999, and is presented in Fig. 15.
Celik and Muneer [75] used generalized regression neural
networks (GRNN) to predict solar radiation on the tilt surface in
Iskenderun, Turkey. The model utilizes input parameters as global
solar irradiation on a horizontal surface, declination and hour
angles. The MAPE,R2 are 14.9Wh=m2 , 98.7%, respectively
Chatterjee and Keyhani [76] used ANN to estimate total solar
radiation (SR) on tilt surface, taking the input parameters (latitude,
ground reectivity and 12 month irradiance values). The output
layer contains ve neurons corresponding to four quarterly optimum tilt angles and total solar radiation on a tilted surface. The
activation function used in the hidden layer is hyperbolic tangent
and linear in the output layer. The LM algorithm has been used for
training and the best validation performance is obtained with
minimum RMSE i.e. 3.2033 at epoch7.The ANN estimates the
optimum tilt angle with 31 accuracy also can be used for estimating optimum tilt angles
Rizwan et al. [77] used a generalized neural network (GNN),
and a modied approach of articial neural network (ANN) to
estimate solar energy in India from 1986-2000 based on different
meteorological and climatological parameter such as sunshine per
hour, temperature ratio, clearness index, latitude, longitude and
altitude and is shown in Table 20. The results of the GNN model

790

S. Mohanty et al. / Renewable and Sustainable Energy Reviews 56 (2016) 778796

Fig. 13. Comparison of predicted and measured values by using membership function [72].

Fig. 14. Mean relative error for the array area for the four testing sites [73].

Fig. 15. Measured and predicted data using soft computing approaches for Bhubaneswar and Vishakhapatnam [74].

and the fuzzy-logic-based model considering the same input


parameters can be compared on the basis of mean relative error.
The mean relative error in the estimation of global solar energy is
found around 4%whereas the same using fuzzy logic is 6%.
Yacef et al. [78] generates a comparative study between Bayesian Neural Network (BNN), Classical Neural Network (NN) and
empirical models for estimating the daily global solar irradiation
(DGSR) from 1998 to 2002 at Al-Madinah (Saudi Arabia) (i.e.
presented in Table 21). Four input parameters have been used such
as air temperature, relative humidity, sunshine duration and
extraterrestrial irradiation. After prediction Bayesian Neural Network (BNN) shows a better prediction than other examined
models (NN structures and empirical models).The performances of
different models are measured in terms of RMSE, MBE and MAE.
Mohamad et al. [79] used Recurrent Neural Networks for Predicting Global Solar Radiation based on Climatological Parameters
(presented in Fig. 16. and Table 22) such as air temperature,
humidity, sunshine duration, and wind speed from 1995 and 2007.

The obtained results by the RNN-based system are compared to


those obtained by other empirical and neural based systems.
The model shows an under estimation between January and
August, and an over estimation between September and
December.
3.3. Radial Basis Function Network (RBFN)
A radial basis function network is an articial network whose
activation function is a radial basis function. The output of the
network is a combination of radial basis functions of the inputs
and neuron parameters. A radial basis function (RBF) network
contains three layers such as an input layer, a hidden layer with a
non-linear RBF activation function and a linear output layer. The
second layer /hidden layer perform a nonlinear mapping from the
input space into a higher dimensional space by using a Gaussian or
some other kernel function. Output layer:-The nal layer performs
a weighted sum with a linear output.

S. Mohanty et al. / Renewable and Sustainable Energy Reviews 56 (2016) 778796

791

Table 20
Absolute Relative Error for Newdelhi, Jodhpur, Nagpur and Shillong Using Generalized Neural Network and Fuzzy logic [77].
Relative error %
Generalized Neural Network

Minimum
Average
Maximum

Fuzzy Logic

Newdelhi

Jodhpur

Nagpur

Shillong

Newdelhi

Jodhpur

Nagpur

Shillong

1.9048
3.2891
4.8768

1.7739
3.1526
4.4953

2.5011
4.6463
6.1574

3.5189
4.8286
6.5876

4.3452
5.1145
7.0771

3.504
5.174
7.124

4.523
5.432
6.913

4.8745
5.6703
7.1864

Table 21
Comparative study between Bayesian Network and Classical Network based upon statistical error [78].
Model

Training Set

Classical NN(3inputs- 20 hidden units)


Bayesian NN(3 inputs- 20 hidden units)
Bayesian NN(2 inputs-20 hidden units)
Bayesian NN(2 inputs- 2 hidden units)

Test Set

RMSE

MBE

MAE

RMSE

MBE

MAE

6.0024
8.2493
7.5815
15.059

0.1144
0.0747
0.0509
0.3704

4.115
4.843
4.667
5.052

17.05
12.79
9.318
8.417

4.697
3.835
3.313
3.065

7.096
6.351
5.938
5.909

Fig. 16. Exact and Estimated monthly Average Global solar radiation [79].

Table 22
MBE, RMSE and Architecture of Models [79].

Table 23
Comparative study between developed RBF-models and conventional regression
models [80].

