Prediction of Daily Global Solar Radiation Using Neural Networks With Improved Gain Factors and RBF Networks

Abstract

Solar radiation data play an important role in solar energy study. These data are not available for location of interest due to lack of meteorological stations. Therefore, it is all important to forecast solar radiation for a site using different climatic variables. Methods in practice like neural networks with back propagation algorithm as training function, firefly algorithms and time series models suffer with slow convergence rates, high computational times and lack of recognizing the non-linear series respectively. Hence, to overcome these drawbacks, the present work proposes three novel approaches for forecasting the daily global solar radiation (DGSR) of 10 Indian cities. Simple Artificial Neural Network (ANN), ANN with forward unity gain and ANN with Regression Networks are considered in estimating the DGSR. The data set consisting of the minimum temperature , maximum temperature , average temperature , wind speed , relative humidity , precipitation , extraterrestrial radiation and sunshine hours are considered as inputs to the proposed approach. Statistical indicators like coefficient of determination , root mean square error , mean bias error and mean absolute percentage error are evaluated to determine the efficiencies of the proposed approaches. Results show that forecasting of DGSR is superior as compared to other approaches.