Graph convolutional network-based security-constrained unit commitment leveraging power grid topology in learning

Abstract

Security-constrained unit commitment (SCUC) is a complex optimization problem in power system operation, which is computationally intensive. To bring significant time-savings, this paper presents a graph convolutional network (GCN)-based SCUC approach (GCN-SCUC) using the information of power grid topology. Instead of tackling the mixed integer linear programming (MILP)-based SCUC (MILP-SCUC), the GCN learner predicts the unit decisions first, and then the MILP-SCUC problem is transformed into a continuous convex one. Numerical experiments are performed on the modified IEEE-30 and IEEE-118 systems to verify the feasibility of our approach both in terms of accuracy and computation time. Moreover, compared with the state-of-the-art MILP-SCUC, the proposed approach achieves speedups of between 13x and 17x on different testing examples with high accuracy.