基于改进K近邻-密度峰值聚类的多模块PCA过程监测  

Multi-block PCA Process Monitoring Based on Modified K-nearest Neighbors-Density Peak Clustering

作　　者：王竞志 徐琛[1] 陶洪峰[1] 杨慧中[1] WANG Jingzhi;XU Chen;TAO Hongfeng;YANG Huizhong(Key Laboratory of Advanced Process Control for Light Industry,Ministry of Education,School of Internet of Things Engineering,Jiangnan University,Wuxi 214122,China)

机构地区：[1]江南大学物联网工程学院教育部轻工过程先进控制重点实验室,江苏无锡214122

出　　处：《控制工程》2025年第4期664-673,共10页Control Engineering of China

基　　金：国家自然科学基金资助项目(62103167);江苏省自然科学基金资助项目(BK20210451);中央高校基本科研业务费专项资金资助项目(JUSRP51733B)。

摘　　要：聚类方法考虑了变量之间的复杂关系,已被广泛应用于多模块过程监测中。但传统方法仍需要一些先验知识,如聚类个数等,限制了其应用范围。因此,提出了一种基于改进K近邻-密度峰值聚类的多模块主元分析(principal component analysis,PCA)过程监测方法。首先,引入信息距离,基于变量之间的信息熵构造信息距离矩阵;然后,计算决策值和其下降趋势变化幅值,通过寻找极大值的方式自动确定聚类个数,即模块个数;最后,通过K近邻优化的分配策略完成模块划分,并建立基于PCA的过程监测模型,所有模块的监测结果通过贝叶斯推论进行融合。将所提方法应用在数值仿真和田纳西-伊斯曼(Tennessee-Eastman,TE)过程仿真中,仿真结果验证了所提方法的有效性。Clustering method has been widely used in the field of multi-block process monitoring by considering the complex relationship between variables.However,the conventional methods still need some prior knowledge,such as the number of clusters,which limits their application.Therefore,a multi-block principal component analysis(PCA)process monitoring method based on modified K-nearest neighbors-density peak clustering is proposed.Firstly,the information distance is introduced,and the information distance matrix is constructed based on the information entropy between variables.Secondly,the decision values and their decreasing trend change amplitude are calculated,and the number of clusters,that is,the number of blocks,is automatically determined by finding the maximum.Finally,the allocation strategy optimized by K-nearest neighbor is used to complete the block division,the process monitoring model based on PCA is established,and monitoring results of blocks are fused by Bayesian inference.The proposed method is applied to numerical simulation and Tennessee-Eastman(TE)process simulation,and the simulation results verify the effectiveness of the proposed method.

关 键 词：多模块 过程监测 信息距离 聚类 主元分析 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程] TH213.3[自动化与计算机技术—控制科学与工程]