为实现汽油中所含组分含量的快速测定, 对93号、 97号汽油, 芳烃、 烯烃、 苯、 甲醇、 乙醇等几类物质, 以及往汽油中添加几类物质后的410个汽油混合物进行拉曼光谱检测。 将获取的原始拉曼光谱经过有效波段提取、 平滑去噪、 基线扣除、 数据归一化等一系列预处理过程, 最终提取出每个汽油混合样品光谱中所含的33个特征峰信息, 依据现行的国标检测方法, 以气相色谱法测定的汽油中各组分含量值为基础, 结合化学计量学多重回归分析方法, 建立了汽油组分含量测定模型。 经过比较, 使用多输出最小二乘支持向量回归机(MLS-SVR)建立的模型优于偏最小二乘(PLS)模型。 MLS-SVR模型对汽油中芳烃、 烯烃、 苯、 甲醇、 乙醇测定精度均较好, 预测均方根误差(RMSEP)分别为0.27%, 0.30%, 0.16%, 0.17%, 0.12%; 相应的相关系数(r)为0.999 2, 0.998 4, 0.998 5, 0.992 6, 0.996 8。 通过对未知混合汽油样品的测定, 证明了该方法具有较好的推广预测精度, 预测均方根误差不超过0.5%, 能够满足工业中的测量需求。 拉曼光谱结合多输出最小二乘支持向量机为汽油组分测定提供了一种高精确、 快捷、 方便的测定方法。

For the purpose of the rapid prediction of every composition in gasoline, the Raman spectra of the gasoline brand 93 and 97, a batch of one-one mixtures with aromatic, olefin, ben, methanol and ethanol with different ratios are measured, 410 mixture samples were measured totally in this research. The obtained Raman spectra were preprocessed by a series of processing, they were data smoothing, baseline deduction and spectral normalized, etc. After that 33 characteristic peaks were extracted to be the eigenvalues for the whole Raman spectra. According to the current national standard test method, the values of every composition were measured by the gas chromatography. By using the eigenvalues as inputs, and actual contents of aromatic, olefin, ben, methanol and ethanol got from gas chromatography as outputs, two mathematical models of multi-output least squares support vector regression and partial least squares combination with multiple regression analysis were established to predict the values of the above compositions of a sample, respectively. The predicting results were compared with the values calculated from the gas chromatography measurement results and the mixture proportions, the multi-output least squares support vector regression has a better effects, and the obtained root mean square error of prediction for aromatic, olefin, ben, methanol and ethanol are 0.27%, 0.27%, 0.22%, 0.17%, 0.14%; the correlation coefficients are 0.999 3, 0.998 5, 0.998 6, 0.992 3, 0.993 5, respectively. This model is also applied to the detection of the unknown sample, the root mean square error of the prediction for the results does not exceed 0.5%, which can achieve the measurement requirements in the industry. Results show that the Raman spectra analysis technology based on multi-output least squares support vector regression can be a precise, fast and convenient new method for gasoline composition detection, and can be applied to the quality control of the gasoline production process, transportation, storage of the gasoline.