近红外光谱结合Boosting-PLS快速检测奶粉中蛋白质含量  

Fast measurement of protein content in milk powder by NIR combined with boosting-PLS

作　　者：张华秀[1] 李晓宁[1] 范伟[1] 梁逸曾[1] 

机构地区：[1]中南大学化学化工学院中药现代化研究中心,湖南长沙410083

出　　处：《计算机与应用化学》2010年第9期1197-1200,共4页Computers and Applied Chemistry

基　　金：国家自然科学基金(批准号:20875104)

摘　　要：采用1种基于Boosting理论的回归建模算法Boosting-偏最小二乘法(BPLS),建立了奶粉中蛋白质含量的近红外模型。先用Kernard-Stone法构建样本训练集和预测集,继对所有样本的近红外光谱进行中心化处理,用BPLS算法进行建模,并对收缩因子v与迭代次数m这2个重要参数进行了优化,当收缩因子为0.9,迭代次数为882时,所建模型的预测结果最好,预测均方根误差(RMSEP)为0.315 9,明显优于偏最小二乘法。结果表明:BPLS算法具有提高模型的预测精度的显著优势,可实现奶粉中蛋白质含量的快速、无损测定。A new regression algorithm named Boosting partial least squares （BPLS）, based on boosting theory, was used to establish the NIR model of protein in milk powder. The training and testing sets were partitioned by Kernard-Stone algorithm before preprocessing the spectroscopy by mean centering. The model was built by BPLS and the parameters, such as shrinkage value （v） and the iteration number （m） were optimized. The best model showed satisfactory prediction when the shrinkage value was 0.9 and the iteration number was 882. The root mean square error prediction （RMSEP） is 0.315 9 which is much better than PLS. This method is suitable for determining the protein content of milk powder rapidly and nondestruetively.

关 键 词：近红外光谱 奶粉 蛋白质 Boosting—PLS 

分 类 号：TQ015.9[化学工程] TP391.9[自动化与计算机技术—计算机应用技术]