2022/09/07 by Antoni M. Sieminski, Sieminski, Antoni M., Carl Donovan +1
Engineering · #Applications (stat.AP) #Energy Load and Power Forecasting #FOS: Computer and information sciences #Power Systems Fault Detection #Underground infrastructure and sustainability
paper · pdf · doi:10.48550/arxiv.2209.03495
openalex publication_date 2022/09/07 · openalex created_date 2022/09/10 · openalex updated_date 2026/07/28
Overhead distribution lines play a vital role in distributing electricity, however, their freestanding nature makes them vulnerable to extreme weather conditions and resultant disruption of supply. The current UK regulation of power networks means preemptive mitigation of disruptions avoids financial penalties for distribution companies, making accurate fault predictions of direct financial importance. Here we present predictive models developed for a UK network based on gradient-boosted location, scale, and shape models, providing spatio-temporal predictions of faults based on forecast weather conditions. The models presented are based on (a) tree base learners or (b) penalised smooth and linear base learners -- leading to a Generalised Additive Model (GAM) structure, with the latter category of models providing best performance in terms of out-of-sample log-likelihood. The models are fitted to fifteen years of fault and weather data and are shown to provide good accuracy over multi-day forecast windows, giving tangible support to power restoration.