建模集选择对近红外光定量检测模型的影响  

Impact of modeling set selection on near infrared quantitative model

作　　者：李朵 李佩佩 龙若兰 冯丹 孙菁[1] LI Duo;LI Peipei;LONG Ruolan;FENG Dan;SUN Jing(Qinghai Key Laboratory of Qinghai-Tibet Plateau Biological Resource,Northwest Institute of Plateau Biology,Chinese Academy of Sciences,Xining 810008,China;Savaid Medical School,University of Chinese Academy of Sciences,Beijing 100049,China)

机构地区：[1]中国科学院西北高原生物研究所,青海省青藏高原特色生物资源研究重点实验室,西宁810008 [2]中国科学院大学存济医学院,北京100049

出　　处：《分析试验室》2023年第1期85-89,共5页Chinese Journal of Analysis Laboratory

基　　金：青海省科研基础条件创新平台专项(2020-ZJ-T05);中国科学院仪器设备功能开发技术创新项目(2022gl09)资助。

摘　　要：以藏药全缘叶绿绒蒿为例,考察了建模集样本数量、总黄酮含量范围、标准差(SD)值3个因素对近红外光(NIR)定量检测模型效果的影响。结果显示:随着样本数量的增加,模型效果先降低后升高,最后趋于稳定,在样本数量为120时效果最佳,达到240后趋于稳定;相同样本数量条件下,总黄酮含量高低对模型没有影响,但含量范围宽时模型效果更好;在总黄酮含量范围相同时,SD值与模型效果无直接联系,而在样本量相同、总黄酮含量范围不同时,SD值越大模型效果越好。研究结果表明,在进行NIR建模时,并不是建模样本数量越多模型就越好,在选择建模样本时要求样本的含量范围尽可能宽,且要有一定的离散性。In this study, Meconopsis integrifolia(Maxim.) Franch. was taken as the research object. The effects of the number, content range and standard deviation(SD) of the modeling set on near infrared(NIR) qualitative and quantitative model performance were discussed using root mean square error(RMSE) and related coefficient(R) values as evaluation indexes. The results showed that the model effect first decreased, then increased, and finally tended to be stable with the increase of the number of samples.The model worked best at the modeling number of 120, and tended to be stable basically when the sample number reached 240. As a result, the content had no effect on the model performance under the same number of modeling set, but the model worked better when the content range was wide. The influence analysis of SD to model indicated that within the same sample concentration, the SD value of the sample set was not directly related to the model effect. When it came to the same sample number with different SD, the performance of the model increased with the increase of SD values. In conclusion, this study indicated that in NIR modeling, the model performance was not enhanced with the increasing of sample number. When selecting modeling samples, the content range of the samples should be as wide as possible, and there should be a certain degree of dispersion.

关 键 词：NIR定量检测模型 样本数量 总黄酮含量范围 标准差(SD)值 

分 类 号：O657[理学—分析化学]