The traditional PID controller in vector control of permanent magnet synchronous motor (PSMS) is difficult in linear calibration and cannot adapt well to the complex working condition environment, which will lead to the decline of system performance, system stability, slower system response, lower system control accuracy, lower motor service life, and lower system safety. Considering that deep reinforcement learning has a strong adaptive ability in the face of complex environments, a control method is proposed to optimize the PID parameters using (deep Q-network) DQN. The algorithm is implemented in MATLAB and the PID parameters are rectified, which are substituted into Simulink to simulate the PMSM vector control model, obtaining a PID parameter with an overshoot of 11.2%, a performance enhancement of 0.015 s in the regulation time, and a better performance with the addition of load. The simulation results show that the DQN-PID algorithm has better precision and accuracy than the PID tuning.