摘要
为了降低样本间的自相关性对支持向量机(SVM)检测性能的影响,提出一种基于高斯和非高斯双子空间SVM(DSSVM)的故障检测方法。首先运用Kolmogorov-Smirnov(KS)检验原始数据中过程变量的正态分布特性,将过程变量划分为高斯子空间和非高斯子空间,并建立基于PCA的高斯子空间和ICA的非高斯子空间故障检测模型;分别对主元矩阵和独立元矩阵引入时滞特性和时差输入特性,该特性的引入能够降低样本间的自相关性;最后将引入时滞和时差特性的矩阵进行组合,运用SVM模型对其进行故障检测和监视。将该方法运用于多变量数值仿真和田纳西-伊斯曼工业过程,并与PCA、ICA、SVM和基于变量分布特征的统计过程检测方法(VDSPM)比较,仿真实验结果进一步验证了该算法的有效性。
In order to reduce the influence of autocorrelation between samples on the detection performance of support vector machine(SVM),a fault detection of SVM method based on Gaussian and non-Gaussian double space SVM(DSSVM)was proposed.First,Kolmogorov-Smirnov(KS)was used to test the normal distribution characteristics of process variables in the original data,and the process variables were divided into Gaussian subspace and non-Gaussian subspace.The fault detection model based on PCA in Gaussian subspace and that based on ICA in non-Gaussian subspace was established.Then,the time-delay characteristics and time difference input characteristics were introduced to the principal component matrix and the independent component matrix,which could reduce the autocorrelation between samples.Finally,the matrix with time delay and time difference characteristics was combined,and SVM was used to detect and monitor the fault.The method is applied to a multivariable numerical simulation and the Tennessee Eastman industrial process.Compared with PCA,ICA,SVM and statistical process monitoring method based on variable distribution(DSSVM),the simulation results further verify the effectiveness of the algorithm.
作者
郭金玉
李涛
李元
GUO Jinyu;LI Tao;LI Yuan(College of Information Engineering, Shenyang University of Chemical Technology, Shenyang 110142, China)
出处
《大连工业大学学报》
CAS
北大核心
2021年第6期460-468,共9页
Journal of Dalian Polytechnic University
基金
辽宁省教育厅科学研究经费项目(LJ2019007).
关键词
故障检测
KS检验
主元分析
独立元分析
支持向量机
fault detection
KS test
principal component analysis
independent component analysis
support vector machine
