对直接标准化算法的改进及其应用

由于各种仪器之间存在差异,主机上建立的定量模型用于从机会导致预测结果出现较大偏差。目前主要通过有标样方法和无标样方法来减小预测偏差。该文对现有标样方法中的直接标准化算法进行改进,在转移矩阵的建立过程中,对从仪器数据矩阵进行主成分分解,以预测均方差为判定标准,确定最终的转移矩阵。并以玉米和烟草数据为对象,测试了该法的有效性。玉米样品含有2种成分:水分和蛋白质;烟草样品含有4种成分:还原糖、总糖、总氮和总碱。结果表明,对于玉米样品中的2种成分,采用改进的方法可显著提高预测的准确度;对于烟草中的4种成分而言,采用改进的方法可获得稳健的预测结果。

An Improved Direct Standardization Algorithm and Its Applications

One of the problems in near infrared quantitative analysis is that the calibration model built on one instrument can not be directly used on other instruments because there are many differences between a master instrument and a slave instrument.A quantitative model built on the master instrument may result in a great deviation when it was used on the slave instrument.There are many ways to solve this problem,which can be divided into two categories,e.g.standard sample methods and non-standard sample methods.Direct standardization(DS) algorithm is one of standard sample methods.An improvement for the direct standardization algorithm was made by a principal analysis on the data matrix generated in a slave instrument,determining the principal component number by root mean square error of prediction(RMSEP) and calculating the transfer matrix by the principal number.This number was determined by the relationship between RMSEP and the principal component number.When the RMSEP reaches a minimum,the corresponding principal component number can be used to calculate the transfer matrix.The improved method was validated by using both corn data and tobacco data.The corn data has two components,water and protein and the tobacco data has four components which are sugar,total sugar,total nitrogen and total plant akaloid.The results showed that the prediction accuracy of the corn data was improved greatly when the improved DS algorithm was used to compare with direct forecast and standard DS algorithm.The prediction accuracy of the tobacco data was relatively robust too.