Gated Graph Attention Networks for Predicting Duration of Large Scale Power Outages Induced by Natural Disasters

Chenghao Duan; Chuanyi Ji; Anwar Walid; Scott Ganz

cs.LGMar 14, 2026

Gated Graph Attention Networks for Predicting Duration of Large Scale Power Outages Induced by Natural Disasters

Authors: Chenghao Duan, Chuanyi Ji, Anwar Walid, Scott Ganz

Organizations: School of Electrical and Computer Engineering, Georgia Institute of Technology, Atlanta, GA, USA · Department of Electrical Engineering, Columbia University, New York, NY, USA · School of Business, University of California, Riverside, Riverside, CA, USA

Abstract

The occurrence of large-scale power outages induced by natural disasters has been on the rise in a changing climate. Such power outages often last extended durations, causing substantial financial losses and socioeconomic impacts to customers. Accurate estimation of outage duration is thus critical for enhancing the resilience of energy infrastructure under severe weather. We formulate such a task as a machine learning (ML) problem with focus on unique real-world challenges: high-order spatial dependency in the data, a moderate number of large-scale outage events, heterogeneous types of such events, and different impacts in a region within each event. To address these challenges, we develop a Bimodal Gated Graph Attention Network (BiGGAT), a graph-based neural network model, that integrates a Graph Attention Network (GAT) with a Gated Recurrent Unit (GRU) to capture the complex spatial characteristics. We evaluate the approach in a setting of inductive learning, using large-scale power outage data from six major hurricanes in the Southeastern United States. Experimental results demonstrate that BiGGAT achieves a superior performance compared to benchmark models.

Figures & tables

Fig. 1: Spatial heterogeneity and dependency of power outages from six hurricanes: (a) Florence, (b) Irma, (c) Laura, (d) Michael, (e) Sally, (f) Zeta. Dark lines: wind boundaries (64, 50, and 34 knots). Clusters on regions with different impacts: Red - significantly impacted counties; Yellow - less impacted counties. Clustered by wind and outage duration.

Events	Short	Medium	Long
Florence	$68$	$21$	$9$
Irma	$158$	$127$	$29$
Laura	$76$	$18$	$17$
Michael	$127$	$31$	$18$
Sally	$62$	$7$	$1$
Zeta	$179$	$44$	$12$

TABLE I: Data summary: number of counties for three classes of outage durations

Fig. 2: Overall BiGGAT structure: Node features are fed into the Bimodal Embedding, then passed through the Gated Graph Attention mechanism. The aggregated messages are transformed to outputs by a Linear Readout layer.

Events	Highest n	Global n-hop Moran’s I
Florence	$3$	$0.273^{*}(6.806)$
Irma	$6$	$0.070^{*}(4.723)$
Laura	$5$	$0.112^{*}(3.499)$
Michael	$3$	$0.053^{*}(1.998)$
Sally	$2$	$0.110^{*}(1.864)$
Zeta	$4$	$0.154^{*}(6.049)$

TABLE II: Global n-hop Moran’s I of peak outage number

Fig. 3: Model performance comparison for (a) Florence, (b) Irma, (c) Laura, (d) Michael, (e) Sally, and (f) Zeta. Each subplot shows classification accuracy, Macro F1, and balanced accuracy for XGB, RF, GAT, BiGAT, and BiGGAT.

Models	Accuracy	Balanced Acc	Macro F1
XGB	$79.7\%$	$67.8\%$	$0.657$
RF	$81.5\%$	$69.4\%$	$0.682$
GAT	$83.7\%$	$68.1\%$	$0.678$
BiGAT	$85.1\%$	$70\%$	$0.686$
BiGGAT	$85.9\%$	$78.3\%$	$0.762$

TABLE III: Event Average performance of models

Explore similar work

Empowering Power Outage Prediction with Spatially Aware Hybrid Graph Neural Networks and Contrastive Learning
Apr 6, 2026Xuyang Shen, Zijie Pan, Diego Cerrai +4Power Systems Graph Neural Networks
University of Connecticut · Western Connecticut State University · Eversource Energy
Inductive Power Grid Cascading Failure Analysis with GRU-Gated Graph Attention
May 7, 2026Tianxin Zhou, Xiang Li, Haibing LuGraph Attention Networks Graph Representation Learning
Department of Computer Science & Engineering, Santa Clara University, Santa Clara, USA · Dept. of Information Systems & Analytics, Santa Clara University, Santa Clara, USA
PowerModelsGAT-AI: Physics-Informed Graph Attention for Multi-System Power Flow with Continual Learning
Feb 24, 2026Chidozie Ezeakunne, Jose E. Tabarez, Reeju Pokharel +1Physics-Informed GNNs Power Systems
Los Alamos National Laboratory, Los Alamos, NM, USA · Department of Physics, University of Central Florida, Orlando, FL 32816 USA

Apr 6, 2026J/K move · Enter open · S save

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

Xuyang Shen, Zijie Pan, Diego Cerrai, Xinxuan Zhang+3

University of Connecticut · Western Connecticut State University · Eversource Energy

Power Systems Graph Neural Networks

Open paper AskPDF arXiv

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 features (e.g., land cover, infrastructure) and event-specific dynamic features (e.g., wind speed, precipitation) via Spatially Aware Hybrid Graph Neural Networks (SA-HGNN). Next, we leverage contrastive learning to handle the imbalance problem associated with different types of extreme weather events and generate location-specific embeddings by minimizing intra-event distances between similar locations while maximizing inter-event distances across all locations. Thorough empirical studies in four utility service territories, i.e., Connecticut, Western Massachusetts, Eastern Massachusetts, and New Hampshire, demonstrate that SA-HGNN can achieve state-of-the-art performance for power outage prediction.

Abstract

Explore similar work

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

Inductive Power Grid Cascading Failure Analysis with GRU-Gated Graph Attention

PowerModelsGAT-AI: Physics-Informed Graph Attention for Multi-System Power Flow with Continual Learning