A Multi-Layer Learning Architecture for Optimal Power Sharing Between Electric Vehicles and Microgrids

Abstract

A multi layer learning architecture is proposed for optimizing the distribution of electric vehicle power among microgrids. The architecture includes three layers: prediction, coordination, and real-time control. Using a prediction layer to forecast demand loads, available renewable energy generation rates, and vehicle availability, the coordination layer solves a constrained optimization problem to allocate energy resources at multiple charging nodes. In turn, the real-time control layer enforces compliance with feasibility constraints while compensating for forecast inaccuracies via adaptive control and reinforcement learning.The performance of the proposed architecture was evaluated through simulation studies based on representative microgrid use cases demonstrating improved use of renewable energy resources, reduced peak demand, and increased economic returns relative to traditional operational strategies. The proposed architecture is scalable, preserves privacy, accommodates heterogeneous fleets of EVs by providing support for both grid services and the mobility needs of users.Key performance indicators for the proposed architecture include; (1) total energy cost; (2) peak shaving; (3) state/charge deviation; and (4) latency. Sensitivity analyses were conducted to determine the robustness of the proposed architecture to forecasting error and size of the fleet being utilized (i.e., whether large or small fleets are affected) and communication overhead across networked agents.