A hybrid ARIMA–ANN method to forecast daily global solar radiation in three different cities in Morocco

The aim of this paper is to propose and introduce an innovative hybrid model to forecasted the daily global solar radiation (DGSR) in three different cities in Morocco. The DGSR depends on several meteorological parameters and it is difficult to determine its behaviour. The used data have been collected from three different cities with different meteorological data. In this context, we compared 3-year forecasting performance of two methods widely used in forecasting research with a hybrid method. The first model uses the artificial neural network (ANN) and the second one, known as autoregressive integrated moving average (ARIMA), is based on both statistic regression and time series analysis. The third model is based on the combination between ANN and ARIMA methodologies. The three methods have been trained within the same experimental data, thereby allowing a much needed homogeneous correlation that is completely missing in the existing research. The forecasted DGSR generated by the hybrid ARIMA–ANN approach shows a high correlation with experimental results and a relatively small error rate. The forecast of DGSR obtained by hybrid ARIMA–ANN method is compared with the single ANN and ARIMA methods and experimental data. In order to be more accurate, an empirical error is measured and contrasted to check the significance of the expected outcomes and the accuracy of the model. In order to assure results reliability, statistical errors are computed and compared to verify the validity of the forecasted results and the performance of the model. The compared results through statistical error are very accurate compared with a single model in term of R², MBE, RMSE, NRMSE, MAPE, TS and Sd. Generally, the obtained value using the hybrid model is the most adequate which can adjust the experimental data with satisfactory precision. In the end, the ARIMA–ANN model can be used in forecasting photovoltaic power or temperature.