Identification Of Marine Oil Spill By Logistic Regression Based On Fluorescence Characteristics Of Oil

The identification of oil species has become an important part of oil spill analysis.Although China has carried out a lot of oil spill identification researches,the identification technology still need to further improve,especially in the case of fast and accurate identification of oil spill.At present,most of the oil spill identification is the qualitative spectrum research,but there are few studies on quantitative identification of oil spill.Therefore,based on this subject,to establish the logistic regression model of fuel oil and crude oil,which is with highly targeted and accurate.This study collected twenty two kinds of crude oil and twelve kinds of fuel oil,and weathering experiment for 30 days.The fluorescence characteristics of oil samples were detected by constant-wavelength synchronous fluorescence method at A?=30nm.The fluorescence spectra for thirty-four kinds of oil samples were analyzed,and the fluorescence intensity values at 280±2nm,300±2nm,320±2nm and 380±2nm were extracted.The modified cosine similarity coefficient of I280nm,I300nm,I332nm,I380nm were between 0.91 and 0.98 before and after weathering,indicating that the short-term weathering has little effect on the fluorescence intensity.Therefore,the model established with unweathered oil samples is also suitable for short-term weathering oil samples.In this study,the oil samples I280nm,I300nm,I332nm,I380nm before weathering were analyzed by normal analysis,variance inflation factor and factor analysis.The results of normal analysis showed that I28nm,I300nm,I332nm,I380nm of crude oil and fuel oil do not accord with the normal distribution.The results of variance inflation factor showed that I380nm is a certain correlation with I280nm,I300nm,I332nm.The information was overlapped and removed.Factor analysis showed that the fluorescence information of I280nm,I300nm,I332nm,I380nm extracted from two common factors,and characterized by 980%,99%,98%,97%of the original information.This study was based on the logistic regression,using the information of variance inflation factor analysis and the extractive common information of factor analysis for modeling.The logistic regression model based on variance expansion factor analysis showed that the accuracy of identification of oil samples before and after weathering was 93%and 98%;The logistic regression model based on factor analysis showed that the accuracy of identification of oil samples before and after weathering was 81.5%and 84%.The logistic regression model based on the variance expansion factor analysis is effective.The model was used to identify fifteen kinds of non-modeling oil samples that before and after weathering,and the accuracy of identification was 93%.In this study,the logistic regression model for identifying oil species is established,which provides a theoretical basis for the development of portable fluorescence detectors used for real-time and on-line identification of oil spills.