System

Architecture

RMSE

MBE

1
2
3
4

MLP
MLP
MLP
MLP

0.0659
0.0680
0.0731
0.0520

0.0085
 0.0080
 0.0003
 0.0006

Models
4-6-1
4-12-1
5-4-1
5-9-1

The input can be modeled as a vector of real numbersx A n .The


output of the network is then a scalar function of the input
vector, : n -, and is given by

N
X

ai j j x  ci j j

RBF Models
H G f t; S
H G f t; S; T
H G f t; S; T; RH
H G f t; T; RH
Conventional regression Model
H G =Ho  0:3824 1:278S=So


2
H G =Ho 0:1166  0:2202S=So 1:0723 S=So
H G =Ho 0:6369 0:037T=T max
H G =Ho 0:7556  0:1353RH=RH max

r%

RMSE

98.21
98.80
98.72
91.16

0.03748
0.01310
0.03241
0.04512

97.28
97.48

0.0512
0.4410

89.50
86.59

0.1215
0.2518

i1

where N is the number of neurons in the hidden layer, ci is the


center vector for neuroni, and ai is the weight of neuron iin the
linear output neuron. The major difference between RBF networks
and back propagation networks is the single hidden layer with RBF
activation function instead of using the sigmoid or S-shaped
activation function as in back propagation.
Mohamed Benghanem et al. [80] used Radial Basis Function
network (RBF) for modeling and predicting the daily global solar
radiation data using meteorological parameter such as air temperature, sunshine duration, and relative humidity, and is shown
in Table 23. The data were recorded from 1998 to 2002 at AlMadinah (Saudi Arabia) by the National Renewable Energy
Laboratory. Based upon the number of inputs four RBF-models
have been developed for predicting the daily global solar radiation.
After prediction the result shows that the RBF-model which uses

the sunshine duration and air temperature as input parameters


gives accurate results as the correlation coefcient in this case is
98.80%.
Naderian et al. [81] used two types of neural networks for
simulating Global solar radiation based on input parameters such
as monthly mean air temperature, maximum air temperature,
minimum air temperature, and relative humidity and sunshine
hours. The data from 1990 to 2006 were used to train the networks while the measured data from 2007 to 2010 were used for
validating the trained networks. To nd the best network structure, several networks were designed and the number of neurons
and hidden layers are changed. One hidden layer with 12 neurons
was found to be the best designed network, which will have an
absolute fraction of variance (R2 ) is 90.81% and mean absolute
percentage error (MAPE) of 8.95%.

792

S. Mohanty et al. / Renewable and Sustainable Energy Reviews 56 (2016) 778796

Jianwu Zeng [82] predicts short-term solar power using a radial


basis function (RBF) neural network-based model. The model uses
a novel two-dimensional (2D) representation for hourly solar
radiation prediction by using historical transmissivity, sky cover,
relative humidity and wind speed as the input. The performance of
the RBF neural network is compared with that of two linear
regression models, i.e. an autoregressive (AR) model and a local
linear regression (LLR) model. The Result shows the RBF neural
network has better performance than the AR and LLR models in
terms of the prediction accuracy.
Mishra et al. [83] estimates direct solar radiation by using radial
basis functions (RBF) and ve MLP networks. The network uses the
following input parameters latitude, longitude, mean sunshine per
hour duration, relative humidity ratio, rainfall ratio and month for
predicting direct solar radiation of eight stations in India. The
Prediction result shows MLP performed better than RBF, as the
RMSE value of RBF network is 7-29% and for the MLP is (0.8-5.4%).

4. Pros and cons of different models described in literature


In traditional way, the approaches used for prediction are:
(a) The empirical approach and (b) The dynamical approach.
Generally Empirical models are based entirely on data. These
models have uncertainty in terms of prediction. Empirical
approaches for nding solar radiation are a function of sunshine
duration only. However in humid regions or coastal region apart
from sunshine duration temperature and humidity are factors
which affect the solar radiation. The dynamical approach is only
useful for modeling large-scale solar radiation prediction but the
disadvantage is it may not predict short-term radiation.
But for local scale & short term solar radiation prediction, the
soft computing approaches are used, which perform nonlinear
mapping between inputs and output.
Main advantage of using articial neural network is,1) requires
less formal statistical training 2) ability to implicitly detect complex nonlinear relationships between dependent and independent
variables 3) ability to detect all possible interactions between
predictor variables 4) and the availability of multiple training
algorithms. Disadvantages are its 1) black box nature 2) greater
computational burden 3) proneness to over tting, and the
empirical nature of model development.
Radial basis function (RBF) is a networks having single hidden
layer with RBF activation function.Radial basis function networks
have advantages of (1) easy design, (2) good generalization,
(3) strong tolerance to input noise, and (4)online learning ability.
This paper presents a review on different approaches of designing
and training RBF networks.
The advantage of using ANFIS is, uses a combination of least
squares and back propagation algorithm for estimation of activation function. ANFIS is based on conventional mathematical tools
which combine the properties of fuzzy logic and neural networks
to form a hybrid intelligent system that enhances the ability to
learn and adapt automatically, produce good result.

