当前位置: 首页 > 期刊 > 《分析化学》 > 2006年第5期
编号:11014205
主成分分析支持向量回归建模方法及应用研究
http://www.100md.com 《分析化学》 2006年第5期
主成分分析,,主成分分析,支持向量回归,近红外光谱,摘要,关键词,1引言,2原理与算法,3实验部分,4结果与讨论,References
     摘 要

    将主成分分析(PCA)用于近红外光谱的特征提取,并与支持向量回归(SVR)相结合,实现了主成分分析支持向量回归(PCASVR)用于近红外光谱定量分析的建模方法。与单纯的SVR方法相比,不仅提高了运算速度,而且提高了模型的预测准确度。将PCASVR方法用于烟草样品中总糖和总挥发碱含量的测定,所得结果的预测均方根误差分别为1.323和0.0477;回收率分别为91.8%~112.6%和88.9%~120.2%。

     关键词 主成分分析,支持向量回归,近红外光谱

    Principal Component AnalysisSupport Vector Regression and Its Application In Near Infrared Spectral Analysis

    Hou Zhenyu1,2, Cai Wensheng1,3, Shao Xueguang3

    (Department of Chemistry, University of Science and Technology of China, Hefei 230026)

    (Department of Chemical Engineering, Henan Institute of Science and Technology, Xinxiang 453003)

    (Department of Chemistry, Nankai University, Tianjin 300071)

    Abstract A new method for quantitative prediction of total sugar (TS) and total volatile alkali (TVA) content in tobacco samples from near infrared (NIR) spectrometry was proposed. The method is a combination of principal component analysis (PCA) and support vector regression (SVR), which extracts features from the NIR spectra using PCA at first and then builds a nonlinear model using SVR. Therefore, the new method is fast in computation and accurate in prediction. 110 NIR spectra was used for investigation of method, taking 60 as calibration set, 20 as test set and 30 as prediction set. Results show that the RMSEPs (root mean square error of prediction) are 1.323 and 0.0477, and the recoveries are 91.8%-112.6% and 88.9%-120.2% for TS and TVA, respectively. ......

您现在查看是摘要页,全文长 8333 字符