A new calibration model transferring strategy maintaining the predictive abilities of NIR multivariate calibration model applied in different batches process of extraction.

• In this work, a new calibration model transfer strategy which combined direct orthogonal signal correction with slope and bias correction (DOSC-SBC) was proposed and applied to maintain the predictive abilities of the original calibration model for new source batches. • The original calibration model can be used for predicting the transformed test samples without updating the model's coefficients though DOSC-SBC. • It will get good results in the multivariate calibration model transfer and be applicable to online or inline monitoring the preparation process of TCM. Extraction process plays an important role in the pretreatment of Flos Lonicerae Japonicae for Qingkailing oral liquid and is also one of the most popular preparation processes of traditional Chinese medicine (TCM). In order to monitor the content of API in the extraction solution, a PLS model was developed based on NIR spectroscopy. However, due to the raw materials' natural variability introduced into the extraction process of TCM or variations in spectral measurement conditions, it was unable to foresee the changes in the spectral response for new source batches, which could render a calibration model invalid. The development of a reliable multivariate calibration model for each new source batch is usually time-consuming and costly. In this work, a new calibration model transfer strategy which combined direct orthogonal signal correction with slope and bias correction (DOSC-SBC) was proposed and applied to maintain the predictive abilities of the original calibration model for new source batches. Performance of this model transfer strategy has been validated by two new individual batches which raw materials were from different sources, and the predictive relative errors of the samples from the above new batches decreased from 15.67% and 19.82% to 8.35% and 6.80% respectively after using DOSC-SBC strategy, demonstrating the good performance of the proposed model transfer strategy. [ABSTRACT FROM AUTHOR]