Permanent Magnet Synchronous Motor Speed Control by Attention Mechanism Reinforcement Learning

The permanent magnet synchronous motor (PMSM) is widely utilized across various domains due to its high power density, extensive speed range, and reliability. However, the performance of traditional PID controllers is influenced by system parameters, necessitating manual parameter tuning through trial and error to achieve the desired control outcomes. This process requires significant human effort and time investment. To address these challenges, an improved deep deterministic policy gradient (PPO) self-adaptive controller has been proposed for PMSM control. This controller leverages reinforcement learning along with attention mechanisms. It combines deep neural networks, the Actor-Critic algorithm, and attention mechanisms to achieve autonomous learning and self-tuning of system parameters, ultimately leading to optimal control. Simulation experiments demonstrate that the improved PPO model controller offers distinct advantages over PID controllers in reaching the target speed. It can bring the motor to the target speed more rapidly without noticeable speed transients while also exhibiting a degree of generalization, allowing it to perform well under various conditions.