Classification and determination of alcohol in gasoline using NIR spectroscopy and the successive projections algorithm for variable selection

Abstract

A methodology for the classification and determination of alcohol (methanol/ethanol) in gasoline using near-infrared reflectance spectrometry and variable selection was proposed. Methanol gasoline and ethanol gasoline were prepared in the laboratory and gasoline (93#) was acquired from a local gas station. Partial least squares (PLS) multivariate calibrations were used to predict methanol/ethanol content. Principal component analysis was used for spectrum classification, obtaining a desirable classification accuracy. Using this strategy, it was feasible to classify alcohol gasoline rapidly. Concerning the multivariate calibration models, the results show that PLS, successive projections algorithm (SPA)-PLS and genetic algorithm (GA)-PLS models are good for predicting methanol and ethanol contents in gasoline; the respective root-mean-square errors of prediction were 0.216 (PLS), 0.163 (SPA-PLS) and 0.210 v/v% (GA-PLS) for methanol gasoline, corresponding to 0.348, 0.235 and 0.203 for ethanol gasoline. The results obtained in this investigation suggest that the proposed methodology is a promising alternative for the determination of alcohol content in gasoline.