Modeling Method of Small Sample Data and Optimization in Ultrasonic Extraction Process of Botanicals

Aiming at the small sample problem in the dual-frequency ultrasonic extraction process of liquorice, this paper proposes a virtual sample generation method to expand small sample data. This method uses Box-Behnken Design (BBD) to design experiments, in order to collect original small sample data for modeling. Based on this forecasting model, data is generated. Moreover, by adding the Firework-Differential Evolution Algorithm with Levy Flight (L-FW-DE), reasonable and effective virtual samples are obtained and selected in the iterative process to expand the sample set. Synthetic samples, consist of virtual samples and original small samples, are applied to establish a prediction model for dual-frequency ultrasonic extraction of liquiritin from liquorice. Then, the optimal process parameters are obtained by an optimization algorithm. Finally, the experiment of extracting liquiritin by dual-frequency ultrasound verifies that under these optimized parameters, the extraction rate is the highest. Results of the experiments show that this method can efficiently generate reasonable virtual samples, and the synthetic sample composed of small samples and virtual samples can enhance the accuracy of the prediction model. The reliable optimal process parameters which are obtained by the forecasting model can provide guidance for the extraction process under the small sample condition.