Probabilistic reliability assessment method for urban power grids based on Bayesian networks

Abstract

Urban power grids are exposed to coupled equipment ageing, weather stress, peak loading, distributed-generation uncertainty, and communication failures, whereas conventional reliability indices are usually calculated from fixed annual averages. This paper proposes a hierarchical evidence-updated Bayesian network for probabilistic reliability assessment and automated decision support. The model fuses network topology, component failure and repair records, Supervisory Control and Data Acquisition (SCADA)/ Advanced Metering Infrastructure (AMI) events, weather and loading forecasts, automation availability, and confidence-weighted thermal or visible-light inspection evidence. Beta- binomial parameter updating and soft-evidence junction-tree inference generate 24-h interruption probabilities, posterior root-cause importance, and annual System Average Interruption Frequency Index (SAIFI), System Average Interruption Duration Index (SAIDI), Customer Average Interruption Duration Index (CAIDI), Average Service Availability Index (ASAI), and expected energy not supplied. Candidate switching, maintenance, and mobile-resource actions are evaluated through intervention inference. A modified IEEE 33-bus urban feeder with 32 line sections, 33 load points, 10,480 customers, and six operating scenarios is used for validation against analytical assessment, a static Bayesian network, and a 200,000-year sequential Monte Carlo reference. The proposed method achieves a mean absolute probability error of 0.028 percentage points and a 2.1% SAIDI error with a 0.41 s inference time. Under the compound storm-aging- communication scenario, coordinated actions reduce SAIDI from 712.1 to 382.7 min/customer-year and Expected Energy Not Supplied (EENS) from 87.61 to 47.08 MWh/year. The framework therefore provides an interpretable reliability-to-control interface suitable for urban distribution automation and condition-monitoring systems.