Alternative Learning Architecture for Solving AC-OPF via Supervised Relaxation and Cross Encoder

Doan Thanh Hien, Keunju Song, Kibaek Kim, Hongseok Kim

Keywords: AC Optimal power flow, alternative learning, compact learning, deep neural network, semi-supervised learning

TL;DR: We propose a semi-supervised framework that combines original and relaxed AC-OPF problems using compact learning and dynamic Lagrange multipliers.

Abstract: As power systems evolve, efficient AC optimal power flow solutions are increasingly critical, yet traditional methods face challenges in speed and scalability. This paper introduces ExpressOPF, an ML-based framework that alternates between the original AC-OPF problem and its relaxed form. By integrating Lagrange multipliers into a compact neural network, ExpressOPF enforces constraints while improving accuracy and inference speed. Tested on 162-, 300-, 1354-, and the real-world 4492-bus system operated by Korea Power Exchange (KPX), ExpressOPF achieves under 1\% cost deviation from MATPOWER, 100,000$\times$ speedup, 75\% model compression, and 40\% lower GPU memory use—while maintaining over 99\% constraint satisfaction. These results highlight its potential for real-time, resource-efficient AC-OPF at scale, with future work aimed at grid reliability and multi-area operations.

Submission Number: 13