【摘要】 目的将改进BP网络应用于重组类人胶原蛋白工程菌高密度发酵过程分析。方法通过添加动量项和可变学习速度的方法对传统BP网络算法进行改进。结果确定了5-9-1的网络结构,选择15组发酵实验数据对其进行训练。改进的方法对网络的收敛起到明显效果。得到的发酵过程BP网络模型收敛性和预测性能较高,平均相对误差仅2.42%。结论该模型较传统动力学模型误差更小,更接近实验过程。更多还原

【Abstract】 Aim To apply the improved Backpropagation(BP) network in the process of high density fermentation of recombinant escherichia coli producing human-like collagen.Methods The BP network was improved by adding momentum and using variable learning rate.Results 15 Teams of data were used to train the network,whose architecture was 5-9-1.The ways by which the network was improved were clearly working.The average relative error of the BP network model was only 2.42%,and its astringencies and predictions were nice.Conclusion The model predictions were in better agreement with the experimental data than the kinetic models.更多还原