Reducing Temperature Calibration Error in Multivariate Analysis of Fluorescence Spectra

Abstract

A method has been demonstrated to reduce temperature calibration error by integrated using principal component analysis, hierarchical cluster analysis and searching combination moving window interval projection to latent structures for fluorescent spectra of Er-doped 98MgCaSrBaYAl2F14-2Ba(PO3)2 and Yb-doped CaF2. The consecutive and consistent use of these multivariate methods for outliers detection, forming training and test datasets and variable selection is shown to allow more than twofold reducing the root-mean-square error of temperature calibration in comparison with the application of projection to latent structures without variable selection.