一种基于变因子梯度型PAST的递推子空间辨识算法  

A modified method of recursive subspace identification algorithms based on variable factor gradient subspace tracking

作　　者：黄金峰 张合新[1] 张植 

机构地区：[1]第二炮兵工程大学301教研室,西安710025 [2]第二炮兵驻二零零厂军代室,北京100854

出　　处：《控制与决策》2012年第8期1226-1230,共5页Control and Decision

摘　　要：针对传统递推子空间算法采用固定遗忘因子而存在易受噪声干扰等问题,提出一种新的变因子递推子空间辨识算法.该算法首先引入变因子构造和更新输入输出Hankel矩阵以及观测向量;然后利用变因子改进梯度型子空间跟踪算法估计系统的广义能观测矩阵,并由广义能观测矩阵来估计系统矩阵;最后运用系统矩阵的特征值空间欧氏距离信息实现变因子更新步骤,使算法具有自适应性.将所提出算法应用于一类时不变系统和慢变系统模型,数值仿真结果表明该算法跟踪速度较快且跟踪效果良好.A new recursive subspace identification algorithm is proposed for the problems existed in traditional algorithms that using fixed factor can easily disturbed by noise. Firstly, input-output Hankel matrices and observation vectors are updated with variable factor. Then, modified algorithm of gradient subspace tracking is designed for recursive estimation of extended observation matrix, which is used to estimate system matrices. Finally, with the help of eigenvalues＇ Euclidean-distance of matrix A, the approach of changing forgetting factor is realized, which ensures the self-adaptation of proposed algorithm. The simulation results show that the tracking performance of new algorithm is faster and better than that of traditional algorithm.

关 键 词：子空间辨识 递推算法 梯度型子空间跟踪 变因子 

分 类 号：TP273[自动化与计算机技术—检测技术与自动化装置]