Comprehensive evaluation of deep reinforcement learning for permanent magnet synchronous motor current tracking and speed control applications

Permanent Magnet Synchronous Motors (PMSMs) are indispensable in industrial applications, requiring precise control to ensure optimal performance. Traditional model-based methods, such as Proportional-Integral (PI) control and Model Predictive Control (MPC), face inherent limitations in robustness and adaptability under complex conditions. Deep Reinforcement Learning (DRL), as a model-free, data-driven approach, offers a transformative solution for PMSM control. This study proposes a DRL-based current control strategy and systematically evaluates the performance of three representative DRL algorithms: Deep Q-Network (DQN), Proximal Policy Optimization (PPO), and Advantage Actor-Critic (A2C) in PMSM control tasks. Key contributions include hyperparameter sensitivity analysis, transfer learning for improved training efficiency, and the application of DRL to multi-objective speed control under varying operational scenarios. Experimental results reveal the hyperparameter sensitivities of different DRL algorithms and provide theoretical insights. The findings demonstrate that transfer learning significantly improves DRL training efficiency and control performance. DRL outperforms traditional controllers in current and speed control, achieving superior dynamic response, tracking accuracy, and adaptability to complex conditions. This study offers new insights into the application of DRL in industrial PMSM control and serves as a reference for its further optimization and practical deployment.