基于缺失变量估计误差的工业过程监测方法  

Industrial process monitoring based on estimation error of missing variables

作　　者：宋励嘉 童楚东[1] SONG Li-jia;TONG Chu-dong(Faculty of Electrical Engineering&Computer Science,Ningbo University,Ningbo 315211,China)

机构地区：[1]宁波大学信息科学与工程学院,浙江宁波315211

出　　处：《高校化学工程学报》2019年第1期167-173,共7页Journal of Chemical Engineering of Chinese Universities

基　　金：国家自然科学基金(61503204;61773225);浙江省自然科学基金(LY16F030001)

摘　　要：传统主元分析(principal component analysis, PCA)算法在实施投影变换挖掘潜在特征成分时,不同投影变换方向对原始数据不同测量变量赋予了不同的权重值。而从故障检测的角度出发,各个测量变量的重要性程度是等同的。因此,传统PCA方法的过程监测性能还有待商榷。为克服这一缺陷,借鉴基于机理模型故障检测方法生成误差的思路,提出一种基于缺失变量估计误差的工业过程监测方法。首先,通过逐一假设每个测量变量缺失后,利用PCA模型中处理缺失变量的迭代算法(iterativealgorithm,IA)推测出相应缺失变量的估计值。然后,以缺失变量的实际值与估计值之间的误差作为被监测对象实施在线故障检测。通过逐一假设变量缺失的方式,等同地对待了每个测量变量。而利用误差实施监测则可以在缺失变量前提下,通过监测误差的变化反映出该缺失变量对PCA模型中特征成分的影响程度。最后,通过在TE过程上的对比验证了该方法的优越性与可行性。When traditional principal component analysis(PCA)algorithm extracts hidden feature components by implementing projecting rotation,different projecting directions assign variables with different weights.However,all measured variables have equal importance under fault detection point view.Therefore,performance monitored by traditional PCA methods requires further improvement.An industrial process monitoring method based on estimation error of missing variable was proposed following the idea of residual generation utilized in the first-principal model based fault detection.By assuming each measured variable was missed in sequence,the estimation of the corresponding missing variable was calculated by iterative algorithm(IA)that can handle missing variable issues in the PCA model.All measured variables were equally weighted through assuming missing variables one by one.The monitored residuals can reflect the variations in deviation between actual and estimated feature components.Finally,superiority of the feasibility of the proposed method is validated by TE processes.

关 键 词：主成分分析 故障检测 缺失变量 误差生成 

分 类 号：TP277[自动化与计算机技术—检测技术与自动化装置]