Estimation of global solar radiation using ANN over Turkey

The main objective of the present study is to develop an artificial neural network (ANN) model based on multi-nonlinear regression (MNLR) method for estimating the monthly mean daily sum global solar radiation at any place of Turkey. For this purpose, the meteorological data of 31 stations spread over Turkey along the years 2000–2006 were used as training (27 stations) and testing (4 stations) data. Firstly, all independent variables (latitude, longitude, altitude, month, monthly minimum atmospheric temperature, maximum atmospheric temperature, mean atmospheric temperature, soil temperature, relative humidity, wind speed, rainfall, atmospheric pressure, vapor pressure, cloudiness and sunshine duration) were added to the Enter regression model. Then, the Stepwise MNLR method was applied to determine the most suitable independent (input) variables. With the use of these input variables, the results obtained by the ANN model were compared with the actual data, and error values were found within acceptable limits. The mean absolute percentage error (MAPE) was found to be 5.34% and correlation coefficient (R) value was obtained to be about 0.9936 for the testing data set.

► ANN model was developed for estimating the monthly mean daily sum global solar radiation. ► Monthly meteorological variables during years 2000–2006 for 31 cities in Turkey were used. ► The ANN model based on the Stepwise regression analysis showed a better estimation. ► The model yields more accurate estimating with lower error and higher correlation values.