Spatially Generalized Model for Predicting Power Outages Caused by Tropical Cyclones

Tropical cyclones have the potential to cause significant damage to the electrical power system, which requires a timely response from utilities. The power restoration process can be optimized when there is an accurate spatiotemporal forecast of outages in advance of a tropical cyclone making landfall. There are numerous models available for making predictions. One such model is the Spatially Generalized Model (SGHOPM) developed by Guikema et al. (2014). It has an advantage over similar models in that it only uses publicly available data to make outage predictions. The SGHOPM makes predictions at the census tract level using a Random Forest model. In addition to the forecast winds, the other predictor variables describe the elevation, land cover, soil, precipitation, and vegetation characteristics in each census tract. The addition of these variables improves the statistical accuracy of the Random Forest model predictions. The SGHOPM is trained and validated using historical outage data from several landfalling tropical cyclones. The cross-validation procedure is used to determine an optimal set of covariates for the Random Forest model.