5. Application of solar radiation


Solar energy is having several applications mainly in thermal
and electrical spheres. Thermal solar systems are used for water
heating, cooling and heat generation process. Solar power is the
conversion of sunlight into electricity, either directly using photovoltaic (PV), or indirectly using concentrated solar power (CSP).
So prediction of solar radiation for a particular location is necessary. Several methods have been used for solar radiation

prediction. This section describes the prediction of solar radiation


and its application to thermal and photovoltaic.
5.1. Application to Photovoltaic system
Salman Quaiyum, Shahriar Rahman and Saidur Rahman [84]
show an application of articial neural network to predict solar
radiation from a dataset collected for a period of nine years from
1992 to 2001.After that, these forecasted values of solar radiation
are used to size standalone PV systems for different locations in
Bangladesh. The result found indicates that establishment of
regional hub or sub-grid will be much more appropriate for
harnessing.
Mahendra Lalwani1, Kothari and Mool Singh [85] describe the
optimal sizing of solar array and battery in a stand-alone photovoltaic (SPV) system, under the conditions of a xed tilt angle and
continuous size variations of solar array and battery. The optimal
sizes of the solar array and battery were to nd it at the minimum
cost of the system, under the specic load demand and the
coveted LPSP.
Khatib et al. [86] shows a new method for determining the
optimal sizing of stand-alone photovoltaic (PV) system in terms of
optimal Sizing of PV array and battery storage. The MATLAB tting
tool is used to t the sizing curves. The data considered for optimal
sizing of the PV array and battery is based in ve cities in Malaysia.
The result shows the designed example for a PV system installed
in Kuala Lumpur, gives satisfactory optimal sizing results.
Saberian et al. [87] Presents a solar power modeling method
using articial neural networks (ANNs).Two methods, i.e. general
regression neural network (GRNN) and feed forward back propagation (FFBP) algorithm have been used to model a photovoltaic
panel output power and approximate the generated power based
on input parameters maximum temperature, minimum temperature, mean temperature, and irradiance; The FFBP neural network
gives better modeling performance Compared to GRNN.
Khaled Bataineh and Doraid Dalalah [88] show a design for a
stand-alone photovoltaic (PV) system for providing required
electricity for a single residential household in rural areas in Jordan. The reliability of the system is quantied by the loss of load
probability. The results shows that using the optimal conguration
for electrifying remote areas in Jordan is benecial and suitable for
long-term investments, especially if the initial prices of the PV
systems are decreased and their efciencies are increased.
M. A [89] used neural networks and genetic algorithms for
sizing of stand-alone photovoltaic system. The author used total
solar radiation data of 40 locations in Algeria to determine the ISOreliability (sizing) curves of a SAPV system (CA, CS).The result
shows a correlation coefcient of 98%.
Guda, H. A, and Aliyu U. O [90] show the design of a standalone photovoltaic power system for a general residential building
in Bauchi, located in Nigeria. Here a photovoltaic power system
can be used to provide an alternative and inexhaustible source of
electrical power to homes through the direct conversion of solar
irradiance into electricity.
Sanusi Y.K, Abisoye S. G, and Awodugba A. O. [91] used articial
neural network for predicting the optimal sizing parameters of
stand-alone photovoltaic system in remote areas based on geographical coordinates. The statistical analysis shows MBE ranged
from 0.046 to  0.078, RMSE ranged from 0.046 to 0.085 and MPE
ranged from -1.262 to 0.749.As the MBE, RMSE and MPE are very
small,these show a good t between measured and ANN model
sizing parameters. So this model is used to predict the PV-array
area and the storage capacity of isolated sites in Nigeria where
solar radiation data is not always available.
Mellit [92] used the Radial basis function networks (RBFN) to
model the optimal sizing curve of stand-alone photovoltaic (PV)

S. Mohanty et al. / Renewable and Sustainable Energy Reviews 56 (2016) 778796

system based on minimum input. The result shows a correlation


coefcient of 98% was reached between the actual and
RBFN model.
Markvart et al. [93] gives a description of a sizing procedure
based on the observed time series solar radiation. The sizing curve
is determined from climatic cycles of low daily solar radiation.
Mohamed Benghanem et al. [80] used Radial Basis Function
network (RBF) for modeling and predicting the daily global solar
radiation data using meteorological parameter such as air temperature, sunshine duration, and relative humidity. The data were
recorded from 1998-2002 at Al-Madinah (Saudi Arabia) by the
National Renewable Energy Laboratory. Based upon the number of
inputs Four RBF-models have been developed for predicting the
daily global solar radiation. After prediction, the result shows that
unlike other models the RBF-model which uses the sunshine
duration and air temperature as input parameters, gives accurate
results as the correlation coefcient in this case is 98.80%.
Chen Qi and Zhu Ming [94] proposed a one-diode equivalent
circuit-based simulation model for a stand-alone photovoltaic
system. The behavior of PV module can be estimated by changing
irradiance intensity, ambient temperature and parameters of the
PV module. The model is capable of resulting in Maximum Power
Point Tracking (MPPT) which can be used in the dynamic simulation of stand-alone PV systems. The main aim of this paper is the
implementation of a PV model in the form of masked block, and by
the model it can predict the electrical output of an arbitrary
module using a one-diode equivalent circuit or a maximum power
point tracking (MPPT) circuit, It is connected to the module the IV
characteristics of PV module with different combination.
Mathew et al. [95] suggested an optimal sizing procedure and
tracking for standalone photovoltaic system of thondamuthur
region, one of the remote areas of India. The PV array can be tilted
at 30, 30, 20 and 20 in every three months to receive maximum
global solar energy. Optimization of PV array tilt angle can reduce
the number of site visits, minimize shading effect and increase
radiation Capturing efciency without any tracking system as it
increases the initial cost, losses and maintenance cost. On comparing both the numerical and intuitive method, the cost estimation results show that the intuitive method is more expensive than
numerical method.
Suchitra et al. [96] shows Optimization of a PV-Diesel hybrid
Stand-Alone System using Multi-Objective Genetic Algorithm.
Generally a hybrid system is the combination of two or more
different sources and it is more efcient. Few/ more renewable
sources like wind, fuel cells, and biomass units are combined to
the diversity of energy production which will reduce the contribution of any one source to meet the load.
Vikrant Sharma and S.S. Chandel [97] evaluated the performance of a 190 kWp solar photovoltaic power plant installed at
Khatkar-Kalan, India. The nal yield, reference yield and performance ratio, are varied from 1.45 to 2.84 kW h/kWp-day, 2.29 to
3.53 kW h/kWp-day and 5583% respectively. The annual average
performance ratio, capacity factor and system efciency are 74%,
9.27% and 8.3% respectively. The average annual measured energy
and annual predicted energy yield of the plant are 812.76 kW h/
kWp and 823 kW h/kWp using PVSYST. The estimated energy
yield is in close agreement with measured results with an uncertainty of 1.4%. The total estimated system losses due to irradiance,
temperature, module quality, array mismatch, ohmic wiring and
inverter, are found to be 31.7%. The result shows maximum energy
is generated during the month of March, September, and October
and minimum in January.

