Genetic Multivariate Calibration Methods for Near Infrared (nir) Spectroscopic Determination of Complex Mixtures

Abstract

The simultaneous determination of ternary mixtures of methylene chloride, ethyl acetate, and methanol using near infrared (NIR) spectroscopy and 4 different genetic algorithms based multivariate calibration methods was demonstrated. The 4 genetic multivariate calibration methods are genetic partial least squares (GPLS), genetic regression (GR), genetic classical least squares (GCLS) and genetic inverse least squares (GILS). The sample data set contains the NIR spectra of 63 ternary mixtures and covers the range from 900 to 2000 nm in 2 nm intervals. Of these 63 spectra, 42 were used as the calibration set, and 21 were reserved for the prediction purposes. Several calibration models were built with the 4 genetic algorithm based methods for each component that makes up the mixtures. Overall, the standard error of calibration (SEC) and the standard error of prediction (SEP) were in the range of 0.22 to 2.5 (% by volume (v/v)) for all the 4 methods. A comparison of genetic algorithm selected wavelengths for each component and for each method was also included.