Weather-Based Damage Prediction Models for Electric Distribution Networks

From thunderstorms to hurricanes, electric distribution networks are subject to a wide range of warm weather storm events. Tropical Storm Irene (2011) and Hurricane Sandy (2012) are two events in recent memory that disrupted over half of The Connecticut Light and Power Company’s (CL&P) service territory, which left some customers without power for up to eleven days. This research study investigates a damage prediction framework for both thunderstorms and hurricanes that combines two generalized linear models to probabilistically determine the occurrence and extent of damages, known as trouble spots, to the overhead power distribution network. The models are inputted with high-resolution weather simulations from the Weather and Research Forecasting (WRF) Model along with distributed information on CL&P’s infrastructure, tree canopy density, and land cover data. The models were subjected to cross validation based on 30 major storm cases including the two tropical storms (Storm Irene and Hurricane Sandy), and exhibited a median percent error less than 30% for predicting the counts of trouble spots per event. Additionally, we explore an operational example of these models by using forecasts from 48 and 24 hours ahead of landfall by Hurricane Sandy to demonstrate how a real-time damage prediction system might operate