基于LNS和主成分分析的局部离群因子故障检测  

Local outlier fault detection based on LNS and principal component analysis

作　　者：曾静 解晓兵 李元 ZENG Jing;XIE Xiaobing;LI Yuan(School of Infoimation Engineering,Shenyang University of Chemical Technology,Shenyang 110142,China)

机构地区：[1]沈阳化工大学信息工程学院,沈阳110142

出　　处：《黑龙江大学自然科学学报》2022年第3期337-344,共8页Journal of Natural Science of Heilongjiang University

基　　金：国家自然科学基金资助项目(61503257,61673279);国家重点研发计划重点专项基金资助项目(2018YFB2003704)。

摘　　要：针对复杂工业过程中的多模态和非线性等问题,提出了一种新的故障检测方法。采用局部邻域标准化(Local neighbor standardization,LNS)方法对多模态数据集进行标准化处理,利用主成分分析(Principal component analysis,PCA)将数据划分为主成分子空间(Principal component subspace,PCS)和残差子空间(Residual subspace,RS),使用局部离群因子(Local outlier factor,LOF)方法分别在这两个子空间进行故障检测。LNS方法可将多模态数据归一化为单模态数据,使PCA能够更准确地划分主成分子空间和残差子空间,LOF方法能够增强PCA处理非线性数据能力,同时能弥补自身单监控统计量的不足。采用LNS-PCA-LOF方法对非线性数值例子和青霉素发酵过程进行了仿真,与PCA、K近邻故障检测(FD-KNN)和LOF等方法相比,验证了所提方法的有效性。Aiming at the multi-mode and nonlinear problems of complex industrial processes,a new fault detection method is proposed.The Local neighbor standardization(LNS)method is used to standardize the multi-mode data sets.Principal component analysis(PCA)is used to divide the data into the Principal component subspace(PCS)and the Residual subspace(RS),and the Local outlier factor(LOF)method is used to perform faults in these two subspaces respectively detection.The LNS method normalizes the multi-mode data into single-modal data,so that the PCA can more accurately divide the principal component subspace and the residual subspace.The LOF method can enhance the ability of PCA to process nonlinear data,while making up for the deficiency of its own single-monitoring statistics.The method LNS-PCA-LOF is used to simulate in a nonlinear numerical example and penicillin fermentation process,and the effectiveness of the proposed method is verified by comparison with PCA,K-nearest neighbor fault detection(FD-KNN)and LOF.

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

分 类 号：TP277[自动化与计算机技术—检测技术与自动化装置]