稀疏偏最小二乘方法用于光谱波长选择及定量分析  

The application of sparse partial least squares method to speatral wavelength selection and quantitative analysis

作　　者：陈月东[1] 

机构地区：[1]江苏省盐城技师学院,江苏盐城224002

出　　处：《计算机与应用化学》2014年第2期239-243,共5页Computers and Applied Chemistry

摘　　要：针对光谱数据局部效应显著、变量间的严重共线性等特征,实施基于转换权向量约束优化的稀疏偏最小二乘回归新技术。它通过在特征变量提取的优化目标函数中加入转换权向量的罚函数,一并执行光谱波长选择和特征提取,随后再构建特征变量与性质变量间的校正模型。其中,罚函数中的最佳罚因子和校正模型中的最优PLS成分数,经由各自取值范围内一定数量试验水平的两因素全面试验设计与校正模型精度调控的交叉验证方式确定。最后,通过面粉生面团切片的近红外光谱数据的试验应用研究,结果显示该技术光谱数据波长选择和特征提取稳健,去噪明显,并显著提高了光谱数据定量校正模型的预测能力。According to the significant local effects of spectral data and the characteristics of serious collinearity between variables, the new technology of sparse partial least squares regression based on conversion rights to constrained optimization is implemented. By adding openalty function of conversion option vectors in optimization objective function extracted from feature variable, it implement wavelength selection of feature extraction, and then build the calibration models between feature variables and natural variables. By the experimental application study of the near-infi＇ared spectroscopy of flour doughs pieces, the results showes that the wavelenght selection and feature of this technology is steady and obvious in de-noising. And it significantly improve the ability of predicting ability of spectral data correction model.

关 键 词：偏最小二乘 约束优化 近红外光谱 波长选择 定量分析 

分 类 号：TQ015.9[化学工程]