Learning to Accelerate Globally Optimal Solutions: Applications in the AC Optimal Power Flow Problem

Fatih Çengil

The enforcement of thousands of nonlinear quadratic constraints makes the Quadratic Convex (QC) relaxation of AC Optimal Power Flow (AC-OPF) prohibitively expensive under varying load conditions. Yet, in any given instance, only a small fraction of these constraints is active at the optimum. We therefore propose a novel two-step approach to identify the minimal “active” subset of quadratic constraints needed to preserve bound quality.
First, an optimality-based exact policy formulates this as a bilevel max–min problem: the outer maximization selects a fixed-size subset of constraints, while the inner minimization solves the corresponding reduced QC relaxation. By applying polyhedral outer approximations and invoking strong duality on the inner problem, we convert the bilevel program into a single-level mixed-integer formulation whose global solution yields the optimal active-constraint set.
Second, to achieve scalability, we train a machine-learning surrogate that maps load-profile features to the active-constraint mask learned by the exact policy. On IEEE cases up to 793 buses, our method achieves up to 79% speed-up (67% on average) with negligible impact on the relaxation quality.

Short Bio

Fatih Çengil received his B.S. in Industrial Engineering from Middle East Technical University, his M.Sc. from Özyeğin University, and his Ph.D. in Industrial Engineering from the University of Arkansas. His research focuses on optimization and machine learning, particularly their applications in energy systems, and includes research experience at Los Alamos National Laboratory. Prior to academia, he worked in industry at Vestel Electronics and Philip Morris International on production and process improvement initiatives. Following his Ph.D., he served as an Instructional Assistant Professor at Texas A&M University, teaching courses in supply chain management, logistics, inventory systems, and quality. His work bridges advanced optimization research with practical industrial applications.

Venue

Friday, June 12, 2026, 4.00 pm

IE Building, Halim Doğrusöz Auditorium (IE 03)