Spatio-Temporal Power Outage Risk Prediction for Interdependent Urban Electricity and Drainage Networks Under Rainstorm Disasters

Increasing power outage events triggered by rainstorm disasters have highlighted the urgent need for power outage risk prediction to enhance emergency preparedness. However, due to the tight risk interdependencies between power distribution networks (PDNs) and stormwater drainage networks (SDNs) under rainfall variability, the accurate and efficient power outage risk prediction remains challenging. This paper proposes a spatio-temporal power outage risk method for interdependent PDNs and SDNs triggered by rainstorm disasters to identify high-risk areas over time. Firstly, a multi-network cascading failure model is formulated to characterize the escalated superposition effect between urban waterlogging in SDNs and power outages in PDNs. A generalized probability density evolution function (GPDEF) with probabilistic partial differential equations (PDEs) is formulated for time-dependent risk modeling of both direct waterlogging-induced power outages and indirect nodal load curtailments in response to rainfall volatility. Furthermore, a hybrid physics-embedded and data-driven method is proposed to efficiently predict spatio-temporal power outage risks, in which the probability-constrained neural network is utilized to solve the formulated GPDEF with probabilistic PDEs. Simulation results demonstrate that the proposed method reduces power outage occurrence time prediction errors by up to 19.4 minutes and improves spatial prediction accuracy by over 12.20%.