太阳能和风能预测综述：从单站点到多站点范式

Abstract The energy sector is undergoing a radical transformation towards an electricity generation system composed mainly of renewable energy sources. Although they offer several advantages compared to fossil fuels (such as scarcity and import-dependency), their stochastic nature makes them unreliable without an adequate storage system. Since electrical networks are characterized by complex dynamics, numerous methods for predicting renewable electricity production have been developed over the years. Among them, machine learning and deep learning methods in this field can be considered successful tools. In this review, we offer an overview of the methodologies applied to the task of renewable energy source forecasting, focusing on solar and wind. We classified methods depending on the number of sites and spatio-temporal information involved in the prediction. Recent research has demonstrated how the processing of simultaneous information coming from multiple plants allows the predictive system to take advantage of both temporal and spatial knowledge of the plants generating the related time series. Moreover, we have analyzed in detail the datasets employed in these experiments, to offer a clear and unified view of the experimental setups and the difficulty in producing a benchmark to compare methods. The purpose of this review is to offer the reader an updated view of the most modern renewable energy forecasting systems by comparing methodologies and approaches used in state-of-the-art research and providing a critical analysis of them.