Machine Learning Assisted Stochastic Unit Commitment: A Feasibility Study

Abstract

Stochastic unit commitment is an effective model for generation scheduling, in the presence of substantial uncertainty. However, effectiveness comes with substantial computational cost. Generally, stochastic unit commitment needs more time than other standard methods such as deterministic, to solve the unit commitment problem. In this paper, the results of initial feasibility studies are presented aiming to find out if using machine learning-based models can facilitate solving stochastic unit commitment problems. A real-world, large-scale test case is used to demonstrate the capabilities and shortcomings of machine learning algorithms in reducing the calculation time without sacrificing accuracy. Our feasibility study reveals that while it is unlikely to train a machine learning model to solve the problem as a standalone platform, it is possible to use a trained machine learning model to assist in accelerating the solution of the stochastic model.