Affiliations: School of Automotive Engineering, Faculty of Vehicle Engineering and Mechanics, State Key Laboratory of Structural Analysis for Industrial Equipment, Dalian University of Technology, Liaoning Province, China
Correspondence:
[*]
Corresponding author. Jing Lian, Automotive Engineering Department, Dalian University of Technology, Dalian, China. E-mail: [email protected].
Abstract: The efficiency and control accuracy of Interior Permanent Magnet Synchronous Motor (IPMSM) are the main factors affecting performance. Manual calibration has the disadvantage of high work intensity, long calibration period and high technical requirement, which leads to low calibration accuracy and motor efficiency. Thus, a novel calibration method based on Deep Deterministic Policy Gradient (DDPG) and Long Short-Term Memory (LSTM) is proposed. By constructing a deep reinforcement learning network, the self-optimization of the optimal working point under any working condition is realized, and the MAP for IPMSM in full speed-torque range is obtained. The method can be used to quickly realize the optimal matching of d-q axis current with arbitrary stator current. It focuses on solving the problem of motor overheating caused by long adjustment time of manually calibrated MAP when the motor is overloaded, to realize fast calibration in overload area. Moreover, the method reduces the dependence on the motor parameters and increases the adaptability of the calibration MAP data to the operating conditions. The simulation and bench test indicate that the method can meet the response requirements of motor torque, and results reveal that the motor efficiency is greatly improved.
