High renewable energy penetration creates significant operational challenges for power systems, especially during extreme weather that disrupts supply–demand balance. This study introduces a framework that integrates extreme scenario identification, data augmentation, and power balance optimization. It defines extreme wind speed events, such as sudden drops, surges, and persistent anomalies, and uses a sliding-window algorithm to extract these events from historical meteorological data. To address the scarcity of extreme samples, a new data augmentation method combines the Wasserstein Generative Adversarial Network with Gradient Penalty (WGAN-GP) and iterative distribution shifting. This approach focuses the generated data on distribution tails while preserving diversity and temporal consistency. An optimization model, which includes various generation resources, energy storage, and load shedding, is developed to assess system flexibility under extreme conditions. Case studies on the projected 2030 Northeast China Power Grid show that the augmentation method expands extreme scenario datasets from 150 to 1000 samples, maintains extremity and temporal consistency, and reveals that wind curtailment rises sharply above 70% renewable share, with storage systems providing key flexibility in high-output scenarios.
Huang et al. (Mon,) studied this question.