Study On Crude Oil And Fuel Oil Species Identification Technology Based On Multivariate Statistical Analysis

With the rapid development of China’s economy and the expanding of energydemand gap,the number of China’s annual imports of petroleum products from theinternational market continues to expand. And imports cover the increasingly widerange of places. Because China’s import of crude oil implements a quota system,whilethe fuel oil implements an automatic licensing system.So many criminals importcrude oil in the name of fuel oil,seriously disrupt the market order.Although previousstudies have established prediction models for the identification of crude oil and fueloil with n-alkanes, pristine and phytane.But original model should detect manyindexes,and the nature of indexes are unstable.Then the accuracy of the model shouldbe improved.Therefore, in order to better carry out species identification of crude oil/fuel oil and standardize the market order,finding more effective indexes to establish aquicker and more accurate model is very important.In this study,twenty-six fuel oil samples and twenty-five crude oil samplesaround the world are chosen to detect eight typical polycyclic aromatic hydrocarbons(PAHs) by gas chromatography coupled to mass spectrometry(GC-MS).By runningSPSS16.0software,principal component analysis(PCA),discriminatory analysis andlogistic regression analysis are made and six models are established to discriminatecrude oil and fuel oil.This paper also does a comparative analysis of the accuracy.As aresult,the classification models for crude oil and fuel oil established in this paper arequick and accurate.They can be applied to the import of crude oil and fuel oilsupervision.And the nature of PAHs are stable and less susceptible to weathing,so themodels can also provide a guide for the oil spill work.The conclusions of this study are as follows:1. Two discriminatory analysis models of crude oil and fuel oil species prediction are established.They are Fisher discriminatory analysis and Bayes discriminatoryanalysis.The accuracy of these two models are both94.1%.Also a logistic regressionanalysis model is established with the accuracy of100%.These three models can bewell used to discriminate crude oil and fuel oil,especially logistic regression analysismodel’s accuracy reaches100%.And the goodness of fitness test results show that thelogistic regression analysis model can well explain the fact data.2.This study combines the principal component analysis (PCA) with thediscriminatory analysis and logistic regression analysis, establishes the PCA Fisherdiscriminatory analysis model, PCA Bayes discriminatory analysis model and PCAlogistic regression analysis model.The accuracy of PCA Fisher discriminatoryanalysis and PCA Bayes discriminatory analysis is84.3%,and the accuracy of PCAlogistic regression analysis is86.3%.The results show that the general discriminatoryanalysis is better than the PCAdiscriminatory analysis.The innovations of this study are as follows:This study is based on multivariate statistical analysis with SPSS16.0statisticalanalysis software. It optimizes the original models using n-alkanes, pristine andphytane as indexes. Eight kinds of more stable polycyclic aromatichydrocarbons(PAHs) are firstly chosen as indexes to establish discriminatory analysisand logistic regression analysis models for crude oil and fuel oil classificationprediction.And this study also gives the functions and guidelines.The models not onlyreduce the indexes but also improve the accuracy.So they can make up for thedeficiencies of the original models.And PAHs are less susceptible to weathing,so themodels also can provide a guide for the oil spill work.