Wind-induced transmission line interruption fragility models: An adaptive GAN-augmented probabilistic classification approach for extremely unbalanced data

Weather-induced outages pose a significant threat to power grid reliability, with transmission systems particularly vulnerable to environmental stressors. Despite numerous tools developed to address this issue, the persistent challenge of weather-related interruptions highlights the need for an accurate fragility model for transmission line interruptions. This paper proposes a novel data-driven approach to model wind-induced transmission line fragility, addressing critical gaps in current methodologies. Our model integrates a novel synthetic data generation approach that creates highly informative synthetic data points, enhancing the representation of rare events. Additionally, we develop an advanced active learning framework that efficiently selects the most relevant instances from large, imbalanced datasets for model training. We further enhance model interpretability through comprehensive sensitivity analysis using SHAP (SHapley Additive exPlanations) values. Results on unseen testing data show significant improvement compared to conventional methods, achieving a 5% improvement in accuracy (from 0.89 to 0.94) in predicting wind-induced transmission line interruptions. Notably, it shows a 16% accuracy improvement (from 0.64 to 0.80) when applied to highly uncertain cases, highlighting its capabilities in high-uncertainty situations. Sensitivity analysis reveals wind gust and mean sea level pressure as the most critical factors influencing interruptions, while also uncovering complex temperature effects where, in a subset of situations, temperature has a significant impact on the interruption probability of lines. This advanced fragility model can offer valuable insights for both real-time dispatch decisions and long-term risk-informed planning, contributing to enhanced power grid resilience in the face of increasing weather-related challenges.