摘 要:文章使用自回归因子分析模型(ARFA)对数据样本进行动态过程建模,分析了卡尔曼滤波和EM 算法在估计ARFA 模型中回归矩阵参数A 和载荷矩阵参数C 的方法。在此基础上,提出了一种使用变分贝叶斯EM(VBEM)故障检测方法,对ARFA 模型参数A 和C 进行推断和动态过程故障检测。仿真实验结果表明,在ARFA 模型下,VBEM 方法对下文所述的阶跃信号、斜坡信号等四类故障的检测效果要优于EM 方法对该类故障的检测效果,并且降低了平均迭代次数。
关键词:自回归因子分析模型;EM 算法;变分贝叶斯EM;故障检测
中图分类号:TP277;TP391.9 文献标识码:A 文章编号:2096-4706(2020)24-0001-06
Parameter Derivation and Fault Detection of ARFA Model Based on VBEM
LI Jijun,ZHANG Zhijie,LI Qicao,DONG Zijian
(School of Electronic Engineering,Jiangsu Ocean University,Lianyungang 222005,China)
Abstract:In this paper,the autoregressive factor analysis(ARFA)model is used to model the dynamic process of data samples,and the methods of Kalman filter and EM algorithm to estimate the regression matrix parameter A and load matrix parameter C in ARFA model are analyzed. On this basis,a fault detection method based on variational Bayes EM(VBEM)is proposed to infer ARFA model parameters A and C and detect dynamic process faults. The simulation results show that under ARFA model,the detection effect of VBEM method is better than that of EM method for four kinds of faults,such as step signal and slope signal,and the average number of iterations is reduced.
Keywords:autoregressive factor analysis model;EM algorithm;variational Bayes EM;fault detection
基金项目:江苏省研究生科研创新计划项目(KYCX20_2941)
参考文献:
[1] 何潇,郭亚琦,张召,等. 动态系统的主动故障诊断技术 [J]. 自动化学报,2020,46(8):1557-1570.
[2] 王世林. 基于多元统计分析的工业过程故障检测研究 [D].北京:华北电力大学(北京),2018.
[3] 刘春燕. 基于PCA 的过程工业故障检测 [D]. 绵阳:西南科技大学,2017.
[4] KU W,STORER R H,GEORGAKIS C. Disturbancedetection and isolation by dynamic principal component analysis [J].Chemometrics and Intelligent Laboratory Systems,1995,30(1):179-196.
[5] LI W H,QIN S J. Consistent dynamic PCA based on errors-invariablessubspace identification [J].Journal of Process Control,2001,11(6):661-678.
[6] KIM D,LEE I B. Process monitoring based on probabilisticPCA [J].Chemometrics and Intelligent Laboratory Systems,2003,67(2):109-123.
[7] 杨沛武. 概率PCA 多元统计方法在过程监控中的应用研究 [D]. 无锡:江南大学,2008.
[8] LI G,QIN S J,ZHOU D H. A New Method of DynamicLatent-Variable Modeling for Process Monitoring [J].IEEE Transactionson Industrial Electronics,2014,61(11):6438-6445.
[9] WEN Q J,GE Z Q,SONG Z H. Data-based linear Gaussianstate-space model for dynamic process monitoring [J].AlChE Journal,2012,58(12):3763-3776.
[10] 周乐. 基于概率的工业过程数据建模与故障检测 [D]. 杭州:浙江大学,2015.
[11] QIN J S. Statistical Process Monitoring:Basics and Beyond [J].Journal of Chemometrics,2003,17(8-9):480-502.
[12] BOX G E P. Some Theorems on Quadratic Forms Applied inthe Study of Analysis of Variance Problems,I. Effect of Inequality ofVariance in the One-Way Clas-sification [J].The Annals of MathematicalStatistics,1954,25(2):290-302.
[13] JAAKKOLA T S. Tutorial on variational approximationmethods [M]//OPPER M,SAAD D. Advanced Mean Field Methods:Theory and Practice.Massachusetts:MIT Press,2001:129-159.
[14] JORDAN M I,GHAHRAMANI Z,JAAKKOLA T S,et al. An Introduction to Variational Methods for Graphical Models [J].Machine Learning,1999,37(2):183-233.
[15] BEAL M J. Variational algorithms for approximate Bayesianinference [D].London:University College London,2003.
作者简介:
李吉俊(1995—),男,汉族,河南邓州人,硕士研究生在读,研究方向:故障检测、图像处理;
章智杰(1996—),男,汉族,江苏连云港人,硕士研究生在读,研究方向:图像处理;
李其操(1997—),男,汉族,浙江诸暨人,硕士研究生在读,研究方向:图像处理;
通讯作者:
董自健(1973—),男,汉族,江苏连云港人,教授,博士研究生,研究方向:检测与控制、通信技术。