摘要
针对现代工业过程的非线性和多分布问题,提出一种基于Jarque-Bera test的故障检测方法。首先,对标准化后的原始数据进行Jarque-Bera test,将变量划分为两个部分;其次,对所有的JB统计量做-ln处理,并采用正态置信概率权值进一步划分,从而使原始变量空间划分为正态和非正态分布的两个子空间;再次,在两个子空间分别基于KPCA和kNN算法提取过程的非线性特征,建立故障检测模型;最后,通过数值案例和TE过程仿真实验验证了所提方法的有效性。
In view of the nonlinear and multi-distribution in modern industrial processes,a fault detection method based on JB test(Jarque-Bera test,J-B test)was proposed.Firstly,having J-B test carried out on the original data after standardization and the variables divided into two parts;and then,having all JB statistics processed by-ln and the weights of normal confidence probability used to further divide the original variable space into two subspaces of normal and non-normal distribution,and having the nonlinear features of the process extracted based on KPCA and kNN algorithm respectively in two subspaces to establish a fault detection model;and finally,having the effectiveness of the proposed method verified by numerical examples and the simulation experiments of Tennessee Eastman(TE)process.
作者
郭小萍
俞巷天
李元
GUO Xiao-ping;YU Xiang-tian;LI Yuan(College of Information Engineering,Shenyang University of Chemical Technology)
出处
《化工自动化及仪表》
CAS
2022年第2期165-174,共10页
Control and Instruments in Chemical Industry
基金
国家自然科学基金项目(61490701,61673279)
辽宁省教育厅科学研究项目(LJ2020021)。
关键词
Jarque-Bera检验
故障检测
多分布过程
非线性
核主元分析
K近邻
Jarque-Bera test
fault detection
multi-distributed process
nonlinearity
kernel principal component analysis
k-nearest neighbor
