Assessing deep reinforcement learning control of permanent magnetassisted synchronous reluctance machines

Abstract

This paper presents an application of deep reinforcement 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.