On Crude Oil And Fuel Oil Species Identification Technology Based On Chemometrics

In recent years, classification identification of crude oil and fuel oil has become a difficult problem that needs to be solved urgently for import and export oil inspection and supervision. To meet the need of the national economic development, China has to import a large quantity of crude and fuel oil from the international market every year. However, policies for importing crude oil and fuel oil are different in China:a strict quota system that only a few of large enterprises have got the import qualification is applied for the import of crude oil, while an automatic licensing system is applied for the import of fuel oil, i.e., enterprises have rights to legally import fuel oil, according to their own requirement. Since crude oil and fuel oil are similar in appearance, some traders smuggle crude oil in the name of fuel oil. These smuggling acts seriously disrupt the trade order, and bring bad influence to the national economic security of China, which have not been solved up to now. In this study, we detected masses of hydrocarbons that can be used as oil fingerprints. Furthermore, by using hydrocarbon in the oil samples as indicators, species identification models have been developed for identifying crude oil and fuel oil, by means of multivariate statistical analysis.In this study, a group of 138 samples of crude and fuel oil collected from different oil origins around the world have been used to detect 31 kinds of hydrocarbon, which can be used as oil fingerprints, including 8 kinds of polycyclic aromatic hydrocarbons, i.e., n-alkanes (n-C9～n-C29, including pristan and phytane), 2-mythylnaphthalene,1-mythylnaphthalene,2,6-dimethylnaphthalene, 2,3,5-trimethynaphthalene,o-biphenylenemethane,2,2-biphenylylene sulfide, phenanthrene and chrysene, by means of gas chromatography coupled to mass spectrometry (GC-MS). From the chromatogram and data, it can be seen that the content and carbon chain of hydrocarbon in different oil samples are different, but the difference between crude oil and fuel oil is not significant. It is found from P-P diagrams that the distributions of all hydrocarbon are not normal (Gaussian distribution). By means of the analyses of robust statistical techniques, we found that the median values of the contents of n-alkanes (n-C9-n-C29, including pristan and phytane) in crude and fuel oil are generally higher than those of polycyclic aromatic hydrocarbons by several times to dozens times; the median content values of n-alkanes in crude oil are mostly larger than those in fuel oil, while the median values of polycyclic aromatic hydrocarbons in crude oil are mostly lower than those in fuel oil. The box-plot figures show that the dispersion values of the n-alkanes in crude oil are relatively small, while, for other hydrocarbons, there are about 10% of sample points that deviate from the mean levels; in fuel oil, the abnormal samples of all indicators are in the range of 1-3, which are relatively smaller than those in crude oil. Overall, from the original data of the hydrocarbons and the basic statistics mentioned above, differences between the contents of n-alkanes and polycyclic aromatic hydrocarbons can be found, but this cannot further identify the specie difference between crude oil and fuel oil.In this study, the relationship and characteristics of hydrocarbons in the oil samples are further analyzed by means of multivariate statistical analysis. From correlation analysis, we found that the correlations between all n-alkanes are statistically significant, while the correlations between the 8 kinds of polycyclic aromatic hydrocarbons are only 50% significant; in n-alkanes, only pristan, phytane and short n-alkanes are significantly correlated to polycyclic aromatic hydrocarbons. Based on principal component analysis (PCA), the dimensions of the array of 108×31 (sample points × numbers of the detected hydrocarbons) were reduced, and the results showed that for the 108 oil samples, in the space of PCl-PC2 only-20% of sample points that are mostly crude oil are off the mean level, while the two species of crude oil and fuel oil are not separately clustered. Therefore, if we only use n-alkanes and polycyclic aromatic hydrocarbons as variables in PCA, we cannot distinguish crude oil and fuel oil.Based on the above analyses,9 kinds of identification models have been developed in this study, by using the Bayes discriminatory analysis (BDA) and Fisher discriminatory analysis (FDA) and binary logical regression analysis (BLRA), and by means of 31 kinds of hydrocarbons including n-alkanes and polycyclic aromatic hydrocarbons. The accuracy of model identification to the oil species by using BLRA is better than by using BDA or FDA, and is up to 100% if using the identification model based on n-alkanes or if using the identification model based on the 31 kinds of hydrocarbons. If we use all hydrocarbons as identification indicators in the model, the accuracy by means of BDA and FDA can get to over 80%. This study indicates that the more kinds of hydrocarbons are used in the model, the accuracy of the identification is higher. Hence, more hydrocarbons, such as long n-alkanes (carbon chain above 30), sterane, and terpanes, will be used in the identification models in the future study, which might improve the model performance.Overall, it is the first time in this study to collect, detect and contents of n-alkanes and polycyclic aromatic hydrocarbons that posse the property of fingerprints in oil. Characteristics of hydrocarbons were analyzed by means of multivariate statistical analysis including correlation analysis, PCA, and factor analysis. Models were developed to identify the crude oil and fuel oil, based on discriminatory analysis and binary logical regression analysis, which have achieved accurate identification of crude oil and fuel oil.