一种基于NARX神经网络的振动主动控制方法
An Active Vibration Control Method Based on NARX Neural Network
宋春生 1熊学春 2陈泊远 2杜刚2
作者信息
- 1. 武汉理工大学 机电工程学院,武汉 430070;武汉理工大学 湖北省磁悬浮工程技术研究中心,武汉 430070
- 2. 武汉理工大学 机电工程学院,武汉 430070
- 折叠
摘要
针对主被动混合隔振系统中次级通道的非线性因素和时变特性,设计一种基于有源非线性自回归神经网络(Nonlinear Auto-regressive With Exogenous Inputs Neural Network,NARX-NN)的次级通道系统辨识的方法,并成功应用于振动主动控制系统中.首先,使用NARX神经网络对次级通道进行辨识得到准确的次级通道模型;其次,采用FIR滤波器重构初级通道的输出,从而获得作动器的输出信号,基于重构得到数据对辨识的网络进行在线学习,可以避免由白噪声激励在系统中带来的随机振动对控制效果的影响;最后搭建仿真模型以及实验平台,仿真结果表明,该控制算法可以克服次级通道的时变性导致的次级通道失真问题;实验结果表明,该算法对15、20 Hz的线谱分别取得30.1、40.4 dB的能量衰减效果,能够有效地实现振动主动控制.
Abstract
In order to study the nonlinear factors and time-varying characteristics of the secondary path in the active-passive hybrid vibration isolation system,a system identification method of the secondary path based on the nonlinear auto-regressive with exogenous inputs neural network(NARX-NN)is proposed and successfully applied to active vibration control systems.First of all,the NARX neural network is used for identifying the secondary path and obtaining an accurate secondary path model.Secondly,the FIR filter is applied to reconstruct the output of the primary path and obtain the output signal of the actuator.The data obtained by the reconstruction is used for online learning of the identified network,which can avoid the influence of random vibration caused by white noise injection on the vibration isolation system.Finally,a simulation model and an experimental platform are built.The simulation results show that this control algorithm can overcome the time-varying influence of the secondary path on the active control effect.The experiment results show that the algorithm achieves the energy attenuation effect of 30.1 and 40.4 dB on the line spectra of 15 and 20 Hz respectively,which can effectively realize the active vibration control.
关键词
振动与波/Fx-LMS前馈控制/NARX神经网络/振动主动控制/在线系统辨识Key words
vibration and wave/Fx-LMS feedforward control/NARX neural network/active vibration control/online system identification引用本文复制引用
基金项目
国家自然科学基金(51879209)
中央高校基本科研业务费专项(2021IVA117)
出版年
2024
国家科技期刊平台
NSTL
万方数据