引用本文:王晶,刘鹏阳,卢山,周萌,陈晓露.基于CE-Louvain分解和动态递归SVDD的分布式过程监测[J].控制理论与应用,2025,42(8):1650~1658.[点击复制]
WANG Jing,LIU Peng-yang,LU Shan,ZHOU Meng,CHEN Xiao-lu.Distributed process monitoring based on CE-Louvain decomposition and dynamic recursive SVDD[J].Control Theory & Applications,2025,42(8):1650~1658.[点击复制]
基于CE-Louvain分解和动态递归SVDD的分布式过程监测
Distributed process monitoring based on CE-Louvain decomposition and dynamic recursive SVDD
摘要点击 2580  全文点击 144  投稿时间：2023-06-18  修订日期：2025-04-02
查看全文  查看/发表评论  下载PDF阅读器   HTML
DOI编号  10.7641/CTA.2024.30424
  2025,42(8):1650-1658
中文关键词  非线性动态过程  过程监测  CE-Louvain分解  支持向量数据描述
英文关键词  nonlinear dynamic processes  process monitoring  CE-Louvain decomposition  support vector data descrip tion
基金项目  国家自然科学基金项目(61973023,62273007,62003220), 深圳市自然科学基金项目(20220813001358001)资助.
作者单位E-mail
王晶 北方工业大学电气与控制工程学院 jwang@ncut.edu.cn 
刘鹏阳 北方工业大学电气与控制工程学院  
卢山* 深圳职业技术学院智能科学与工程研究所 lushan@szpu.edu.cn 
周萌 北方工业大学电气与控制工程学院  
陈晓露 北京大学工学院  
中文摘要
      针对全厂过程的复杂非线性动态特征,本文提出了一种分布式的过程监测方法.它包括两个主要内容:基 于copula entropy Louvain(CE-Louvain)的过程分解和基于动态递归支持向量数据描述(DR-SVDD)的故障检测. 首 先, 根据机理知识将全厂过程中的变量初步映射为和过程结构相对应的无向图模型,引入CE来描述无向图中不同 节点(即过程变量)之间的权重,并基于将CE-Louvain算法精细分解为合理的子块.然后,针对每个子块提出了基于 DR-SVDD的分布式故障检测方法以提高故障检测率.最后,利用贝叶斯融合推理方法得到全局过程监测结果.提出 的方法在Tennesse-Eastman(TE)过程中得到了验证.
英文摘要
      A distributed process monitoring method is proposed for the complex nonlinear dynamic characteristics of the plant-wide process. It consists of two main elements: process decomposition based on copula entropy Louvain (CE Louvain) and fault detection based on dynamic recursion support vector data description (DR-SVDD). Firstly, the variables in the plant-wide process are initially mapped into an undirected graph model corresponding to the process structure based on the mechanistic knowledge, and CE is introduced to describe the weights between different nodes (i.e., process variables) in the undirected graph, and the CE-Louvain algorithm is finely decomposed into reasonable subblocks based on the CE Louvain algorithm. Then, a DR-SVDD-based distributed fault detection method is proposed for each sub-block to improve the fault detection rate. Finally, the global process monitoring results are obtained by using Bayesian fusion inference method. The proposed method is validated in the Tennessee-Eastman (TE) process.