文献类型：期刊文献

英文题名：Angle-Based Multiblock Independent Component Analysis Method with a New Block Dissimilarity Statistic for Non-Gaussian Process Monitoring

作者：Huang, Jian[1];Yan, Xuefeng[1]

机构：[1]E China Univ Sci & Technol, Minist Educ, Key Lab Adv Control & Optimizat Chem Proc, Shanghai 200237, Peoples R China

年份：2016

卷号：55

期号：17

起止页码：4997

外文期刊名：INDUSTRIAL & ENGINEERING CHEMISTRY RESEARCH

收录：;EI(收录号：20162102414842);WOS:【SCI-EXPANDED(收录号:WOS:000375521200022)】；

基金：The authors gratefully acknowledge support from the following foundations: the 973 project of China (2013CB733600) and the National Natural Science Foundation of China (21176073).

语种：英文

外文关键词：Process monitoring - Data description - Gaussian distribution - Gaussian noise (electronic)

摘要：In recent years, the multiblock method has attracted substantial attention. Conventional multiblock methods divide an entire data set into several blocks, and the monitoring in each block is conducted separately. The multiblock method highlights local information but ignores the information among different blocks. In this paper, we propose an angle-based multiblock independent component analysis (MBICA) method and create a new block dissimilarity (BD) statistic to measure the changes between blocks. Hierarchical clustering is adopted to cluster variables with small angles into a block. ICA models are then built into each block. Support vector data description (SVDD) is introduced to yield a final monitoring decision. The changes of blocks are determined by the differences between the angles of the monitored data and the benchmark data, leading to BD statistics. The proposed MBICA-BD method is applied to the Tennessee Eastman process. The simulation results demonstrate the superiority of the MBICA-BD method.