Learning Constraint Surrogate Model for Two-stage Stochastic Unit Commitment

The increasing penetration of renewable energy sources introduces significant uncertainty in power system operations, making traditional deterministic unit commitment approaches computationally expensive. This paper presents a machine learning surrogate modeling approach designed to reformulate the feasible design space of the two-stage stochastic unit commitment (TSUC) problem, reducing its computational complexity. The proposed method uses a support vector machine (SVM) to construct a surrogate model based on the governing equations of the learner. This model replaces the original 2|L| * |S| transmission line flow constraints, where |S| is the number of uncertainty scenarios and |L| is the number of transmission lines with |S| much less than |L|, with a significantly reduced set of 1 * |S| linear inequality constraints. The approach is theoretically grounded in the polyhedral structure of the feasible region under the DC power flow approximation, enabling the transformation of 2|L| line flow limit constraints into a single linear constraint. The surrogate model is trained using data generated from computationally efficient DC optimal power flow simulations. Simulation results on the IEEE 57-bus and 118-bus systems demonstrate SVM halfspace constraint accuracy of 99.72% and 99.88%, respectively, with TSUC computational time reductions of 46% and 31% and negligible generation cost increases (0.63% and 0.88% on average for IEEE 57- and 118-bus systems, respectively). This shows the effectiveness of the proposed approach for practical power system operations under renewable energy uncertainty.