Artificial neural network based prediction of monthly global solar radiation in Indian stations

Abstract

The prediction of solar radiation (SR) plays a vital role for many applications in renewable energy research. Several parametric and non-parametric machine learning algorithms are being used to predict the SR. In this present study it is proposed to predict monthly global solar radiation (GSR) on a horizontal surface, based on meteorological and geographic variables using different artificial neural network (ANN). ANN based model is applied to 23 Indian locations having different climatic conditions to find most influencing input parameters for SR prediction. The input parameters identified are latitude (Lat), longitude (Long), maximum temperature (T max ), minimum temperature (T min ), altitude (Alt), sunshine hours (SH), extraterrestrial radiation (ER) and clearness index (CI) for different cities of India. In order to predict the accurate SR in ANN, three neural networks such as Multilayer Perceptron (MLP), Radial Basis Function (RBF) and Generalized Regression Neural Network (GRNN) have been selected. Each network consists of five models. Performance of the models has been evaluated using Mean Absolute Percentage Error (MAPE), Mean Square Error (MSE) and Root Mean Square Error (RMSE). The most accurate result for the networks RBF, MLP and GRNN are 7.82%, 8.89% and 8.81% in that RBF obtain highly accurate result among others. MAPE yielded promising results and it has been observed that latitude, longitude, altitude, minimum temperature, maximum temperature are the most relevant input parameters to obtain SR.