Implementation of deep reinforcement learning in permanent magnet synchronous motors control: A review

Permanent Magnet Synchronous Motors (PMSMs) are recognized for high efficiency, torque-to-inertia ratio, and robust properties, making them ideal for the rapid development of electric vehicles, robotics, and the aerospace industry. Recently, Deep Reinforcement Learning (DRL) algorithms have gained significant attention in the control domain due to their independence from plant models and advanced decision-making capabilities. These features make DRL highly suitable for addressing challenges in PMSM control such as load disturbances, speed tracking, and parameter variations. This review explores recent DRL techniques applied to PMSM speed, current, and torque control. Discrete and continuous algorithms, including Deep Q-Network (DQN), Deep Deterministic Policy Gradient (DDPG), and Twin Delayed DDPG (TD3), are examined in terms of their basic principles, practical implementations, and the benefits they provide in overcoming challenges in PMSM control. In addition, to demonstrate the efficiency of DRL, the review provides a summary and comparison of DRL applied to optimize classical control methods elaborated within various PMSM control strategies. Comparisons of DRL implementations in PMSM control are highlighted to validate their real-time applicability in experiments, and potential areas for future research and improvement are outlined.