From Forecast to Action: A Deep Learning Model for Predicting Power Outages During Tropical Cyclones

ABSTRACT Power outages caused by tropical cyclones (TCs) pose serious risks to electric power systems and the communities they serve. Accurate, high‐resolution outage forecasting is therefore critical for both proactive mitigation planning and real‐time emergency response. Most existing outage prediction models operate in open‐loop or event‐level settings and cannot update forecasts as storm conditions and system states evolve. To address this limitation, we propose the SpatioTemporal Outage ForeCAST (STO‐CAST) model, a spatiotemporal deep learning framework that performs state‐dependent, observation‐updated rolling inference throughout TC events. STO‐CAST enables outage forecasts to evolve in response to updated meteorological projections and newly observed outage information during runtime. The model integrates static environmental and infrastructure attributes with dynamic meteorological and outage sequences and produces hourly outage forecasts at a 4 km by 4 km resolution. STO‐CAST supports dual‐horizon forecasting, providing short‐term nowcasting with a 6‐hour lead time for real‐time situational awareness and long‐term forecasting with a 60‐hour lead time to inform proactive planning and resource staging. A case study of Typhoon Muifa (2022), evaluated under a Leave‐One‐Storm‐Out framework, demonstrates the model's operational value, including its ability to track evolving outage hotspots and to provide diagnostic insight through error decomposition that distinguishes the effects of model limitations, meteorological uncertainty, and observation gaps. Overall, STO‐CAST offers a scalable and interpretable framework to support risk‐informed emergency response and enhance power system resilience under intensifying TC threats.