摘要： 甲醇汽油和乙醇汽油都为清洁能源，但甲醇汽油和乙醇汽油的优缺点各有不同，其中甲醇、乙醇的含量决定了汽油性能的优劣，对甲醇汽油和乙醇汽油进行判别区分以及醇类汽油中醇含量进行定量测定非常重要。通过中红外光谱技术对醇类汽油的类型进行判别并对其含量进行定量分析。首先通过对比分析甲醇汽油和乙醇汽油的中红外光谱图，采用随机森林（random forest, RF）对甲醇汽油和乙醇汽油样品进行判别；在建立甲醇汽油和乙醇汽油样品定性判别模型之后，分别建立甲醇汽油和乙醇汽油的定量测定模型，从而精确测定汽油中对应的醇的含量。为减小在实验过程中实验仪器振动、噪声等原因导致的光谱漂移、光散射等现象，对中红外光谱进行预处理。首先采用不同预处理，如(savitzky-golay，S-G)卷积平滑、多元散射校正(multiplicative scatter correction, MSC)、标准正态变量变换(standard normal variable transformation, SNV)、导数(derivatives)等方法进行校正，分别建立适合甲醇汽油和乙醇汽油的检测模型。预处理后的数据分别建立甲醇汽油、乙醇汽油的最小二乘支持向量机（least square support vector machine, LS-SVM）模型。采用随机森林（random forest, RF）对甲醇汽油和乙醇汽油样品进行判别，发现当决策树个数为61时，判别正确率达到98.28%。对于LS-SVM模型，比较建模结果可知：无论是甲醇汽油还是乙醇汽油，标准正态变量变换（SNV）预处理效果最好，经SNV校正处理后建立的甲醇汽油甲醇含量测定LS-SVM模型的预测相关系数Rp为0.9519，均方根误差(root mean square error of prediction，RMSEP)为1.766 3；经过标准正态变量变换后建立的乙醇汽油乙醇含量测定LSSVM模型的预测相关系数Rp为0.951 5，均方根误差RMSEP为1.770 3。该研究可为甲醇汽油、乙醇汽油的定性判别和其含量测定提供技术参考和理论依据，为甲醇汽油产业提供测量醇类汽油检测的新方法，具有较为重要的现实意义，也为其他类型的化工产品的检测奠定了基础。
关键词：中外光谱；醇类汽油；最小二乘支持向量机；随机森林
Abstract：Methanol gasoline and ethanol gasoline are both clean energy sources, but the advantages and disadvantages of them are different. Among them, the content of methanol or ethanol determines the performance of gasoline. Therefore, it is of great significance to qualitatively distinguish methanol gasoline and ethanol gasoline and quantitatively determine the alcohol content in alcohol gasoline. In this paper, the types of alcohol gasoline and its content were identified and quantitatively analyzed by mid-infrared spectroscopy. Firstly, by comparing and analyzing the mid-infrared spectroscopy of methanol gasoline and ethanol gasoline, Random Forest (RF) was used to discriminate methanol gasoline and ethanol gasoline samples. After establishing the qualitative model of methanol gasoline and ethanol gasoline, the quantitative determination model of methanol gasoline and ethanol gasoline is established to accurately determine the corresponding alcohol content in gasoline. In order to reduce the spectrum drift and light scattering caused by vibration and noise of the experimental instrument during the experiment, the mid-infrared spectrum was pretreated. In the process of analysis, different pre-treatment methods are first used for correction, such as S-G convolution smoothing, Multivariate Scattering Correction (MSC), Standard Normal Variable (SNV), derivatives (1st derivative and 2nd derivative), and then, Least Square Support Vector Machine (LSSVM) models of methanol gasoline and ethanol gasoline were established respectively. It was found that the discriminant accuracy is up to 98.23% when the number of decision trees was 61. Secondly, for the LS-SVM model, the comparison of modeling results showed that for both methanol gasoline and ethanol gasoline, SNV pre-treatment had the best effect. The predictive correlation coefficient Rp of LSSVM model after the transformation of standard normal variables for methanol content determination of methanol gasoline was 0.951 9 and RMSEP was 1.766 3. In the same situation, ethanol gasoline was 0.951 5 and 1.770 3, respectively. This research can provide technical reference and theoretical basis for the qualitative discrimination and content determination of methanol gasoline and ethanol gasoline. The detection technology can provide a new method for the measurement of alcohol gasoline in the methanol gasoline industry, which has important practical significance and lays a foundation for the detection of other types of chemical products.
Key words：Mid-infrared spectroscopy; Alcohol gasoline; Least square support vector machine; Random forest
胡 军，刘燕德，郝 勇，孙旭东，欧阳爱国. 甲醇汽油、乙醇汽油定性判别及其醇含量测定模型研究[J]. 光谱学与光谱分析, 2020, 40(05): 1640-1644.
HU Jun, LIU Yan-de, HAO Yong, SUN Xu-dong, OUYANG Ai-guo. Qualitative Discrimination and Quantitative Determination Model Research of Methanol Gasoline and Ethanol Gasoline. SPECTROSCOPY AND SPECTRAL ANALYSIS, 2020, 40(05): 1640-1644.
[1] LIU Yan-de, HU Jun, TANG Tian-yi, et al(刘燕德, 胡 军, 唐天义, 等). Spectroscopy and Spectral Analysis(光谱学与光谱分析), 2019, 39(2): 459.
[2] OUYNAG Ai-guo, LIU Jun, WANG Ya-ping, et al（欧阳爱国，刘 军，王亚平，等）. Laser and Infrared（激光与红外）, 2012, 42(8): 901.
[3] DING Hong-yuan, HUANG Rong-hua, WANG Zhao-wen, et al(丁红元, 黄荣华, 王兆文，等). Journal of Huazhong University of Science and Technology·Natural Science Edition(华中科技大学学报·自然科学版), 2012, 6: 113；2013, 1: 59.
[4] JIN Pan-pan, ZHANG Teng, YANG Xiao-ping, et al（金盼盼, 张 腾, 杨晓平, 等）. Auto Mobile Science and Technology（汽车科技）, 2012, (5): 66.
[5] LIU Yan-de, WU Ming-ming, SUN Xu-dong, et al（刘燕德, 吴明明, 孙旭东, 等）. Transactions of the Chinese Society of Agricultural Engineering(农业工程学报), 2016, 6: 289.
[6] HAN Zhong-zhi, WAN Jian-hua, LIU Kang-wei, et al（韩仲志, 万剑华, 刘康炜, 等）. Chinese Journal of Analysis Laboratory(分析试验室), 2015, 11: 1268.
[7] YAO Jie, DAI Lian-kui, LIN Yi-ling(姚 捷, 戴连奎, 林艺玲). Chinese Journal of Light Scattering（光散射学报）, 2013, 25(1): 59.
[8] OUYANG Ai-guo, LIU Jun(欧阳爱国, 刘 军). Journal of Southwest China Normal University·Natural Science Edition（西南师范大学学报·自然科学版）, 2012, 37(9): 98.
[9] LI Yan-ru, SUN Rui-qing, WANG Yong-miao（李雁如, 孙瑞卿, 王永苗）. Modern Chemical Industry(现代化工), 2013, 33(2): 113.
[10] Hutengs C, Vohland M. Remote Sensing of Environment, 2016, 178: 127.
[11] Jog A, Carass A, Roy S, et al. Medical Image Analysis, 2017, 35: 475.
[12] Silva M P F D, Brito L R E, Honorato F A, et al. Fuel, 2014, 116(1): 151.
[13] Zhang L, Li G, Sun M, et al. Infrared Physics & Technology, 2017, 86: 116.
