Statistical Forecasting of Hurricane Power Outages

• Activity Number: 131
• Type: Contributed
• Date/Time: Monday, August 5, 2013 : 8:30 AM to 10:20 AM
• Sponsor: Section on Physical and Engineering Sciences
• Author(s): Seth Guikema*+ and Roshanak Nateghi and Steven Quiring
• Companies: Johns Hopkins University and Johns Hopkins University and Texas A&M University
• Keywords: hurricane ; reliability ; forecasting ; power outage ; data mining

Abstract

Hurricanes regularly cause widespread loss of power along the U.S. coastline. Having accurate forecasts of total outages and the spatial pattern of outages before hurricane landfall is critical for utility preparedness decision making. In this paper we summarize the development of the Hurricane Power Outage Model, a statistical power outage forecasting model developed over the past seven years, and we present results of the model for Hurricanes Irene (2011) and Sandy (2012). The current version of the model is an ensemble tree-based data miner, developed through multiple holdout validation testing. We present two versions of the model, one based on private utility data and applicable to the service area of a particular utility, the other a "broad area" model using only public data and applicable to the entire coastline. The utility-specific model achieves average out of sample MAE of ~2%. The broad area model provides useful guidance for emergency preparedness decisions, though with higher prediction error. This paper discusses the model development and validation process, the prediction errors of the two models, and the variables that are most influential in the predictions.