Development of Time Optimal Current Trajectories for Permanent Magnet Synchronous Motors (PMSM) Under Voltage and Current Limit using Reinforcement Learning

This paper presents development of a time optimal current vector trajectories of permanent magnet synchronous motor (PMSM) to improve dynamic performance at voltage and current limits. In order to change the torque of a PMSM drive, it can be controlled directly through the current. To increase the motor’s torque dynamics, a high rate change of current vector is required, and the torque dynamics can be degraded under voltage limit conditions, which is particularly prominent in the high-speed operating range of the motor due to high back-emf voltage. Changing the torque can be represented as altering the current vector from initial point to terminal point. By optimizing the trajectory that connects these two points, torque dynamics of a PMSM drive can be improved. A sequential current vector change can be modeled by Markov Decision Process (MDP). In this paper, Q-learning is used to learn the trajectory in MDP environment and apply the learned trajectory to validate its effectiveness in a PMSM simulation model.