SOFT COMPUTING BASED ON AR-ANFIS AND AR-ANN FOR MODELING AND PREDICTING HALF HOUR GLOBAL SOLAR RADIATION

Abstract

In this paper, we are interested in the exploration of the possibility to develop models for predicting the half hour global solar radiation. For this purpose, we applied a method which combines the artificial neural networks (ANN) based on the multilayer perceptron (MLP) and the adaptive neuro fuzzy inference system (ANFIS) with an autoregressive process (AR). In this basis, two models called AR-ANN and AR-ANFIS are developed to analyze and predict half hour global solar radiation time series measured during three years in the area of Agdal at Marrakesh, Morocco. To evaluate the performance of the developed models, a comparison with real measurements is achieved. In term of some statistical criteria, the developed models are able to generate halfhour global solar radiation time series with a variance accounting around 97% and an MSE around 0.25%. Consequently they can successfully be an excellent tool for the management of daily global solar radiation in case of lack of measurements in a given region having similar climate as Marrakesh. Deep comparison shows that AR-ANFIS model is slightly more accurate than the AR-ANN model.