Dynamic modeling of the effects of vegetation management on weather-related power outages

Abstract

This paper develops a machine learning outage prediction model (OPM) to serve as a simulation framework capable of quantifying the reduction in damages to the distribution electric grid due to vegetation management for storm events. The model covers the Eversource Energy distribution grid territory in Connecticut and uses a random forest model with input variables for vegetation, vegetation management, land cover, drought, elevation, weather and electrical infrastructure to predict outages for each circuit (the operational units of the power distribution network). The model is trained on 165 storms from the years 2005 to 2019. The results show that over the last five years of the study (2015–2019) the annual reduction in trouble spots in the electric grid due to enhanced tree trimming is between 25.7 and 42.5% and there is good matching between increased trouble spot reduction and increased vegetation management. Further, we demonstrate improved outage predictions when including vegetation management data as an input variable, with a 4.1% reduction in mean absolute percentage error in leave-one-storm-out cross-validation. This framework could be used to examine varying vegetation management scenarios and the results should be useful for decision makers such as utilities, municipalities and regulators in optimizing vegetation management and broader grid resilience enhancement plans.