Virtual Sample Generation And Screening And Application In Optimization Of Process Parameters Of Plant Medicinal Materials

The active ingredients of plant medicinal materials are widely used in medicine and food.The extraction of active ingredients is conducive to the maximum utilization of plant medicinal materials,so it has strong practicability and broad development prospects.In the process of extracting active ingredients from medicinal materials by ultrasound,the value of process parameters has a significant impact on the extraction rate of active ingredients.Appropriate process parameters are beneficial to extract the active ingredients from plant medicinal materials with maximum efficiency.How to choose the appropriate process parameters is an important issue for the extraction of active ingredients.At present,the intelligent method has become a major solution to the optimization of process parameters.By constructing the prediction model of process parameters and optimizing on the basis of the model,the best parameters of extraction process are obtained.However,the experiment of extracting active ingredients is time-consuming and laborious,and the data needs to be collected through manual test.There are some difficulties in data acquisition,making a small sample problem when building a prediction model.Small samples mean a small number of samples and insufficient information.The model trained with small samples has problems of insufficient learning and low prediction accuracy.Yet,a high-precision prediction model is the basis for obtaining accurate process parameters.If the prediction model is not accurate enough,the obtained process parameters will be affected.Virtual sample generation based on full factorial design(FFD)and virtual sample generation based on grasshopper optimization algorithm(GOA)are proposed to solve this problem.As the generated virtual samples need further optimization and screening,this paper proposes a virtual sample screening method based on the connection weights of extreme learning Machine(ELM).High quality virtual samples are generated and selected effectively.After obtaining experimental data,support vector regression(SVR)is selected as the prediction model.Virtual samples generated in different ways and screened are combined with small samples into synthetic samples,which are used to train the model.And then,analyze and compare the results.The results show that the model trained by synthetic samples composed of screened virtual samples and small samples has the best effect.At the same time,the prediction model is established with the idea of integration to further improve the accuracy of the model.On the basis of the integration model,the optimal combination of process parameters is obtained by grasshopper optimization algorithmIn this paper,the active ingredients liquiritin and chlorogenic acid are extracted from licorice and honeysuckle by dual-frequency ultrasound,and corresponding extraction process parameters are optimized.Through experimental verification,the effectiveness of the virtual sample generation and screening method proposed in this paper is proved,and the extraction process parameter optimization of the active ingredients of the two medicinal materials is realized.It provides a reference for optimizing the process parameters of ultrasonic extraction of plant medicinal materials.