全文下载:2012050474
邹凌伟, 田学民*
(中国石油大学(华东) 信息与控制工程学院,山东 青岛 266580)
摘要: 针对核主成分分析(KPCA)和主成分分析(PCA)的一些不足,提出一种基于集成主成分分析的故障检测方法。该方法将PCA与KPCA结合,利用KPCA描述过程的非线性信息并提取核主成分,再利用PCA对原始信息和核主成分一同提取线性主成分,通过构造统计量T2和SPE(或Q)进行故障检测。在TE(Tennessee-Eastman)过程上的仿真研究表明,本文提出的方法较PCA和KPCA有更高的故障检测精度。
关键词: 故障检测; 主成分分析; 核主成分分析; 特征提取
中图分类号: TP 277文献标志码: A
A Fault Detection Method Based on Integrated PCA
ZOU Ling-wei, TIAN Xue-min
(CollegeofInformationand ControlEngineering,ChinaUniversityofPetroleum,Qingdao266580,China)
Abstract: A kind of fault detection method based on Integrated Principal Component Analysis is proposed for some weakness of KPCA and PCA. This method combines PCA and KPCA, firstly the nonlinear information of process is described and the kernel principal component is extracted by KPCA, and then combining the original information and kernel principal component as a whole extracted linear principal component by PCA, The T2 and Q statistics are constructed for fault detection. The simulation results on TE process show that the proposed method can detect process faults more accurate than traditional PCA and KPCA.
Key words:fault detection; principal component analysis; kernel principal component analysis; feature extraction
收稿日期:2012-05-30
基金项目: 国家自然科学基金项目(51104175),山东省自然科学基金项目(ZR2011FM014).
作者简介: 邹凌伟(1988—),男,硕士研究生.*通信联系人.