Hybrid short-term wind power forecasting model using theoretical power curves and temporal fusion transformers

Wind energy penetration has radically increased in the last decade constituting one of the main renewable energy resources of the energy transition. However, its intermittent nature necessitate the development of accurate Wind Power Forecasting (WPF), essential in several applications, including grid reliability and cost minimization. Despite advancements in this sector, a notable gap remains in integrating physics-informed (PI) approaches with transformer-based architectures. This study proposes a novel hybrid WPF model that integrates the Temporal Fusion Transformer (TFT) with theoretical power curve modeling techniques. The integration of manufacturer specifications, Numerical Weather Prediction (NWP) data and PI equations ensures robust and reliable input to the TFT. The proposed approach is validated using real-world datasets from two distinct wind turbines operating in different geographical locations. To comprehensively evaluate forecasting performance, a modified Forecast Skill Index (FSI) is introduced, FSI-WPF, benchmarking accuracy against the theoretical power curve rather than a persistence model. Experimental results demonstrate that the proposed method significantly outperforms conventional forecasting models, achieving up to a 60% reduction in Root Mean Squared Error (RMSE) and an R2 score of up to 99.47%. This study advances the integration of PI modeling with deep learning architectures, paving the way for more accurate and reliable WPF.