第 29 卷第 8 期
2012 年 8 月
控 制 理 论 与 应 用
Control Theory & Applications
Vol. 29 No. 8
Aug. 2012
非非非线线线性性性迭迭迭代代代学学学习习习控控控制制制问问问题题题的的的延延延拓拓拓修修修正正正牛牛牛顿顿顿法法法
文文文章章章编编编号号号: 1000−8152(2012)08−1063−06
亢京力
(中国航天科工集团 信息系统工程重点实验室, 北京 100854)
摘要: 对于非线性迭代学习控制问题, 提出基于延拓法和修正Newton法的具有全局收敛性的迭代学习控制新方
法. 由于一般的Newton型迭代学习控制律都是局部收敛的, 在实际应用中有很大局限性. 为拓宽收敛范围, 该方法
将延拓法引入迭代学习控制问题, 提出基于同伦延拓的新的Newton型迭代学习控制律, 使得初始控制可以较为任
意的选择. 新的迭代学习控制算法将求解过程分成N 个子问题, 每个子问题由换列修正Newton法利用简单的递推
公式解出. 本文给出算法收敛的充分条件, 证明了算法的全局收敛性. 该算法对于非线性系统迭代学习控制具有全
局收敛和计算简单的优点.
关键词: 迭代学习控制; 延拓法; 修正Newton法; 全局收敛; 非线性系统
中图分类号: TP273 文献标识码: A
A new iterative learning control algorithm of
extension-updated Newton method for nonlinear systems
KANG Jing-li
(Science and Technology on Information Systems Engineering Laboratory,
China Aerospace Science and Technology Corporation, Beijing 100854, China)
Abstract: A new algorithm based on extension method and updated Newton method with global convergence for
nonlinear iterative learning control problem is proposed. Since classical Newton-type iterative learning schemes are local
convergence, conditions of local convergence can be hardly satisfied in practice. In order to widen the range of convergence,
extension method is introduced to iterative learning control problem. A new Newton-type iterative learning control scheme
based on homotopy extension is presented, in which the initial control can be chosen arbitrarily. The solving process is
subdivided to N subproblem by the new algorithm. The exchange column update Newton method is employed to solve
the subproblem by simple recurrent formula. Sufficient conditions for global convergence of this algorithm are given and
proved. The implementation of the new algorithm has advantage of guaranteeing global convergence and avoiding complex
calculation for nonlinear iterative learning control.
Key words: iterative learning control; extension method; updated Newton method; global convergence; nonlinear
systems
1 Introduction
Iterative learning control is a control strategy that
needs to improve the control performance of every iter-
ative process by operating in a repetitive mode. In it-
erative learning control systems, information from pre-
vious executions of the task is used in an attempt to
generate the updated control iteration and the tracking
error between the output trajectory and desired trajec-
tory tends to zero. Such systems include robot arm
manipulators, disk drive, chemical batch reactors and
other nonlinear industry. Since iterative learning con-
trol was originally introduced by Arimoto
[1]
, signifi-
cant developments in iterative learning research area
has stimulated considerable interests in various update
algorithms for linear and nonlinear systems
[2]
. In re-
cent years, the study of iterative learning control has put
more and more emphases on nonlinear systems which
are the most often seen cases in practice
[3–5]
. Newton-
type iterative learning control schemes are one of the
important and effective schemes which have the advan-
tage of improving the convergence speed for nonlinear
systems. In [6], a new nonlinear iterative learning con-
trol algorithm used a special form of Newton method in
continuous time domain. Xu and Tan
[7]
provided P-type
learning and Newton-type learning method for non-
affine nonlinear systems. The proposed P-type iterative
learning control scheme has the simple form required
a prior system knowledge, while Newton-type itera-
tive learning control schemes have faster convergence
by incorporating a varying learning gain. For discrete
Received 9 May 2012; revised 8 July 2012.
This work was supported by the National Natural Science Foundation (NNSF) of China (No. 61004056).