基于软Actor-Critic算法的强化学习控制器改进交错并联DC-DC升压变换器电压调节

This paper proposes the use of a soft Actor-Critic (SAC) algorithm based reinforcement learning (RL) controller as the only primary controller to improve the dynamic performance of the output voltage of the three-phase interleaved DC-DC boost converter. The advantages of maximum entropy learning are discussed, and the principles of the SAC algorithm are elucidated. Design schemes for neural networks (NNs) and reward functions are provided. The SAC based RL agent is trained offline and the stability analysis is conducted at the operating point. The agent is deployed on a physical platform for testing. Comparative analysis with existing methods demonstrates the effectiveness of this approach in improving voltage control capability in the interleaved converter while exhibiting strong robustness to variations in converter parameters, reference values, and loads.