CMStatistics 2022: Start Registration
View Submission - CMStatistics
Title: Probabilistic forecasting of weather-driven faults on electricity distribution networks Authors:  Daniela Castro-Camilo - University of Glasgow (United Kingdom) [presenting]
Jethro Browell - University of Glasgow (United Kingdom)
Abstract: Electricity networks are exposed to the weather, and severe weather may cause faults that result in power cuts. Predicting the occurrence of faults in a region on time scales from hours to days ahead can increase preparedness and accelerate the response to weather-related faults, and ultimately reduce the duration of power cuts. Furthermore, these predictions should quantify uncertainty so that planners may assess risk and distribute limited resources accordingly. We present a method for probabilistic fault prediction that leverages ensemble numerical weather prediction and methods from extreme value theory for discrete processes. Data describing network topology and vulnerability, such as elevation and proximity to vegetation, are combined with meteorological data to model the occurrence of faults, which is stochastic and may be heavy-tailed. In addition, forecasts of future weather conditions are required, and associated uncertainty is quantified via ensemble numerical weather prediction, which requires statistical post-processing. Finally, we discuss the communication of the resulting complex forecast information to decision-makers.