应用近红外光谱结合化学计量学方法实现了对甲醇柴油中的甲醇含量的定量分析。 以实验室配制的32种不同浓度［浓度范围为2%～25.8%(φ)］的甲醇柴油溶液为研究对象, 在4　500～7　000 cm-1光谱范围内, 建立偏最小二乘(PLS)、 支持向量机(SVM)和最小二乘支持向量机(LS-SVM)三种定量分析模型。 在建立SVM模型时, 经过比较分析, 径向基函数(radial basis function, RBF)作为SVM模型的核函数时可以获得更高的预测精度。 最终获得甲醇含量的PLS, SVM和LS-SVM三种模型的预测相关系数RP分别为0.985 9, 0.990 3, 0.998 9, 预测均方根误差RMSEP分别为0.405 2, 0.356 3, 0.062 4, 可以看出, 三种预测模型都可以达到很好的效果, 最优的预测模型是使用LS-SVM建模。 研究结果表明, 利用近红外光谱法结合化学计量学方法对甲醇柴油中甲醇含量的检测具有可行性, 并可以达到很好的效果。 采用近红外光谱技术结合化学计量方法对甲醇柴油中甲醇含量进行定量分析, 也为近红外光谱技术快速无损检测甲醇柴油甲醇含量提供参考和应用价值。

Near infrared spectroscopy combined with chemometric resolution method was applied to analyze methanol content in methanol diesel quantitatively. In this experiment, 32 samples of methanol diesel in the range of 2%~25.8% (φ) were used for spectroscopy collection and analysis. Three quantitative analysis models of partial least squares (PLS), support vector machine (SVM) and least squares support vector machines (LS-SVM) were established in the range of 4 500~70 000 cm-1. The Radial Basis Function (Radial Basis Function, RBF) was the optimal kernel function of the SVM model compared three different kinds of prediction model during process of building the SVM model. The research results showed correlation coefficient (RP) and the root mean square error of prediction (RMSEP) of PLS, SVM and LS-SVM were 0.985 9, 0.990 3, 0.998 9 and 0.405 2, 0.356 3, 0.062 4 respectively. It was obviously that the three kinds of prediction model could obtain better results especially with LS-SVM modeling. The research results indicated that near infrared spectroscopy combined with chemometrics resolution method could achieve better results when used for detecting methanol content in methanol diesel. Moreover, using near infrared spectroscopy combined with chemometric resolution method to analyze methanol content in methanol diesel quantitatively may have reference value and application potentiality for near infrared spectroscopy quick and non-destructive test of methanol content in methanol diesel.