基于光谱技术建立的多元校正模型通常条件下只适用于同一台仪器、 相同的测试条件及同批次或同类别的样品。 在仪器、 测试环境、 样品发生变化后, 已建光谱模型不再适配, 需要进行模型转移。 模型转移是限制光谱技术推广应用的关键技术瓶颈, 模型转移是否成功直接影响到可见-近红外光谱技术的推广应用, 为此, 综述了其研究现状, 并探讨了其未来发展方向。 首先, 将模型转移问题分成了两类: 第一类是相同样品在不同仪器或不同测试环境（不同温度/不同湿度）等条件下产生的模型不适配问题; 第二类是不同批次、 不同物理形态、 不同种类间产生的模型不适配问题。 这两类问题性质不同, 解决第一类模型转移, 能够保证同源样品的准确性和稳定性; 解决第二类, 能够实现光谱模型在不同样品间的自动传递和匹配应用。 然后, 梳理了常用的模型转移算法并进行了分类, 包括模型更新、 基于光谱校正算法、 基于结果校正算法等, 并列举了每个类别的模型转移算法的应用。 模型更新是一种重新计算模型系数最直接的方法, 通过扩展和调整模型来满足新的变化; 基于光谱校正算法是通过算法计算转移矩阵, 实现对光谱的校正; 基于结果校正算法是通过算法计算预测结果和实际结果系数, 从而实现预测结果的校正。 最后, 指出未来应着重研究第二类模型转移问题, 并且要寻找能够实现机器自动校正的模型转移, 从根本上解决模型转移这一限制光谱速测应用的主要技术瓶颈。

Generally, the multivariate calibration model based on spectroscopy is only for the same instrument, the same test conditions and the same batch or similar samples. However, with the increasing demand for spectral application, the problem that different samples cannot share the spectral model has become the fundamental technical bottleneck limiting spectral technology application. In the visible near-infrared spectrum analysis, after the change of the instrument, the test environment and the sample, the established spectral model is no longer suitable. So the model transfer is needed to solve this kind of problem. The model transfer is the key technology bottleneck to limit the application of spectral technology. Therefore, this paper summarizes the current research situation and discusses future development direction. First of all, the model transfer problem is divided into two categories: the first is the model mismatch of the same sample under different instruments or different test environments, called the first type of model transfer; the second is the model mismatch between different samples, called the second type of model transfer. These two kinds of problems are different in nature. To solve the first type of model transfer can ensure the accuracy and stability of homologous samples. And to solve the second type can realize the automatic transfer and matching application of spectral model between different products. Then, the commonly used model transfer algorithms are sorted and classified, including model updating, spectrum based correction algorithm, result based correction algorithm, and the application of each category of model transfer algorithm is listed. Model updating is the most direct method for recalculating model coefficients, which can meet the new changes by expanding and adjusting the model. Spectrum based correction algorithm is based on the calculation of the transfer matrix to achieve spectral correction. Result based correction algorithm is based on the calculation of pre-test results and actual results coefficients, so as to achieve the correction of prediction results. Finally, it is pointed out that the second type of model transfer should be studied in the future, especially the automatic model transfer by machine, so as to realize the real spectral velocity measurement.