基于递归神经网络的超声波电机力矩迟滞辨识  

Torque Hysteresis Identification of Ultrasonic Motor Based on Recurrent Neural Network

作　　者：傅平[1,2] FU Ping(Department of Physics and Telecommunication Engineering,Minjiang University,Fuzhou 350121,China;Fujian Provincial Education Department Key Laboratory of Advanced Motion Control,Minjiang University,Fuzhou 350121,China)

机构地区：[1]闽江学院物理学与电子信息工程学院,福州350121 [2]闽江学院福建省教育厅先进运动控制重点实验室,福州350121

出　　处：《微特电机》2022年第11期8-15,共8页Small & Special Electrical Machines

基　　金：国家自然科学基金项目(51277091);福建省自然科学基金资助项目(2020J01841)。

摘　　要：超声波电机的输入输出变量呈现强非线性特性,其速度-力矩之间存在迟滞,且随着驱动频率和负载等因素改变,用通常的辨识难以取得满意的效果。针对超声波电机的速度-力矩迟滞,使用基于递归神经网络和李亚普诺夫稳定性的方法,可以在一定程度上反映电机的迟滞特性。整个系统使用基于半实物仿真的超声波电机测试平台,其中迟滞辨识采用递归神经网络辨识器(RNNI)。由于RNNI的参数可以在线进行调整,因此当电机输入输出参数发生变化时,通过改变RNNI的参数可以实现不同迟滞特性的辨识,同时利用李亚普诺夫稳定性方法进行RNNI的参数调整。实验结果表明,递归神经网络辨识器通过改变神经网络参数对超声波电机迟滞可以进行有效的辨识,MSE小于6×10^(-4),不同负载下辨识误差小于0.11。The input and output variables of ultrasonic motor show strong nonlinear characteristics,and there exits hysteresis between torque and velocity of motor.The satisfactory results are difficult to be obtained by conventional identification.A control method based on recursive neural network and Lyapunov stability was proposed to reflect the nonlinear hysteresis of ultrasonic motor to a certain extent.The whole system included the ultrasonic motor test platform based on the hardware-in-the-loop simulation,and the hysteresis was identified by the recursive neural network identifier(RNNI).Because recurrent neural network could adjust its online parameter values when operation conditions of the motor changed,control parameters could be adjusted by changing the parameters of neural network identification and the parameters could be adjusted by Lyapunov stability.Experimental results show that the hysteresis of ultrasonic motor can be identified effectively by changing the parameters of neural network.The MSE was less than 6×10^(-4),and the identification error was less than 0.11 under different loads.

关 键 词：超声波电机 递归神经网络 李雅普诺夫稳定性 半实物仿真 迟滞 

分 类 号：TM355[电气工程—电机]