基于近红外光谱成品油性质检测方法与算法的研究
发布时间：2018-11-22 17:39
【摘要】：近红外光谱分析技术的广泛应用推动了成品油性质检测技术的快速发展,建立一个预测精度高、可靠性高、稳定性好的检测模型是近红外检测技术的首要目标。为实现这一目标,本文设计了基于近红外光谱的成品油性质检测算法,针对模型的预测精度和结果可靠性问题进行深入研究。本文的第一章综述了课题研究背景,以及成品油性质检测的研究现状;第二章首先介绍偏最小二乘法的基本原理,给出了基于偏最小二乘法的成品油性质检测方法流程,主要包括采集近红外光谱、选择特征谱段、光谱预处理、选择相似样本、建立偏最小二乘模型、性质预测和结果分析七个部分;最后对检测过程中存在的问题进行分析。第三章设计了基于主成分分析以及性质间相关性分析的校正集异常样本剔除方法,分析异常样本对模型预测精度的影响,介绍主成分分析的基本原理,给出校正集异常样本剔除方法的详细步骤,并以某炼化企业93#汽油研究法辛烷值的检测为案例,对样本的异常原因做出了详细分析。第四章首先分析成品油性质检测精度的影响因素,包括温度和噪声干扰等;设计了基于光谱温度修正的检测精度提升方法,介绍基于分段直接标准化算法的光谱转移函数构造过程,以某炼化企业95#汽油研究法辛烷值的检测为案例,给出详尽分析;另外,设计了基于离散小波变换和快速傅里叶变换算法的检测精度提升方法,介绍离散小波变换和快速傅里叶变换的基本原理,给出基于离散小波变换和快速傅里叶变换算法成品油性质检测方法的详细步骤,并给出了某炼化企业95#汽油研究法辛烷值的检测案例及分析。第五章设计了基于样本分布集中度和模型预测能力的成品油性质检测结果可信度评价方法,介绍常用检测模型的评价指标,给出成品油性质检测结果可信度评价方法的详细步骤,并以某炼化企业95#汽油研究法辛烷值的检测为案例,做出详细分析。根据实验结果分析,本文设计的基于近红外光谱的成品油性质检测算法,能够满足模型预测精度高、结果可靠性高的要求,为炼化企业提升了经济效益。
[Abstract]:The wide application of Near-infrared spectroscopy (NIR) technology has promoted the rapid development of oil quality detection technology. It is the primary goal of NIR detection technology to establish a detection model with high prediction accuracy, high reliability and good stability. In order to achieve this goal, an algorithm based on near infrared spectroscopy (NIR) is designed to detect the properties of oil products. The prediction accuracy and reliability of the model are studied in detail. The first chapter of this paper summarizes the research background of the subject, and the research status of oil quality testing. In the second chapter, the basic principle of partial least square method is introduced, and the method flow of oil quality detection based on partial least square method is presented, which mainly includes collecting near infrared spectrum, selecting characteristic spectrum, spectrum pretreatment, selecting similar samples. The partial least square model is established, the property prediction and the result analysis are seven parts. Finally, the existing problems in the detection process are analyzed. In the third chapter, we design a method of eliminating abnormal samples based on principal component analysis and property correlation analysis, analyze the influence of abnormal samples on the prediction accuracy of the model, and introduce the basic principle of principal component analysis. This paper gives the detailed steps of the method of eliminating abnormal samples in correction set, and takes the detection of octane number of gasoline research method of 93# in a refinery enterprise as an example, and makes a detailed analysis of the abnormal reason of the sample. In the fourth chapter, the factors affecting the precision of oil quality detection, including temperature and noise interference, are analyzed. The method of improving the detection precision based on spectral temperature correction is designed, and the construction process of spectral transfer function based on subsection direct standardization algorithm is introduced. Taking the detection of octane number of 9 gasoline research method in a refinery enterprise as an example, the detailed analysis is given. In addition, the detection accuracy enhancement method based on discrete wavelet transform and fast Fourier transform is designed, and the basic principles of discrete wavelet transform and fast Fourier transform are introduced. Based on discrete wavelet transform (DWT) and fast Fourier transform (FFT), this paper presents the detailed steps of testing the properties of refined oil products, and gives a case study and analysis of the octane number of 9 gasoline research method in a refinery enterprise. In chapter 5, the reliability evaluation method based on sample distribution concentration and model prediction ability is designed, and the evaluation indexes of common test models are introduced. The detailed steps of the reliability evaluation method for the testing results of the oil product properties are given, and taking the detection of the octane number of the 9 gasoline research method in a refinery enterprise as an example, the detailed analysis is made. According to the analysis of the experimental results, the proposed algorithm based on near infrared spectroscopy can meet the requirements of high prediction accuracy and high reliability of the model, and improve the economic benefit for the refinery and chemical enterprises.
