2025/02/17 by Gabriel Rodrigo de Souza Dantas, Dantas, Gabriel, Jethro Browell +1 · 1 citation
Computer Science · Engineering · #Applications (stat.AP) #Artificial intelligence #Computational Physics and Python Applications #Computer science #Electric power system #Electrical engineering #Energy Load and Power Forecasting #Engineering #Environmental science #FOS: Computer and information sciences #Geography #Integrated Energy Systems Optimization #Meteorology #Physics #Power (physics) #Probabilistic forecasting #Probabilistic logic #Term (time) #Wind power #Wind power forecasting
paper · open access · doi:10.48550/arxiv.2502.11960
openalex publication_date 2025/02/17 · openalex created_date 2025/10/10 · openalex updated_date 2026/08/05
This paper presents a method for probabilistic wind power forecasting that quantifies and integrates uncertainties from weather forecasts and weather-to-power conversion. By addressing both uncertainty sources, the method achieves state-of-the-art results for lead times of 6 to 162 hours, eliminating the need for separate models for short- and mid-term forecasting. It also improves short-term forecasts during high weather uncertainty periods, which methods based on deterministic weather forecasts fail to capture. The study reveals that weather-to-power uncertainty is more significant for short-term forecasts, while weather forecast uncertainty dominates mid-term forecasts, with the transition point varying between wind farms. Offshore farms typically see this shift at shorter lead times than onshore ones. The findings are supported by an extensive, reproducible case study comprising 73 wind farms in Great Britain over five years.