ML-driven optimization for the 'Unit Commitment' problem in power systems

The "Unit Commitment" (UC) problem is essential for optimal power generation in power system dispatch and operations. However, solving this mixed-integer nonlinear programming problem efficiently is becoming increasingly challenging, particularly with the growing integration of renewable energy sources. Current solvers often struggle, causing inefficiencies in power system optimization.

This project leverages machine learning to enhance the scalability and efficiency of UC solutions. We will implement graph neural networks to find near-tight linear approximations of nonlinear constraints and to partition the problem into smaller subproblems for easier management. These advancements aim to enable faster and more accurate power system optimization and will be developed in consultation with electricity dispatchers, software solutions providers, as well as other key stakeholders.