A Data-Driven Decision Support Tool for Anticipating Tropical Storm Impacts to the United States Power Grid

Abstract

Power outages caused by tropical storms significantly impact the United States economy and citizens every year. If the impacts of these storms could be better anticipated, they could be more effectively mitigated. While there has been some research focused on predicting the impacts of tropical storms on the power grid, they have not yet resulted in effective decision making support tools that can effectively improve storm response. This paper attempts to improve upon previous work by describing a novel predictive modeling approach designed to accurately forecast tropical storm related power outages across the Continental United States. This framework considers a wide range of multidisciplinary data derived from numerical weather prediction, US Census Bureau products, and various other earth datasets to describe the environmental and infrastructural conditions during storms. A rigorous evaluation of 38 historical tropical storms shows that model performance compares favorably to established, smaller-scale outage models and demonstrates a greater than 50% reduction in Mean Squared Error over the most similar large-scale outage model in the literature. This modeling approach has potential to inform decision makers about the overall severity of storm impacts, as well as which regions are more likely to be more effected by persistent power outages. This information could allow emergency managers to prepare more effectively and reduce the felt impacts of these storms.

Keywords

• Power outage
• tropical storm
• predictive analytics
• machine learning