基于单隐层神经网络的压电执行器非线性特征建模策略研究  

On Nonlinear Feature Modeling Strategy of Piezoelectric Actuator Based on Single Hidden Layer Neural Network

作　　者：盖冉翔 汪旭东[1,2] 刘旭辉 姚兆普[1,2] GE Ranxiang;WANG Xudong;Liu Xuhui;Yao Zhaopu(Beijing Institute of Control Engineering,Beijing 100094,China;Beijing Engineering Research Center of Efficient and Green Aerospace Propulsion Technology,Beijing 100094,China)

机构地区：[1]北京控制工程研究所先进空间推进技术实验室,北京100094 [2]北京市高效能及绿色宇航推进工程技术研究中心,北京100094

出　　处：《空间控制技术与应用》2021年第4期103-108,共6页Aerospace Control and Application

基　　金：国家重点研发资助项目(2020YFC2201002,2020YFC2201101)。

摘　　要：研究了压电致动器的几种主要非线性特征的联合建模方案,其中包括迟滞、蠕变和温度漂移.提出了一个利用NARMAX模型的级联单隐层前馈神经网络以消除迟滞的影响,利用信息标准和误差缩减比算法确定对模型误差影响最大的几项回归因子作为网络的输入节点.实验表明,利用多网络泛化和正则化策略,网络在测试数据上的绝对误差可以下降到不高于±0.1μm;通过将运行时间、温度传感器测量值和激励电压频率三项数据加入输入节点,可补偿蠕变和温度漂移导致的非线性因素,将最终在测试集上的绝对误差限制在±0.01μm之内,且对于不同的激励电压频率具有良好的泛化能力.本文的研究成果对于多非线性耦合的压电执行器建模有一定的借鉴意义.The joint modeling scheme of several main nonlinear characteristics of piezoelectric actuator is studied,including hysteresis,creep and temperature drift.A cascaded single hidden layer feed-forward neural network based on NARMAX model is proposed to eliminate the influence of hysteresis.The information criteria and error reduction ratio algorithm are used to determine the regression factors which have the greatest influence on the model error as the input nodes of the network.The experimental results show that the absolute error of the network on the test dataset can be reduced to no more than±0.1μm via the multi-network generalization and regularization.By leveraging the running time,the measured value of the temperature sensor and the frequency of input voltage into the input nodes,the nonlinear factors caused by creep and temperature drift can be compensated,and the final absolute error on the test set can be limited to±0.01μm.The network has good generalization ability for different excitation voltage frequencies.The research results can be used as reference in the modeling of multiple nonlinear coupling piezoelectric actuators.

关 键 词：压电执行器 多非线性耦合 NARMAX模型 神经网络模型 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]