基于正交信号回归法对中试在线近红外定量模型的模型传递研究

模型传递使在特定条件下建立的模型能够应用于新的样品状态、环境条件或仪器状态。本研究在中药金银花中试在线水提过程中，以水提液中的绿原酸含量为研究对象，建立绿原酸含量的近红外定量模型。针对由金银花来源不同带来的模型失效问题，从光谱背景校正的角度出发，提出以KS算法挑选待传递样本中的代表性样本，结合正交信号回归(OSR)光谱背景校正方法，对不同来源样本的近红外光谱进行光谱背景校正，并深入探讨OSR方法实现近红外定量模型在不同来源中试样本间传递的应用原理。经模型传递后，模型对新批次样本预测的RSEP由14.91%下降到7.11%，RPD由2.95上升到5.36，预测准确度明显提高。实验证明，选择代表性样本的KS算法结合OSR光谱背景校正的模型传递方法不仅能减小不同来源原料药之间的偶然误差，同时消除了中试制剂过程和算法本身的系统误差，因此能够有效地改善因样本来源差异而造成的模型失效现象。该研究充分阐释了OSR模型传递方法的应用原理，以光谱背景校正与挑选代表性样本回归的方法实现了近红外定量模型在不同来源原料药中试样本之间的模型传递，增强了模型应对原料药批次间变动的能力，提高了模型的稳健性，为多来源原料药中试在线制剂过程中有效成分含量的快速实时监测提供了有效方法。

Model Transfer of On-Line Pilot-Scale Near Infrared Quantitative Model Based on Orthogonal Signal Regression

Model established under a certain condition can be applied to the new samples, environmental conditions or instrument status through the model transfer. In the process of pilot-on-line water extraction of Flos Lonicerae Japonicae, the content of chlorogenic acid is measured with High Performance Liquid Chromatography (HPLC) as a reference method, and a NIR quantitative model of chlorogenic acid is established by partial least square regression (PLSR)。In order to solve the problem of model’s failure to predict accurately the content of chlorogenic acid in the samples of Flos Lonicerae Japonicae from different sources, the KS algorithm is used to select the representative samples from samples to be transferred, orthogonal signal regression (OSR)algorithm is used to correct the NIR spectral background of the samples from different sources. And deeply discussing how the OSR worked in the model transfer from different sources. After the model transferred, the RSEP of transferred model predicting the new batch samples decreases from 14.91% to 7.11%, RPD rises from 2.95 to 5.36, indicating the obvious improvement of prediction accuracy. The results show that the model transfer method which combines the KS algorithm with OSR can diminish the spectral background variation between the samples of different sources effectively, because it not only reduces the accidental errors of the spectral background from the pharmaceutical raw materials from different sources, but also eliminates the system errors in the preparation process of pilot-on-line water extraction and the OSR algorithm. Based on this, it could correct model failure caused by different sample sources. This paper explains the application principle of OSR. Make NIR model to be transferred between the pilot samples which medicinal raw material come from different sources by spectral background and selecting the representative samples regression. Strengthen the model’s adjustment with the batch variations of medicinal raw material and improve the robustness of the NIR quantitative model. It will provide a method for the rapid and on-line detection of the active ingredient content of multi-sources samples during the process of pilot-on line water extraction, and promote the application of NIR quantitative model in the preparation process of TCM.