基于特征向量提取的核主元分析法  

Kernel Principal Component Analysis Based on Feature Vector Selection

作　　者：吴洪艳[1,2] 黄道平[1] 

机构地区：[1]华南理工大学自动化科学与工程学院,广州510640 [2]湛江师范学院信息科学与技术学院,湛江524048

出　　处：《计算机科学》2009年第7期185-187,255,共4页Computer Science

基　　金：广东省科技计划资助项目(2003B50301)资助

摘　　要：核主成分分析(KPCA)是非线性化工过程故障检测与诊断时常用的多变量统计控制方法之一。从两个方面改进了KPCA的故障检测性能。为了提高KPCA方法故障检测的准确率,提出了基于小波的KPCA故障检测方法。当样本数大时,采用基于几何考虑的特征向量提取(FVS)算法,降低了KPCA计算的复杂性,缩短了计算时间。Ten-nessee Eastman process仿真给出了所提出的方法的有效性。Kernel principal component analysis（KPCA） is one of multivariate statistical control methods for solving nonlinear chemical process fault diagnosis. In this paper, it improves KPCA from two aspects. First, in order to improve the accuracy of KPCA for fault detection, a new method combined with wavelet was developed. Second, feature vector selection（FVS） scheme was adopted to reduce the computational complexity of KPCA whereas preserve the geometrical structure of the data. Tennessee Eastman process（TEP） simulations were carried out to show the given approach＇s effectiveness in process monitoring performance.

关 键 词：故障诊断 核主成分分析 特征向量提取 小波变换 

分 类 号：TP277[自动化与计算机技术—检测技术与自动化装置]