【摘要】：越來越多的人選擇汽車出行,這必會帶來了不少問題,如日益減少的石油資源和石油需求量增加之間的矛盾,為了緩解矛盾,盡快尋找石油燃料的替代產(chǎn)品。乙醇柴油是柴油替代產(chǎn)品的一種,然而,不同廠商生產(chǎn)的乙醇柴油品質(zhì)參差不齊,不利于乙醇柴油的推廣使用,因此,需要一種便捷的手段對乙醇柴油的品質(zhì)進行檢測。本文采取了光譜診斷技術(shù)對乙醇柴油的主要指標(biāo)做了研究。建立乙醇柴油品質(zhì)指標(biāo)的準(zhǔn)確可靠定量分析模型,具體結(jié)論如下:1.以乙醇柴油為實驗?zāi)繕?biāo),利用近紅外光譜(near infrared spectroscopy,NIR)技術(shù)對乙醇柴油的乙醇含量、密度、粘度進行定量分析,采用五種預(yù)處理方法對光譜數(shù)據(jù)進行處理,并建立了最小二乘支持向量機、主成分回歸和偏最小二乘回歸三種模型。結(jié)果表明:在多元散射校正-平滑預(yù)處理下,最小二乘支持向量機對乙醇柴油密度、粘度、乙醇含量的建模效果最好,相關(guān)系數(shù)分別Rp是0.995,0.995和0.995;RMSEP分別是6.8×10-4,1.13×10-2和0.5714×10-1。2.以乙醇柴油為實驗?zāi)繕?biāo),利用中紅外光譜(mid-infrared spectroscopy,MIR)技術(shù),對乙醇柴油進行光譜采集與分析。對乙醇柴油MIR原始數(shù)據(jù)進行不同的預(yù)處理,并對光譜數(shù)據(jù)進行波段篩選,分別建立了乙醇柴油乙醇含量、密度、粘度PLSR模型,得出以下主要結(jié)論:綜合比較八種變量篩選方法,發(fā)現(xiàn)UVE-SPA-CARS-PLS對乙醇含量的建模效果最好,模型預(yù)測集的Rp、RMSEP分別為0.978、0.825。變量篩選較原始光譜建立的模型來說,不僅模型輸入數(shù)量減少,預(yù)測效果也有所提高。3.利用拉曼光譜技術(shù),對乙醇柴油進行光譜采集與分析,對乙醇柴油拉曼光譜原始數(shù)據(jù)進行不同的預(yù)處理,并對光譜數(shù)據(jù)進行波段篩選,分別建立了乙醇柴油乙醇含量、密度、粘度PLSR模型,得出以下主要結(jié)論:發(fā)現(xiàn)SPA-CARS-PLS對乙醇含量的建模效果最好,模型預(yù)測集的Rp、RMSEP分別為0.978、0.825。波段篩選出的波長變量以及建模的結(jié)果為以后設(shè)計便攜式中紅外光譜儀打下基礎(chǔ)。
[Abstract]:More and more people choose to travel by car, which will bring a lot of problems, such as the contradiction between decreasing oil resources and increasing oil demand. In order to alleviate the contradiction, find alternative products of petroleum fuel as soon as possible. Ethanol diesel is one of the alternative products of diesel fuel. However, the quality of ethanol diesel produced by different manufacturers is not uniform, which is not conducive to the promotion and use of ethanol diesel. Therefore, a convenient method is needed to detect the quality of ethanol diesel. In this paper, the main indexes of ethanol diesel oil were studied by spectral diagnostic technique. The accurate and reliable quantitative analysis model of ethanol diesel quality index was established, and the concrete conclusion was as follows: 1. The ethanol content, density and viscosity of ethanol diesel oil were quantitatively analyzed by using near infrared spectroscopy (near infrared spectroscopy,NIR) technique. Five pretreatment methods were used to process the spectral data. Three models, namely least squares support vector machine, principal component regression and partial least squares regression, are established. The results show that the least square support vector machine (LS-SVM) has the best modeling effect on the density, viscosity and ethanol content of ethanol diesel under the condition of multivariate scattering correction and smoothing pretreatment. The correlation coefficient Rp is 0.995 and 0.995 respectively, and the correlation coefficients are 6.8 脳 10 ~ (-4) ~ 1.13 脳 10 ~ (-2) and 0.5714 脳 10 ~ (-1) ~ (2), respectively. Taking ethanol diesel oil as the experimental object, the spectral acquisition and analysis of ethanol diesel oil were carried out by using mid-infrared spectroscopy (mid-infrared spectroscopy,MIR) technology. The MIR raw data of ethanol diesel were pretreated with different bands and spectral data were screened. The PLSR models of ethanol content, density and viscosity of ethanol diesel oil were established, and the following main conclusions were obtained: comprehensive comparison of eight methods for screening variables. It was found that UVE-SPA-CARS-PLS had the best effect on the modeling of ethanol content, and the Rp,RMSEP of the model prediction set was 0. 978 / 0. 825 respectively. Variable screening is more effective than the original spectral model. Not only the input number of the model is reduced, but also the prediction effect is improved. The spectral data of ethanol diesel oil were collected and analyzed by Raman spectroscopy. The original data were pretreated and the spectral data were screened. The ethanol content and density of ethanol diesel oil were established. The main conclusions of viscosity PLSR model are as follows: it is found that SPA-CARS-PLS has the best effect on modeling ethanol content, and the Rp,RMSEP of model prediction set is 0.978 鹵0.825, respectively. The wavelength variables selected from the band and the modeling results lay the foundation for the later design of the portable mid-infrared spectrometer.
【學(xué)位授予單位】：華東交通大學(xué)
【學(xué)位級別】：碩士
【學(xué)位授予年份】：2017
【分類號】：O657.3;TQ517

