利用穷举法建立多维化学指纹量化模型鉴别海上原油的研究

On approach to identifying marine oil spill by establishing a quantitative model of multidimensional chemical fingerprint through exhaustion method

为了解决一维油指纹鉴别风化油种准确率不高的问题,提出多维化学指纹量化模型。由于海上溢油种类主要为原油或燃料油,以7种中东原油、8种非中东原油共15种油样为研究对象,对构成多维化学指纹的3个参数正构烷烃、荧光和镍钒比风化前后的变化特征进行解析,并借助化学计量学方法对正构烷烃n-C16~n-C35进行主成分分析,提取出4个主成分;同步荧光光谱利用db7小波基进行6层离散小波变换,在d3下提取出5处荧光信息。结合镍钒比采用穷举法优化出正构烷烃和荧光具有代表性的参数作为建模变量。从4个主成分中筛选出第三主成分,从5处荧光信息中筛选出(280±2) nm的小波系数。以筛选出的3个变量建立Fisher判别模型。结果表明,建立的模型具有变量少、信息量大、操作方便等特点,对15种建模油样鉴别准确率达到100%,对非建模和风化后16种油样判别准确率达到93.8%,比现有报道溢油鉴别方法的准确率提高了0.7%,能够有效鉴别油种。

The paper is intended to examine and trace the influence or impact of the experimental methods and the personal subjective discriminative identification of the oil spill through the qualitative analysis of the traditional single-D chemical fingerprinting. The reason of our choice for the singleness and limitations of the single-dimensional chemical fingerprinting lies in that it is often the case that misjudgment or recognition failure,especially incapable identification of the marine oil spill as a result of repetitive weathering. However, a quantitative identification based on the multi-dimensional chemical fingerprinting proposed by us in this paper can help to solve such a hardnut problem quite easily. To be exact,seeing that the marine oil spills is mainly referring to the crude oil or the marine fuel,a total of 15 oil samples from 7 Middle Eastern crude oils and 8 non-Middle East crude oils have to be used as the research samples. That is,first of all,the changing features of N-alkanes,the fluorescence and the ratio of nickel to vanadium in the oil samples pre-&-pro weathering can be suggested for analysis,with the multi-D chemical fingerprinting made of the 3 above mentioned parameters. By means of chemometrics,it would be possible for the 4 main components to be able to get extracted from the N-alkanes n-C16-nC35 through the principal component analysis. And,then,synchronized fluorescence spectra can be gained through analyzing the 6-layer discrete wavelet transform on the db7 wavelet basis,with 5 kinds of discrete wavelet information being extracted at D3. And,then,it would be possible to optimize the representative parameters of N-alkane and fluorescence in combination with the ratio of nickel as the modeling variables to vanadium through the exhaustive method. The 3 rd principal component can then be chosen from the 4 principal components by selecting the wavelet coefficient at( 280 ± 2) nm from the 5 kinds of discrete wavelet information. And,finally,a Fisher discriminant model can be set up by the 3 optimized variables,which can be featured by less variables. And,eventually,great amount of information and convenient operation can be used to identify the discriminant accuracy entirely for the 15 oil samples adopted in the given model,with 93. 8% for the 16 oil samples being non-modeled and weathered,which should be 0. 7 percentage point higher than the identification method as has been reported.