目的 通过BP神经网络结合正交试验多指标优化四物汤水提工艺。方法 以加水量、提取时间、提取次数为考察因素，在R语言环境下用熵权法计算5-羟甲基糠醛、绿原酸、咖啡酸、芍药苷、阿魏酸、毛蕊花糖苷、洋川芎内酯A、藁本内酯8种有效成分多指标综合得分作为评价指标，先采用正交试验设计，再建立BP神经网络模型，通过网络训练，预测四物汤最优水提工艺。结果 优化得到的四物汤水提工艺条件为加8倍量水，提取3次，每次1 h，检验样本的网络预测值和实际测量值的相对误差小于1%。结论 建立的数学模型可对四物汤水提工艺进行分析和预测，所得工艺稳定可行，可高效提取四物汤中的有效成分。

Objective To optimize the water extraction process of Siwu Decoction by BP neural network combined with orthogonal experiment. Methods The water amount, the extraction time, and the extraction times were taken as factors. Entropy weight method was used to calculate the comprehensive scores of the multi-indicators of eight active components of 5-hydroxymethylfurfural, chlorogenic acid, caffeic acid, paeoniflorin, ferulic acid, verbascoside, senkyunolide A, and ligustilide in R language environment. Using comprehensive score as an evaluation indicator, the BP neural network model was established by orthogonal experiment design, and the optimal water extraction process of Siwu Decoction was predicted through network training. Results The optimized extraction process of Siwu Decoction was carried out by adding 8 times of water and extracting 3 times for 1 h. The relative error between the network predicted value and the actual measured value of the test sample was less than 1%. Conclusion The established mathematical model can analyze and predict the water extraction process of Siwu Decoction. The obtained process is stable and feasible, and can effectively extract the active ingredients in Siwu Decoction.

四川省重点研发项目（18ZDYF3528）