拉曼光谱结合偏最小二乘的甲醇汽油甲醇含量快速定量分析方法研究  

Study on Rapid Quantitative Analysis Method of Methanol Content in Methanol Gasoline by Raman Spectroscopy and Partial Least Squares

作　　者：李茂刚 闫春华 杜瑶 张天龙[2] 李华[1,2] LI Mao-gang;YAN Chun-hua;DU Yao;ZHANG Tian-long;LI Hua(College of Chemistry and Chemical Engineering,Xi’an Shiyou University,Xi’an 710065,China;Key Laboratory of Synthetic and Natural Functional Molecular of the Ministry of Education,College of Chemistry&Materials Science,Northwest University,Xi’an 710127,China)

机构地区：[1]西安石油大学化学化工学院,陕西西安710065 [2]合成与天然功能分子教育部重点实验室,西北大学化学与材料科学学院,陕西西安710127

出　　处：《光谱学与光谱分析》2021年第7期2099-2104,共6页Spectroscopy and Spectral Analysis

基　　金：国家自然科学基金项目(21873076,21675123,21605123)资助。

摘　　要：甲醇汽油是一种用以替代传统汽油的新型燃料,其品质受到甲醇含量的严重影响。因此,甲醇汽油中甲醇含量的快速分析对其品质把控具有深远意义。基于拉曼光谱(Raman)结合偏最小二乘(PLS)建立了一种甲醇汽油中甲醇含量快速定量分析方法。采用激光拉曼光谱仪对49组甲醇汽油样品的Raman光谱进行采集,并进行光谱解析。比较了五种光谱预处理方法对甲醇汽油原始Raman光谱的预处理效果,并采用变量重要性投影(VIP)对小波变换(WT)预处理后的甲醇汽油Raman光谱数据进行了特征变量提取。其次,采用五折交叉验证(5-flod cross-validation(CV))对PLS校正模型的潜变量数目(LVs)及VIP阈值进行优化。在最优输入变量和模型参数下,分别构建了基于不同输入变量的PLS模型。研究表明,相较于原始光谱-偏最小二乘模型(RAW-PLS)和小波变换-偏最小二乘模型(WT-PLS),变量重要性投影-偏最小二乘模型(VIP-PLS)可以获得更好的分析性能,其预测集决定系数(R^(2)_(p))为0.9604,均方根误差(RMSEP)为0.0341。因此,Raman光谱结合PLS是一种快速准确的甲醇汽油中甲醇含量分析方法。Methanol gasoline is a new fuel to replace traditional gasoline,and its quality is greatly affected by methanol content.Therefore,the rapid analysis and detection of methanol content in methanol gasoline will have far-reaching significance for its quality control.A rapid quantitative analysis method of methanol content in methanol gasoline based on Raman spectroscopy and partial least squares(PLS)was established in this work.Raman spectra of 49 methanol gasoline samples were collected by laser Raman spectrometer,and spectral analysis was carried out.The effects of five spectral pretreatment methods on the raw Raman spectra of methanol gasoline were compared.In addition,variable importance in projection(VIP)was used to extract the Raman spectra’s feature variables preprocessed by wavelet transform(WT).The number of latent variables(LVs)and VIP threshold of the PLS calibration model was optimized by 5-flod cross-validation(CV).Under the optimal input variables and model parameters,PLS models based on different input variables were constructed.The results show that compared with RAW-PLS and WT-PLS,VIP-PLS can achieve better analysis performance,with the determination of the prediction set(R^(2)_(p))of 0.9604 and root mean square error of prediction set(RMSEP)of 0.0341.Therefore,Raman spectroscopy combined with PLS is a fast and accurate method for analysing methanol content in methanol gasoline.

关 键 词：甲醇汽油 拉曼光谱 变量重要性投影 偏最小二乘 

分 类 号：O657.37[理学—分析化学]