Title: Sampling Restoration Decisions in Electric Power Distribution Systems
Abstract: The distribution restoration problem is a multi-stage, mixed-integer program that must be solved under critical regimes to minimize load shedding after a disaster which disconnects a distribution feeder's customers from its energy sources. To alleviate the computational challenges of solving this problem, and provide a means of representing and assimilating uncertainty in operating conditions, we apply Sampling Decisions, a recently developed generative AI method, to generate restoration patterns consistent with an optimization-aligned probability distribution. We show on a small feeder that Sampling Decisions improves solve times by fixing integer decisions, and further that the recovered restoration patterns achieve costs comparable to a ground-truth solution.
Bio: Robert Ferrando is a postdoctoral associate in the Computational Intelligence and Modeling Group (A-1) at Los Alamos National Laboratory. He received his BS in Mathematics from Macaulay Honors College, City University of New York, College of Staten Island, and his MS and PhD from the University of Arizona’s Graduate Interdisciplinary Program in Applied Mathematics. His current research focus is on optimization and control of power systems under uncertainty, ensuring that grid operations are resilient and robust to extreme events, including both natural and human disasters.