A Data-driven Generative Framework for Wind Turbine Power Prediction across Unseen Wind Sectors

This paper presents a two-stage generative framework for wind turbine power prediction (WTPP) under long-term uncertain wind conditions. By modeling at the sector level and training on five years of SCADA data, the method predicts sectoral power curves in unseen years through a temporal hold-out validation scheme. Compared with the physics-based model on the same test years, the proposed approach achieves lower errors while maintaining interpretability. The results highlight improved generalization and interpretability, providing a practical tool for power forecasting and grid operation.