Abstract

Power outages caused by severe storms produce enormous economic losses and societal disruptions. Infrastructure hardening for a more resilient power grid can reduce weather-induced outages but necessitates accurate simulations of intervention efficacy. Data-driven models have been developed to forecast outages but struggle with extreme events where data might be limited. Meanwhile, physics-based models can predict failures under strong winds but have limited scope in lower wind ranges. Despite the two models’ complementary benefits, studies investigating their integration have been limited. In the present study, the physical attributes of the infrastructure system are incorporated into data-driven models by coupling structural fragilities of the pole-wire overhead power distribution system with machine learning (ML) techniques applied in an outage prediction model for the northeastern United States. The ML model is used to calibrate the physics-based fragility curves, which are subsequently used to predict outages for high-impact events where empirical data may be limited. The results of this hybrid physics-based and data-driven (HPD) model indicate modeling improvements using hybrid over strictly data-driven approaches for extreme events. Root mean square error improvements of 48% are exhibited for high-impact event outage prediction. The hybrid model was then utilized to counterfactually assess the impacts of grid hardening activities such as pole replacement, pole class upgrade, improved pole chemical treatment, and undergrounding in reducing pole failures over the last fifteen years. The results indicate selected strategies targeted to the oldest 5% of infrastructure could have reduced over 100 (33% of all) pole failures annually across the state of Connecticut.

Full Text
Published version (Free)

Talk to us

Join us for a 30 min session where you can share your feedback and ask us any queries you have

Schedule a call