793

5.2. Application to thermal


Amir Hematian, YahyaAjabshirchi and Amir Abbas Bakhtiari
[98] present an experimental analysis of at plate solar air collector. The absorber of solar collector made of steel plate having an
area of 2  1 m2 and thickness of 0.5 mm in the form of window
shade has been developed for increasing the air contact area. The
surface of absorbent plate was covered by black paint. For insulating the collector, the glass wool with the thickness of 5 cm was
used. The experiments on the efciency were conducted for a
week during which the atmospheric conditions were almost uniform and data was collected from the collector. The results show
the collector efciency in forced convection was lower, but the low
temperature difference between the inlet and outlet of the collector decreased its heat loss. Also the average air speed in forced
convection was about 21% higher than the natural convection.
The thermal performance of a solar water heating system with
at plate collectors is carried by researchers Ayompe et al. [99] in
the past.
Cruz-Peragon et al. [100] used a general methodology to validate a collector model, with undetermined associated complexity,
serves to characterize the device by means of critical coefcients,
such as the lm convection transfer coefcient, plate absorptance
or emittance.
ANN based approach has been extensively used for obtaining
performance of a solar Output temperature and the performance
of at plate solar collector in literatures Farkas et al. [101], Sozan
et al. [102], and Tariq et al. [103], can be obtained by using ANN
based approach.
Farahat, Sarhaddi, and Ajam [104] present an exergetic optimization of at plate solar collectors to determine the optimal
performance and design parameters of solar to thermal energy
conversion systems. By increasing the incident solar energy per
unit area of the absorber plate, the energy efciency increases.
The modeling of a domestic water heating system has been
used Kalogirou et al. [105] and [106]. Use of MLP and ANFIS are
available in the literatures (Farzad Jafarkazemi et al. [107], which
are used to estimate the performance of at plate solar collectors.
Solar irradiance, ambient temperature, collector tilt angle and
working uid mass ow rate are used as input and the efciency is
presented at the output.
Experimental based study with soft computing has been carried out earlier for solar air heaters described in Karim and Hawlader [108].
The main drawback of at plate absorber air collectors is the
low heat transfer coefcient which shows lower thermal efciency. So the area available for heat transfer should not be greater
than the projected area of the absorber, otherwise the absorber
becomes unnecessarily hot which in turn leads to higher heat loss
[109,110].
Zelzouli et al. [111] present the modeling of a solar collective
heating system to predict the system performances. Two systems
are proposed for this: (1) Solar Direct Hot Water, which is composed of at plate collectors and thermal storage tank, (2) Solar
Indirect Hot Water in which we added an external heat exchanger
of constant effectiveness to the rst system. For the 1st system, the
maximum average water temperature within the tank in a typical
day in summer and annual performances are calculated by varying
the number of collectors connected in series. For the 2nd, the
detailed analysis of water temperature within the storage and
annual performances by varying the mass ow rate on the cold
side of the heat exchanger and the number of collectors in series
on the hot side. It is shown that the stratication within the storage is strongly inuenced by mass ow rate and the connections
between collectors are explained here. The optimization of the

794

S. Mohanty et al. / Renewable and Sustainable Energy Reviews 56 (2016) 778796

mass ow rate on cold side of the heat exchanger is seen to be an


important factor for the energy saving.
Dr. Saad T. Hamidi Mohamaad A. Fayath [112] predicts the
thermal characteristics for open designer shape of solar collector
of at plate of area 2.34 m2, connected to water tank of 8 l
capacity. The results of this research is to obtain hot water at
average temperatures up to 520 C at mid-day during February
month, as the water temperature is at its lowest value in this
month in Baghdad city, with an average efciency of the system up
to 53.6%.This predictive study is compared with a previous measurement work and conrmed that the results match well.
Cuadros et al. [113] used a simple procedure to size active solar
heating schemes for low-energy
building design (1) to estimate the climate variables, (2) to
compare the efciencies of solar heat collectors, and (3) to size
certain installations for domestic hot water, (4) radiant ooring, or
(5) heating of buildings. The values of the climate variables the
monthly means of the daily values of solar radiation, maximum
and minimum temperatures, and number of hours of sun are
determined from data available in the FAOs CLIMWAT database.

