Optimal Generation Cost (OGC) is a key factor in the efficient operation of DC Microgrid (DC MG). OGC in Microgrids (MGs) is influenced by the Optimal Power Flow (OPF) problem, which is inherently non-linear and non-convex. Traditional methods, often operating on different time scales, struggle with real-time implementation and rely on precise system models, making them vulnerable to uncertainties and disturbances. Recent advances in Deep Reinforcement Learning (DRL) offer alternative data-driven methods to handle these complexities. This study proposes a distributed optimal control framework for DC MGs using the Multi-Agent Soft Actor-Critic (MASAC) algorithm. The approach learns optimal control policies without requiring accurate system models, enabling Soft Actor-Critic (SAC) agents to interact continuously with the MG environment. Simulation results demonstrate that the MASAC-based controller outperforms previous methods, such as the Twin Delayed Deep Deterministic Policy Gradient (TD3) algorithm, achieving lower generation costs and improved performance under uncertainties.
