Predicting solar radiation with RBF neural networks

Abstract

In this paper radial basis function neural networks are applied to the prediction of global solar radiation. The networks are employed as one-step-ahead predictors of the solar radiation time series and iterated over time to obtain longer term predictions. Several models are compared varying the input dimension, the network size and the time series sampling rate. An empiric rule is proposed for network input selection. All networks are trained using one data set and evaluated for prediction performance on unseen data. Predictor performance is assessed taking root mean square measures of the error over the prediction horizon. The aim of this work is to select a model to be used in a climate simulator for an hydroponic greenhouse.