6. Conclusion
The prediction or estimation of solar radiation using soft
computing approaches is reviewed extensively in this paper. Solar
radiation data is essential for solar system design, power generation and solar energy research. A number of predictive models
based on soft computing applications, such as Multi layer perceptron, radial basis function, generalized regression, genetic
algorithm, back propagation, Leven bergMarquardt have been
reviewed here. The following meteorological (sunshine Duration,
Temperature, Humidity, Clearness index, Global solar radiation,
Extraterrestrial radiation) and Geographical parameters (Latitude,
Altitude, Longitude) are used for prediction. Results are obtained
either by simulation or by using statistical analysis. The model
gives good accuracy by minimizing the error. Hence this methodology may be adopted to predict solar radiation data in remote
areas or the places where measuring instruments are not available.

References
[1] Angstrom A. Solar and terrestrial radiation. Q J R Meteorol Soc 1924;50:121
6.
[2] Page J.K. The estimation of monthly mean values of daily total short wave
radiation on-vertical and inclined surfaces from sun shine records for latitudes 400N400S. In: Proceedings of the United Nations Conference on New
Sources of Energy, 98. 1961. p. 37890.
[3] Prescott J. Evaporation from water surface in relation to solar radiation. Trans
R Soc S Aust 1940;64:1148.
[4] Benson RB, Paris MV, Sherry JE, Justus CG. Estimation of daily and monthly
direct, diffuse and global solar radiation from sunshine duration measurements. Sol Energy 1984;32:52335 View at Scopus.
[5] Falayi EO, Adepitan JO, Rabiu AB. Empirical models for the correlation of
global solar radiation with meteorological data for Iseyin, Nigeria. Int J Phys
Sci 2008;3:2106 View at Scopus.
[6] Augustine C, Nnabuchi MN. Correlation between Sunshine Hours and Global
Solar Radiation in Warri, Nigeria. Abakaliki, Nigeria: Department of Industrial
Physics, Ebonyi State University; 2009. p. 10.
[7] Medugu DW, Yakubu D. Estimation of mean monthly global solar radiation in
YolaNigeria Using angstrom model. Adv Appl Sci Res 2011;2:41421.
[8] Sanusi Yekinni K, Abisoye Segun G. Estimation of Solar Radiation at Ibadan,
Nigeria. J Emerg Trends Eng Appl Sci 2011;2:7015 ISSN: 2141-7016.
[9] Sa S, Zeroual A, Hassani M. Prediction of global daily solar radiation using
higher Order statistics. Renew Energy 2002;27:64766.
[10] Taha Ahmed Tawk Hussein. Estimation of Hourly Global Solar Radiation in
Egypt Using Mathematical Model. Int J Latest Trends Agric Food Sci 2012:2.
[11] Ghobadi G, Gholizadeh B, Motavalli S. Estimating global solar radiation from
common meteorological data in sari station Iran. Intl J Agric Crop Sci
2013;5:26504.

