BP神经网络和遗传算法在乳酸菌发酵参数优化中的应用  

Optimization of fermentation variables by using artificial neural network Optimization of fermentation variables by using articial neural network model and genetic algorithm

作　　者：高爱同[1] 毕珂[1] 齐育平[1] 蒋冬花[1] 

机构地区：[1]浙江师范大学化学与生命科学学院,金华321004

出　　处：《应用与环境生物学报》2014年第1期112-116,共5页Chinese Journal of Applied and Environmental Biology

基　　金：国家自然科学基金项目(31070008和31270061)资助~~

摘　　要：为提高短乳杆菌L2菌株γ-氨基丁酸(GABA)的产量,建立了一个反映因素与产量之间的非线性关系模型.运用Plackett-Burman设计、中心组合试验设计(CCD)对MRS培养基组成和培养条件进行了优化,筛选出4个影响发酵的关键因素:蛋白胨、葡萄糖、谷氨酸钠、初始pH.在此基础上,采用误差反向传播神经网络(BPN)和遗传算法(GA)确定了4个关键因素的适宜参数:蛋白胨21.185 g/L,葡萄糖3.857 g/L,谷氨酸钠48.948 g/L,初始pH 4.05.最终使短乳杆菌L2菌株的GABA产量达到了27.765 g/L,比原始MRS培养基的13.452 g/L提高了106.4%.研究表明利用BPN-GA方法进行发酵条件优化是一种行之有效的途径.In order to improve the γ-aminobutyric acid （GABA） production ofLactobacillus brev＆ strain L2 and establish a model reflecting the nonlinear relationship between the factor and yield, the Plackett-Burman （PB） design and central composite design （CCD） were used to optimize the medium component and culture condition. By analyzing the statistical regression, we found peptone, glucose, MSG and initial pH were the most important factors. On this basis, error back propagation neural network （BPN） and genetic algorithm （GA） were applied to determine the optimum fermentation parameters as peptone 21.185 g/L, glucose 3.857 g/L, MSG 48.948 g/L and initial pH 4.05. Ultimately, the GABA production of strain L2 was up to 27.765 g/L, more than doubling the original yield （13.452 g/L）, which indicated that using BPN-GA method to optimized fermentation conditions is an effective way.

关 键 词：短乳杆菌L2 γ-氨基丁酸(GABA) Plackett—Burman(PB)设计 误差反向传播神经网络(BPN) 遗传算法(GA) 

分 类 号：TQ920.1[轻工技术与工程—发酵工程]