Soft computing methods for predicting daily global solar radiation

Abstract

In this study, four artificial intelligence (AI) technologies, including extreme learning machine (ELM), a novel hybrid model which combines the least squares support vector machine with particle swarm optimization algorithm (PSO-LSSVM), back propagation neural network (BPNN) model and generalized regression neural network (GRNN) model, were developed for estimating daily global solar radiation (GSR) for Zhengzhou of China. Six meteorological variables are selected as the evaluating indices, while the daily GSR is the output. Three statistical indices were employed to measure and evaluate the prediction performance of these four AI models. By comparison of the results of these four AI models and an empirical model, it is concluded that the developed hybrid model which combines the least squares support vector machine with particle swarm optimization algorithm outperforms the other four models and it can be used for forecasting the daily GSR with high accuracy.