Research On Multi-source Information Fusion Modeling Method Of Traditional Chinese Medicine Salvia Miltiorrhiza Extraction Process

Chinese traditional medicine production process is the specific formatting stage of the inherent quality of TCM products.Each unit of the process of traditional Chinese medicine pharmaceutical will have an effect on quality of traditional Chinese medicine,and traditional Chinese medicine extracting process is the inevitable stage and the most basic stage of TCM.Ensure the batches' consistency of the quality of TCM products,the beginning of treatment process of the Chinese native medicine preparation-extraction process must are strictly controlled.Mechanism of traditional Chinese medicine extraction process is complex.Currently,real-time monitoring of a number of major ingredients or active ingredients in the process of extraction can be achieved by using process analysis technique such as near infrared method.but how to effectively fusion the process parameter information and the quality monitoring information,and the research of product quality prediction model and control method of extraction process is insufficient.Therefore,based on the ethanol extractionprocess of traditional Chinese medicine salvia miltiorrhiza as the research object,and studied the near infrared spectroscopy in real-time analysis its content changing of fat-soluble effective components.Then,near infrared quantitative process model is established,and the process model is updated and multi-source information fusion model is set up.It lays the foundations for the quality control of alcohol extraction process of salvia miltiorrhiza.The article main research content as follows:First salvia miltiorrhiza of different batches were extracted in the small scale of experiments,and use near infrared spectrum technology to real-time acqure NIR data in the process of salvia miltiorrhiza extract,and apply the HPLC analysis technique to obtain the concentration of index component data such as tanshinone ? A,Cryptotanshinone and tanshinone ? to quantitative analysis,and establish the model of least squares of tanshinone forecasting model.In order to improving the performance of near infrared quantitative prediction model to respectively investigate first derivative(lstd),second derivative(2ndd),multiplicative scatter correction(MSC),standard normal variate(SNV),wavelet denoising,(WDS)and Savitzky-Golay Smoothing(SG)spectral preprocessing methods,and using synergy interval partial further squares(siPLS)to select feature spectraLThe result respectively showed that:RMSECV was 0.0053 g·L-1,RMSEP was 0.0058 g·L-1,RPD was 3.06(tanshinone II A);RMSECV was 0.0020 g·L-1,RMSEP was 0.0093 g·L-1,RPD was 2.05(cryptotanshinone);RMSECV was 0.0048 g·L-1,RMSEP was 0.0045 g·L-1,RPD was 3.33(tanshinone I).In general,diffefence of Chinese medicine raw materials quality,production environment,the operating personnel changes and the instrument response fluctuations will lead to the degradation of predict performance of NIR analysis model,so use finite batch samples to establish the quantitative model that lead to larger prediction deviation.Therefore,in order to improve the reliability and robustness of near infrared quantitative analysis in the process of traditional Chinese medicine pharmaceutical application,so need to update and maintain the established model.And try to use SIC(simple interval calculation)algorithm and the RS(random selection)to update model,and evaluate model updating as a result.After the corresponding results of SIC updation were that:RMSECV was 0.0068 g·L-1,RMSEP was 0.0054 g·L-1,RPD was 3.14(tanshinone ? A);RMSECV was 0.0027 g·L-1,RMSEP was 0.0027 g·L-1,RPD was 3.07(cryptotanshinone);RMSECV was 0.0051 g·L-1,RMSEP was 0.0051 g·L-1,RPD was 4.26(tanshinone ?).After the corresponding results of RS updation were that:the RMSECV was 0.0064 g·L-1,RMSEP was 0.0068 g·L-1,RPD was 2.50(tanshinone ? A);RMSECV was 0.0025 g·L-1,RMSEP was 0.0025 g·L-1,RPD was 3.18(cryptotanshinone);RMSECV was 0.0051 g·L-1,RMSEP was 0.005 g·L-1,RPD was 3.72(tanshinone ?).Similarly,to choose different batches salvia miltiorrhiza and different concentration alcohol extraction process as the research object,and to set up the tanshinone ? A,tanshinone? index component of NIR