后向变量选择偏最小二乘法用于近红外光谱定量校正模型的建立  

The Calibration Model of Nearinfrared Reflectance Spectroscopy Determined by the Method of Backward Variable Selection PLS Regression(BVSPLS)

作　　者：吴功煌[1] 史新元[2] 吴志生[2] 乔延江[2] 

机构地区：[1]首都医科大学,北京100069 [2]北京中医药大学

出　　处：《数理医药学杂志》2010年第3期257-260,共4页Journal of Mathematical Medicine

基　　金：北京市科技计划资助课题(D02050040040111)

摘　　要：目的:对小麦中蛋白质含量以及橘叶中的橙皮苷含量建立定量校正模型。方法:将原始数据分为校正集、停止集和测试集,考察不同的停止集样本量对删除变量数的影响,并以相关系数与采用传统的偏最小二乘法(PLS)所建立的模型进行对比。结果:使用于建模的变量数分别减少了79%和89%,同时相关系数分别从0.9016和0.7730提高到了0.9306和0.8657。结论:较之于传统的偏最小二乘法,该方法所需用于建模的变量数大大减少,提高了模型的运行速度和预测能力,并且使所建立的模型更富于解释性。Objective：To build the calibration model of the protein content in wheat and chengpigan content in jueyu.Methods：The original data was divided into calibration set,stop set and test set,investigated the size of stop set impact on the deleted variable numbers,and compared the model by the R-square.Results：The number of varialbes used to model decreased 79% and 89%,simultaneously the R-square improved from 0.9016 and 0.7730 to 0.9306 and 0.8657,respectively.Conclusion：Compared to traditional partial least squares（PLS） method,the BVSPLS decreased the numbers of variables greatly,improved the modeling speed and the predictive ablity,maked the model easy to understand.

关 键 词：近红外 变量选择 偏最小二乘 

分 类 号：R311[医药卫生—基础医学] R285.5