Data-driven predictive models for wind-induced transmission line interruptions

Extreme weather events continually challenge transmission line integrity, highlighting the need for accurate vulnerability assessment of power grid components to ensure grid resilience. This paper proposes data-driven predictive models that explore and establish reliable associations between weather, environmental, and line characteristic data and wind-induced transmission line interruptions. To address the challenges of unbalanced datasets and enhance model accuracy, we propose event-based sampling, a new approach that leverages the stochasticity and physics of hazard effects to identify input subspaces most challenging for the machine learning models to classify, thus improving the overall models’ performance. We trained a diverse set of machine learning models to capture the complex underlying patterns of the datasets. Results of extensive testing studies indicate that the XGBoost model performs the best with a high predictive accuracy of 94.2%. In-depth sensitivity studies using Shapley values highlight the significance of wind gust, line length, and mean sea pressure suggesting that these factors play a substantial role in line interruptions. The developed model primarily serves as a data-driven fragility function for comprehensive weather-related risk assessment of transmission infrastructure, enabling long-term planning for line reinforcement and investment strategies. Additionally, the model shows potential for operational applications such as power outage mitigation and resource allocation during extreme weather events. • New ML models link weather, line, and environmental features to interruptions. • A novel event-based sampling enhances model performance in uncertain regions. • XGBoost model performs best in predicting wind-induced line interruptions. • Shapley analysis quantifies the impact of features on transmission line interruptions. • The models can assist utilities with short and long-term resilience enhancement.