Near infrared spectroscopic determination of diesel fuel parameters using genetic multivariate calibration

Abstract

The use of full spectral region from near infrared spectroscopic analysis does not always end up with a good multivariate calibration model as many of the wavelengths do not contain necessary information. Due to the complexity of the spectra, some of the wavelengths or regions may, in fact, disturb the model-building step. Genetic algorithms are one of the useful tools for solving wavelength selection problems and may improve the predictive ability of conventional multivariate calibration methods. This study demonstrates application of genetic algorithm-based multivariate calibration to near infrared spectroscopic determination of several diesel fuel parameters. The parameters studied are cetane number, boiling and freezing point, total aromatic content, viscosity, and density. Multivariate calibration models were generated using genetic inverse least squares (GILS) method and used to predict the diesel fuel parameters based on their near infrared spectra. For each property, a different data set was used and in all cases the number of samples was around 250. Overall, percent standard error of prediction (%SEP) values ranged between 2.48 and 4.84% for boiling point, total aromatics, viscosity, and density. However, %SEP results for cetane number and freezing point were 11.00% and 14.86%, respectively.