Abstract:
Solar radiation data play an important role in sola
r energy relevant researches. These
data are not available for some locations due to th
e absence of the meteorological stations.
Therefore, solar radiation data have to be predicte
d by using solar radiation estimation
models. This study presents an integrated Artificia
l Neural Network (ANN) approach for
estimating solar radiation potential over Iran base
d on geographical and meteorological
data. For this aim, the measured data of 31 station
s spread over Iran were used to train
Multi-Layer Perceptron (MLP) neural networks with d
ifferent input variables, and solar
radiation was the output. The accuracy of the model
s was evaluated using the statistical
indicators of Mean Absolute Percentage Error (MAPE)
, Root Mean Square Error
(RMSE), and Correlation Coefficient (R); hence, the
best model in each category was
identified. The Stepwise Multi NonLinear Regression
(MNLR) method was used to
determine the most suitable input variables. The re
sults obtained from the ANN models
were compared with the measured data. The
MAPE
and
RMSE
were found to be 2.98%
and 0.0224, respectively. The obtained R value was
about 99.85% for the testing data set.
The results testify to the generalization capabilit
y of the ANN model and its excellent
ability to predict solar radiation in Iran.
Keywords
:
ANN, Meteorological data, Multi non-linear regress
ion, Solar radiation.
