一种基于核主元分析的非参数控制限算法  

Algorithm for non-parametric confidence bounds based on kernel principal component analysis

作　　者：赵忠盖[1] 刘飞[1] 

机构地区：[1]江南大学自动化研究所,江苏无锡214122

出　　处：《系统工程学报》2008年第1期96-100,共5页Journal of Systems Engineering

基　　金：国家高技术研究发展计划(863计划)资助项目(2007AA04Z198);新世纪优秀人才支持计划资助项目(NCET-05-0485)

摘　　要：实际过程中,监控指标值往往不满足确定的概率分布,给控制限的计算带来了困难,现有的计算方法计算复杂、精度差.根据高维空间良好的分类特性,提出一种基于核主元分析(XPCA)的监控指标控制限计算方法,将监控指标值映射到高维空间,再在高维空间中对其映射数据进行监控,实现了控制限的间接求取,避免了复杂的计算.In the real industrial practice, monitoring indices do not meet any decided distribution usually, so it is difficult to compute, while the existing methods suffer from the complexity of calculation and bad precision. Due to the well classified characteristic in high-dimensional space, a novelty method based on kernel principal component analysis （KPCA） is proposed to calculate the confidence limit of monitoring indices. Through monitoring the mapped values of indices in high-domensional space based on PCA, the method indirectedly achieves the calculation of confidence limit and need not more workload.

关 键 词：非线性主元分析 核主元分析(KPCA) 非参数控制限 SPE指标 

分 类 号：TP206.3[自动化与计算机技术—检测技术与自动化装置]