A Reinforcement Learning Controller Based on Double DQN for DC microgrids with Constant Power Loads

The presence of Constant Power Loads (CPLs) in DC microgrid poses significant challenges due to their nonlinear and time-varying characteristics. Additionally, the unpredictable fluctuations in renewable energy sources further complicate the stability of the system. To address these issues, this paper proposes a robust model-free reinforcement learning control method based on double Deep Q-Network (DQN). The proposed control approach is designed to tackle the dynamic nature of CPLs and adapt to changes in the system environment. Specifically, it focuses on stabilizing the DC-link voltage, which is crucial for maintaining the overall stability of the microgrid. In this study, the performance of the double DQN control method is evaluated under various scenarios, including significant changes in load power, source voltage, and power generation conditions. Through simulation studies, we demonstrate the effectiveness of the proposed approach in controlling and stabilizing the DC-link voltage despite the dynamic nature of the system as well as the uncertainties imposed by CPLs and renewable energy sources.