Work

A Bayesian hierarchical model for power system resilience assessment

Year: 2026

Type: article

Abstract: Extreme weather events and operational failures now cause longer, more widespread, and far less predictable power outages. Quantifying these impacts requires models that capture both observed outage patterns and the hidden variability that drives disruption and recovery processes. Existing approaches emphasize outage occurrence and risk assessment using weather-dependent variables, but they often depend on uncertain forecast inputs and offer limited insight into customer-level impacts. In addition, such approaches treat outages as independent events and overlook the complex interactions between weather, infrastructure vulnerabilities, and operational factors. This paper presents a Bayesian hierarchical model that integrates the EAGLE-I power outage dataset (2014-2023) with National Weather Service metadata to predict both weather-related and non-weather-related power outages. The proposed approach captures stochastic environmental and operational influences through hierarchical latent effects, without relying on explicit weather prediction. Spatial and temporal dependence across regions is captured through multilevel structures that jointly model outage duration and customer impact within a unified probabilistic formulation. Full posterior uncertainty quantification supports resilience assessment, synthetic scenario generation, imputation of incomplete records, and statistically consistent forecasts under data sparsity. The model further enables fair benchmarking across diverse geographic and operational environments and provides a practical tool for utilities and policymakers to anticipate outage risks, guide infrastructure decisions, and strengthen adaptive strategies under evolving climate and grid conditions.

Source: International Journal of Electrical Power & Energy Systems

DOI: 10.1016/j.ijepes.2026.111829

Authors: Chandan Chaudhary, Yansong Pei, Mohammed Ben-Idris, Joydeep Mitra

Institution: Michigan State University

Language: English

Topic: Power System Reliability and Maintenance

Subfield: Safety, Risk, Reliability and Quality

Field: Engineering

Domain: Physical Sciences

SDGs: Industry, innovation and infrastructure, Climate action

Open Access status: gold

Estimated APC paid: $3,920