油品荧光特性的Logistic回归分析鉴别海上溢油的研究

Study on identification of marine oil spill using logistic regression based on petroleum fluorescence characteristics

用恒波长同步荧光法检测了280±2 nm、300±2 nm、332±2 nm 3个特征波长下的22种原油和12种燃料油的荧光强度,并以此作为变量,建立了快速定量鉴别海上溢油的原油和燃料油的Logistic回归模型。Hosmer-Lemeshow拟合优度检测值2.43小于15.51、其对应P值0.97大于0.05,说明建模数据的信息被充分提取和利用。风化前后油样的受试者工作特征曲线(ROC曲线)分析显示该鉴别方法准确性较高,曲线下边面积(AUC)大于0.9,灵敏度和特异性大于0.85。用15种非建模油样对所建立的模型进行验证,计算结果与实际情况吻合度较高,鉴别正确率达93%。此外,该模型也适用于油种风化后的鉴别。

Fluorescence intensity of twenty two crude oils and twelve marine fuels were detected by a constant wavelength synchronous fluorescence spectrometry at three feature wavelengths 280 - 2 nm,300 - 2 nm and 332 - 2 nm, A logistic regression model with fluorescence and wavelength as variables was established, which was used for rapid quantitative identification of crude oil and marine fuel spilled on the sea. Hosmer-Lemeshow goodness of fit test of 2.43 less than 15.51 and P of 0. 97 greater than 0. 05, these indicated that the extraction and the utilization of modeling data were sufficient. Receiver operating characteristic curves of oil samples before and after weathering showed that the model had high accuracy for oil species distinguishing, AUC was greater than 0. 9, the sensitivity and the specificity were greater than 0.85. Fifteen oil samples except modeling were verified by the model, the calculation results matched with the actual truth very well, the identification accuracy reached 93%. In addition, The model was also applicable to the identification of oil species after weathering.