Extracting resilience events from utility outage data based on overlapping times and locations

Abstract

To study power system resilience with real data, it is necessary to group individual power outages recorded by utilities into events in which outages cluster and overlap due to extreme weather. We show how to automatically group utility outage data into resilience events based on their time and location. Each outage is represented as a cylinder in three-dimensional space, with a disk centered at the outage location in the geographic plane and a vertical extent corresponding to a limited outage duration, so that two outages overlap in time and space if their cylinders intersect. The grouping algorithm can be implemented as a graph whose nodes are the outages and whose edges represent the overlaps of outages in time and space, so that events are the connected components of the graph. Extending time-based grouping to both time and location is particularly useful when extracting events from outage data collected across a wide area, as it prevents unrelated outages from being incorrectly merged into anomalous events solely due to temporal overlap. We propose a metric to tune the parameters of the grouping algorithm to minimize anomalous events. The grouping of outages into events works with both detailed utility outage data and web-scraped EAGLE-I outage data. Results are validated against NOAA storm event records and DOE-417 reports. The automatically extracted events from utility data closely match documented major weather events.