Evaluation of the global solar irradiation from the artificial neural network technique

Abstract

In this study, many experiments were carried out to assess the influence of the some input parameters on performance of the multilayer perceptron techniques, which is architectures one of the artificial neural network (ANN). To estimate the daily global solar irradiation (GSI) on horizontal surface, we have developed eight models by using four weather input parameters collected from a radiometric station installed at Ghardaia site (southern of Algeria), such as sunshine duration, daily mean temperature, daily relative humidity, and Solar declination. In order to select the best configuration among the chooses combinations which provides a good accuracy, three statistical formulas (or statistical indicators) have been evaluated, such as the normalized root mean square errors, (nRMSE), normalized mean bias error (nMBE, and coefficient of determination R²). We noted that the ANN model provides best performance compared to the developed empirical models. results of the nMBE, nRMSE, and R² are about 0.048%, 4.422%, and 98.00%, for ANN and 0.0772%, 5.5381%, and 97.22% for the quadratic model. Moreover, it was proved that the sunshine duration is important parameter, for predicting the GSI.