Forecasting Storm-Related Power Outages in Connecticut and Massachusetts

Our modern society demands reliable electric power to energize our daily lives. The interaction of severe weather and vegetation is among the biggest contributors to power outages on overhead lines in New England. In this session, we will present recent research related to the development and implementation for our outage prediction models in Connecticut and Western Massachusetts. Our outage models were inputted with numerical weather prediction simulations (>120 storms) using the Weather Research and Forecast (WRF) model, and augmented with land use aggregated around overhead lines and utility-specific infrastructure data. A host of tree-based models (decision tree, random forest, boosted tree, ensemble decision tree) were used to relate the explanatory variables to the historic outage data, and we found that the ensemble decision tree had the best error performance metrics. Although model calibration and validation were acceptable, we have found that certain combinations of weather-related variables yield more stable outage predictions than our previously published work. By utilizing different combinations of weather-related variables (i.e. sustained wind variables only; gust variables only; inclusion of all wind and precipitation variables), we can best present the complete range of outage predictions to emergency preparedness functions, which can lead to efficient pre-storm allocation of restoration crews and resources.