Accelerating Stochastic Unit Commitment with Multifidelity Scenario Bundling

Abstract

Conference Title: 2025 57th North American Power Symposium (NAPS)

Conference Start Date: 2025 Oct. 26

Conference End Date: 2025 Oct. 28

Conference Location: Storrs, CT, USA

Day-ahead commitment of generators, or unit commitment, is crucial for operating power grids. The problem of unit commitment can be modeled as a stochastic program, for which the uncertainties typically manifest in loads or renewable generation. However, stochastic unit commitment tends to be computationally challenging and therefore expensive to solve. In this paper, we present a modified progressive hedging algorithm that utilizes multifidelity modeling to rapidly generate high-quality solutions for large-scale stochastic unit commitment problems. Our methodology allows for the decomposition of the main problem into subproblems that bundle multiple high and low fidelity scenarios. This form of multi-fidelity bundling enables information sharing across subproblems that accelerates progressive hedging. We demonstrate the efficacy of our multifidelity scenario bundling approach on a RTS-GMLC unit commitment exemplar.