Design and Analysis of a Torque Controller for an IPMSM using Reinforcement Learning

This paper introduces a novel method for torque control in electric drive systems using reinforcement learning (RL). The traditional field-oriented control (FOC) in an interior permanent magnet synchronous machine (IPMSM) is replaced with an RL-based controller developed using the Twin Delayed Deep Deterministic Policy Gradient (TD3) algorithm in Simulink. The RL agent is trained to determine optimal voltage values for the IPMSM in the dq-plane, considering observations and reward functions tailored to the torque control problem. The reward function incorporates constraints to prevent training on negative torque and speed values and exceeding maximum motor values. Comparative simulation results and analysis demonstrate the performance of the RL-based torque controller, emphasizing its advantages over FOC in terms of adaptability and torque response without requiring a precise mathematical model of the system. This research contributes to the advancement of electric drive control, showcasing the potential of RL algorithms to enhance torque behavior and adaptability in electrical drives.