引用本文:牛善帅,王军政,赵江波,沈伟.基于命令滤波和神经网络的电液伺服系统有限时间反步控制[J].控制理论与应用,2026,43(2):249~258.[点击复制]
NIU Shan-shuai,WANG Jun-zheng,ZHAO Jiang-bo,SHEN Wei.Finite-time backstepping control of electro-hydraulic servo system based on command filter and neural network[J].Control Theory & Applications,2026,43(2):249~258.[点击复制]
基于命令滤波和神经网络的电液伺服系统有限时间反步控制
Finite-time backstepping control of electro-hydraulic servo system based on command filter and neural network
摘要点击 150  全文点击 21  投稿时间：2024-02-28  修订日期：2025-04-11
查看全文  查看/发表评论  下载PDF阅读器   HTML
DOI编号  10.7641/CTA.2024.40122
  2026,43(2):249-258
中文关键词  命令滤波  神经网络  有限时间稳定  反步控制  电液伺服系统
英文关键词  command filter  neural network  finite-time stability  backstepping control  electro-hydraulic servo system
基金项目  国家自然科学基金项目(62173038), 群体协同与自主实验室开放基金课题项目(QXZ23013202)资助.
作者单位E-mail
牛善帅 北京理工大学 自动化学院 niushanshuai@163.com 
王军政* 北京理工大学自动化学院 3120185008@bit.edu.cn 
赵江波 北京理工大学自动化学院  
沈伟 北京理工大学自动化学院  
中文摘要
      为提升电液伺服系统的跟踪控制性能, 本文考虑系统中包括参数不确定性、未建模动态以及外部扰动在内 的集总不确定性, 提出一种基于命令滤波和神经网络的有限时间反步控制方法. 该方法利用Levant微分器作为命令 滤波器, 获取虚拟输入变量与虚拟控制律差分信号的导数, 不仅避免了标准反步控制中的“复杂性爆炸”问题, 还通 过重构实现了对非匹配不确定性的估计; 同时, 与传统基于神经网络的反步控制需要多个神经网络相比, 该方法仅 利用1个神经网络来逼近匹配不确定性, 避免了多个神经网络所导致的控制器的复杂性和脆弱性. 另外, 通过引入 由分数指数幂函数和多项式函数构成的分段函数反馈以加速收敛, 系统实现有限时间稳定的同时避免了奇点问题. 最后, 搭建实验平台开展对比实验验证新方法的有效性和优越性.
英文摘要
      To improve the tracking control performance of electro-hydraulic servo systems, considering the lumped uncertainties, including parameter uncertainty, unmodeled dynamics, and unknown disturbances, a finite-time backstepping control method based on command filter and neural network is proposed. A Levant differentiator is used as the command filter to obtain the derivative of the differential signal between the virtual input variable and the virtual control law, which not only avoids the “explosion of complexity” in standard backstepping control, but also estimates unmatched uncertainty through reconstruction; Compared with traditional neural network-based backstepping control that requires multiple neural networks, this method only uses one neural network to approximate the matched uncertainty, avoiding the complexity and fragility of controllers caused by multiple neural networks. By introducing piecewise feedback functions composed of a fractional exponential power function and a polynomial function to accelerate convergence, the system achieves finite time stability while avoiding singularity problems. Finally, an experimental platform is established to conduct comparative experiments, and the effectiveness and superiority of the proposed new method are verified.