Delay Compensated Reinforcement Learning Predictive Control for Power Converters

This paper proposes a delay compensated reinforcement learning predictive control (DC-RLPC) solution for power converters. More precisely, an actor critic-based intelligent agent is integrated into the predictive control framework, establishing a model-free approach that eliminates the reliance on physical information. It enhances the robustness of model predictive control by mitigating the effects of unknown or unmodeled dynamics. Furthermore, a novel delay compensated Bellman equation is proposed, and the corresponding RL training algorithm is developed to address the digital delay in control systems. It lays the foundation for future practical applications of RL. Finally, the effectiveness of the proposed DC-RLPC is validated through numerical examples.