Data-Driven Deep Reinforcement Learning Control: Application to New Energy Aircraft PMSM

As the power of the new energy aircraft, the performance of permanent magnet synchronous motor (PMSM) directly determines the reliability and stability of the flight state. The complex coupling characteristics (strong nonlinearity, time-varying parameters, multiple working modes) essentially exist in the PMSM speed control system, which makes it difficult for traditional control methods to meet application requirements. Aiming at the PMSM speed control, this paper proposes a novel data-driven deep reinforcement learning method. The system operating data is utilized, and a deep deterministic policy gradient (DDPG) agent is established to fully characterize the system data relationship. Instead of the speed loop and current loop in traditional PID control, DDPG intelligent controller directly outputs the quadrature axis voltage. The problem of poor torque dynamic performance caused by the slow response speed of the current loop in the traditional speed control system is effectively solved. Simulation results demonstrate that, under three different typical operating conditions, the proposed method can meet high requirements of PMSM control system. Compared with the traditional PID controller, the DDPG controller shows excellent dynamic and steady-state characteristics, and has stronger robustness and stability.