Researchers develop statistical model to better understand cascading failures in networks
Cascading network outages are a serious problem. The failure in one component can lead to overloading in others, causing widespread disruptions to the power grid. In order to address such situations, a protection system acts to isolate the initial fault, automatically removing one or more transmission lines in less than a minute. Scientists have studied the detailed modeling of specific mechanisms of protective operation, but these studies have focused principally on small transmission system networks.
Researchers from Iowa State University and the U.S. Department of Energy’s Argonne National Laboratory have taken a different approach. Gaining access to detailed protection data across an entire transmission system is difficult. Instead, the researchers extract outage patterns from historical utility data showing how line outages propagate in two transmission networks in less than one minute. They compute statistical properties such as the size and frequency of the outage configurations and then create a model that can generate new patterns, including rare large patterns, consistent with those statistics.
“The ability to generate representative outage patterns from any given starting line outage is important because the historical outages alone will not cover the full range of credible possibilities when assessing the risk of future outages,” said D. Adrian Maldonado, an assistant energy systems scientist in Argonne’s Mathematics and Computer Science division.
Why focus on the fast time scale? The patterns formed at the fast time scale are generally small, connected subnetworks that are mostly trees, whereas those formed at slower time scales tend to be disconnected sets of lines.
Why use historical utility data? The 19 years of outage data from the Bonneville Power Administration and 12 years from the New York State Independent System Operator provide an approach based on real, rather than simulated, data (see Fig. 1).
How can one be sure that the new patterns are valid? To validate their model, the researchers use the Wasserstein distance as a metric to quantify how close the generated patterns are to the observed patterns. They also statistically test whether the degree sequences of the generated and observed patterns can be considered samples from the same probability distribution.
“We don’t try to reproduce a pattern exactly but to generate new patterns that are statistically consistent with the observed patterns,” Maldonado said.
The research team envisions that their new statistical modeling approach will help improve simulations used for risk assessment and potentially inform the evaluation of new protection strategies, taking into consideration not only common cases of network outages but also rare configurations whose likelihood and structure are difficult to capture with traditional methods yet can have a significant impact.
For the full study, see the paper: Ian Dobson, D. Adrian Maldonado, and Mihai Anitescu, “The statistical spread of transmission outages on a fast protection time scale based on utility data,” preprint arXiv:2407.15059v1, 2024.
Argonne National Laboratory seeks solutions to pressing national problems in science and technology by conducting leading-edge basic and applied research in virtually every scientific discipline. Argonne is managed by UChicago Argonne, LLC for the U.S. Department of Energy’s Office of Science.
The U.S. Department of Energy’s Office of Science is the single largest supporter of basic research in the physical sciences in the United States and is working to address some of the most pressing challenges of our time. For more information, visit https://energy.gov/science.