一种改进CPLS算法及其在过程监控中的应用  

Improved CPLS Algorithm and Its Application in Process Monitoring

作　　者：李庆华 潘丰 赵忠盖 

机构地区：[1]江南大学轻工过程先进控制教育部重点实验室,江苏无锡214122

出　　处：《系统仿真学报》2018年第2期622-628,共7页Journal of System Simulation

基　　金：国家自然科学基金(61273131)

摘　　要：CPLS(Concurrent PLS)对PLS分解的过程变量和质量变量的残差和主元进行进一步的提取,从而将变量投影到五个子空间,并由此构建了对过程变量和质量变量信息的完整监控框架。但是,在CPLS中,假设残差为可以求解的确定值,而残差本质上为随机分布量。因此,采用随机模型及其基于随机模型的监控更能反应残差的特性。在基于CPLS的过程监控中,采用因子分析(FA)算法对PLS中的残差进行分析,建立了基于FA的改进CPLS模型,并构建了符合正态分布假设条件的监控指标,提高了监控指标与建模指标的一致性。Concurrent PLS （CPLS） further extracts information from the residuals of input variables and quality variables drawn by PLS, thus the raw data are projected into five subspaces. The process monitoring based CPLS provides a whole framework for the monitoring of input variables and quality variables. The model for residuals is developed by a deterministic manner while the residuals are inherently stochastic; therefore a probabilistic model is more proper for describing their features. This paper introduces factor analysis （FA） into CPLS, in which FA instead of PCA is used to analyze the residuals to develop the improved CPLS model, and the monitoring indices for checking the validity of variables satisfying Gaussian distribution are built to improve the consistence between the modeling objective and the monitoring indices.

关 键 词：CPLS 因子分析 期望最大化(EM)算法 过程监控 

分 类 号：TP391.9[自动化与计算机技术—计算机应用技术]