Establishment and Verification of Analytical Models Applied to NIRS Determination of Moisture and Sulfur in Crude Oil

Moisture and sulfur in 120 samples of crude oil were determined by the standard methods given in GB/T 11133-2015 and GB/T 17040-2008, respectively. And the transmission NIR spectra of the 120 samples were collected under the optimized condition. 4 abnormal samples were rejected by the algorithm of Leverage. Analytical model for prediction of moisture contents was established under the following conditions:① method of Savitzky-Golay was applied for wave-filtering pretreatment of the NIR spectra; ② spectal section in the range of 6 200 to 8 200 cm^-1 was selected for the model establishing; ③ number of principle components was 6; and ④ the algorithms of PLS and alternate verification were used in the final model-establishment. Conditions for establishing the analytical model for prediction of sulfur contents were as follows:① method of 2nd derivative-Norris Derivative was applied to the pretreatment of the NIR spectra; ② spectral section between 4 400 to 4 700 cm^-1 was selected for model establishing; ③ number of principle components was 6; and ④ the algorithms of PLS and alternate verification were adopted in the final model-establishment. Good correlationships between values of prediction and of determination of both the moisture and sulfur contents were obtained. And values of R_c² and RMSEC found were 0.989 9 and 0.084 2 respectively (for the model of moisture prediction), and 0.996 3 and 0.069 6 respectively (for the model of sulfur prediction), showing that those two models can be used effectively for the prediction of moisture and sulfur contents in the crude oil. Ten unknown samples of crude oil were analyzed by NIRS under the prescribed condition and contents of moisture and sulfur were predicted by the 2 models. It was found that the predicted values of the 2 components in the 10 sample were in consistent with the values of the 2 components in these samples obtained by the standard analytical methods, giving relative deviations less than 10%.