Distributed Secondary Control Based Cooperative Deep Reinforcement Learning of Microgrids

The inherent diversity and complexity of distributed generators in DC microgrids render primary control insufficient for achieving effective voltage restoration. A secondary control leveraging multi-agent reinforcement learning is introduced to concurrently tackle both voltage restoration and current equalization challenges. Within this framework, the compensation parameters of the controller are formulated as the action sets of the intelligent agents, thereby reformulating the voltage restoration and current equalization issues into an optimization based control problem. A dual-stage reward function-driven reinforcement learning approach is further developed to dynamically adjust the control strategy using voltage and current error metrics, effectively reducing bus voltage fluctuations and current equalization time.