Power Outage Prediction for Adverse Weather Events

• Activity Number: 126 - New Development in Reliability Models and Innovative Applications
• Type: Contributed
• Date/Time: Monday, July 30, 2018 : 8:30 AM to 10:20 AM
• Sponsor: Section on Physical and Engineering Sciences
• Author(s): Seth Guikema* and Steven Quiring and Ken Buckstaff and Mike Beck and Alan Bowman and Brent McRoberts and Roshanak Nateghi
• Companies: University of Michigan and Ohio State University and First Quartile Consulting and Beck Consulting and Clarkson University and Texas A&M University and Purdue University
• Keywords: probabilistic prediction; predictive modeling; power outage; extreme weather; ensemble models

Abstract

A wide variety of weather conditions can cause power outages, from winter storms and wind storms to prolonged heat events. Being able to estimate physical damage and the number of person-hours needed to restore electric power based on a weather forecast has the potential to help utilities restore power more quickly and efficiently. This paper presents an integrated model for estimating weather-related equipment damage and restoration needs. The model is based on historic outage, weather, and restoration data and provides fully probabilistic estimates of physical damage and person-hours needed for restoration. It is used operationally at different lead times, based on weather forecasts with up to a five-day lead time. The model is a three-stage model. In the first stage, an ensemble of classifiers is used to estimate the probability of weather-induced outages in the utility service territory on a given day. In the second stage, four conditional quantile regression forests are used to estimate the probability density function for the number of damaged assets in each of four asset classes. In the final stage, a linear regression model estimates person hours for needed for restoration.