基于多数据结构的集成质量监控方法  

Integrated Quality Monitoring Method Based on Multiple Data Structure

作　　者：薛敏 杨健 谭帅 侍洪波 XUE Min;YANG Jian;TAN Shuai;SHI Hongbo(Key Laboratory of Advanced Control and Optimization for Chemical Processes,Ministry of Education,East China University of Science and Technology,Shanghai 200237,China)

机构地区：[1]华东理工大学化工过程先进控制和优化技术教育部重点实验室

出　　处：《华东理工大学学报（自然科学版）》2019年第6期938-945,共8页Journal of East China University of Science and Technology

基　　金：国家自然科学基金（61374140）;国家自然科学基金青年基金（61403072）

摘　　要：考虑到工业过程中不同数据结构特征的提取方式可能会影响质量监控性能,提出了一种融合过程数据集全局与局部结构特征的集成质量监控(Ensemble Learning based Multiple Data Structures Quality Monitoring,E-MDSQM)方法。首先,构建偏最小二乘(Partial Least Square,PLS)、邻域保持回归(Neighborhood Preserving Regression,NPR)、局部全局主成分回归(Local and Global Principal Component Regression,LGPCR)3种基础模型,分别描述过程数据的全局结构、局部拓扑及局部全局混合结构信息;然后,基于一种新的监控指标,采用遗传优化算法求得最优权重,集成融合各统计量并确定控制限;最后,通过田纳西-伊斯曼(Tennessee-Eastman Process,TE)过程仿真,评估集成模型的监控效果,并与PLS、NPR、LGPCR 3种基础算法比较,实验结果表明该集成模型取得了较好的综合效果。The extraction method of different data structure features may affect the quality monitoring performance in industrial process. Hence, this paper proposes an integrated quality monitoring(E-MDSQM) method,which integrates the global and local structure features of the process data set method. Firstly, three basic models of partial least square(PLS), neighborhood preserving regression(NPR), and local principal component regression(LGPCR) are constructed to describe the global structure of the data, the local topology, and the local global hybrid structure information, respectively. And then, by introducing a new monitoring index, the genetic optimization algorithm is used to obtain the optimal weight, and the integration statistics are integrated, and the control limits are determined. Finally, the simulation via the Tennessee-Eastman process(TE) process is made to evaluate the monitoring effect of the integrated model and the comparisons with the three basic algorithms of PLS, NPR and LGPCR are also undergone. It is shown from these the experimental results that the integrated model can achieve better comprehensive effect.

关 键 词：故障检测 质量监控 流形学习 集成学习 

分 类 号：TP277[自动化与计算机技术—检测技术与自动化装置]