摘要
目的优选自拟清瘟方制备工艺。方法以料液比(因素A)、提取时间(因素B)、提取次数(因素C)为考察因素,以绿原酸、木犀草苷、靛蓝、靛玉红含量,干膏得率及其综合评分为考察指标,采用L_(9)(3^(4))正交试验法优化制备工艺;再以正交试验中的三因素水平为输入,以6种考察指标为输出,使用多层感知器(MLP)神经网络、反向传播(BP)神经网络、径向基(RBF)神经网络、遗传算法(GA)-BP神经网络进一步优化制备工艺。结果正交试验得出2种最佳方案(A_(2)B_(1)C_(2)和A_(1)B_(1)C_(2)),料液比1∶10或1∶8(g/mL),提取时间30 min,提取次数2次。但不同指标的最佳方案不同。绿原酸为A_(1)B_(1)C_(2),木犀草苷为A_(2)B_(1)C_(2)。预测发现,使用MLP神经网络算法时,平均误差率较大;相较于RBF下神经网络,BP神经网络算法可获得更小的误差和更优的决定系数,但仍无法解决局部最优解的问题。因此,采用GA-BP神经网络算法优化,可使决定系数最大化,误差平方和最小化,最优参数为料液比0.099(g/mL),提取2次,每次29.97 min。结论通过使用优化后的GA-BP神经网络,并结合传统正交试验设计,能准确提高工艺参数的优化程度,解决正交试验易出现局部最优解的难题,并降低数据处理难度。
Objective To optimize the preparation process of self-prepared Qingwen Formula.Methods The preparation process was optimized by the L_(9)(3^(4))orthogonal test with the solid-liquid ratio(factor A),extraction time(factor B)and extraction times(factor C)as the investigation factors,and with the contents of chlorogenic acid,luteoloside,indigo,indirubin,dry extract yield and their comprehensive score as the investigation indicators.The preparation process was further optimized by the multilayer perceptron(MLP)neural network,back propagation(BP)neural network,radial basis function(RBF)neural network,genetic algorithm(GA)-BP neural network with the three factor and their levels in the orthogonal test as the input,and with the six investigation indicators as the output.Results Two optimal schemes(A_(2)B_(1)C_(2)and A_(1)B_(1)C_(2))were obtained by the orthogonal test:the solid-liquid ratio was 1∶10 or 1∶8(g/mL),and extracting twice,each time for 30 min.But the optimal scheme of different indicators was different,that of chlorogenic acid was A_(1)B_(1)C_(2),while that of luteoloside was A_(2)B_(1)C_(2).It was found that the average error rate was high by the MLP neural network.It was also found that compared with those by the RBF neural network,the error was smaller and the coefficient of determination(R2)was better by the BP neural network,but it still could not solve the problem of local optimal solution.Therefore,the R2 could be maximized and the sum of squared error could be minimized by the GA-BP neural network,and the optimal parameters were as follows:the solid-liquid ratio was 0.099(g/mL),and extracting twice,each time for 29.97 min.Conclusion The optimal GA-BP neural network combined with traditional orthogonal test can promote the optimization of process parameters,solve the problem of local optimal solution in the orthogonal test,and decrease the difficulty of data processing.
作者
马诗瑜
何敬成
詹陆川
林伟杰
林思濠
胡小刚
卞晓岚
MA Shiyu;HE Jingcheng;ZHAN Luchuan;LIN Weijie;LIN Sihao;HU Xiaogang;BIAN Xiaolan(Ruijin Hospital,Shanghai Jiao Tong University School of Medicine,Shanghai,China 200023;Shunde Hospital of Southern Medical University,Foshan,Guangdong,China 528000;Guangdong Provincial People's Hospital,Guangzhou,Guangdong,China 510000;Zhuhai People's Hospital·Zhuhai Hospital Affiliated with Jinan University,Zhuhai,Guangdong,China 519099;School of Pharmacy,Shanghai University of Medicine&Health Sciences,Shanghai,China 201318;Chongqing University Cancer Hospital,Chongqing,China 400030)
出处
《中国药业》
CAS
2023年第12期56-62,共7页
China Pharmaceuticals
基金
上海市2022年度“科技创新行动计划”启明星项目[沪科〔2022〕80号]
2021年上海市“医苑新星”青年医学人才培养资助计划[沪卫人事〔2022〕65号]
上海交通大学“交大之星”计划医工交叉研究基金[YG2022QN015]
2021年度转化医学协同创新中心合作研究项目[TM202103]。
关键词
人工神经网络算法
中药制备
正交试验
工艺优化
数据处理
局部最优解
artificial neural network
preparation of traditional Chinese medicine
orthogonal test
process optimization
data processing
local optimal solution
