Topology transferable physics-informed graph learning for national-scale AC optimal power flow

초록

Alternating current optimal power flow (AC-OPF) has been a challenging problem for decades due to its nonconvexity, but fast and efficient solutions are even more needed because of high penetration of large scale renewable generation and load growth. Recently, neural networks (NN) have gained attention in solving AC-OPF, but it is still in an early stage to be applicable for real and large-scale power system operation with topology-changing characteristics. To this end, we propose GraphOPF that considers topology-transferability, scalability, NN training time, self-supervision, and feasibility altogether. Specifically, we propose edge-aided graph neural networks to handle topology changes and a hard-constrained embedded layer for satisfying power flow equations. We find that the proposed architecture has high sample efficiency that enables fast NN training by using a small amount of training samples (e.g., one tenth samples only). We further demonstrate that graph transfer learning can be applied upon topology changes, which finishes NN training within 1 min for the Korea national grid comprising approximately 4500 buses and 6000 transmission lines. Extensive experiments show that the proposed framework is up to 9 times faster in NN training against the baselines and up to 66 times faster in per-instance inference time compared to mathematical optimization solvers for solving AC-OPF of large-scale PGLib power systems, e.g., GOC 2312, 3970, 4601, and EPIGRIDS 5658, while achieving more than 99% component-wise constraint satisfaction on average.

키워드