Learning optimization proxies for large-scale Security-Constrained Economic Dispatch

The Security-Constrained Economic Dispatch (SCED) is a fundamental optimization model for Transmission System Operators (TSO) to clear real-time energy markets while ensuring reliable operations of power grids. In context growing operational uncertainty, due increased penetration renewable generators and distributed resources, operators must continuously monitor risk in real-time, i.e., they quickly assess the system’s behavior under various changes load production. Unfortunately, systematically solving an problem each such scenario not practical given tight constraints operations. To overcome this limitation, paper proposes learn proxy SCED, Machine Learning (ML) that can predict optimal solution SCED milliseconds. Motivated by principled analysis market-clearing optimizations MISO, novel just-in-time ML pipeline addresses main challenges learning solutions, variability load, output production costs, as well combinatorial structure commitment decisions. A combined classification-plus-regression architecture also proposed, further capture solutions. Numerical experiments are reported on French transmission system, demonstrate approach’s ability produce, within time frame compatible with operations, accurate proxies produce relative errors below 1%.