[2604.04916] Empowering Power Outage Prediction with Spatially Aware Hybrid Graph Neural Networks and Contrastive Learning

Computer Science > Machine Learning arXiv:2604.04916 (cs) [Submitted on 6 Apr 2026] Title:Empowering Power Outage Prediction with Spatially Aware Hybrid Graph Neural Networks and Contrastive Learning Authors:Xuyang Shen, Zijie Pan, Diego Cerrai, Xinxuan Zhang, Christopher Colorio, Emmanouil N. Anagnostou, Dongjin Song View a PDF of the paper titled Empowering Power Outage Prediction with Spatially Aware Hybrid Graph Neural Networks and Contrastive Learning, by Xuyang Shen and 6 other authors View PDF HTML (experimental) Abstract:Extreme weather events, such as severe storms, hurricanes, snowstorms, and ice storms, which are exacerbated by climate change, frequently cause widespread power outages. These outages halt industrial operations, impact communities, damage critical infrastructure, profoundly disrupt economies, and have far-reaching effects across various sectors. To mitigate these effects, the University of Connecticut and Eversource Energy Center have developed an outage prediction modeling (OPM) system to provide pre-emptive forecasts for electric distribution networks before such weather events occur. However, existing predictive models in the system do not incorporate the spatial effect of extreme weather events. To this end, we develop Spatially Aware Hybrid Graph Neural Networks (SA-HGNN) with contrastive learning to enhance the OPM predictions for extreme weather-induced power outages. Specifically, we first encode spatial relationships of both static featur...