Temporal-Feature-Enhanced Reinforcement Learning Control with Adaptive Speed-Loop Gain for PMSM Drives
TL;DR
Comparative simulations with conventional PI control and standard TD3 indicate that the proposed temporal-feature-enhanced reinforcement learning control approach for a PMSM drive system with adaptive regulation of the speed-loop gain improves transient response, reduces speed tracking error, and suppresses torque ripple.
Abstract
The high-performance control of permanent magnet synchronous motor (PMSM) drives is challenged by operating uncertainties and coupled electromechanical dynamics. This paper proposes a temporal-feature-enhanced reinforcement learning control approach for a PMSM drive system with adaptive regulation of the speed-loop gain. The learning agent is integrated into a dual-loop control structure to generate voltage control commands and update the speed-loop gain in real time. A multiobjective learning criterion is designed to balance tracking accuracy, torque smoothness, and gain boundedness. Temporal error features are further embedded in the state representation to capture short-term transient trends with limited additional complexity. Comparative simulations with conventional PI control and standard TD3 indicate that the proposed method improves transient response, reduces speed tracking error, and suppresses torque ripple. These results demonstrate the effectiveness of temporal-feature-enhanced reinforcement learning for coordinated control of PMSM drives.