Improvement Performances of Sensorless Control for PMSM Based on FOC Strategy Using MRAS Observer, Ant Colony Optimization, and RLTD3 Agent

Begining from the need to increase the accuracy of Permanent Magnet Synchronous Motor (PMSM) rotor speed estimation in the case of using of sensorless control, this article shows a PMSM sensorless system control structure that embeds a Field Oriented Control (FOC) type control approach and a Model Reference Adaptive System (MRAS)-type observer. Furthermore, to achieve superior MRAS observer estimation performance, an Ant Colony Optimization (ACO) computational intelligence algorithm is used to perform optimal settling of the PI controller parameters embedded in the MRAS observer. Another method to increase the estimation performance for the MRAS observer is to use a special machine learning like Reinforcement Learning (RL) namely Twin-Delayed-Deep-Deterministic Policy Gradient (TD3)-type agent that can provide additional signals after proper training to obtain a more accurate estimate for the rotor speed of PMSM. The control structures, operating equations and control algorithms are presented. The validation of the proposed improvement methods for PMSM rotor speed estimation is achieved through numerical simulations using Matlab/Simulink programming environment.