[12] Ituen EE, Esen NU, Samuel C. Prediction of global solar radiation using
relative humidity, maximum temperature and sunshine hours in Uyo, in the
Niger Delta Region. Nigeria Adv Appl Sci Res 2012;3:192337.
[13] Marwal VK, Punia RC, Sengar N, Mahawar S. A comparative study of correlation functions for estimation of monthly mean daily global solar radiation
for Jaipur, Rajasthan (India). Indian J Sci Technol 2012:5.
[14] Tolabi, et al. New Technique for Global Solar Radiation Prediction using
Imperialist Competitive Algorithm. J. Basic. Appl. Sci. Res 2013;3:95864.
[15] Khorasanizadeh H, Mohammad K, Jalilyand M. A statistical comparative
study to demonstrate the merit of day of the year-based models for estimation of horizontal global solar radiation. Energy Convers Manag
2014;87:3747.
[16] Al-Rawahi NZ, Zurigat YH, AI-Azri NA. Prediction of Hourly Solar Radiation on
Horizontal and Inclined Surfaces for Muscat/Oman. J Eng Res 2011;8:1931.
[17] Almorox J, Bocco M, Willington E. Estimation of daily global solar radiation
from measured temperatures at Canada de Luque, Crdoba, Argentina.
Renew Energy 2013;60:3827.
[18] Kaplanis SN. New methodologies to estimate the hourly global solar radiation; Comparisons with existing models. Renew Energy 2006;31:78190.
[19] Gairaa K, Bakelli Y. A Comparative Study of Some Regression Models to
Estimate the Global Solar Radiation on a Horizontal Surface from Sunshine
Duration and Meteorological Parameters for Ghardaia Site, Algeria, ISRN.
Renew Energy 2013:111.
[20] Kaplanis S, Kaplani E. Stochastic prediction of hourly global solar radiation
for Patra, Greece. Appl Energy 2010;87:374858.
[21] Yohanna JK. A model for determining the global solar radiation for Makurdi,
Nigeria. Renew Energy 2011;36:198992.
[22] Mejdoul R. the Mean Hourly Global Radiation Prediction Models Investigation in Two Different Climate Regions in Morocco. Int J Renew Energy Res
2012:2.
[23] Tu rk Tog rul I, Tog rul H. Global solar radiation over Turkey: comparison of
predicted and measured data. Renew Energy 2002;25:5567.
[24] Li H. Estimating daily global solar radiation by day of year in China. Appl
Energy 2010;87:30117.
[25] Jiang Y. Estimation of monthly mean daily diffuse radiation in China. Appl
Energy 2009;86:145864.
[26] Adaramola MS. Estimating global solar radiation using common meteorological data in Akure, Nigeria. Renew Energy 2012;47:3844.
[27] Bulut H, Bu yu kalaca O. Simple model for the generation of daily global
solar-radiation data in Turkey. Appl Energy 2007;84:47791.
[28] Musa B, Zangina U, Aminu M. Estimation Global Solar radiation of Maiduguri,
Nigeria using Angstrom model. ARPN J Eng Appl Sci 2012:7.
[29] Premalatha1 N, ValanArasu A. Estimation of global solar radiation in India
using articial neural network. Int J Eng Sci Adv Technol 2012;2:171521.
[30] Emad A, El-Nouby Adam M. Estimate of Global Solar Radiation by Using
Articial Neural Network in Qena,Upper Egypt. J Clean Energy Technol
2013;1:14850.
[31] Khatib T, Mohamed A, Mahmoud M, Sopian K. Estimating Global Solar
Energy Using Multilayer Perception Articial Neural Network. Int J Energy
2012:6.
[32] Lubna B, Mohammad A, Eman A. Hourly Solar Radiation Prediction Based on
Nonlinear Autoregressive Exogenous (Narx) Neural Network. Jordan J Mech
Ind Eng 2013;7:118.
[33] Szen A, Arcakliolu E. zalpM, KanitEG.Use of articial neural networks for
mapping of solar potential in Turkey. Appl Energy 2004;77:27386.
[34] Szen A, Arcakliolu E, zalp M. Estimation of solar potential in Turkey by
articial neural networks using meteorological and geographical data.
Energy, Convers Manag 2004;45:303352.
[35] Rajesh K, Aggarwal RK, Sharma JD. New Regression Model to Estimate Global
Solar Radiation Using Articial Neural Network. Adv Energy Eng 2013;1:66
72.
[36] Voyant C, Darras C, Muselli M, Paoli C. Bayesian rules and stochastic models
for high accuracy prediction of solar radiation. Appl Energy 2014;114:218
26.
[37] Maamar L., Salah H., Nawal C. Predicting global solar radiation for North
Algeria. International Conference on Renewable Energies and Power Quality
(ICREPQ14).
[38] AI-AlawiSM AI-HinaiHA. An ANN. Based approach for predicting global
radiation in locations with no direct measurement instrumentation. Renew
Energy 1998;14:199204.
[39] Hasni A., SehliA., DraouiB., BassouA., AmieurB.. Estimating global solar
radiation using articial neural network and climated at ainthesouth- westernregionofAlgeria.EnergyProcedia2012; 18:5317.
[40] Lu N, Qin J, Yang K, Sun J. A simple and efcient algorithm to estimate daily
global solar radiation from geostationary satellite data. Energy 2011;363179
88 2011;36.
[41] Yildiz BY, ahin M, enkal O, Pestemalci V, Emraholu NA. Comparison of
two solar radiation models using articial neural networks and remote
sensing in Turkey. Energy Sources, PartA 2013;35:20917.
[42] Ouammi A, Zejli D, Dagdougui H, Benchrifa R. Articial neural network
analysis of Moroccan solar potential. Renew Sustain Energy Rev
2012;16:487689.
[43] Sivamadhavi V, Samuel Selvaraj R. Prediction of monthly mean daily global
solar radiation using Articial Neural Network. J. Earth Syst. Sci
2012;121:150110.

