初态学习下的迭代学习控制  

Iterative learning control with initial state learning

作　　者：孙明轩[1] 

机构地区：[1]浙江工业大学信息工程学院,杭州310032

出　　处：《控制与决策》2007年第8期848-852,共5页Control and Decision

基　　金：教育部留学回国人员科研项目启动基金;国家自然科学基金项目(60474005)

摘　　要：提出一种新的初态学习律,以放宽常规迭代学习控制方法的初始定位条件.它允许一定的定位误差,在迭代中不需要定位在某一具体位置上,使得学习控制系统具有鲁棒收敛性.针对二阶LTI系统,给出了输入学习律及初态学习律的收敛性充分条件.依据收敛性条件,学习增益的选取需系统矩阵的估计值,但在一定建模误差下,仍能保证算法的收敛性.所提出的初态学习律本身及其收敛性条件均与输入矩阵无关.By novel initial state learning, the assumption on initial repositioning is relaxed for the conventional iterative learning control. It is usually assumed that at the beginning of each trial, the initial state is reset to the desired one without repositioning errors. The learning schemes under consideration are of robust convergence, which allow initial repositioning errors and initial states not to be specified positions. Sufficient conditions for the convergence are given for the second-order LTI systems, by which learning gains can be chosen. The learning schemes can overcome imperfect knowledge about system dynamics to achieve complete tracking, though the initial state learning laws are independent of the input matrix.

关 键 词：迭代学习控制 初态学习 初始条件 收敛性 

分 类 号：TP273[自动化与计算机技术—检测技术与自动化装置]