【学位授予单位】：东南大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：O657.33;TE622
本文编号：2350084
[Abstract]:The wide application of Near-infrared spectroscopy (NIR) technology has promoted the rapid development of oil quality detection technology. It is the primary goal of NIR detection technology to establish a detection model with high prediction accuracy, high reliability and good stability. In order to achieve this goal, an algorithm based on near infrared spectroscopy (NIR) is designed to detect the properties of oil products. The prediction accuracy and reliability of the model are studied in detail. The first chapter of this paper summarizes the research background of the subject, and the research status of oil quality testing. In the second chapter, the basic principle of partial least square method is introduced, and the method flow of oil quality detection based on partial least square method is presented, which mainly includes collecting near infrared spectrum, selecting characteristic spectrum, spectrum pretreatment, selecting similar samples. The partial least square model is established, the property prediction and the result analysis are seven parts. Finally, the existing problems in the detection process are analyzed. In the third chapter, we design a method of eliminating abnormal samples based on principal component analysis and property correlation analysis, analyze the influence of abnormal samples on the prediction accuracy of the model, and introduce the basic principle of principal component analysis. This paper gives the detailed steps of the method of eliminating abnormal samples in correction set, and takes the detection of octane number of gasoline research method of 93# in a refinery enterprise as an example, and makes a detailed analysis of the abnormal reason of the sample. In the fourth chapter, the factors affecting the precision of oil quality detection, including temperature and noise interference, are analyzed. The method of improving the detection precision based on spectral temperature correction is designed, and the construction process of spectral transfer function based on subsection direct standardization algorithm is introduced. Taking the detection of octane number of 9 gasoline research method in a refinery enterprise as an example, the detailed analysis is given. In addition, the detection accuracy enhancement method based on discrete wavelet transform and fast Fourier transform is designed, and the basic principles of discrete wavelet transform and fast Fourier transform are introduced. Based on discrete wavelet transform (DWT) and fast Fourier transform (FFT), this paper presents the detailed steps of testing the properties of refined oil products, and gives a case study and analysis of the octane number of 9 gasoline research method in a refinery enterprise. In chapter 5, the reliability evaluation method based on sample distribution concentration and model prediction ability is designed, and the evaluation indexes of common test models are introduced. The detailed steps of the reliability evaluation method for the testing results of the oil product properties are given, and taking the detection of the octane number of the 9 gasoline research method in a refinery enterprise as an example, the detailed analysis is made. According to the analysis of the experimental results, the proposed algorithm based on near infrared spectroscopy can meet the requirements of high prediction accuracy and high reliability of the model, and improve the economic benefit for the refinery and chemical enterprises.
【学位授予单位】：东南大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：O657.33;TE622
【参考文献】
相关期刊论文 前10条
1 李敬岩;安晓春;田松柏;杨星明;;柴油十六烷值快速分析技术研究[J];石油炼制与化工;2016年05期
2 侯培国;李宁;常江;王书涛;宋涛;;SG平滑和IBPLS联合优化水中油分析方法的研究[J];光谱学与光谱分析;2015年06期
3 范文;王萍;袁悦;孙红跃;;基于SVM分类可信度的暴雨/冰雹分类模型[J];北京工业大学学报;2015年03期
4 夏巧生;;非线性偏最小二乘建模方法及在近红外光谱建模上的应用[J];计算机与应用化学;2014年01期
5 张军;姜黎;陈哲;余谦;梁静秋;王京华;;基于近红外光谱技术成品汽油分类方法的研究[J];光谱学与光谱分析;2010年10期
6 陈斌;邹贤勇;朱文静;;PCA结合马氏距离法剔除近红外异常样品[J];江苏大学学报(自然科学版);2008年04期
7 冯尚坤;徐海菊;;基于BP神经网络的啤酒酒精度近红外光谱快速检测[J];红外技术;2008年01期
8 吕慧英;任玉林;刘名扬;;PLS-ANN算法-NIR光谱非破坏性Norvasc药物有效成分的定量分析[J];高等学校化学学报;2007年05期
9 田高友,袁洪福,褚小立,刘慧颖,陆婉珍;结合小波变换与微分法改善近红外光谱分析精度[J];光谱学与光谱分析;2005年04期
10 闵顺耕,李宁,张明祥;近红外光谱分析中异常值的判别与定量模型优化[J];光谱学与光谱分析;2004年10期
相关会议论文 前3条
1 陆婉珍;袁洪福;徐广通;刘伟;陈英;;现代近红外光谱技术在石油产品分析中的应用[A];中国分析测试协会科学技术奖发展回顾[C];2015年
2 陈朝晖;冯新泸;史永刚;刘足票;;短波近红外光谱法测定汽油辛烷值的机理探讨[A];全国第六届分子振动光谱学术报告会文集[C];1990年
3 冯新泸;史永刚;;柴油分子结构的近红外光谱表征及其与凝点的关系[A];全国第六届分子振动光谱学术报告会文集[C];1990年
相关硕士学位论文 前1条
1 高俊;近红外光谱分析技术在油品分析中的应用研究[D];南京工业大学;2005年
，本文编号：2350084
本文链接：https://www.wllwen.com/kejilunwen/huaxue/2350084.html
教材专著
