Improvement Performance of FOC for PMSM Based on Reinforcement Learning TD3 Agent Current Controller

This paper proposed an improved filed-oriented control (FOC) strategy for permanent magnet synchronous motors (PMSMs) by integrating a reinforcement learning-driven twin delayed deep deterministic policy gradient (TD3) agent as the current controller to address the nonlinearities inherent in motor drive systems and to enhance overall control effectiveness. The reward function is designed with a focus on penalizing current tracking errors, control effort, control signal smoothing, and integral current error. Furthermore, the TD3 agent is trained at multiple operating points, allowing it to effectively handle the nonlinearities associated with FOC for PMSMs. This hybrid control configuration minimizes computational complexity, improves adaptability, and mitigates the sensitivity to parameter tuning. Simulation results validate these objectives, demonstrating successful speed and current control across various operating points and disturbance scenarios.