High-performance servo control of PMSM under low-speed rotation conditions based on reinforcement learning

Achieving high-precision servo control performance of permanent magnet synchronous motors (PMSM) under low-speed rotation conditions is a prerequisite for establishing high-performance satellite laser communication links. The key technology lies in the accurate description of the nonlinear characteristics of the PMSM rotational state. This paper designs a deep deterministic policy gradient (DDPG) nonlinear controller capable of handling continuous action spaces, utilizing gradient descent methods to train both the critic and actor neural networks separately, thereby achieving precise fitting of nonlinear mappings. Through Simulink simulations, the results demonstrate that compared to traditional proportional-integral linear controllers, the DDPG control exhibits shorter response and settling times, smaller tracking errors when following reference low-speed signals; it also shows faster q-axis current response and smaller d-axis current fluctuations when torque is applied, effectively improving the PMSM servo control performance under low-speed operating conditions.