학술논문
基于局部Fisher判别分析的复杂化工过程故障诊断 / Fault diagnosis of complex chemical process based on local Fisher discriminant analysis
Document Type
Academic Journal
Source
计算机应用研究 / Application Research of Computers. 35(4):1122-1129
Subject
Language
Chinese
ISSN
1001-3695
Abstract
为了提高复杂化工过程中故障检测和分类能力,提出基于局部Fisher判别分析(local Fisher discriminant analysis,LFDA)的复杂化工过程故障诊断方法.首先计算训练数据的局部类内和类间离散度矩阵,寻找LFDA的投影方向;其次把训练数据和测试数据向投影向量上投影,提取特征向量;最后计算特征向量间的欧氏距离,运用KNN分类器进行分类.把提议的LFDA方法应用到Tennessee Eastman (TE)过程,监控结果表明,LFDA的效果好于FDA和核Fisher判别分析(kernel Fisher discriminant analysis,KFDA),说明LFDA方法在分类及检测不同类的故障方面具有高准确性及高灵敏度的优势.
In order to improve the ability of fault detection and classification of complex chemical process,this paper proposed a fault diagnosis method of complex chemical process based on LFDA.Firstly,it calculated the local within-class and betweenclass scatter matrix of training data to find the projection direction.Secondly,it projected the training and test data into the projection vector for extracting the feature vector.Finally it calculated the Euclidean distances between feature vectors,and used KNN for classification.It applied the proposed method to the TE process.The monitoring results show that LFDA is better than FDA and KFDA,and LFDA method has the advantages of high accuracy and high sensitivity in classification and fault detection of different classes.
In order to improve the ability of fault detection and classification of complex chemical process,this paper proposed a fault diagnosis method of complex chemical process based on LFDA.Firstly,it calculated the local within-class and betweenclass scatter matrix of training data to find the projection direction.Secondly,it projected the training and test data into the projection vector for extracting the feature vector.Finally it calculated the Euclidean distances between feature vectors,and used KNN for classification.It applied the proposed method to the TE process.The monitoring results show that LFDA is better than FDA and KFDA,and LFDA method has the advantages of high accuracy and high sensitivity in classification and fault detection of different classes.