Online Cooperative Optimal Power Scheduling of Multiple Microgrids via Hierarchical Imitation Learning

This paper looks into the issue of optimal power scheduling of multiple microgrids using hierarchical imitation learning. The system is designed to be a hierarchical learning model towards a two-level microgrid community (MGC) structure. The upper-level MGC agent uses an imitation learning algorithm to schedule exchange power between different microgrids, while lower-level microgrid agents are controlled by individual energy management systems using mixed-integer linear programming (MILP). In this paper, we focus on achieving economic dispatch in a large microgrid community while maintaining the privacy of the local microgrids. A simulation study of hierarchical imitation learning is provided based on an MGC system. Our results show the outstanding performance of the designed algorithm with a cost close to the centralized optimal results, about 10% improvement compared to the offline method, and very fast execution, which would be suitable for online power scheduling.