化工学报 ›› 2020, Vol. 71 ›› Issue (3): 1254-1263.DOI: 10.11949/0438-1157.20190893

• 过程系统工程 • 上一篇    下一篇

基于故障判别增强KECA算法的故障检测

韩宇1,2,李俊芳1,2,高强1,2(),田宇1,2,禹国刚2,3   

  1. 1.天津理工大学电气电子工程学院,天津 300380
    2.天津市复杂系统控制理论及应用重点实验室,天津 300384
    3.天津理工大学工程训练中心,天津 300384
  • 收稿日期:2019-08-01 修回日期:2019-10-17 出版日期:2020-03-05 发布日期:2020-03-05
  • 通讯作者: 高强
  • 基金资助:
    国家自然科学基金项目(61502340);天津市自然科学基金项目(18JCQNJC01000);天津市高等学校创新团队培养计划资助项目(TD12-5015)

Fault detection based on fault discrimination enhanced kernel entropy component analysis algorithm

Yu HAN1,2,Junfang LI1,2,Qiang GAO1,2(),Yu TIAN1,2,Guogang YU2,3   

  1. 1.School of Electrical and Electronic Engineering, Tianjin University of Technology, Tianjin 300384, China
    2.Tianjin Key Laboratory for Control Theory & Applications in Complicated Systems, Tianjin 300384, China
    3.Tianjin University of Technology Engineering Training Center, Tianjin 300384, China
  • Received:2019-08-01 Revised:2019-10-17 Online:2020-03-05 Published:2020-03-05
  • Contact: Qiang GAO

摘要:

基于核熵主成分分析方法的统计模型仅利用正常工况下数据进行建模,而忽略了监控系统数据库中一些已知类别的先前故障数据。为了利用先前故障数据中包含的故障信息来增强故障检测性能,提出了一种故障判别增强KECA(fault discriminant enhanced kernel entropy component analysis, FDKECA)算法。该法通过采用无监督学习和监督学习方法建立模型,同时监测非线性核熵主成分(kernel entropy component, KEC)和故障判别成分(fault discriminant component, FDC)两类数据特征。此外,利用贝叶斯推理将相应的监视统计信息转换为故障概率,并通过加权两个子模型的结果来构建基于总体概率的监视统计量。通过数值仿真和田纳西伊斯曼(Tennessee Eastman, TE)过程仿真实验,证明和传统KECA相比,FDKECA算法能够有效利用故障数据提高故障检测率。

关键词: 过程控制, 过程系统, 故障检测, 核熵成分分析, 混合模型

Abstract:

The statistical model based on the kernel entropy principal component analysis method only uses the data under normal operating conditions for modeling, and ignores some known categories of previous failure data in the monitoring system database. This paper proposed a fault discriminant enhanced the kernel entropy component analysis algorithm by using fault information contained in the previous fault data to enhance the fault detection performance. The model adopted unsupervised learning and supervised learning methods to monitor two types of data features: nonlinear kernel entropy component and fault discriminant component. In addition, the proposed algorithm employed Bayesian inference to convert the corresponding monitoring statistics into failure probability. The weighted probability of the two sub-models generated the monitoring statistics of the overall probability. Through numerical simulation and Tennessee Eastman process simulation experiments, the FDKECA algorithm is more effective than the traditional KECA since the improvement of the fault detection rate.

Key words: process control, process systems, fault detection, kernel entropy component analysis, mixing model

中图分类号: