AI4OPT's Work On Dual Proxies Recognized

Georgia Tech Ph.D. student Guancheng Qiu (Ivan) received the Best Paper Award in the Energy Systems Division at the 2025 IISE Annual Conference for his work titled “Dual Conic Proxy for Semidefinite Relaxation of AC Optimal Power Flow.” The paper, co-authored with Mathieu Tanneau and Pascal Van Hentenryck, introduces an innovative approach to solving challenging power systems problems using a machine learning technique called dual proxies.

Exploring Dual Proxies for AC-OPF

The research explores the AC Optimal Power Flow (AC-OPF) problem, a central task in power systems operations that involves finding the most cost-effective way to dispatch electricity across a network while satisfying physical and operational constraints. Solving AC-OPF often involves semidefinite programming (SDP), a powerful mathematical method that can be computationally expensive. This paper proposes learning dual feasible solutions—termed “dual proxies”—to approximate the solution to these SDPs more efficiently.

A Foundational AI4OPT Contribution

Dual proxies offer a novel way to approximate the dual solutions of SDPs using a conic neural architecture that preserves feasibility, making them not only accurate but also reliable. By learning in the dual space, this method achieves near-optimal results while avoiding the complexity of solving the full SDP in real time. The approach is generalizable, with potential applications far beyond power systems.

Recognition at a Premier Engineering Venue

The award from the Institute of Industrial and Systems Engineers (IISE) highlights the paper’s impact on the future of energy systems modeling and optimization. The IISE Annual Conference is a leading venue for industrial and systems engineering research, and the recognition underscores NSF AI4OPT’s leadership in AI-driven solutions for infrastructure resilience and energy efficiency.