Adaptive deep reinforcement learning-based control strategy for high-performance permanent magnet synchronous motor drive systems

• Datum: 2026
• Hauptverfasser: Dukkipati, S., Nagendra, S. S., Kumar, B. H., Parimalasundar, E.
• Format: Artikel
• Sprache: Englisch
• Veröffentlicht: National Technical University "Kharkiv Polytechnic Institute" and Аnatolii Pidhornyi Institute of Power Machines and Systems of NAS of Ukraine 2026
• Schlagworte: deep reinforcement learning; permanent magnet synchronous motor; deep deterministic policy gradient; twin delayed deep deterministic policy gradient; adaptive motor control; actor-critic algorithm
• Online Zugang: https://eie.khpi.edu.ua/article/view/358720
• Назва журналу: Electrical Engineering & Electromechanics

Introduction. In recent days, electric vehicles, robotics and in many control system applications, permanent magnet synchronous motors (PMSMs) are widely utilized. Problem. Due to non-linear behavior of system, external interferences and frequent changes in parameters, conventional control techniques like direct torque control, field-oriented control and PI control, frequently experience decline in performance. Goal. This paper presents a new deep learning based reinforcement learning (RL) PMSM control approach that makes use of the twin delayed deep deterministic policy gradient (TD3) and deep deterministic policy gradient (DDPG) algorithms. These algorithms utilize actor-critic architectures to learn optimal control policies in a model-free manner, enabling adaptive and intelligent motor control. Methodology. A MATLAB/Simulink-based simulation framework is developed to train and evaluate the proposed deep reinforcement learning (DRL) based controllers against conventional PI controllers. Performance metrics, including speed tracking accuracy, torque ripple minimization are analyzed. Results. The results demonstrate that DRL-based controllers exhibit superior adaptability, robustness, and dynamic performance under varying load and speed conditions in contrast to traditional control methods. Notably, the comparative analysis reveals that the TD3 algorithm outperforms DDPG by mitigating overestimation bias, resulting in smoother torque output and more stable control actions. Scientific novelty. This paper illustrates the capability of DRL for advanced PMSM control. Practical value. Paving the way for real-time implementation in modern electric drive systems. References 25, tables 3, figures 12.