Premium
Comparison of multivariate statistical methods for dynamic systems modeling
Author(s) -
Barceló Susana,
VidalPuig Santiago,
Ferrer Alberto
Publication year - 2011
Publication title -
quality and reliability engineering international
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.913
H-Index - 62
eISSN - 1099-1638
pISSN - 0748-8017
DOI - 10.1002/qre.1102
Subject(s) - multivariate statistics , goodness of fit , autocorrelation , process (computing) , computer science , series (stratigraphy) , partial least squares regression , least squares function approximation , statistics , data mining , mathematics , machine learning , paleontology , biology , operating system , estimator
In this paper two multivariate statistical methodologies are compared in order to estimate a multi‐input multi‐output transfer function model in an industrial polymerization process. In these contexts, process variables are usually autocorrelated (i.e. there is time‐dependence between observations), posing some problems to classical linear regression models. The two methodologies to be compared are both related to the analyses of multivariate time series: Box‐Jenkins methodology and partial least squares time series. Both methodologies are compared keeping in mind different issues, such as the simplicity of the process modeling (i.e. the steps of the identification, estimation and validation of the model), the usefulness of the graphical tools, the goodness of fit, and the parsimony of the estimated models. Real data from a polymerization process are used to illustrate the performance of the methodologies under study. Copyright © 2010 John Wiley & Sons, Ltd.