Rotor position estimation method of PMSM based on recurrent neural network

Abstract

To obtain an accurate rotor position is an important part of sensorless control of permanent magnetic synchronous machine (PMSM) since PMSM is usually operated under speed varying conditions. However, the existing methods cannot provide a satisfactory tracking performance under harmonic polluted condition in wide operating range, which degrade the position detecting accuracy. In this paper, a recurrent neural network (RNN) based phase detection method is used for rotor position estimation of PMSM. To tackle the nonlinear problem in wide operating range of PMSM, RNN is used as a nonlinear dynamic system to establish the mapping relationship of the voltages of stator and rotor position. The proposed method can also provide frequency-adaptive filtering capability to remove the affect of harmonic. To validate the method, a comparison simulations with conventional SRF-PLL method are carried out under speed varying and harmonic disturbances conditions. The results confirm the effectiveness of the proposed method.

Keywords

• Sensorless control
• Permanent magnetic synchronous machine
• Rotor estimation