quantitative model in the different concentration alcohol extraction process of same batches.The performance parameters of model respectively were that:RMSECV was 0.0078 g·L-1,RMSEP was 0.0044 g·L-1,RPD was 4.00(tanshinone ? A);RMSECV was 0.0079 g·L-1,RMSEP was 0.0079 g·L-1,RPD was2.58(tanshinone ?).Has also set up tanshinone ?A and tanshinone ? index component model of NIR quantitative model in the different concentration alcohol extraction process of different batches.The modeling results were that:the RMSECV was 0.0079 g·L-1,RMSEP was 0.0058 g·L-1,RPD was 3.08(tanshinone ? A);RMSECV was 0.0108 g·L-1,RMSEP was 0.0108 g·L-1,RPD was 3.18(tanshinone ?).At the same time,two methods aboved were used to update.After the corresponding results of SIC updation were that:the RMSECV was 0.0081 g·L-1,RMSEP was 0.0042 g·L-1 RPD was 4.01(tanshinone ? A);RMSECV was 0.0106 g·L-1,RMSEP was 0.0106 g·L-1,RPD was 3.24(tanshinone ?).After the corresponding results of RS updation were that:the RMSECV was 0.0072 g·L-1,RMSEP was 0.0048 g·L-1,RPD was 3.50(tanshinone ?A);RMSECV was 0.0102 g·L-1,RMSEP was 0.0102 g·L-1,RPD was 3.01(tanshinone I).The results showed that the updated model's performance is better than the initial model,and SIC updating methods is better than that of the RS method in most cases.Second,multi-source information fusion model of two extraction process of salvia miltiorrhiza is studied.To fusion the raw material quality attribute information,the process parameters and the process state information of one fried and two fried stage in the process of ethanol extraction of salvia miltiorrhiza,and used to predict the quality of the product.The adjustable parameters of selected for ethanol amount of first fried,ethanol concentration of first fried,ethanol amount of second fried and ethanol concentration of second fried.Then,the D-optimum design was used to arrange 24 experiments.To collect 5 batches salvia miltiorrhiza from different regions as different material properties input,and randomly were assigned to different group.Using NIR technology to monitor the extracting process and acqure process NIR spectra as state variables,and the unfolded two-dimensional spectrum matrix were processes with 6 kinds of pretreatment methods.Then,to combine the pretreatmented spectrum with the quality rate of tanshinone ? A,cryptotanshinone and tanshinone ? to aquire its PLS scores values respectively,and takeing 10 before scoring value again,operation variables and material attribute fusion to establish the multi-source information fusion model Then,used the HPLC to measure tanshinone ? A,cryptotanshinone and tanshinone ? content of combined liquid of one fried and two fried.Start with the operating variables as independent variables,with reference to the index composition to set up the conventional PLS model,and the corresponding model results were that:the RMSECV was 0.5444 mg·g-1,RMSEP was 0.2710 mg·g-1,RPD was 0.83(tanshinone ?A);RMSECV was 0.3031 mg·g-1,RMSEP was 0.3031 mg·g-1,RPD was 0.99(cryptotanshinone);RMSECV was 0.3119 mg·g-1,RMSEP was 0.3119 mg·g-1,RPD was 1.12(tanshinone ?).Finally,to make operating variables,material properties and process quality parameters fusing as independent variables,with indicating composition as reference value to establish model,the modeling result respectively were:the RMSECV was 0.1728 mg·g-1,RMSEP was 0.0.0317 mg·g-1,RPD was 6.91(tanshinone ? A);RMSECV was 0.1534 mg·g-1,RMSEP was 0.1534 mg·g-1,RPD was 4.02(cryptotanshinone);RMSECV was 0.1171 mg·g-1,RMSEP was 0.1171 mg·g-1,RPD was 4.76(tanshinone I).Results showed that the multi-source data fusion model with good performance than the conventional models.In conclusion,salvia miltiorrhiza extraction process as the study stage,the NIR model have built successfully,and quantitative prediction model suggests that NIR spectra and quality information have the correlation,and offer the foundation for fusion modeling;Model updating,try the SIC algorithm and the RS method to applied to model updating of traditional Chinese medicine extraction process.It provides a reference for Chinese medicine extraction process control;Multi-source data fusion model provides a certain reference significance for control and detection extraction process of traditional Chinese medicine.