Study On Building And Transfering Of Calibration Models For The Contents Of Two Consitutents In Liquorice Based On Near Infrared Spectroscopy

In recent years,near infrared spectroscopy(NIRS)technology has been widely used in Traditional Chinese Medicine(TCM),agriculture and other industries,by virtue of its high efficiency and nondestructive testing.Licquorice is regarded as Chinese herb and its determination of the content was complicated and time-consuming.Therefore,much attention has been paid to the NIRS technology with simple and rapid determination by researchers.A reliable qualitative and quantitative model for TCM could be established by use of NIRS combined with chemometrics technology to fast and accurate prediction for unknown samples.However,the NIRS region was serious overlaped,and effective information was masked,leading that the effect of model directly established by the NIRS was bad;Thus,chemometrics methods could be useded for pretreating and selecting the NIRS,improving the forecase and accuracy of model.On the other hand,the model built on instrument 1 that spent a lot of time and energy can’t be directly applied to instrument 2 for the prediction of spectra,and the transmission effect between the models was unsatisfied.Model transfer can solve this problem.In this paper,Quantitative model with model transfer was used for predicting the contents of liquorice(LI)and glycyrrhizic acid(GL)in liquorice based on NIRS and chemometrics methods,which is important that rapid quality evaluation of licorice and the efficient utilization of models by NIRS.The main content of this paper is as follows:1.The spectral pretreatment was optimized.The content of LI and GL about 40 batches of liquorice from different manufacturers and types was determined by HPLC method based on pharmacopoeia of the People’s Republic of China 2015;NIRS manufacturersed by three companies were adapted to obtain NIRS spectrometers.Standard normal variable(SNV),derivative smoothing(DS)and standardization methods were investigated.The pretreatment of NIRS and the content of the two consitutents were applied to construct the quantitative model by the partial least square(PLS),and the performance of the model evaluated with determination coefficient(R~2),root mean square error(RMSE),average absolute difference value(MAE),and relative analysis error(RPD)by internal cross validation(CV)and leave-one-out methods.The results showed that the best model was acquired by SNV.2.Optimal variable was selected.Particle swarm optimization algorithm(PSO),genetic algorithm(GA),variable importance in projection(VIP),Monte Carlo uninformative variable elimination(MCUVE),competitive adaptive competition reweighted sampling(CARS),bootstrapping soft shrinkage(BOSS)and random frog(RF)were used to filter the NIRS with SNV pretreatment.The models established and evaluated the same as content 1,compared with the original quantitative model.The results show that the model was optimized best by GA method.3.Model transfer was preferred.Instrument 1 was considered as the master instrument,and the other instruments were used as slave instruments.Direct standardization(DS),slope/deviation(SB),the spectral space transformation(SST),procrustes analysis(PA),and multiple scatter correction(MSC)were used to transfer models built by the master instrument.The results show that the effect of model transfer with DS was successful.Model transfer was based on GA variable selection method,which could reduced the number of variables and improved the computing efficiency as well as the robustness of the model,solving currency of model.4.Chemometrics could be applied to experimental design.Combining the D-optimal design combined with response surface method,the extraction for Naodesheng was optimizated.The experimental results show that the weighted content for the total saponins,flavonoids and ferulic acid was the highest under that ethanol dosage,the ethanol concentration and reflux extraction time were 10.81 times,80%and 45minutes,respectively.