Reinforcement Learning-Based PI Control Strategy for Single-Phase Voltage Source PWM Rectifier

To address issues of lagging dynamic response and complex parameter tuning in traditional double-loop control of single-phase voltage source PWM rectifiers, this research proposes a hierarchical intelligent control strategy integrating reinforcement learning (RL). Firstly, the mathematical model in the d-q rotating coordinate system is established by analyzing the circuit topology of a single-phase VSR. Subsequently, a double-loop control structure comprising a voltage outer loop and a current inner loop is developed: the current inner loop adopts DQ feedforward decoupling control to achieve independent conditions; the voltage outer loop innovatively employs a single-neuron PI controller based on reinforcement learning, which optimizes control parameters in real-time via a deep deterministic policy gradient (DDPG) algorithm, thus forming an adaptive hierarchical control system. Finally, the effectiveness of the strategy is validated through simulation models built on the Matlab/Simulink platform. Simulation results demonstrate superior dynamic performance of the proposed method under load mutation conditions, significantly improving dynamic response quality and steady-state performance.