Multiagent Power Flow Control for Plug-and-Play Battery Energy Storage Systems in DC Microgrids

Multiagent reinforcement learning has proven remarkably effective at finding near-optimal solutions to complex non-linear control problems when compared to classical schemes. Such problems typically arise when considering power management problems related to advanced power distribution applications, such as micro/smart grids, smart buildings, electric vehicles, and vehicle-to-grid applications. The achievement of balanced synchronized charge/discharge of energy storage systems in real-time is often a critical factor in fulfilling optimized power flow and enhancing battery health and usable lifetime. It is also critical to reducing power losses, supporting energy/power balance, and integration of renewable/intermittent energy. This paper proposes a control adaptation to optimize the power flow of battery energy systems in a DC autonomous microgrid. Multiagent neighbor-to-neighbor information related to the variation in the load participation and measured state of charge is locally exploited to optimize the balance of power storage. The results confirm accurate synchronization of the charge/discharge and enhanced balanced output voltage under an excessive continuous load variation. In addition, for different expectations of real operation, regarding battery capacities, initial states of charge, environmental impacts, and degradation. Furthermore, the independence of the microgrid operation from the number of battery energy storage systems is verified through plug-and-play insertions and removals.