【參考文獻】

相關(guān)期刊論文 前10條

1 歐陽愛國;唐天義;周鑫;劉燕德;;最小二乘支持向量機結(jié)合中紅外光譜測定甲醇柴油甲醇含量[J];發(fā)光學(xué)報;2016年10期

2 歐陽愛國;唐天義;黃志鴻;劉燕德;;甲醇柴油品質(zhì)的拉曼光譜檢測[J];激光與光電子學(xué)進展;2016年11期

3 于雷;洪永勝;周勇;朱強;徐良;李冀云;聶艷;;高光譜估算土壤有機質(zhì)含量的波長變量篩選方法[J];農(nóng)業(yè)工程學(xué)報;2016年13期

4 于雷;洪永勝;耿雷;周勇;朱強;曹雋雋;聶艷;;基于偏最小二乘回歸的土壤有機質(zhì)含量高光譜估算[J];農(nóng)業(yè)工程學(xué)報;2015年14期

5 王元忠;趙艷麗;張霽;金航;;紅外光譜結(jié)合統(tǒng)計分析對不同產(chǎn)地瑪咖的鑒別分類[J];食品科學(xué);2016年04期

6 張冰;鄧之銀;鄭靖奎;王曉萍;;拉曼光譜技術(shù)的汽油組分含量測定[J];光譜學(xué)與光譜分析;2015年06期

7 郭文川;董金磊;;高光譜成像結(jié)合人工神經(jīng)網(wǎng)絡(luò)無損檢測桃的硬度[J];光學(xué)精密工程;2015年06期

8 陳玉鋒;莊志萍;魏林博;張曉靜;于吉民;;激光拉曼光譜內(nèi)標(biāo)法直接測定甲醇含量[J];理化檢驗(化學(xué)分冊);2015年04期

9 丁希斌;張初;劉飛;宋星霖;孔汶汶;何勇;;高光譜成像技術(shù)結(jié)合特征提取方法的草莓可溶性固形物檢測研究[J];光譜學(xué)與光譜分析;2015年04期

10 李江波;郭志明;黃文倩;張保華;趙春江;;應(yīng)用CARS和SPA算法對草莓SSC含量NIR光譜預(yù)測模型中變量及樣本篩選[J];光譜學(xué)與光譜分析;2015年02期

相關(guān)博士學(xué)位論文 前2條

1 劉淼;智能人工味覺分析方法在幾種食品質(zhì)量檢驗中的應(yīng)用研究[D];浙江大學(xué);2012年

2 淡圖南;基于光譜分析的燃油組分檢測技術(shù)研究[D];浙江大學(xué);2011年

相關(guān)碩士學(xué)位論文 前2條

1 黃志鴻;基于近紅外、中紅外和拉曼光譜法甲醇柴油品質(zhì)檢測研究[D];華東交通大學(xué);2016年

2 覃趙軍;微生物發(fā)酵的光鑷?yán)庾V法監(jiān)測與分析[D];廣西師范大學(xué);2013年

，

本文編號：2204341