
第22卷第2期 2014年2月
文章编号
1004924X(2014)02-0296-08
光学精密工程
Optics and Precision Engineering
Vol.22No.2
Feb, 2014
近红外光谱稀疏分量分析检测柴油品质参数
周扬1,2*,戴曙光2,葛丁飞1
(1.浙江科技学院信意与电子工程学院,浙江杭州310023; 2.上海理工大学光电信息与计算机工程学院,上海200093)
摘要:由于光谱盲源分离中的独立分量分析方法(ICA)在案油品控参数近红外光谱定量分析时预测效果不理想,稳定性不高,本文提出了一种在稀疏特性下的盲源分离近红外光谱分析思路-—近红外光谱稀疏分量分析法,并用该方法预测了柴油沸点、密度、芳烃总量等品控参数。首先利用柴油校正集光谱样本训练元余字典并完成光谱在该字典下的稀疏变换,接着完成混合矩阵估计,最后用混合矩阵与柴油品控参数建立回归预测模型。针对混合矩阵估计中光谱稀疏度不为一时聚状特征模棚导致无法确定案类数的间题,提出将AP聚类算法应用于聚类过程。实验表明,近红外光谱稀疏分量分析法对柴油沸点、密度、芳烃总量预测的相关系数(R)、预测均方根误差(RMSEP)分别达到了98.91%,99.68%, 99.43%和2.84.0.88×10-3,0.59.性能优于ICA及全谱偏最小二乘(PLS)等传统方法。该方法可作为一种柴油品控参数检测的有效盲源分离定量分析方法,并可推广于其它光谱检测领域。
关键词:近红外光谱荣油检测;独立分量分析法;稀疏分量分新法;源分离
中图分类号:0657,33;U473.12
文献标识码:A
doi;10. 3788/(OPE, 20142202. 0296
Detection of diesel qualityparametersby nearinfrared
spectroscopy with sparse component analysis
ZHOU Yang'-2·,DAI Shu-guang', GE Ding-fei
(1.College of Information and Electronic Engineering
Zhejiang Uniersity of Science and Technology,Hangzhou 310023,China;
2.School of Optical-Electrical and Computer Engineering,Shanghai 200093,China)
*Correspondingauthor,E-mail:zybuaa@163.com
Abstract: Independent Component Analysis(ICA) in a blind source separation method can not obtain i-deal prediction results and excellent measuring stability in detecting diesel quality parameters by near infrared spectroscopy. Therefore, a blind source separation method based on sparse characteristics of near-infrared spectroscopy is proposed. The method is named Near Infrared Spectroscopy with Sparse Component Analysis(NIFS SCA) and is used for the prediction of boiling points, density and total ar omatics for the diesel. This method firstly trains the redundant dictionary by spectral samples and fin ishes the sparse transformation for the calibration sample. Then, it estimates the mixing matrix, and establishes the regression model between mixing matrix and diesel quality parameters, As the cluste
收稿日期:2013-05-06;修订日期:2013-05-30.
基金项目:国家自然科学基金资助项目(No.51376162);渐江省教育厅资助项目(No.Y201327869);浙江科技学院交
叉预研专项(No,2013JC07Y)