【作者】 淡图南；

【导师】 戴连奎；

【作者基本信息】 浙江大学 ， 控制科学与工程， 2011， 博士

【摘要】 燃油组分和品质直接关系到其燃烧效率、发动机寿命以及尾气排放造成的环境污染,现有的标准检测方法存在成本高、速度慢、检测过程复杂等局限性。光谱分析技术具有快速、无损、低成本、安全可靠等许多现有的标准方法不可比拟的优势。本文采用近红外和拉曼光谱分析技术,结合样本分类、线性和非线性校正模型、异常样本剔除等方法,建立了汽柴油组分和品质指标的定量分析模型,并开发了相应的快速检测仪,具体研究内容包括：1、提出了一种结合支持向量机(SVM)分类模型的甲醇汽油组分近红外光谱检测方法。通过测量甲醇汽油的近红外光谱,并采用微分和标准正态变换对谱图进行了预处理；对甲醇汽油和不含甲醇的成品汽油建立了SVM分类模型,将甲醇汽油和成品汽油区分开；然后对于甲醇汽油,通过谱图相关性分析找到甲醇含量对应的特征波段,并建立了最小二乘支持向量机(LSSVM)定量分析模型。实验结果表明,采用上述近红外光谱法检测甲醇汽油中的甲醇含量可以达到很高的预测精度,预测复相关系数R2达到0.9926,均方预测误差(RMSEP)为0.58%(体积)。2、将结合偏最小二乘(PLS)特征提取的LSSVM算法(PLS-LSSVM)应用于柴油品质的近红外光谱检测。基于预处理后的柴油近红外光谱,分别建立了柴油十六烷值、硫含量、密度、50%回收温度的PLS-LSSVM定量校正模型,并与常用的PLS、LSSVM等方法进行了比较。实验结果表明,PLS-LSSVM模型集成了PLS与SVM模型的优势,能够减少样品待测属性与光谱之间非线性程度的影响。3、采用光栅色散型拉曼光谱,结合异常样本检测技术,建立了汽油芳烃含量、烯烃含量和氧含量的PLS定量分析模型。通过剔除个别异常样本,有效地提高了预测模型精度,得到了较好的预测效果。其中汽油芳烃含量模型R2达到0.9965, RMSEP为0.3；烯烃含量模型R2达到0.9274,RMSEP为0.52；氧含量模型R2达到0.9843,RMSEP为0.083。实验结果表明：采用拉曼光谱分析技术可以有效的解决汽油族组成的定量分析问题,其分析精度显著高于近红外光谱法,同时也适用于汽油生产过程中的在线分析。4.自主开发研制了基于近红外光谱的便携式甲醇汽油快速检测仪。针对目前市场上缺乏有效的甲醇汽油品质快速检测仪的现状,自行开发研制了基于光栅色散型近红外光谱仪的便携式甲醇汽油快速检测仪,并已应用于多个甲醇汽油生产企业。现场实际应用结果表明：该仪器对工业现场生产的甲醇汽油和成品汽油的分类正确,甲醇含量检测精度高,重复性好,操作和维护简便,能够满足国内甲醇生产企业、油库油站等用户对甲醇汽油甲醇含量快速检测的需求。更多还原

【Abstract】 The composition and quality of engine fuel directly influence its combustion efficiency, engine life and emissions which could cause the environmental pollution. The existing standard testing methods for engine fuel are generally slow, complex and high-cost. Spectral analysis technology is fast, nondestructive, low cost and safe, which makes it better than many existing standard methods. This thesis applies near infrared (NIR) and Raman spectroscopy in quantitative analysis of engine fuel composition and quality indices by combining with sample classification, calibration modeling and outlier detection techniques. Moreover, a fast NIR analyzer for methanol-gasoline is developed. Detailed research contents include:1. A detection method for methanol-gasoline composition is proposed based on NIR spectroscopy with support vector machine (SVM) classification model. Firstly, the NIR spectra of the methanol gasoline is measured and pre-processed by using derivative and standard normal variate. Secondly, an SVM classification model to distinguish methanol gasoline and regular gasoline without methanol is built. Thirdly, the feature wavelength of methanol content is found by correlation analysis between the spectra and methanol content. Finally, a least square SVM (LSSVM) quantitative model for the methanol content is established. Experimental results show that using NIR spectroscopy to detect the methanol content in methanol-gasoline can achieve high predictive accuracy. The corresponding multiple correlation coefficient of prediction (R2) is up to 0.9926; the root mean square error of prediction (RMSEP) is 0.58% (volume).2. The LSSVM algorithm with PLS feature extraction (PLS-LSSVM) is applied to the NIR spectroscopy detection of diesel quality. Based on the pre-processed diesel NIR spectrum, PLS-LSSVM quantitative models of diesel quality properties, such as cetane number, sulfur content, density and 50% recycling temperature, are respectively established. Compared with common-used PLS, LSSVM and other models, the results show that the PLS-LSSVM model integrates the benefits of PLS and SVM algorithms, which can reduce the influence of nonlinear degree between the properties and spectra of samples.3. By using a dispersive Raman spectrometer, PLS quantitative analysis models of gasoline aromatics content, olefin content and oxygen content are established combining with outlier detection. By removing the outlier samples, the prediction model accuracy is effectively improved. The RMSEP of aromatics, olefin and oxygen content are 0.30,0.52 and 0.083 respectively, and the corresponding R2 are 0.997, 0.927 and 0.984 respectively. Experimental results show the effectiveness of Raman spectroscopy to analyze the hydrocarbon group of gasoline, and the analysis accuracy is significantly higher than that of NIR spectroscopy.4. A portable rapid NIR analyzer for methanol gasoline is developed because of the strong absorption of methanol in NIR spectra. To meet the need of rapid detection of the methanol content, a portable fast analyzer based on dispersive NIR spectrometer is independently developed. It has been applied in several production enterprises of methanol gasoline. Practical application results show that this instrument can correctly classify methanol gasoline and regular gasoline, and it can precisely detect the methanol content. It is easy to operate and maintain this instrument. All these advantages make it to satisfy the needs of the methanol-gasoline production enterprises and other users in China.更多还原