Presentations

Abstract

Accurate forecasts of the power output of wind farms are essential for efficient integration into electricity markets and power systems. The state of the art in wind power forecasting is particularly rich, with spatiotemporal models among the mainstream approaches for short-term horizons up to 6 hours. Using data from geographically distributed wind farms, in an area that can span several tens of kilometers, has been shown to improve forecast accuracy by up to 20\% for such horizons. However, data from neighboring wind farms is rarely shared, as they are typically owned by competing stakeholders concerned about protecting their competitive advantage.  To address this challenge, we investigate confidentiality/privacy-preserving collaborative forecasting methods that enable operators to benefit from shared data without revealing sensitive information. Building on our previous works with privacy-preserving tree-based and Kolmogorov-Arnold Networks (KAN) models, we conduct an in-depth comparison of these two privacy-preserving paradigms using the GEFCOM2014 dataset. While both approaches provide collaborative privacy-preserving learning opportunities, they offer different strengths: tree-based methods inherently preserve privacy and handle missing data robustly, whereas KAN models emphasize interpretability, allowing insights into which farms contribute the most to predictive performance.  In this contribution, we compare both methods using fundamental operational criteria: accuracy, resilience, interpretability, and computational efficiency. The goal is to provide a practical guide for applying privacy-preserving (PP) forecasting in wind power generation. Our results indicate that both models achieve comparable accuracy, but with different trade-offs that make them suitable for different collaborative settings. These findings highlight that privacy-preserving collaboration can unlock the benefits of spatio-temporal forecasting, while safeguarding commercial interests, thereby supporting a broader integration of wind power in the energy mix.