引用本文:霍本岩,李祯,刘艳红,赵新刚,张宇波.面向上肢康复的改进Hammerstein模型与最优迭代学习控制[J].控制理论与应用,2026,43(3):587~595.[点击复制]
HUO Ben-yan,LI Zhen,LIU Yan-hong,ZHAO Xin-gang,ZHANG Yu-bo.Improved Hammerstein model and optimal iterative learning control for upper limb rehabilitation[J].Control Theory & Applications,2026,43(3):587~595.[点击复制]
面向上肢康复的改进Hammerstein模型与最优迭代学习控制
Improved Hammerstein model and optimal iterative learning control for upper limb rehabilitation
摘要点击 511  全文点击 76  投稿时间：2023-12-29  修订日期：2025-12-18
查看全文  查看/发表评论  下载PDF阅读器   HTML
DOI编号  10.7641/CTA.2024.30836
  2026,43(3):587-595
中文关键词  功能性电刺激  Hammerstein模型  迭代学习控制  上肢康复系统
英文关键词  functional electrical stimulation  Hammerstein model  iterative learning control  upper limb rehabilitation system
基金项目  国家重点研发计划项目(2022YFB4703203),国家自然科学基金青年科学基金项目(62103376)资助.
作者单位E-mail
霍本岩* 郑州大学电气与信息工程学院 huoby@zzu.edu.cn 
李祯 郑州大学电气与信息工程学院  
刘艳红 郑州大学电气与信息工程学院  
赵新刚 中国科学院沈阳自动化研究所机器人与智能系统全国重点实验室  
张宇波 郑州大学电气与信息工程学院  
中文摘要
      功能性电刺激(FES)是一种临床上常用的康复技术,但目前基于FES的康复设备需要由康复医师根据经验 控制,控制精度难以保证.本文使用参数递推辨识方法建立上肢肌骨系统的Hammerstein模型,进而提出了基于数据 驱动Hammerstein模型的最优迭代学习控制方法. 首先,分别使用神经网络(NN)、传递函数表示肌肉收缩过程中的 招募曲线(IRC)与线性激活动态(LAD),建立预训练数据集并对模型参数进行递推辨识,从而建立上肢肌骨系统的 改进Hammerstein模型;然后,根据模型的IRC与LAD分别设计NN非线性控制器与参数最优迭代学习控制器 (POILC), 得到NN与POILC串联控制器(NPOILC);最后,搭建上肢康复系统实验平台,招募5位受试者对所提出的方 法进行实验验证.实验结果表明:相较于传统的PD型迭代学习控制器和多项式与POILC串联控制器,本文提出 的NPOILC最终迭代的均方根误差分别由8.02?与9.74?降低至4.87?,分别下降了39.3%与50.0%;收敛迭代数分别由 平均5.0次与4.3次减至3.4次,分别下降了32.0%与20.9%. 实验结果验证了本文方法的有效性,实现了上肢康复系统 的建模与控制,且具有较高的控制精度与较快的收敛速度.
英文摘要
      Functional electrical stimulation (FES) is a widely used rehabilitation technique clinically. However, currently FES-based devices require control by rehabilitation physicians based on their experience, making it difficult to guarantee control accuracy. To solve this problem, this paper employs a recursive identification method of parameters to establish a Hammerstein model of the upper limb musculoskeletal system. And then, an optimal iterative learning control method based onthis model is proposed. Initially, the neural network (NN) and a transfer function are used to represent the isometric recruitment curve (IRC) and linear activation dynamics (LAD) of muscles, respectively. A pre-trained dataset is established and the model parameters are recursively identified based on the dataset, establishing an improved Hammerstein model for the upper limb musculoskeletal system. Subsequently, a NN controller and a parameter optimal iterative learning controller (POILC)aredesigned based ontheIRCandLADofthemodel,respectively, resulting in the NN-POILCcascadedcontroller (NPOILC). Finally, an experimental platform is built, and five subjects are recruited to verify the proposed methods. The experimental results show that compared with PD-ILC and the Polynomial-POILC, the proposed NPOILC reduces the RMS error of the final iterations from 8.02? and 9.74? to 4.87?, marking a reduction of 39.3% and 50.0%, respectively. The experimental results verify the effectiveness of the method proposed in this paper, achieving modeling and control of the upper limb rehabilitation system, with higher control accuracy and faster convergence rate.