Forecasting mesoscale distribution of surface solar irradiation using a proposed hybrid approach combining satellite remote sensing and time series models

Abstract

A new hybrid forecasting tool is developed in this study which makes use of satellite remote sensing data of surface solar irradiation coupled to a Double Exponential Smoothing time series model. The prediction capabilities of the Double Exponential Smoothing model are reported to be higher than the ARMA and NAR-Neural Network. The mean absolute percentage error of this hybrid system is revealed to be the lowest (4.89%) on average for 5 consecutive days-ahead forecasts over the years 2013–2015, with the smallest standard deviation reported throughout the year, characteristic of a highly stable and robust model (3.83 W/m2). Exploring the performance of the model for the best and worst case scenarios reveal that high prediction accuracies on both spatial and temporal scales are achievable with strong positive linear correlations of the orders of 0.928 and 0.894 respectively, averaged over the 5 days forecasts. The performance of the hybrid system is found to be higher as compared with benchmark accuracy reached by several other models employed in literature. Finally, the use of the hybrid forecasting tool developed in providing energy and grid management facilities for the island of Mauritius is also presented.