Daily global solar radiation forecasting over a desert area using NAR neural networks comparison with conventional methods

This paper presents a solar radiation forecasting method using nonlinear autoregressive neural networks (NAR). NAR predicts a clearness index that is used to forecast global solar radiations. The NAR model is based on the feed forward multilayer perception model with two inputs and one output. Data of three years (2012-2014) of global solar radiation time-series for Ghardaïa site (desert area), south Algeria have been used to develop the model. A comparison with Box-Jenkins (ARMA) method was done, and the proposed approach was found to be more efficient and accurate. The forecasted values are compared with the measured data and the accuracy of the models is judged based on the statistical analysis such as root mean square error (RMSE) and his normalized value (nRMSE), mean bias error (MBE) and his normalized value (nMBE) and the mean percentage error (MPE). The obtained results showed an improvement of the NAR model over ARMA in term of mean absolute error (MPE) of 23.89% and a decrease in RMSE values of about 15.50% while the coefficient correlation was found to be 0.91.