Reinforcement Learning-based Energy Management Strategy for Flexible Hybrid ac/dc Microgrid

The introduction of the three-phase dual-purpose dc-ac/dc power converter (PC), a recent addition to the universal converter family, holds significant promise for improving the reliability and flexibility of modern microgrids (MGs). This paper presents a novel energy management strategy for a flexible hybrid ac/dc MG architecture featuring dual-purpose dc-ac/dc power converters. These converters offer the capability of both dc/ac and dc/dc power conversion, allowing for connection to either the ac bus or the dc bus through simple reconfiguration, thereby providing enhanced modularity and operational adaptability to various MG conditions. Leveraging model-free reinforcement learning (RL) algorithms, the proposed strategy optimizes the operation mode of the dual-purpose converters to minimize the interlinking converter utilization based on historical MG operational data. The feasibility of the proposed MG architecture and the performance of the RL-based energy management strategy are verified by simulation results.