Transient Recovery of Energy Storage Balance in DC Microgrids with MARL-Based Power Control

Multi-Agent Reinforcement Learning (MARL) - based primary -secondary-control has been shown to exhibit high-performance in Microgrid-based power and energy applications. However, in highly dynamic environments, such as within vehicle-to-grid applications, loss of performance can occur. Specifically, transient loss of accuracy in the synchronization of energy-storage-balance after dynamic topology changes is a known defect, which can overload batteries and reduce stability margins. In this work a newly developed methodology for transient recovery and fault ride-through in battery-based DC-Microgrids is developed and validated. Specifically, an enhancement to MARL-based control utilizing a planned policy with compensation for the DC infrastructure influence is developed, and regional assessment of energy-flow efficiency is examined. The real-time results with quasi-random battery insertion and removal under realistic environmental conditions confirms a reduction in transient recovery time (0.66-13.366%), coupled with enhanced voltage stability (2.637-3.24%) and smoothness (2.9739-3.8462%), better load steadiness (6.666-37.091%), energy saving (2.94%), energy-flow balance enrichment (6.468 %), and raised efficiency (2.626%).