结合前馈调参与迭代学习的数据驱动控制方法
A data-driven control method combining feedforward tuning and iterative learning control
摘要点击 30  全文点击 26  投稿时间:2019-06-27  修订日期:2019-12-31
查看全文  查看/发表评论  下载PDF阅读器
DOI编号  10.7641/CTA.2019.90484
  2020,37(6):1367-1376
中文关键词  迭代前馈调参  迭代学习控制  数据驱动控制  基函数
英文关键词  iterative feedforward tuning  iterative learning control  data-driven control  basis function
基金项目  
作者单位E-mail
薄雨蒙 中国科学院大学,中国科学院长春光学精密机械与物理研究所 bym619@mail.ustc.edu.cn 
曹明生 中国科学院大学,中国科学院长春光学精密机械与物理研究所  
高慧斌 中国科学院长春光学精密机械与物理研究所  
中文摘要
      在前馈控制中, 需要尽可能的去除前馈控制器对系统模型的需求, 同时保证高精度和鲁棒性. 本文提出了 一种数据驱动的将迭代前馈调参与迭代学习控制进行结合的方法, 通过引入基函数参数化的前馈控制器和输入整 形滤波器, 使用梯度下降法求解最优系统前馈控制器, 消除期望轨迹引入的扰动; 通过迭代学习控制, 消除系统重复 性扰动, 进一步提高控制精度. 算法具有不依赖系统模型, 高精度, 适用于变轨迹任务的优点. 文中给出了相应的仿 真, 并应用到一个直线电机系统, 通过实验验证了算法的有效性.
英文摘要
      In feedforward control process, requirement of system model should be moved as much as possible, meanwhile, achieve high control precision and extrapolation ability. In this paper, a data-driven algorithm combining iterative parameter tuning and iterative learning control is proposed. Reference-induced disturbance is compensated by introducing parameterized feedforward controller and input shaping filter with basis function, and using gradient-descent method to calculate optimal system feedforward controller. Repetitive disturbance in system is eliminated by introducing iterative learning control to achieve further improvement of control precision. The proposed algorithm has the advantage of modelfree, high precision and high extrapolation ability. Simulation results are provided in this paper. Experiments of proposed algorithm is carried on a linear motor system to validate effectiveness and extrapolation ability of algorithm.