Day-Ahead Wind Power Ramp Events Forecasting Method for Extreme Weather Based on TimeGAN and Diffusion Model

Accurate prediction of wind power ramp events is critical for ensuring the reliability and stability of power systems. However, most existing research on wind ramp events forecasting exhibits certain limitations, including: 1) Extreme weather is marked by high meteorological variability, leading to unstable wind power output and frequent ramp events. However, its low occurrence frequency limits the availability of representative samples, while numerical weather prediction (NWP) suffers from large errors under such conditions. These factors together make ramp event prediction particularly challenging. 2) The spatiotemporal correlation has not been fully considered, which restricts the improvement of forecasting accuracy. To address these limitations, this study introduces a day-ahead wind ramp events forecasting method considering extreme weather and spatiotemporal correlation information. First, cold wave events are precisely identified and the corresponding weather scenario dataset is expanded to enhance sample diversity and improve model robustness. Subsequently, a diffusion model is employed to correct NWP data, thereby providing more accurate input features for subsequent forecasting. Building upon this, the model further integrates spatiotemporal correlations among different power stations with local meteorological features of the target station, enabling refined extraction of predictive information for wind power ramp events. Finally, separate forecasting models are constructed for cold wave and conventional weather conditions to enhance adaptability and prediction accuracy. Validation on a real-world dataset in China confirms that our method outperforms the best existing benchmark by 2.7% in accuracy and 3.6% in F1-score.