化工学报 ›› 2023, Vol. 74 ›› Issue (9): 3865-3878.DOI: 10.11949/0438-1157.20230501

• 过程系统工程 • 上一篇    下一篇

基于最大信息系数的分组支持向量数据描述故障检测

张逸豪(), 王振雷()   

  1. 华东理工大学能源化工过程智能制造教育部重点实验室,上海 200237
  • 收稿日期:2023-05-23 修回日期:2023-08-30 出版日期:2023-09-25 发布日期:2023-11-20
  • 通讯作者: 王振雷
  • 作者简介:张逸豪(1999—),男,硕士研究生,sdzyh998710@163.com
  • 基金资助:
    国家自然科学基金重大项目(61890930-3);国家自然科学基金项目(62173145);中央高校基本科研业务费专项(222202317006);高等学校学科创新引智计划项目(B17017)

Fault detection using grouped support vector data description based on maximum information coefficient

Yihao ZHANG(), Zhenlei WANG()   

  1. Key Laboratory of Smart Manufacturing in Energy Chemical Process, Ministry of Education, East China University of Science and Technology, Shanghai 200237, China
  • Received:2023-05-23 Revised:2023-08-30 Online:2023-09-25 Published:2023-11-20
  • Contact: Zhenlei WANG

摘要:

工业过程的众多变量之间往往存在着复杂的相关关系,传统的故障检测模型通常会忽略不同变量间相关性的差异,对不同相关关系的变量采用相同的特征提取方法,从而导致检测效果欠佳。针对以上问题,提出了一种基于最大信息系数的分组支持向量数据描述故障检测模型,首先计算变量间的最大信息系数矩阵,按照相关性的不同对变量进行分组,再通过最大信息系数为模型混合核函数中高斯核与多项式核的权重分配提供理论指导,从而分别为各组建立不同的支持向量数据描述检测模型,完成最大信息系数与支持向量数据描述的紧密结合,最终实现分布式故障检测。通过仿真对比,验证了该模型的可行性与有效性。

关键词: 故障检测, 最大信息系数, 变量分组, 核函数, 支持向量数据描述

Abstract:

There are often complex correlations among many variables in the industrial process. Traditional fault detection models often ignore the differences in correlation between different variables and use the same feature extraction method for variables with different correlation relationships, resulting in poor detection performance. In response to the above issues, this article proposes a fault detection model based on maximum information coefficient grouping support vector data description. Firstly, the maximum information coefficient matrix between variables is calculated, and the variables are grouped according to different correlations. Then, the maximum information coefficient provides theoretical guidance for the weight allocation of Gaussian and polynomial kernels in the mixed kernel function of the model. Thus, different support vector data description detection models are established for each group, achieving a close combination of maximum information coefficient and support vector data description, and ultimately achieving distributed fault detection. The feasibility and effectiveness of the model were verified through simulation comparison.

Key words: fault detection, maximum information coefficient, variables grouping, kernel function, support vector data description

中图分类号: