基于变量选择比自适应迭代法的近红外光谱变量选择方法研究  

Variable Selection Method of Near Infrared Spectrum Based on Variable Selection Ratio Adaptive Iterative Method

作　　者：文鹏 宦克为[1] 赵环 王迪 WEN Peng;HUAN Kewei;ZHAO Huan;WANG Di(School of Physics,Changchun University of Science and Technology,Changchun 130022;Jilin Branch of China Mobile Construction Co.,Ltd.,Changchun 130112)

机构地区：[1]长春理工大学物理学院,长春130022 [2]中移建设有限公司吉林分公司,长春130112

出　　处：《长春理工大学学报（自然科学版）》2024年第1期23-28,共6页Journal of Changchun University of Science and Technology(Natural Science Edition)

基　　金：吉林省科技发展计划项目(20210101158JC)。

摘　　要：近红外光谱分析技术(NIRS)存在信号弱、谱带重叠等问题,为了提高模型预测精度,提出了变量选择比自适应迭代法(PSAI)。通过蒙特卡洛法(MCS)采样,从样本中采集不同样本子集,利用偏最小二乘法(PLS)计算出每个子回归模型以及每个变量回归系数的平均值和标准差并得到初始权重,进而选取出最佳特征变量。结果表明,变量选择比自适应迭代法与自助软收缩法、无信息变量消除法以及竞争自适应重加权采样法相比,小麦蛋白数据预测精度分别提升了14%、21.4%、4.1%;牛奶蛋白数据预测精度分别提升了25%、43.3%、8.7%。所以变量选择比自适应迭代法对于简化预测模型,提高模型预测精度是可行的。Near infrared spectral analysis technology(NIRS)has problems such as weak signal and spectral band overlap.In order to improve model prediction accuracy,variable proportional selection adaptive iteration(PSAI)is proposed.By Monte Carlo method(MCS)sampling,different sample subsets were collected from the samples,and partial least square method(PLS)was used to calculate the mean value and standard deviation of each sub-regression model and each variable regression coefficient and get the initial weight,and then select the best characteristic variable.The results showed that the prediction accuracy of wheat protein data was improved by 14%,21.4%and 4.1%,respectively,compared with the adaptive iteration method,the self-shrinking method,the non-informative variable elimination method and the competitive adaptive reweighted sampling method.The prediction accuracy of milk protein data was improved by 25%,43.3%and 8.7%,respectively.Therefore,variable selection than adaptive iteration method is feasible to simplify the prediction model and improve the prediction accuracy of the model.

关 键 词：变量选择 权重 自适应迭代 加权自助采样 

分 类 号：O657.33[理学—分析化学]