基于BP神经网络的产絮菌发酵参数的优化

Optimization of flocculant-producing bacteria fermentation parameters by BP neural network

摘要:

为优化对产絮菌克雷伯氏菌(Klebsiella. sp.)的发酵参数,首先用正交实验量化絮凝菌发酵参数对絮凝率和产率的影响度,确定温度、摇床转数、pH 3个影响度较高的发酵参数作为神经网络的输入,絮凝率和产率为输出,同时设计了训练样本．经过反复训练,建立了准确度高、误差小的预测模型,并利用该模型实现发酵参数双指标的全局优化,其最优发酵条件为温度33 ℃、摇床转数141 r／min、pH 7.90,经验证此条件下实际絮凝率和产率分别为92.67％和2.180 9 g／L,絮凝率提高了4.08％,产率提高了14.36％,使生物絮凝剂产量低的问题得以初步改善．利用所建立模型预测生物絮凝菌F+在发酵罐中的发酵过程,仿真误差较小．

Abstract:

To optimize the fermentation parameter of Klebsiella. sp, orthogonal experiments were used to quantify the degree of the fermentation parameters of flocculant-producing bacteria MFX on the flocculation rate and the yield, which indentified temperature, shaking revolution and pH value as the input of neural network, flocculation rate and yield as the output of neural network, and then a training sample was designed. After repeat training, a prediction model with high accuracy and small error was established by which the optimum fermentation conditions: the temperature of 33 ℃, the stirring speed at 141 r／min and pH value of 7.90, were obtained. Meanwhile, the actual flocculation rate and the yields were 92.67％ and 2.180 9 g／L, respectively. The flocculation rate increased 4.08％, and the yields of bio-flocculant increased 14.36％, which improved the yield of bio-flocculant. The model was used to predict the fermentation process of flocculant-producing bacteria F+in the fermenter, the simulation error was small, which was the basis for the prediction and control of industrial fermentation process.