引用本文:王云龙,王泽政,王永富,赵晶.带有干扰观测器的线控转向系统复合自适应神经网络控制[J].控制理论与应用,2021,38(4):433~443.[点击复制]
WANG Yun-long,WANG Ze-zheng,WANG Yong-fu,ZHAO Jing.Composite adaptive neural network control for steer-by-wire systems with disturbance observer[J].Control Theory and Technology,2021,38(4):433~443.[点击复制]
带有干扰观测器的线控转向系统复合自适应神经网络控制
Composite adaptive neural network control for steer-by-wire systems with disturbance observer
摘要点击 2016  全文点击 746  投稿时间：2020-06-11  修订日期：2020-09-20
查看全文  查看/发表评论  下载PDF阅读器
DOI编号  10.7641/CTA.2020.00340
  2021,38(4):433-443
中文关键词  线控转向  神经网络  干扰观测器  李雅普诺夫稳定性  硬件在环
英文关键词  steer-by-wire (SbW)  neural network  disturbance observer  Lyapunov stability theory  hardware-in-loop
基金项目  国家自然科学基金项目(51775103)资助.
作者单位E-mail
王云龙 东北大学 yunlongw@stumail.neu.edu.cn 
王泽政 北京航空航天大学  
王永富* 东北大学 1810131@stu.neu.edu.cn 
赵晶 东北大学  
中文摘要
      考虑车辆线控转向(SbW)系统存在不确定动态特性以及外界干扰影响. 本文提出一种带有干扰观测器的复 合自适应神经网络实现SbW系统的精确建模与稳定控制. 首先, 利用神经网络在线逼近系统不确定动态, 避免控制 器设计中使用到系统模型的先验知识. 然后, 结合系统的跟踪误差与建模误差提出一种新的复合自适应学习率来 更新神经网络的权值, 从而加快跟踪误差的收敛速度. 最后通过设计干扰观测器补偿系统受到摩擦力矩、回正力矩 与神经网络逼近误差的影响, 提高了系统的抗干扰能力. 李雅普诺夫稳定性理论证明了闭环系统的跟踪误差信号 一致最终有界. 数值仿真与硬件在环实验结果验证了该控制方法的有效性和优越性.
英文摘要
      Steer-by-Wire (SbW) systems are usually affected negatively by uncertain dynamics and external disturbance. This paper proposes a composite adaptive neural network with disturbance observer to achieve the accurately modeling and stable control for SbW system. Firstly, the neural network is adopted to approximate system’s uncertain dynamics such that the prior knowledge of uncertain dynamics can be avoided. Secondly, a novel composite adaptive learning law, which is constructed by the tracking error and modeling error, is designed to update the weight of neural network and improve the convergence of tracking error. Finally, a disturbance observer is proposed for the compensation of friction torque, self-aligning torque and the neural network approximated error, which enhances the anti-interference performance of SbW system. Lyapunov stability theory proves that the tracking error is uniformly ultimately bounded. Numerical simulation and hardware-in-loop experiment show the effectiveness and superiorities of the proposed control method.