Enhancing TSO-DSO coordination: a machine learning approach for efficient operations planning

This paper explores Machine Learning-Guided Optimization (MLGO) as a framework for integrating Distribution System Operator (DSO) network-connected energy storages into Transmission System Operator (TSO) market models. The framework first simulates storage bidding behaviour under grid constraints using a mixed-integer programming model, then trains a neural network to learn the complex bidding patterns. The trained model is reformulated into linear constraints and embedded within a Quadratically Constrained AC Optimal Power Flow problem. The case study considers complex bidding strategies for storage assets in New York Independent System Operator's (NYISO) energy and ancillary services markets, focusing on the Southeast New York region with high renewable penetration. The results show that MLGO can reduce computation time by 50% compared to traditional optimization approaches while maintaining solution quality, making it particularly valuable for day-ahead market clearing with distributed energy resources.