Description

U.S. Independent System Operators (ISOs) rely heavily on Mixed-Integer Linear Programming (MILP) to support critical operational decisions, including Security-Constrained Unit Commitment (SCUC) and Security-Constrained Optimal Power Flow (SCOPF). These tools determine which generators run, how much power they produce, and how to maintain system reliability. While MILP models are computationally efficient and scalable, they require significant simplifications of the physical power system. This creates an accuracy–solvability tradeoff: to ensure fast computation, important nonlinear dynamics and uncertainty effects are simplified or ignored, resulting in economic inefficiencies.

This project explores a new paradigm: Learning to Optimize for large-scale Mixed-Integer Nonlinear Programming (MINLP) problems in Unit Commitment. By combining machine learning with structured optimization, the goal is to solve large-scale nonlinear problems efficiently ...