永磁同步电机神经网络谐波注入转矩脉动抑制方法

Torque Ripple Suppression Method of PMSM with Neural Network Harmonic Injection

摘要:

为了抑制永磁同步电机转矩脉动，提出一种基于偏差电流输入的自适应线性神经网络谐波注入方法。该方法以交轴和直轴电流偏差作为神经网络环节输入，应用 2 个自适应线性神经网络环节拟合电压谐波补偿分量，并将其注入到电流环控制器输出，用以抵消交轴电压和直轴电压中的谐波分量，进而抑制转矩脉动。为验证本方法有效性，对比进行了 3 种控制方法的仿真和实验。结果表明，与传统双闭环控制相比， 本方法可将转矩脉动降低约 97%。相比于传统神经网络谐波注入法，本方法可获得相近的转矩脉动抑制效果， 并且控制结构更为简单，参数整定方便，易于工程实现，算法运行时间缩短约 60%。

Abstract:

In order to suppress the torque ripple of permanent magnet synchronous motor(PMSM)，an adaptive linear neural network(CDI-Adaline)harmonic injection method based on deviation current input is proposed. The method uses the q-axis and d-axis current deviations as the input of neural network，and applies 2 adaptive linear neural network links to fit the voltage harmonic compensation components and injects them into the output of the current loop controller to offset the harmonic components in the q-axis and d-axis voltages，thereby suppressing torque ripples. To verify the effectiveness of this method，simulations and experiments of 3 control methods were conducted and compared. The results show that，compared with the conventional double closed-loop control，this method can reduce torque ripple by about 97%. Furthermore，compared with the conventional neural network harmonic injection method，this method can obtain similar torque ripple suppression effect with simpler control structure，more convenient parameter adjustment and easier engineering implementation，and its algorithm running time is reduced by about 60%.