基于强化学习的主动模型变化死区控制用于含恒功率负载的直流微电网储能系统

The issues of constant power load (CPL) and model mismatch lead to significant current ripples and voltage fluctuations in dc microgrids, presenting a substantial challenge for the control methods of energy storage systems. To address these issues, this article proposes a reinforcement learning based active model variation deadbeat control (RL-AMVDB), which adjusts a few control model parameters in the current control loop. The model variation factors and the states of closed-loop system are formulated as a Markov decision process, allowing the model variation factors to be optimized through RL. The performance of the proposed method is evaluated through both simulation and hardware experiments. Experimental results demonstrate that the proposed method reduces current ripple by 16.6% under model mismatch conditions. Additionally, under sudden CPL changes, the settling time, voltage fluctuations, and current ripple are reduced by 38.4%, 10.2%, and 12.5%–20%, respectively.