Power outages caused by severe weather events, such as hurricanes, tornadoes, and thunderstorms, have major impacts on the electric utility distribution operations. A damage-prediction model which can estimate the number of outages that can occur based upon given weather forecasts can improve the efficiency of such operations by reducing the economic and societal costs associated with restoration efforts.

We have developed a Poisson regression model in a Bayesian hierarchical framework for spatial lattice data to predict number of outages in a fashion that is suitable for use by electric utilities. This model is coupled to the meso-gamma-scale numerical weather prediction forecasts generated by the "Deep Thunder" system developed at the IBM Thomas .J. Watson Research Center. Specific attention is given to building models that incorporate uncertainties in both the weather and outage data from the perspective of multiple spatial resolutions as well as spatial correlation in the areal outage data. An algorithm based upon Markov Chain Monte Carlo (MCMC) Methods is implemented in order to make the Bayesian inference.

The outage prediction model has leveraged historical outage data from an electrical utility company in the northeastern United States. It is being utilized by that company in their distribution operations. We will discuss results to date and how the model is being applied.