Assessing deep reinforcement learning control of permanent magnet-assisted synchronous reluctance machines
Cristina Martín, M. A. González-Cagigal, Álvaro Rodríguez del Nozal and Juan M. Mauricio Department of Electrical Engineering, E.T.S.I., Seville University Seville (Spain) 2026-02-15

Abstract

This paper presents an application of deepreinforcement learning (DRL) for controlling permanent magnet-assisted synchronous reluctance machines (PMA-SynRMs). A model-free DRL agent is trained to control the power converter switching states, aiming to accurately track current references. The DRL-based control scheme is compared against a traditional finite control set model predictive control (FCS-MPC) strategy employing a simplified linear model of the PMA-SynRM. Simulation results demonstrate that the DRL controller achieves superior performance in terms of tracking accuracy and harmonic distortion reduction, effectively handling the machine's inherent nonlinearities. Furthermore, the DRL agent exhibits robustness against measurement errors. The findings highlight the potential of DRL as a viable alternative to conventional model-based control methods for high-performance PMA-SynRM drives, offering improved adaptability, robustness, and operational flexibility.

Key words: Artificial neural networks, deep reinforcement learning, electrical drives, model predictive control, permanent magnet-assisted synchronous reluctance machines.

References

[1] M. D. Nardo et al., “Permanent Magnet Assisted Synchronous Reluctance Machine Design for Light Traction Applications”, IEEE Transactions on Industry Applications (2024), Vol. 60, no. 4, pp. 6079-6091.

[2] S. Vazquez, J. Rodriguez, M. Rivera, L. G. Franquelo and M. Norambuena, “Model Predictive Control for Power Converters and Drives: Advances and Trends,” IEEE Transactions on Industrial Electronics (2017), Vol. 64, no. 2, pp. 935-947.

[3] Li, Z., Su, J., Gao, H., Zhang, E., Kuang, X., Li, C., Bi, G., and Xu, D. “Sensorless Control of PMaSynRM Based on Hybrid Active Flux Observer”, Electronics (2025), Vol. 14, no. 2, 259.

[4] M.A. González-Cagigal, C. Martín, M. Bermúdez, P. Cruz-Romero, “Comparison of nonlinear Kalman filtering schemes for sensorless control of permanent magnet-assisted synchronous reluctance machines”, International Journal of Electrical Power & Energy Systems (2025), Vol. 165.

[5] K. Tan, J. Su, B. Zhong and G. Yang, “A Computationally Efficient Full-Speed Domain Control Method for PMaSynRM Considering Magnetic Saturation”, IEEE Transactions on Power Electronics (2025), early acess.

[6] M. Schenke, W. Kirchgässner and O. Wallscheid, “Controller Design for Electrical Drives by Deep Reinforcement Learning: A Proof of Concept” IEEE Transactions on Industrial Informatics (2020), Vol. 16, no. 7, pp. 4650-4658.

[7] S. Bhattacharjee, S. Halder, Y. Yan, A. Balamurali, L. V. Iyer and N. C. Kar, “Real-Time SIL Validation of a Novel PMSM Control Based on Deep Deterministic Policy Gradient Scheme for Electrified Vehicles”, IEEE Transactions on Power Electronics (2022), Vol. 37, no. 8, pp. 9000-9011.

[8] G. Book et al., “Transferring Online Reinforcement Learning for Electric Motor Control From Simulation to Real-World Experiments”, IEEE Open Journal of Power Electronics (2021), Vol. 2, pp. 187-201.

[9] M. Schenke, B. Haucke-Korber and O. Wallscheid, “Finite-Set Direct Torque Control via Edge-Computing-Assisted Safe Reinforcement Learning for a Permanent-Magnet Synchronous Motor,” IEEE Transactions on Power Electronics (2023), Vol. 38, no. 11, pp. 13741-13756.

[10] Yiming Zhang, Jingxiang Li, Hao Zhou, Chin-Boon Chng, Chee-Kong Chui, Shengdun Zhao, “Comprehensive evaluation of deep reinforcement learning for permanent magnet synchronous motor current tracking and speed control applications”, Engineering Applications of Artificial Intelligence (2025), Vol. 149.

[11] Sutton, R. S. & Barto, A. G. (2018). Reinforcement Learning: An Introduction (2nd ed.). MIT Press.

[12] Hessel, M., Modayil, J., Van Hasselt, H., Schaul, T., Silver, D., & Sutton, R. (2017). Rainbow: Combining Improvements in Deep Reinforcement Learning. In Advances in Neural Information Processing Systems (NeurIPS), 30.