Unveiling the backbone of the renewable energy forecasting process: Exploring direct and indirect methods and their applications

A myriad of techniques regarding renewable energy forecasting have been proposed in recent literature, commonly classified as physical, statistical, machine learning based or a hybrid form thereof. The renewable energy forecasting process is however elaborate and consists of multiple stages, where different approaches from these four categories apply variably, complicating a holistic classification of the process. This paper resolves this by utilizing the fundamental difference between direct and indirect forecasting in terms of model complexity, data availability, spatial and time horizons as the backbone to structure this intricate forecasting process. As such, a significant step towards a generalized framework for renewable energy forecasting is presented. Additionally, a most promising recommendation emerges: leveraging physics-based knowledge from indirect models to enhance training of direct methods.