Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

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

Xuyang Shen, Zijie Pan, Diego Cerrai, Xinxuan Zhang, Christopher Colorio, Emmanouil N. Anagnostou, Dongjin Song

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 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.

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

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 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.

Paper Structure

This paper contains 21 sections, 13 equations, 7 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Related work
  3. Problem statement
  4. Methodology
  5. Dynamic Graph Learning Module
  6. Hybrid Graph Convolutional Module
  7. Contrastive Learning Across Intra and Inter-Event
  8. Optimization Objectives
  9. Experiment
  10. Experiment Setup
  11. Datasets and Preprocessing
  12. Evaluation Protocols and Metrics
  13. Baseline Model
  14. Main Results
  15. Ablation Study
  16. ...and 6 more sections

Figures (7)

  • Figure 1: The framework of SA-HGNN. The dynamic graph learning module learns event-specific adjacency matrices guided by external structure, which are used in the dynamic graph convolution. The hybrid graph convolution includes a dynamic GCN (DGCN) for capturing event-specific patterns and a static GCN (SGCN) that aggregates information from a shared graph. Their outputs are concatenated to form location-wise embeddings. Contrastive learning further refines these embeddings by aligning similar locations within events and separating dissimilar ones across events. A regression module then projects the fused embeddings to predict outage values.
  • Figure 2: Actual outages vs. predicted outages comparison of four models on Connecticut extreme weather data.
  • Figure 3: Comparison of learned location embeddings across different methods on the Connecticut dataset.
  • Figure 4: SA-HGNN CL representation comparison
  • Figure 5: Outage prediction distribution of Hurricane Irene (August 28, 2011) in Connecticut. The ground truth total outage count is 16022 (a). SA-HGNN (b) produces the closest total prediction (14512) compared to XGBoost (21385), GIN (11273), and GAT (13742).
  • ...and 2 more figures