目的 通过响应面法和反向传播（back propagation，BP）神经网络筛选南葶苈子Descurainia sophia配方颗粒最佳提取工艺。方法 通过检测提取液的出膏率、电导率（σ）、表面张力（γ）、折光率（RI）、D₁₀、D₅₀、D₉₀、pH等物性，采用相关性分析与物理指纹图谱筛选出关键物性参数，采用层次分析法（analytic hierarchy process，AHP）-熵权法（entropy weight method，EWM）对提取液各物化特性进行综合评分，并建立BP神经网络与响应面预测模型，优选最佳提取工艺。结果 确定出膏率、σ和D₅₀为关键物性参数；模型预测结果确定南葶苈子最佳提取工艺为提取1 h、12倍量水、提取3次。结论 通过构建BP神经网络预测的最佳提取工艺与响应面结果一致，贴近生产实际，以常规化学成分定量结合关键物性确定最佳提取工艺，可为中药配方颗粒的提取工艺优化提供参考。

Objective Through response surface method and back propagation (BP) neural network, the optimal extraction process of the Nantinglizi (Descurainiae Semen, DS) formula granules was screened. Methods Physical properties such as extract yield, electrical conductivity (σ), surface tension (γ), refractive index (RI), D₁₀, D₅₀, D₉₀, pH, etc. of the extracted liquid were detected, and key physical property parameters were screened by correlation analysis and physical fingerprinting. Analytic hierarchy process was adopted. Analytic hierarchy process (AHP) - entropy weight method (EWM) was used to comprehensively score each index of the extraction solution, and BP neural network and response surface prediction model were established to optimize the extraction process. Results The key physical properties were determined as extract yield, electrical conductivity and D₅₀. The optimal extraction process of DS was 1 h extraction, 12 times of water extraction and three times of extraction. Conclusion The best extraction process predicted by BP neural network is consistent with the results of response surface and closer to production practice. In this study, the optimal extraction process was determined by quantitative analysis of conventional chemical composition combined with key physical properties, which could provide reference for the optimization of extraction process of traditional Chinese medicine formula granules.

R283.6

国家长三角科技创新共同体联合攻关项目（2023CSJGG1700）