用相关系数法匹配色谱图形鉴别海面溢油源

IDENTIFYING THE SOURCE OF SPILLED OIL AT SEA BY MATCHING GAS CHROMATOGRAM PATTERN WITH CORRELATION COEFFICIENTS

用皮尔逊相关系数法匹配色谱图形,根据溢油与可疑油的相似性鉴别溢油源。通过风化实验,研究出六个受风化影响很小的油色谱峰作为信息点,用K-最邻近分类法鉴别非常相似的油品。

Chromatograms peaks of six slightly weathered oil samples were selected as information points C17, Pr, C18, Ph, C19 and C20 in the field experiment. The degree of match between suspected spill oil chromatogram peaks (determined by the Peatson correlative coefficient) and the chromatogram area as a basic element was utilized to identify the spilled oil. If the spilled oil and the suspected source oil come from the same source, their chromatograms will inevita-blly be the same and the correlation coefficient will be equal to or near 1. This method w,orks within the instrument limitation, and therefore will not be influenced by the oil concentration. In this study, four crude oils and four light fuel oil samples were selected. Duplicated chromatograms were run for each sample under the instrumental conditions. After that, all the chromatograms were calculated according to the correlation matrix. The results showed that the correlation coefficients of samples within the batch were all>0.9900, and<=======0.9000 between the batches. The impact of weathering to the oil chromatogram fingerprint is important. After one month weathering experiment on Daqing crude oil and O# liyht fuel oil on the sea, the fluctuation of the weathered oil, especially O# light fuel oil and oil with carbon less than C16, chromatogram was larger, but it can not be recognized by the naked eye. By applying the six selected chromatograms as information point, the unmeathered and weathered oil chromatograms were calculated correlatively. The correlation coefficient was still over 0.9900. Identification of similar oil samples was regarded as difficult. Here, the so-called similar oil is that the correlation coefficients of two different oil samples are 0.9900 and slightly different at the 3rd and 4th after the point which can be effectively recognized by K-nearest neighbor classification. The same oil sample should have maximum similarity, therefore, maximum weighting value is given to the largest correlation coefficient. The method is l/(1-γ)2(γ is correlation coefficient).