摘要
为了有效改进支持向量机(SVM)在工业过程中的故障检测性能,提出一种基于多模型SVM(multi-model SVM,MM-SVM)的多模态过程故障检测方法.首先,运用局部概率密度方法对多模态数据进行预处理,消除多模态数据对故障检测性能的影响;其次,通过改变SVM的核参数建立多个SVM模型进行故障分类;最后,将多个SVM模型的分类结果进行整合,通过概率大小定义数据类别,实现对故障的有效检测.将该方法应用于多模态数值例子和田纳西-伊斯曼多模态过程,并与PCA、KPCA和SVM方法作比较,实验结果进一步验证了该方法的有效性.
To effectively improve the fault detection performance of support vector machine(SVM)in industrial processes,a fault detection method of multi-modal process based on multi-model SVM(MM-SVM)was proposed.Firstly,the local probability density method was applied to preprocess the multi-modal data to eliminate the influence of the multi-modal data on fault detection performance.Then,multiple SVM models for fault classification were established by changing the kernel parameters of SVM.Finally,the classification results of multiple SVM models were integrated,and the data category was defined by the probability to achieve effective fault detection.The proposed method was applied to a multi-modal numerical example and the Tennessee-Eastman multi-modal process.Compared with PCA,KPCA and SVM,the experimental results further verify the effectiveness of the proposed method.
作者
郭金玉
李涛
李元
GUO Jinyu;LI Tao;LI Yuan(Shenyang University of Chemical Technology,Shenyang,110142,China)
出处
《沈阳化工大学学报》
CAS
2023年第6期533-541,共9页
Journal of Shenyang University of Chemical Technology
基金
国家自然科学基金项目(62273242)
辽宁省教育厅项目(JYTMS20231516)。
关键词
支持向量机
核参数
局部概率密度
多模态过程
故障检测
support vector machine
kernel parameters
local probability density
multi-modal process
fault detection
