Hybrid Learning Approaches for the AC-OPF

I. Repiso López, S. Pineda Morente, J. M. Morales González

The AC Optimal Power Flow (AC-OPF) is the cornerstone of power system operations, yet its NP-hard and non-convex nature gives rise to multiple local optima whose impact on solution quality depends critically on the network topology and the loading regime. Following the learning-to-optimize paradigm, this work focuses on the design of hybrid strategies that integrate mathematical programming with data-driven methods to enhance the computational performance of state-of-the-art AC-OPF algorithms. In particular, we combine convex relaxations and more tractable approximations with statistical learning techniques to support the optimization process. These strategies are assessed across IEEE benchmark networks of varying sizes and operating conditions, indicating improved convergence behavior and overall performance, while maintaining interpretability.

Keywords: AC-OPF, Non-convex optimization, Power systems, Hybrid learning approaches, Learning-to-optimize, Statistical learning

SI Optimización y Aprendizaje Estadístico en Energía
September 3, 2026  9:00 AM
Aula 28

M. Fischetti

J. González Díaz, I. Gómez-Casares, P. Belotti, B. Ghaddar

Á. M. González Rueda

S. Pineda, J. M. Morales González