乙醇脱氢酶氨基酸组成和最适温度均匀设计的神经网络模型

A uniform design based neural network model for amino acid composition and optimum temperature of alcohol dehydrogenase

构建了乙醇脱氢酶氨基酸组成和最适温度的神经网络模型,并运用均匀设计优化神经网络结构。结果表明,当最小训练率为0．12,动态参数为0．6,Sigmoid参数为0．98,隐含层结点数为9时,神经网络的拓扑结构为最优。所得样本误差为0．00999,模型对温度预测的平均绝对百分比误差为5．13%,均方根误差为5．12℃,平均绝对误差为3．82℃,优于逐步回归计算的结果。

A link between amino acid composition and the optimal temperature of alcohol dehydrogenase was established in this paper. Back propagation neural network （BPNN） was used as the mathematical tool and the uniform design method was used to optimize the architecture of the BPNN. Results showed that the calculated and predicted temperature fitted the optimum temperature of alcohol dehydrogenase very well and the MAPEs （ Mean Absolute Percent Error） were 4.87% and 3.76% , the MSEs （ Mean Square Error） were 4.94℃ and 4.42℃ , the MAEs （Mean Absolute Error） were 3.61℃ and 3.35℃ , respectively. It was more accurate in fitting and predictions compared with the model based on stepwise regression.