基于可变滑动窗口KLPP的故障检测  

Fault detection based on variable moving window KLPP

作　　者：郭金玉 郭佳燕 李元 GUO Jinyu;GUO Jiayan;LI Yuan(College of Information Engineering,Shenyang University of Chemical Technology,Shenyang 110142,China)

机构地区：[1]沈阳化工大学信息工程学院,辽宁沈阳110142

出　　处：《大连工业大学学报》2023年第6期463-468,共6页Journal of Dalian Polytechnic University

基　　金：国家自然科学基金项目(62273242);辽宁省教育厅科学研究项目(LJ2019007).

摘　　要：为了提高KLPP在故障检测过程中对非线性和时变特性的自适应能力,提出一种基于可变滑动窗口KLPP(VMWKLPP)的故障检测方法。利用训练数据建立KLPP模型,并计算其统计量和控制限;对测试样本块进行检验,通过正常过程的变化来调节窗口的大小,选择最优的窗口大小。滑动窗口来添加新的样本块和丢弃旧的样本块,实现窗口数据样本的实时更新,以进一步更新KLPP模型和控制限。将该方法运用于田纳西-伊斯曼过程中,仿真结果表明,与KLPP和滑动窗口KLPP(MWKLPP)相比,VMWKLPP方法在工业过程监控中具有明显的优越性。To improve the adaptive ability of KLPP to deal with nonlinear and time-varying characteristics in fault detection process,a fault detection method based on variable moving window kernel locality preserving projections(VMWKLPP)is proposed.The KLPP model is built using training data,and its statistic and control limit are calculated.Then,the test sample block is checked.The size of the window is varied according to the change in the normal process,and the optimal window size is selected.The window is moved to add new sample blocks and discard old sample blocks,so as to realize real-time update of window data samples,and further update the KLPP model and control limit.This method is applied to the Tennessee Eastman process,the simulation results show that VMWKLPP has obvious advantages over KLPP and moving window KLPP(MWKLPP)in industrial process monitoring.

关 键 词：故障检测 核局部保持投影(KLPP) 田纳西-伊斯曼过程 可变滑动窗口 

分 类 号：TP277[自动化与计算机技术—检测技术与自动化装置]