为了提高近红外模型的精确度与准确度, 需要定期地对原模型进行修正。常用的方法是在原模型中添加一些包含新信息的新样品, 因此, 样品的选择成为模型维护过程中的关键因素之一。以利用近红外光谱分析法测定油菜籽含油量为例, 向原模型中添加不同偏差的样品建立独立的近红外模型, 并设计相应的验证集对各模型的预测性能进行全面评价。结果表明: 不同偏差的样品对模型预测性能的改善效果有差异, 只有当新样品的偏差与原模型的预测偏差相匹配时, 添加的新样品才能更有效地对原模型进行修正。依据偏差选择样品的新思路为近红外模型的维护提供了一条有效地途径。

In order to improve precision and accuracy of a near infrared model, some samples should be periodically added to the model. One very common way is to select representative samples with some new information, so the choice of sample will be one of key factors in the optimal course of a near infrared model. Using oil content in rapeseed as a case study, the calibrate models were established with samples of different deviations added to the model, and their performances were studied by different test sets. As a result, it has been found that there is a difference among the deviation, which will influence the performances of the corresponding model. Only when deviation of a new sample matches with that of a model, the sample will effectively enhance the performance of the model. In a word, sample selection on the basis of the deviation is a highly efficient way to optimize the model.