Improved Performance for PMSM Control Based on Robust Controller and Reinforcement Learning
Improved Performance for PMSM Control Based on Robust Controller and Reinforcement Learning
TLDR
The Reinforcement Learning Twin-Delayed Deep Deterministic Policy Gradient (RL-TD3) agent is the most suitable for machine learning for process control, and a robust controller whose control quantities are adjusted by a properly created and trained RL- TD3 agent is synthesized.
摘要
This article presents a Permanent Magnet Synchronous Motor (PMSM) control system which retains its performance for a significant variation of the parameters and load torque which represent disturbance for the control system. Classically, the PMSM control system is built in the form of a Field Oriented Control (FOC) control strategy structure built around PI speed (outer loop) and current (inner loop) controllers. We present the design stages and the numerical simulations performed in Matlab/Simulink, which prove the superiority of the robust control, by comparison with the classic FOC-type control structure. Because the Reinforcement Learning Twin-Delayed Deep Deterministic Policy Gradient (RL-TD3) agent is the most suitable for machine learning for process control, we synthesize a robust controller whose control quantities and are adjusted by a properly created and trained RL-TD3 agent. Using this robust combined controller plus RL-TD3 agent, superior performance is achieved in terms of response time and speed ripple.