S. Mohanty et al. / Renewable and Sustainable Energy Reviews 56 (2016) 778796

[44] Linares-Rodriguez A, Antonio Ruiz-Arias J, Pozo-Vzquez D. Generation of


synthetic daily global solar radiation data based on ERA-Interim reanalysis
and articial neural networks. Energy 2011;36:535665.
[45] Mellit A, Mekki H, Messai A, Kalogirou SA. FPGA-based implementation of
intelligent predictor for global solar irradiation. Expert Syst Appl
2011;38:266885.
[46] Kadirgamaa K, Amirruddina AK, Bakara RA. Estimation of Solar Radiation by
Articial Networks: East Coast Malaysia. Energy Procedia 2014;52:3838.
[47] Elmiir HK, Azzam YA, Younes FaragI. Prediction of hourly and daily diffuse
fraction using neural network, as compared to linear regression models.
Energy 2007;32:151323.
[48] Angela K, Taddeo S, James M. Predicting Global Solar Radiation Using an
Articial Neural Network Single-Parameter Model. Advances in Articial
Neural Systems 2011:17.
[49] Sanusi YK, Abisoye SG, Abiodun AO. Application of Articial Neural Networks
to predict Daily solar radiation in Sokoto. Int J Current Eng Technol
2013;3:64752 ISSN 2277-4106.
[50] Lazzs JA, PonceA AP, Marn J. Estimation of global solar radiation over the
City of La Serena (Chile) using a neural network. Appl Sol Energy 2011;47
(1):6673.
[51] Yadav AK, Chandel SS. Articial Neural Network based Prediction of Solar
Radiation for Indian Stations. Int J Comput Appl 2012;50(09758887):50.
[52] Yacef R, Benghanem M, Mellit A. Prediction of daily global solar irradiation
data using Bayesian neural network: a comparative study. Renew Energy
2012;48:14654.
[53] Fazel Ziaei Asl S, Karami A, Ashari G. Daily Global Solar Radiation Modelling
Using Multi-Layer Perceptron (MLP) Neural Networks. World Acad Sci Eng
Technol 2011:55.
[54] Ibeh GF, Agbo GA. Estimation of mean monthly global solar radiation for
Warri- Nigeria(Using Angstrom and MLP ANN model). Adv Appl Sci Res
2012;3:128.
[55] Azeez MAA. Articial neural network estimation of global solar radiation
using meteorological parameters in Gusau Nigeria. Arch Appl Sci Res
2011;3:58695.
[56] Rahimikhoob A. Estimating global solar radiation using articial neural
network and air temperature data in a semi-arid environment. Renew
Energy 2010;35:21315.
[57] Koca A, Oztop H, Varol Y, Ozmen Koca G. Estimation of solar radiation using
articial neural networks with different input parameters for Mediterranean
region of Anatolia in Turkey. Expert Syst Appl 2011;38:875662.
[58] Krishnaiah T, Srinivasa Rao S, Madhumurthy K. Neural Network Approach for
Modelling Global Solar Radiation. J Appl Sci Res 2007;3:110511.
[59] Rehman S, Mohandes M. Splitting global solar radiation into diffuse and
direct normal fractions using articial neural networks. Energy Sources
2012;34:132636.
[60] Senkal O, Tuncay K. Estimation of solar radiation over Turkey using articial
neural network and satellite data. Appl Energy 2009;86:12228.
[61] enkal O, Kuleli T. Estimation of solar radiation over Turkey using articial
neural network and satellite data. Appl Energy 2009;86:12228.
[62] enkal O. Modeling of solar radiation using remote sensing and articial
neural networkinTurkey. Energy 2010;35:4795801.
[63] Vassiliki HM, Dimitrios HM. Solar radiation: Cloudiness forecasting using a
soft Computing approach. Artif Intell Res 2013;2:6980.
[64] Elminir HK, Azzam YA, Younes FI. Prediction of hourly and daily diffuse
fraction using neural network, compared to linear regression models. Energy
2007;32:151323.
[65] Fei W, Zengqiang M, Hongshan Z. Short-Term Solar Irradiance Forecasting
Model Based on Articial Neural Network Using Statistical Feature Parameters. Energies 2012;5:135570.
[66] Ozgur S. Prediction of Hourly Solar Radiation in Six Provinces in Turkey by
Articial Neural Networks. J. Energy Eng 2012:194204.
[67] Santamouris. Modelling the Global Solar Radiation on the Earths Surface
Using Atmospheric Deterministic and Intelligent Data-Driven Techniques. J
Climate 1999:12.
[68] Rahoma WA. Application of Neuro-Fuzzy Techniques for Solar Radiation. J
Comput Sci 2011;7:160511.
[69] Mohammadi, et al. Potential of adaptive neuro-fuzzy system for prediction
of daily global solar radiation by day of the year. Energy Convers Manag
2015;93:40613.
[70] Iqdour R, Zeroual A. A rule based fuzzy model for the prediction of solar
radiation. Revue des Energies Renouvelables 2006;9:11320.
[71] Mohanty S. ANFIS based prediction of monthly average global solar radiation
over Bhubaneswar (state of Odisha).In; International journal of. Ethics Eng
Manag Educ 2014;1:5 ISSN 2348-4748.
[72] Sumithira TR, Nirmal A, Ramesh R. An adaptive neuro-fuzzy inference system (ANFIS) based Prediction of Solar Radiation. J Appl Sci Res 2012;8:346
51.
[73] Mellit A, Kalogirou SA, Shaari S, Salhi H. Methodology for predicting
sequences of mean monthly clearness index and daily solar radiation data in
remote areas: Application for sizing a stand-alone PV system. Renew Energy
2008;33:157090.
[74] Mohanty S., Patra P.K., Sahoo S.S..Comparision and prediction of Monthly
Average Solar Radiation data Using Soft computing approach for Eastern
India.
[75] Celik AN, Muneer T. Neural network based method for conversion of solar
radiation data. Energy Convers Manag 2013;67:11724.

