文摘
Multivariate calibration models are constructed through the use of Gaussian basis functions to extract relevant information from single-beam spectral data. These basis functions are related by analogy to optical filters and offer a pathway to the direct implementation of the calibration model in the spectrometer hardware. The basis functions are determined by use of a numerical optimization procedure employing genetic algorithms. This calibration methodology is demonstrated through the development of quantitative models in near-infrared spectroscopy. Calibrations are developed for the determination of physiological levels of glucose in two synthetic biological matrixes, and the resulting models are tested by application to external prediction data collected as much as 4 months outside the time frame of the calibration data used to compute the models. The calibrations developed with the Gaussian basis functions are compared to conventional calibration models computed with partial least-squares (PLS) regression. For both data sets, the models based on the Gaussian functions are observed to outperform the PLS models, particularly with respect to calibration stability over time.