Wind energy penetration has increased in the last decade, constituting one of the main renewable energy resources of the energy transition. However, its intermittent nature necessitates the development of accurate Wind Power Forecasting (WPF), essential to ensure grid reliability and cost minimisation. This study has unveiled a forecasting model that blends physics with artificial intelligence to improve WPF accuracy.
Conducted within the framework of the CRETE VALLEY project, the study titled Hybrid short-term wind power forecasting model using theoretical power curves and temporal fusion transformers introduces a new model that combines a theoretical power curve model, which reflects how turbines are expected to perform under given wind conditions, with a cutting-edge Temporal Fusion Transformer (TFT) deep learning architecture. By merging physics-based insights with advanced machine learning, the team achieved forecasts that are both more accurate and more reliable.
Tests on real-world data from wind turbines in Turkey and Ireland showed that the hybrid model reduced forecasting errors by up to 60% compared to baseline models, achieving near-perfect accuracy with an R² of 99.5% in some cases.
The study also introduced a new evaluation metric, the Forecast Skill Index for Wind Power Forecasting (FSI-WPF), which benchmarks forecasts against the physical power curve rather than against simple persistence models. This provides a more realistic measure of forecasting “skill.”
The findings highlight the promise of hybrid physics–AI approaches in renewable energy forecasting, offering system operators more dependable tools for integrating wind energy into the grid. Future research will extend the model to whole wind farms and refine the treatment of meteorological uncertainties.
Cover photo by David Bruwer on Unsplash
