A Numerical Procedure for Multivariate Calibration Using Heteroscedastic Principal Components Regression

Abstract

Many methods have been developed to allow for consideration of measurement errors during multivariate data analyses. The incorporation of the error structure into the analytical framework, usually described in terms of the covariance matrix of measurement errors, can provide better model estimation and prediction. However, little effort has been made to evaluate the effects of heteroscedastic measurement uncertainties on multivariate analyses when the covariance matrix of measurement errors changes with the measurement conditions. For this reason, the present work describes a new numerical procedure for analyses of heteroscedastic systems (heteroscedastic principal component regression or H-PCR) that takes into consideration the variations of the covariance matrix of measurement fluctuations. In order to illustrate the proposed approach, near infrared (NIR) spectra of xylene and toluene mixtures were measured at different temperatures and stirring velocities and the obtained data were used to build calibration models with different multivariate techniques, including H-PCR. Modeling of available xylene–toluene NIR data revealed that H-PCR can be used successfully for calibration purposes and that the principal directions obtained with the proposed approach can be quite different from the ones calculated through standard PCR, when heteroscedasticity is disregarded explicitly.

Data-Driven Process Monitoring and Fault Diagnosis: A Comprehensive Survey

“…Bayesian PCA [135], and MLPCA (maximum likelihood PCA) [90,[136][137][138][139] are some of the approaches used to address this issue. It is noteworthy that MLPCA [132] or principal component regression [140] can handle heteroskedastic data (i.e., data with variance dependent on time or operating conditions). An interesting proposal in this regard is the integration of PCA methodology with data reconciliation [141].…”