摘要
针对传统多块故障监测方法建模时,一般仅考虑到过程局部信息的问题,提出一种分层信息提取的多块主元分析(Principal component analysis,PCA)故障监测方法。第一层考虑过程变量之间的互信息,对过程变量进行分块,提取过程局部信息;第二层对每个变量块进一步提取累计误差和二阶差分等特征信息,结合观测值信息共同将每个变量块扩展为3个信息子块;对每个信息子块采用PCA方法进行监测,并基于贝叶斯方法融合所有信息子块的监测结果。所提方法的有效性和性能在田纳西-伊斯曼(Tennessee-Eastman,TE)过程监控中进行了验证和分析。
The traditional multi-block fault monitoring methods only use local information of the process,and other valid information of the data set is ignored.In view of this problem,a multi-block principal component analysis(PCA)fault monitoring method with hierarchical information extraction is proposed.Firstly,this paper calculates the mutual information between the process variables and blocks the process variables to extract the process local information.Second,this paper further digs the feature information of the process data,extracts the cumulative error and second-order difference information for each variable block,and with the observed values information expands each variable block into three information sub-blocks.Each sub-block is monitored by the PCA method,and the monitoring results of all the sub-blocks are integrated based on the Bayesian method.The validity and performance of the proposed method are verified and analyzed in Tennessee-Eastman(TE)process monitoring.
作者
翟超
熊伟丽
Zhai Chao;Xiong Weili(School of Internet of Things Engineering,Jiangnan University,Wuxi 214122,China)
出处
《南京理工大学学报》
EI
CAS
CSCD
北大核心
2020年第4期471-480,共10页
Journal of Nanjing University of Science and Technology
基金
国家自然科学基金(61773182)
国家重点研发计划子课题(2018YFC1603705-03)
江苏高校“青蓝工程”资助项目。
关键词
分层信息提取
互信息
主元分析
故障监测
多块建模
hierarchical information extraction
mutual information
principal component analysis
fault monitoring
multi-block modelling
