ANN-based forecasting of solar irradiation under data clustering: an approach for improved estimation of PV power production

The performance of PV power-related variable forecasting is gaining importance due to the growing needs for PV plants integration in electrical grids and to the pivotal role of forecasting in new energy management systems. Based on such a consideration, this paper proposes a method using non-supervised learning combined with a partially recurrent ANN to obtain an increased accuracy in day-ahead solar irradiation forecasting. The proposed hybrid method is applied to two case studies related to locations with different climatic characteristics and compared to different other methods, including the non-clustering-based solution. The proposed method allows reducing the forecasting normalized root mean squared error up to about 10% and significantly outperforms a baseline method.