改进型人工神经网络优化Iturin A发酵培养基研究  

Optimization of fermentation medium for Iturin A by improved artificial neural network approach

作　　者：付茂红 钟娟[1,3,2] 谭忠元[1,2] 彭文璟[1,3] 杨杰[1,3] 周金燕[1,3] 谭红[1,3] 

机构地区：[1]中国科学院环境与应用微生物重点实验室(成都生物研究所),四川成都610041 [2]中国科学院大学,北京100049 [3]四川省环境微生物重点实验室,四川成都610041

出　　处：《广东农业科学》2015年第17期102-107,共6页Guangdong Agricultural Sciences

基　　金：国家科技支撑计划项目(2011BAE06 B04-18)

摘　　要：为提高伊枯草菌素A(Iturin A)的产量,选择发酵培养基中对Iturin A合成有影响的5个组分作为自变量、以Iturin A产量为响应值,设计5因素10水平的均匀设计试验。以均匀设计试验数据为基础,分别建立二次多项式模型和一种改进型人工神经网络模型来优化发酵培养基,最后通过比较两种模型的优劣选择改进型人工神经网络模型优化培养基组分。结果表明,相比于二次多项式模型,基于相同试验设计的改进型人工神经网路模型有更好的拟合精度和泛化能力,使用人工神经网络模型优化后的培养基发酵48 h后,Iturin A产量为1.121(±0.089)g/L,比二次多项式模型优化的培养基高出13.23%,此时Iturin A发酵培养基的优化组分为葡萄糖、KH2PO4、Mg SO·7H4 2O、酵母膏和大豆蛋白胨总氮浓度分别为42.6、3.62、3.14、0.12、2 g/L。In order to improve the yield of Iturin A, five components of fermentation culture medium influencing the synthesis of Iturin A were chosen as independent variables, the yield of hurin A was used as response value, a uniform design method with 5 factors and 10 levels was designed. Based on the uniform design, a quadratic polynomial model and an improved artificial neural network model were carried out to optimize the culture medium. By comparing the effects of two models, we chose the optimal components of fermentation culture medium predicted by the improved artificial neural network. The results showed that the improved artificial neural network had better fitting precision and generalization capacities than quadratic polynomial model based on the same experimental design. By this improved artificial neural network, the yield of Iturin A reached 1.121 （ ±0.089 ） g/L after 48 hours of fermentation when the concentrations of glucose, KH2PO4, MgSO4·7H20, yeast extract and total nitrogen in soy peptone were 42.6, 3.62, 3.14, 0.12 and 2 g/E, respeetively. The yield increased by 13.23%, compared with the yield optimized by quadratic polynomial model.

关 键 词：伊枯草菌素A 均匀设计 遗传算法 人工神经网络 二次多项式 

分 类 号：S482.2[农业科学—农药学] TQ455[农业科学—植物保护]