Machine Learning-Aided Security Constrained Optimal Power Flow

Though many approaches have been proposed in recent decades to solve the full AC optimal power flow (OPF) problem, efficiently finding the solution still remains challenging due to its highly non-linear and non-convex nature, especially for large scale networks. Machine learning has proven to significantly improve the computational efficiency in many problems. Thus in this paper, a learning augmented optimization approach is developed to solve the security-constrained optimal power flow (SCOPF) problem. More specifically, a multi-input multi-output random forest model is developed to first solve network voltage magnitudes and angles of buses. Then, physics-based network equations are employed to determine the current injection and complex/real power injection at different buses. To evaluate the efficiency of the proposed machine learning-aided algorithm, two benchmark models are adopted: (i) one with the conventional MATPOWER Interior Point Solver, and (ii) the other one with an end-to-end pure machine learning approach. Test results on a 500-bus network show that the proposed machine learning-aided approach has significantly improved the computational efficiency compared to the MATPOWER solver, while all network constraints are successfully satisfied.