郭金玉,李 涛,李 元.基于主元增广矩阵的SVM故障检测[J].深圳大学学报理工版,2021,38(5):543-550.[doi:10.3724/SP.J.1249.2021.05543]
GUO Jinyu,LI Tao,and LI Yuan.SVM based on principal component augmented matrix for fault detection[J].Journal of Shenzhen University Science and Engineering,2021,38(5):543-550.[doi:10.3724/SP.J.1249.2021.05543]
点击复制
基于主元增广矩阵的SVM故障检测
郭金玉,李 涛,李 元

沈阳化工大学信息工程学院,辽宁沈阳110142

控制科学与技术; 支持向量机; 主元分析; 增广矩阵; 故障检测; 自相关性; 田纳西-伊斯曼过程

SVM based on principal component augmented matrix for fault detection
GUO Jinyu, LI Tao, and LI Yuan

College of Information Engineering, Shenyang University of Chemical Technology, Shenyang 110142, Liaoning Province, P.R.China

control science and technology; support vector machine; principal component analysis; augmented matrix; fault detection; autocorrelation; Tennessee-Eastman process

DOI: 10.3724/SP.J.1249.2021.05543

备注

为提高支持向量机(support vector machine, SVM)的故障检测率,提出一种基于主元增广矩阵的SVM(SVM based on principal component augmented matrix, PCAM-SVM)故障检测方法.运用主元分析(principal component analysis, PCA)算法在主元空间中计算得分矩阵,再加入得分的时滞输入特性和时差输入特性,构建增广矩阵.运用正常数据和故障数据的增广矩阵训练SVM模型,获得判别分类函数,再运用SVM模型对测试数据进行分类.PCAM-SVM方法通过构建主元增广矩阵,增加模型输入特性复杂度,有效降低了数据自相关性,提高了SVM的故障检测性能.将该方法应用于多变量动态仿真案例和田纳西-伊斯曼过程,并与PCA、独立元分析(independent component analysis, ICA)、核主元分析(kernel principal component analysis, KPCA)、SVM和PCA-SVM方法比较,验证了PCAM-SVM方法的有效性.
To effectively improve the fault detection rate of the support vector machine(SVM)based method, we propose a fault detection method with SVM based on principal component augmented matrix(PCAM-SVM). Firstly, the principal component analysis(PCA)algorithm is used to obtain the scores of training data in principal component space. Secondly, the input characteristics of time delay and time difference for the scores are added to construct the augmented matrix. Then, the combined augmented matrix on normal data and fault data is used to obtain the discriminant function of SVM model for classification. Finally, the SVM model is used to perform the classification operation on test data. The proposed method increases the complexity of input characteristics of model, reduces data autocorrelation, and enhances the fault detection performance of SVM by constructing principal component augmented matrix. The method has been applied to a multivariate dynamic simulation case and the Tennessee-Eastman process. The simulation results validate the feasibility and effectiveness of PCAM-SVM by comparison with PCA, independent component analysis(ICA), kernel principal component analysis(KPCA), SVM and PCA-SVM.
·