
radiation). The RMSE is 6.74%. The model performs well in cloudy
and clear sky condition.
Kisi [20] investigated fuzzy genetic for solar radiation modeling
of seven cities in Turkey. The authors selected latitude, longitude,
altitude, month as inputs and RMSE is 6.29 MJ/m
2
. It is shown that
the fuzzy genetic method gives better results than the ANN and
ANFIS (adaptive neuro fuzzy inference system) model.
Mostafavi et al. [21] used hybrid genetic programming (GP) and
simulated annealing (SA) called as GP/SA for new formulation of
solar radiation in terms of sunshine, total precipitation, mean
relative humidity, maximum and minimum temperature. The
MAPE varies from 0.103 to 0.214 for Tehran and Kerman cities in
Iran. It is suggested that maximum and minimum temperature are
most influencing variable in prediction. The different ANN inputs
parameters and prediction accuracy are summarized in Table 1.
Based on the literature survey it is found that prediction
accuracy of ANN models get changed with geographical and
meteorological variables as input parameters. For selection of
Table 1
Input variables used in ANN based prediction of solar radiation.
Reference Models and
training algorithm
Input variables to ANN Model ANN Model prediction accuracy Location
Linares-
Rodríguez
et al. [22]
MLP Latitude, longitude, day of the year, daily clear sky global
radiation, cloud cover, total column ozone and water
vapor
RMSE 13.52% for training stations and 14.20%
for testing stations
Spain
Koca et al.
[23]
MLP Latitude, longitude, altitude, months, average
temperature, average cloudiness, average wind velocity
and sunshine duration
Maximum RMSE is 6.9% Seven cities in
Mediterranean
region of
Anatolia, Turkey
Khatib et al.
[24]
MLP Latitude, longitude, day
0
s number and sunshine ratio The MAPE in estimating global and diffuse
radiation are 7.96%, 9.8% respectively
Malaysia
Khatib et al.
[25]
Linear, nonlinear, fuzzy
logic and ANN models
Latitude, longitude, day number and sunshine ratio MAPE of 5.38% (global radiation), 1.53% (diffuse
radiation)
Five sites in
Malaysia
Elminir et al.
[26]
Multilayer feed forward
network
Wind direction, wind velocity, ambient temperature,
relative humidity, cloudiness and water vapor
RMSE are 5.02%, 7.46% and 3.97% for infrared,
ultraviolet and global solar radiation
respectively
Helwan, Aswan
monitoring
stations
Tymvios et al.
[27]
ANN and Angström Theoretical daily sunshine duration, measured daily
sunshine duration, month, daily maximum temperature,
monthly mean value of theoretical sunshine duration,
monthly mean value of measured sunshine duration,
extraterrestrial radiation, monthly mean value of daily
global radiation, total global radiation, daily
extraterrestrial radiation.
The maximum RMSE of ANN model is 10.15 and
in Ångström Model, RMSE is 13.36, showing
ANN model give better results than Ångström
Model
Athalassa in
Cyprus
Alam et al.
[28]
MLP Latitude, longitude, altitude, month of the year, mean
duration of sunshine per hour, rainfall ratio, relative
humidity
RMSE varies from 1.65 to 2.79% India
Jiang [29] Feed-forward back
propagation neural
network and Empirical
Model
Monthly mean daily clearness index, sunshine percentage RMSE in empirical models are 0.783, 0.781
whereas in ANN model is 0.746, showing
accurate estimation of ANN than empirical
models.
China
Mubiru and
Banda
[30]
Feed forward back-
propagation ANN;
Levenberg–Marquardt
(LM)
Annual average of sunshine hours, cloud cover, relative
humidity, rainfall, latitude, longitude and altitude
MAPE, R
2
are 0.3, 97.4% respectively and better
results obtained by ANN than sunshine based
conventional model
Uganda
Şenkal and
Kuleli
[31]
ANN and Physical
model
Latitude, longitude, altitude, month, mean diffuse
radiation and mean beam radiation
RMSE values using the MLP and the physical
model are 54 W/m
2
and 64 W/m
2
(training
cities); 91 W/m
2
and 125 W/m
2
(testing cities),
respectively
Turkey
Jiang Y [32] ANN model and
empirical regression
model
Latitude, altitude and mean sunshine R
2
¼0.97, RMSE¼ 1.4 MJ/m
2
China
Benghanem
et al. [33]
ANN model Different combination of air temperature, relative
humidity, sunshine duration and the day of year
R value of 97.65% is obtained using sunshine
duration and air temperature as inputs to the
ANN model
Al-Madinah
(Saudi Arabia)
Fadare [34] ANN Latitude, longitude, altitude, month, mean sunshine
duration, mean temperature, and relative humidity
The R
2
for training and testing cities are higher
than 90%
195 cities in
Nigeria
Azadeh et al.
[35]
Integrated ANN-MLP Location, month, mean value of maximum temperature,
minimum temperature, relative humidity, vapor pressure,
total precipitation, wind speed and sunshine hours
MAPE is 0.03 and ANN models give better
results than Angstrom model
Iran
Sözen and
Arcak-
lioğlu
[36]
ANN Geographical coordinates, mean sunshine duration, mean
temperature and month
MAPE is less than 3.832%, Turkey
Rahimikhoob
[37]
ANN Maximum and minimum air temperature, extraterrestrial
radiation
RMSE 2.534 MJ/m
2
/day, R
2
88.9% better than
Hargreaves and Samani [38] equation
Ahwaz (Iran)
Hasni et al.
[39]
ANN Air temperature, relative humidity The MAPE, R
2
are 2.9971%, 99.99% south-western
region of Algeria
Yildiz et al.
[40]
Two ANN models Latitude, longitude, altitude, month and meteorological
land surface temperature to first ANN Model.
The R
2
for first, second ANN are 80.41%, 82.37%
respectively for testing station, showing better
estimation of second model than first model
Turkey
latitude, longitude, altitude, month and satellite land
surface temperature to second ANN Model.
Rumbayan
et al. [41]
MLP Month, latitude, wind speed, precipitation, sunshine
duration, humidity and temperature
MAPE is found to be 3.4% with 9 neurons in
hidden layer
Indonesia
A.K. Yadav et al. / Renewable and Sustainable Energy Reviews 31 (2014) 509–519 511