基于局部邻域标准化和核主元分析的故障检测  

FAULT DETECTION BASED ON LOCAL NEIGHBOR STANDARDIZATION AND KERNEL PRINCIPAL COMPONENT ANALYSIS

作　　者：曾静 李磊 李元 Zeng Jing;Li Lei;Li Yuan(School of Information Engineering,Shenyang University of Chemical Technology,Shenyang 110142,Liaoning,China)

机构地区：[1]沈阳化工大学信息工程学院,辽宁沈阳110142

出　　处：《计算机应用与软件》2022年第10期59-63,86,共6页Computer Applications and Software

基　　金：国家重点研发计划“制造基础技术与关键部件”重点专项(2018YFB2003704);国家自然科学基金项目(61503257,61673279)。

摘　　要：针对工业过程的多模态和非线性特性,提出一种基于局部邻域标准化(Local Nelghbor Standardization,LNS)和核主元分析(Kernel Principa Component Analysis,KPCA)相结合的故障检测方法(LNS-KPCA)。通过计算训练数据集中样本之间的距离来确定每一个样本的最近K近邻集合,然后利用该K个近邻集的均值和标准差对当前样本进行标准化处理,以消除过程数据的多分布特征,使得标准化后的数据服从或近似服从同一正态分布,结合核主元分析能够处理非线性过程的特征,在标准化后的数据集中应用KPCA确定T~2和SPE控制限进行故障检测。在非线性数值例子和青霉素发酵过程中进行了仿真研究,并与主元分析(Principal Component Analysis,PCA)、KPCA和K近邻故障检测(FD-KNN)等方法进行对比分析验证了该方法的有效性。A fault detection method(LNS-KPCA)based on local neighbor standardization(LNS)and kernel principal component analysis(KPCA)is proposed for the multi-modal and nonlinear characteristics of industrial processes.The distance between the training data set sample was calculated to determine the K neighbor of each sample set recently,then the mean and standard deviation of the K neighbors set for the current sample were used to standardize processing,to eliminate the process of data distribution,so that the standardization of data obey or similar obey the same normal distribution.The kernel principal component analysis was combined to process the nonlinear characteristics of the process,and KPCA was applied in the standardized dataset to determine the T~2 and SPE control limit for fault detection.In this paper,the nonlinear numerical examples and penicillin fermentation process were simulated and compared with PCA,KPCA and K-nearest neighbor fault detection(FD-KNN),which verified the effectiveness of the proposed method.

关 键 词：局部邻域标准化 核主成分分析 青霉素发酵过程 故障检测 多模态 

分 类 号：TP3[自动化与计算机技术—计算机科学与技术] TP277