PCA-CHMM在化工过程故障诊断中的应用  

Research on integrated PCA-CHMM based framework for fault diagnosis in chemical processes

作　　者：周韶园[1] 王树青[1] 张建明[1] 

机构地区：[1]工业控制技术国家重点实验室浙江大学先进控制研究所

出　　处：《浙江大学学报（工学版）》2005年第10期1475-1480,共6页Journal of Zhejiang University：Engineering Science

基　　金：国家"863"高技术研究发展计划资助项目(2001AA413110);国家自然科学基金资助项目(60421002)

摘　　要：为了解决主元分析(principal component analysis,PCA)在故障分类方面的不足,提出一种结合PCA和连续隐马尔可夫模型(continuous hidden Markov model,CHMM)的故障诊断方法.PCA对测量数据进行优化降维,用少数几个主元表征过程主要变化信息,实现过程特征提取.但是PCA不能对这些过程变化信息进行有效的分类和识别,运用CHMM成熟的时序模式分类能力能够较好地解决这个问题.通过Tennessee Eastman过程仿真验证了基于PCA和CHMM的故障诊断方法的性能.An approach to overcome the poor fault diagnostic performance of principal component analysis （PCA） by combining PCA with continuous hidden Markov model （CHMM） for off-line fault diagnosis was proposed. PCA was employed to reduce the large number of correlated variables to a small number of principal components in an optimal way. These principal components were used as feature sequences, which indicate most of the process operating information. PCA＇s inability to correctly classify various kinds of process operating information can be resolved by using CHMM. Case studies from Tennessee Eastman process show the potential of the proposed method.

关 键 词：主元分析 隐马尔可夫模型 故障诊断 Tennessee Eastman过程 

分 类 号：TP277[自动化与计算机技术—检测技术与自动化装置]