模型传递可使特定条件下建立的近红外模型能够应用于新的样品状态、环境条件或仪器状态。正交信号回归是一类基于“光谱背景校正”的模型传递方法,利用虚拟标准光谱拟合主从批次光谱间的线性关系,将从批次光谱向主批次光谱映射,以实现近红外定量模型的传递,但该方法对虚拟光谱的代表性要求较高,回归过程中易出现较大偏差。因此,该文提出一种直接正交信号校正法(direct orthogonal signal correction,DOSC)联合斜率截距校正算法(slope and bias correction,SBC)(DOSC-SBC)的数据处理方法,针对近红外定量模型对不同批次样本制剂过程中目标成分含量预测准确度较差的问题,分析不同批次样本间因组分差异带来的光谱背景差异和模型预测误差的性质,通过DOSC消除与目标值无关的光谱背景差异,联合SBC算法对不同批次间样本批次间系统误差进行校正,实现近红外定量模型在不同批次间传递。该研究将DOSC-SBC应用于金银花水提和醇沉制剂过程中,模型对新批次样本的预测误差由32.3%,237%降低到7.30%,4.34%,预测准确度显著提高,实现了制剂过程中新批次样本目标成分的快速定量。DOSC-SBC模型传递方法实现了近红外定量模型在不同批次间传递,且该方法不需要标准样品,有利于促进近红外技术在中药制剂过程的应用,为中药生产过程中有效成分的实时监测提供参考。 Model transfer allows near-infrared models built under specific conditions to be applied to new sample conditions, environmental conditions, or instrument status. Orthogonal signal regression is a type of model transfer method based on “Spectral Background Correction ”. Fitting the linear relationship between the spectra of master and slave batches by using the virtual standard spectrum, mapping from the batch spectrum to the master batch spectrum to achieve Near infrared quantitative model of the transfer, but the method of the representative of the virtual spectrum of the higher requirements of the regression process is prone to greater deviation. Therefore, this paper presents a method of data processing based on direct orthogonal signal correction (DOSC) combined with slope and bias correction (SBC) (DOSC-SBC) For different batches of sample preparation process target content of the poor accuracy of the prediction problem, analysis of different batches of samples due to component differences caused by differences in spectral background and the nature of the model prediction error, through the DOSC elimination of the target value has nothing to do Spectral background differences, combined with the SBC algorithm to calibrate the system error between batches of different batches, to achieve near infrared quantitative model in different batches passed. In this study, DOSC-SBC was applied to the honeysuckle water extract and alcohol precipitation preparation. The prediction error of the model for new batch samples was reduced from 32.3%, 237% to 7.30% and 4.34%, and the prediction accuracy was significantly improved. Fast quantitation of target components in new lot samples during the process. DOSC-SBC model transfer method to achieve the near-infrared quantitative model in the transfer of different batches, and the method does not require standard samples will help promote the application of NIR in the process of traditional Chinese medicine preparations for the production of active ingredients in real-time monitoring for reference.