Selection of spectral variables and improvement of the accuracy of calibration of temperature by projection onto latent structures using the fluorescence spectra of Yb$^{3+}$ : CaF$_{2}$

The efficiency of interval methods for selecting variables and the divisive method of interval optimization by a genetic algorithm is considered with the purpose to increase the accuracy of temperature calibration by projection onto latent structures. The fluorescence spectra of Yb$^{3+}$ : CaF$_{2}$ recorded in the 880–1120 nm band with a resolution of approximately 0.2 nm in the temperature range from 66 to 150$^{\circ}$C in 2$^{\circ}$C increments are used as an example. The best result for the root-mean-square error of temperature prediction in the test sample (0.45$^{\circ}$C) was obtained by interval projection onto latent structures by a combination of moving windows. The use of methods for selecting spectral variables improved more than two times the accuracy of temperature calibration.