A predictive evaluation of global solar radiation using recurrent neural models and weather data

This paper presents predictive models based on dynamic recurrent neural networks DRNNs with short term delay units to predict daily solar radiation intensity. The proposed approach aims to evaluate the daily global solar radiation using simple recurrent neural networks (SRNNs) with meteorological data. First, we present a reference model based on a feed-forward multilayer perceptron (MLP), then we present several recurrent models of the same structure but with various number of delay units that memorize the outcomes of the recurrent model to be used in subsequent iterations. The obtained comparative results showed advantage of DRNNs over simple MLPs when we deal with time series meteorological records. The performance of the proposed approach has been evaluated using statistical analysis.