摘要
对于复杂的工业过程,采集到的过程数据能反映出生产过程的内在变化和运行状况。本文提出一种新的多变量统计过程监测策略,数据建模过程包含主元分析(Principal Component Aanlysis,PCA)与正交局部保持投影(Orthogonal Locality PreservingProjection,OLPP)两步。首先利用PCA在不丢失任何信息的前提下将原始数据旋转成不相关的潜变量,然后再作OLPP以提取能表征过程正常数据内在局部近邻结构的特征用于故障检测。利用T^2和SPE(或Q)统计量以及核密度估计方法确定的控制限进行化工过程的在线监测,TE过程仿真实验验证了该混合方法的有效性和优越性。
In this paper, a novel approach for multivariate statistical process monitoring is proposed, which consists of two steps： principal component analysis （PCA） plus orthogonal locality preserving projection （OLPP）. PCA is firstly applied to rotate the original process dataset to tmcorrelated latent variables without losing any information, and then perform OLPP to extract first several components to represent the intrinsic structure of process data from normal operating condition. Hotelling＇s T2 and the squared prediction error （SPE or Q） statistic charts are then presented and the kernel density estimation is utilized to determine a proper control limits for the two statistics. Case study on the Tennessee Eastman benchmark process demonstrates the superiority of PCA plus OLPP based method in fault detection over PCA and LPP based monitoring methods.
出处
《计算机与应用化学》
CAS
CSCD
北大核心
2012年第9期1065-1068,共4页
Computers and Applied Chemistry
基金
国家自然科学基金(21176073)
博士点基金(20090074110005)
曙光计划(09SG29)
新世纪优秀人才(NCET-09-0346)
关键词
故障检测
主元分析
正交局部保持投影
TE过程
fault detection, principal component analysis, orthogonal locality preserving projection, TE process
