引用本文:张成,邓成龙,李元.基于WPG-KNMF的非线性动态过程监控研究[J].控制理论与应用,2025,42(3):569~578.[点击复制]
ZHANG Cheng,DENG Cheng-long,LI Yuan.Research on nonlinear dynamic processes monitoring based on WPG-KNMF[J].Control Theory and Technology,2025,42(3):569~578.[点击复制]
基于WPG-KNMF的非线性动态过程监控研究
Research on nonlinear dynamic processes monitoring based on WPG-KNMF
摘要点击 25  全文点击 2  投稿时间：2023-03-09  修订日期：2024-09-04
查看全文  查看/发表评论  下载PDF阅读器
DOI编号  10.7641/CTA.2023.30121
  2025,42(3):569-578
中文关键词  核非负矩阵分解  非线性过程  动态过程  投影梯度  Wasserstein距离  故障检测
英文关键词  kernel non-negative matrix factorization  nonlinear process  dynamic process  projected gradient  wasserstein distance  fault detection
基金项目  国家自然科学基金项目(62273242), 辽宁省教育厅基本科研项目(LJKMZ20220792), 辽宁省研究生教育教学改革项目(LNYJG2022177)资助.
作者单位E-mail
张成 沈阳化工大学 zhangcheng@syuct.edu.cn 
邓成龙 沈阳化工大学  
李元* 沈阳化工大学 li-yuan@mail.tsinghua.edu.cn 
中文摘要
      针对非线性动态过程故障检测问题,本文提出一种基于Wasserstein距离投影梯度核非负矩阵分解(WPGKN-MF)的故障检测方法.首先,采用投影梯度方法对KNMF的基矩阵和系数矩阵进行更新.其次,在高维特征空间中,使用Wasserstein距离结合滑动窗口方法,构造新的统计量进行故障检测.本文方法将KNMF中迭代方法改进为投影梯度方法,通过KNMF将数据的非线性结构捕获,并结合Wasserstein距离消除样本间自相关性影响.通过一个数值例子和基于工业控制系统执行器诊断方法的开发与应用(DAMADICS)过程的实验数据进行仿真实验,与传统核主成分分析(KPCA)、核非负矩阵分解等方法进行对比,仿真结果验证了本文所提方法的有效性.
英文摘要
      Aiming at the fault detection of nonlinear dynamic processes, a fault detection method based on the Wasserstein distance projection gradient kernel non-negative matrix factorization (WPG-KNMF) is proposed. Firstly, the projection gradient method is used to update the basis matrix and coefficient matrix in kernel non-negative matrix factorization (KNMF). Secondly, the Wasserstein distance combined with the sliding window method is used to construct new statistics for fault detection in high-dimensional feature space. In this paper, the iterative method in KNMF is improved to the projection gradient method. The nonlinear structure of the data is captured by the KNMF and the Wasserstein distance is combined to eliminate the influence of autocorrelation between samples. The proposed approach is tested in a numerical case and in the development and application of methods for actuator diagnosis in industrial control systems (DAMADICS) process. The experimental results indicate that the proposed approach has an advantage over conventional methods, such as the kernel principal component analysis (KPCA) and the kernel non-negative matrix factorization.