Quality characterisation of crude oils by partial least square calibration of NIR spectral profiles

Abstract

The correlation of near infrared (NIR) spectral profiles of crude oils with total boiling point (TBP) assay and density has been studied. Fourteen partial least square (PLS) models were calibrated by a training set of 110 samples. The developed NIR methods have reproducibility limits equivalent to the American Society for Testing and Materials (ASTM) reference procedures. Predictive accuracy of the models was estimated from the root mean square error of prediction (RMSEP) using cross validation. The correlation coefficients were better than 0.98 in these calibrations indicating good model structure. Such a rapid, reliable and inexpensive quantitative method, which is sufficient to assess the potential quality of the crude oil, especially the product yield, is essential in today's competitive refining environment.