Optimizing Meta-Parameters in Continuous Piecewise Direct Standardization

Abstract

When in-line or on-line spectroscopy is performed on a process, different types of variations are present in the measured spectrum. The variation due to the chemistry of the process is the wanted variation, which should be used for monitoring the process. Also unwanted variation is present in the measured spectrum. One example is variation due to changes in the temperature of the sample. This unwanted variation will cause deterioration of the performance of the spectroscopic analyzer. In this paper the properties of continuous piecewise direct standardization (CPDS) for temperature corrections in near-infrared (NIR) spectra are discussed. A procedure to obtain sensible settings of the meta-parameters in the CPDS algorithm has been developed. Initial values of the meta-parameters can be found for each data set by looking at spectra as a function of temperature. Particularly, difference spectra are shown to be useful in order to determine the degree of linearity of the spectral disturbances caused by the temperature changes. A cross-validation procedure around these initial settings is proposed to establish the optimum meta-parameters. This procedure was successfully applied to two NIR data sets. For each data set, it was found that CPDS corrected for temperature disturbances in an independent test data set and maintained the prediction errors on the same low level as for a calibration model built at the calibration temperature.