This project investigates the use of modern machine learning (ML) models for improving Day-Ahead (DA) wind power forecasting in Denmark’s DK2 bidding zone. Using open source weather data several meteorological stations and wind generation data from the "European Network of Transmission System Operators for Electricity" (ENTSO-E), several ML models were trained, validated, and tested. The data pre-processing included grouping by capacity and positions of grid connected turbines, interpolation of missing values and creating weighted values that represent weather conditions in the DK2 region. Results show that the "ensemble" models significantly outperformed both a Naive benchmark and the operational forecasts published by ENTSO-E, with the best performing model achieving a 84% better MAPE score compared to the Naive model, and 25.3% better compared to ENTSO-E. The project demonstrates the usefulness of ML models to enhance forecast accuracy and suggests that Transmission System Operators (TSOs) like Energinet in Denmark could benefit from adopting more advanced forecasting methods. The findings also highlight the importance of feature engineering and careful data processing. Future research should focus on improving weather input precision and exploring turbine-level forecasting to further bridge the gap between research and operational application.
Building similarity graph...
Analyzing shared references across papers
Loading...
Dante Fridlizius Lindberg
Ludvig Nyqvist
Building similarity graph...
Analyzing shared references across papers
Loading...
Lindberg et al. (Wed,) studied this question.