A Multi-Agent Reinforcement Learning-based Secondary Control for Voltage Restoration and Current Sharing in DC Microgrids

Microgrids are designed for sustainable energy generation and the efficient use of Distributed Generators (DGs). DC microgrids face challenges such as current sharing and voltage restoration, commonly addressed using droop control. Despite its simplicity, droop control can lead to voltage deviations and inefficient current sharing. This paper develops a Multi- Agent Reinforcement Learning (MARL) based secondary control method using Deep Q- Networks (DQN) to improve voltage restoration and current sharing in DC microgrids. The proposed method leverages centralized training to utilize comprehensive environmental information, integrating both the voltage restoration and current sharing objectives into each agent's local reward function. Simulation results demonstrate the effectiveness of the MARL approach in addressing the limitations of droop control, ensuring stable current flow and enhancing overall system performance.