A data-driven approach to quantify disparities in power outages

Abstract

This research proposes a data-driven approach to identify possible disparities in a utility’s outage management practices. The approach has been illustrated for an Investor-Owned Utility located in the Midwest region in the U.S. Power outage data for approximately 5 years between March 2017 and January 2022 was collected for 36 ZIP/postal codes located within the utility’s service territory. The collected data was used to calculate the total number of outages, customers affected, and the duration of outages during those 5 years for each ZIP code. Next, each variable was normalized with respect to the population density of the ZIP code. After normalizing, a K-means clustering algorithm was implemented that created five clusters out of those 36 ZIP codes. The difference in the outage parameters was found to be statistically significant. This indicated differential experience with power outages in different ZIP codes. Next, three Generalized Linear Models were developed to test if the presence of critical facilities such as hospitals, 911 centers, and fire stations, as socioeconomic and demographic characteristics of the ZIP codes, can explain their differential experience with the power outage. It was found that the annual duration of outages is lower in the ZIP codes where critical facilities are located. On the other hand, ZIP codes with lower median household income have experienced more power outages, i.e., higher outage counts in those 5 years. Lastly, the ZIP codes with a higher percentage of the White population have experienced more severe outages that have affected more customers.