795

[76] Chatterjee A, Keyhani A. Neural network estimation of micro grid maximum


solar power. IEEE Trans Smart Grid 2012;3:18606.
[77] Rizwan M, Jamil M, Kothari DP. Generalized Neural Network Approach for
Global Solar Energy Estimation in India. IEEE Trans Sustain Energy
2012;3:57684.
[78] Yacef R, Benghanem M, Mellit A. Prediction of daily global solar irradiation
data using Bayesian neural network. Comp Study Renew Energy
2012;48:14654.
[79] Rami El-Hajj M, Mahmoud S, Ali Massoud H. Predicting Global Solar Radiation Using Recurrent Neural Networks and Climatological Parameters World
Academy of Science, Engineering and Technology,International Journal of
Mathematical, Computational. Phys Quantum Eng 2014;8:8.
[80] Benghanem M, Mellit A. Radial Basis Function Network-based prediction of
global solar radiation data: application for sizing of a stand-alone photovoltaic system at Al-Madinah, Saudi Arabia. Energy 2010;35:375162.
[81] Naderian M, Barati H, Golashahi M, Farshidi R. Application of Fully Recurrent
(FRNN) and Radial Basis Function (RBFNN) Neural Networks for Simulating
Solar Radiation. Bull Environ Pharmacol Life Sci 2014;3:1329.
[82] Zeng Jianwu. Short-Term Solar Power Prediction Using an RBF Neural Network IEEE Power and Energy Society General Meeting. 2011. p. 18.
[83] Mishra A, Kaushika ND, Zhang G, Zhou J. Articial neural network model for
the estimation of direct solar radiation in the Indian zone. Int J Sustain
Energy 2008;27(3):95103.
[84] Quaiyum S, Rahman S, Rahman S. Application of Articial Neural Network in
Forecasting Solar Irradiance and Sizing of Photovoltaic Cell for Standalone
Systems in Bangladesh. Int J Comput Appl 2011;32(09758887):516.
[85] Lalwani M, Kothari DP, Singh M. Size optimization of stand-alone photovoltaic system under local weather conditions in India. Int J Appl Eng Res
2011;1:95161.
[86] Khatib T, Mohamed A, Sopian K, Mahmoud M. A New Approach for Optimal
Sizing of Standalone Photovoltaic Systems. Int J Photo Energy 2012;2012:17.
[87] Saberian A, Hizam H, Radzi MAM, Ab Kadir MZA, Mirzaei M. Modelling and
Prediction of Photovoltaic Power Output Using Articial Neural Networks.
Int J Photo Energy 2014:110.
[88] Khaled Bataineh K, Dalalah D. Optimal Conguration for Design of StandAlone PV System. Smart Grid Renew Energy 2012;3:13947.
[89] M. A. Sizing of a stand-alone photovoltaic system based on neural networks
and genetic algorithms: application for remote areas. J Electr Electron Eng
2007;7:45969.
[90] Guda HA, Aliyu U O. Design of a Stand-Alone Photovoltaic System for a
Residence in Bauchi. Int J Eng Technol 2015;5:3444.
[91] Sanusi YK, Abisoye SG, Awodugba AO. Application Of Neural Networks for
Predicting the Optimal Sizing Parameters Of Stand-Alone Photovoltaic Systems. SOP Trans Appl Phys 2014;1:126.
[92] H.A.A.G.A. Mellit A. Benghanem M. Prediction and modelling signals from
the monitoring of stand-alone pv sys- tems using an adaptive neural network model, in Proceedings of the 5th ISES European solar conference,
(Germany) 2004; 224230.
[93] Balouktsis A, Karapantsios T, Antoniadis A, D., Paschaloudis A, Bilalis N.
Sizing stand-alone photovoltaic systems. Int J Photo energy 2006:2006.
[94] Qi C.,Ming Z..Photovoltaic Module Simulink Model for a Stand-alone PV
System, International Conference on Applied Physics and Industrial Engineering 2012;24:94-100.
[95] Mathew, et al. Optimal Sizing Procedure for Standalone PV System for University
Located Near Western Ghats in India. Int J Eng Adv Technol 2014;3(4):2239.
[96] Suchitra, et al. Optimization of a PV-Diesel hybrid Stand-Alone System using
Multi-Objective Genetic Algorithm. Int J Emerg Res Manag Technol 2013;2(5):68
76.
[97] Sharma V., Chandel S.S. Performance analysis of a 190 kWp grid interactive
solar photovoltaic power plant in India Energy, Vol. 55 (15); 2013. p. 47685.
[98] Hematian A, Ajabshirchi Y, Bakhtiari A. Experiimental analysis of at plate
solar air collector efciency. Indian J Sci Technol 2012;5:31837.
[99] Ayompe LM, Duffy A. Analysis of the thermal performance of solar water
heating system with at plate collectors in a temperate climate. Appl Thermal Eng 2013;58:44754.
[100] Cruz-peragon F, Palomar JM, Casanova PJ, Dorado MP, Manzano-Agugliaro F.
Characterization of solar at plate collector. Renew Sustain Energy Rev
2012;16:170920.
[101] I Farkas, Geczy-vigP P. Neural network modelling of at-plate solar Collectors. Comput Electron Agric 2003;40:78102.
[102] Sozen A, Menlik M, Unvar S. Determination of efciency of at-plate solar
collectors using neural network approach. Expert Syst Appl 2008;35:15339.
[103] Tariq O, Salah H, Moussa C, Abdi H. Purpose of neuronal method modelling of
solar collector. Int J Energy Environ 2012;3:918.
[104] Farahat S, Sarhaddi F, Ajam H. Exergetic optimization of at plate solar
Collectors. Renew Energy 2009;34:116974.
[105] Kalogirou SA, Panteliu S, Dentsoras A. Modeling of solar domestic water heating
systems Using articial neural networks. Sol Energy 1999;65:33542.
[106] Kalogirou SA. Prediction of at-plate collector performance parameters using
Articial neural Networks. Sol Energy 2006;80:24859.
[107] Farzad J, Masoud M, Maryam K, Ahmad R. Performance prediction of at-Plate
solar collectors using MLP and ANFIS. J Basic Appl Sci Res 2013;3:196200.
[108] Karim, M.A. and HAwlader M.N.A..Development of solar air collectors for
drying Applications.Energy Conversions and Management45:329-344.
[109] Yeh HM, Lin TT. Efciency improvement of at-plate solar air heaters. Energy.
1996;21:43543.

796

S. Mohanty et al. / Renewable and Sustainable Energy Reviews 56 (2016) 778796

[110] El-Sawi AM, Wi AS, Younan MY, Elsayed EA, Basily BB. Application of
folded sheet metal in at bed solar air collectors. Appl Thermal Eng 2010;30:
86471.
[111] Zelzouli K, Guizani A, Sebai R, Kerkeni C. Solar Thermal Systems Performances
versus Flat Plate Solar Collectors Connected in Series. Engineering
2012;4:88193.

[112] Dr. Saad T. Hamidi Mohamaad A. Fayath, Prediction of thermal characteristics


for solar water Heater pp.18-31.
[113] Cuadros F, Lopez-Rodrguez F, Segador C, Marcos A. A simple procedure to
size active solar heating schemes for low-energy building design. Energy
Build 2007;39:96104.

Potrebbero piacerti anche