化工学报 ›› 2017, Vol. 68 ›› Issue (4): 1459-1465.DOI: 10.11949/j.issn.0438-1157.20161438

• 过程系统工程 • 上一篇    下一篇

基于动态单类随机森林的非线性过程监控方法

曹玉苹, 卢霄, 田学民, 邓晓刚   

  1. 中国石油大学(华东)信息与控制工程学院, 山东 青岛 266580
  • 收稿日期:2016-10-11 修回日期:2016-12-21 出版日期:2017-04-05 发布日期:2017-04-05
  • 通讯作者: 曹玉苹
  • 基金资助:

    国家自然科学基金项目(61273160,61403418,21606256);山东省自然科学基金项目(ZR2014FL016,ZR2016FQ21,ZR2016BQ14);中央高校基本科研业务费专项资金项目(14CX02174A)。

Nonlinear process monitoring using dynamic one-class random forest

CAO Yuping, LU Xiao, TIAN Xuemin, DENG Xiaogang   

  1. College of Information and Control Engineering, China University of Petroleum, Qingdao 266580, Shandong, China
  • Received:2016-10-11 Revised:2016-12-21 Online:2017-04-05 Published:2017-04-05
  • Supported by:

    supported by the National Natural Science Foundation of China (61273160, 61403418, 21606256),the Natural Science Foundation of Shandong Province (ZR2014FL016, ZR2016FQ21, ZR2016BQ14) and the Fundamental Research Funds for the Central Universities (14CX02174A).

摘要:

针对高维化工过程中存在的非线性和动态特性,提出了一种基于动态单类随机森林(dynamic one-class random forest,DOCRF(的过程监控方法。对正常运行状态下的过程数据进行稀疏性分析,根据其反分布产生离群点数据。利用典型变量分析对正常数据进行相关性分析,分别将正常数据和离群点数据投影到典型变量空间,利用典型变量空间数据训练单类随机森林。基于单类随机森林模型根据待检测样本与正常数据的相似度构造监控统计量进行故障检测。在Tennessee Eastman过程的仿真结果表明,所提DOCRF方法总体优于单类支持向量机方法。

关键词: 过程监控, 单类随机森林, 单类支持向量机, 典型变量分析

Abstract:

For the nonlinear and dynamic characteristics of chemical processes, a process monitoring method based on dynamic one-class random forest (DOCRF) is proposed. The sparsity of the process data under normal operating conditions is analyzed. Then, outliers are produced according to the inverse distribution of the normal data. Canonical variate analysis is used to analyze the correlation of normal data, and to project the normal data and outliers into canonical variate space. One-class random forest is trained by using the data in canonical variate space. The monitoring statistic is established according to the similarity between the test sample and the normal data in one-class random forest for fault detection. Simulation results on Tennessee Eastman process showed that the proposed DOCRF method was better than one-class support vector machine overall.

Key words: process monitoring, one-class random forests, one-class support vector machine, canonical variate analysis

中图分类号: