近红外光谱定量分析的模型转移新方法

A new method of model transfer in near infrared spectral quantitative analysis

由于近红外光谱不同仪器间存在差异性,在一台仪器上耗费大量人力物力建立的定量分析模型往往无法直接应用于另一台仪器。为了提高已有模型的通用性,需要对所建模型进行模型转移使其能够适用于更多仪器。本文针对近红外光谱定量分析中的模型转移方法展开研究,利用基于变量投影重要性系数（VIP）的改进叠加偏最小二乘（SPLS）算法（VIP-SPLS）与常用的模型转移方法直接标准化（DS）相结合,提出一种新的近红外光谱分析模型转移方法（VIP-SPLS-DS）,使得主仪器上建立的模型经模型转移后能更精确地预测从仪器上的测试样本。将本文提出的模型转移新方法（VIP-SPLS-DS）应用于真实的近红外光谱标准数据集,对两台从仪器的预测精度相对于已有的PLS-DS模型分别提高了29.3%和30.8%,显著改善了模型转移效果,具有良好的实际应用前景。

There exist differences between different near infrared spectroscopy instruments, so quantitative analysis models built on one instrument cannot be directly applied to the other, however building models often takes a lot of manpower and material resources. In order to improve the generality of the existing model, model transfer for the existing model is needed to make it adapt to a new instrument. Based on the combination of the direct standardization （DS） algorithm and our VIP-SPLS （improved stacked PLS based on variable importance in the projection） algorithm, a new method called VIP-SPLS-DS is proposed to improve the performance of model transfer in near infrared spectral quantitative analysis, and to ensure that after the model transfer, the model built on the master instrument will predict more accurately the test samples from the slave instruments. The proposed method is applied to a commonly used benchmark NIR spectral data for evaluation. The results demonstrate that the precision improvement of VIP-SPLS-DS model transfer method is 29.3% and 30.8% to two different slave instruments when using the existing PLS-DS model transfer method, respectively. The new VIP-SPLS-DS model transfer method presents a promising prospect in practical use.