论文部分内容阅读
将多模型共识偏最小二乘法用于近红外光谱定量分析。利用随机抽取的训练子集建立一系列偏最小二乘模型,选取其中性能较好的部分模型作为成员模型,用这些成员模型来预测未知样品。将该方法用于一组生物样本的近红外光谱与样品中人血清白蛋白、γ-球蛋白以及葡萄糖含量之间的建模研究,并与单模型偏最小二乘法了进行比较。结果 PLS对独立测试集中三种组分进行50次重复预测的平均RMSEP分别为0.1066,0.0853和0.1338,RMSEP的标准偏差分别为0.0174,0.0144和0.0416;而本方法重复预测的平均RMSEP分别为0.0715,0.0750和0.0781,RMSEP的标准偏差分别为0.0033,0.2729×10-4和0.0025。
Multi-model consensus partial least squares method was used for quantitative analysis of near infrared spectroscopy. A series of partial least squares models were set up by using randomly selected training subsets. Part of models with good performance were selected as member models, and these member models were used to predict unknown samples. The proposed method was applied to the modeling of near-infrared spectroscopy in a group of biological samples and to the determination of human serum albumin, γ-globulin and glucose in the sample and compared with the single-model partial least-squares method. Results The average RMSEP of 50 repeated predictions of PLS for the three independent components was 0.1066, 0.0853 and 0.1338 respectively, and the standard deviations of RMSEP were 0.0174, 0.0144 and 0.0416, respectively. The average RMSEP of repeated prediction by PLS was 0.0715, 0.0750 and 0.0781, respectively. The standard deviation of RMSEP was 0.0033, 0.2729 × 10-4 and 0.0025, respectively.