Online Inductances Estimation of the Permanent Magnet Synchronous Machines based on Deep Learning and Recursive Least Square Algorithms

Abstract

This paper presents a novel method to identify in real time d- and q- axes inductances of the permanent magnet synchronous machines (PMSM), which are normally vary during the operation due to the saturation of the magnetic fields. The proposed method is based on the combination of deep learning and recursive least square algorithms. The deep learning model is trained offline in order to compensate the non-linearity effect of the voltage source inverter, while the recursive least square algorithm was employed to estimate online d- and q- axes inductances based on the compensated d- and q- axes stator voltages, measured d- and q- axes stator currents and the operating speed. The proposed methods can overcome the problems associated with the existing methods known as effect of noise, unavailability of accurate information of the inverter. Extensive experimental studies were conducted to evaluate the estimation accuracy and the robustness of  the proposed method in the critical operating conditions including the variation of load torque, operating speed and field-weakening conditions.