应用BP神经网络优化南极磷虾酶解工艺

Optimization of enzymatic hydrolysis of Antarctic krill by BP neural network

为探究南极磷虾酶解过程的影响参数,以水解度为考查指标,实施多因素作用下的酶解实验,建立了基于BP神经网络的南极磷虾酶解工艺模型。使用78个样本对该神经网络模型进行83次迭代后,得到了准确度最优的拟合模型(MSE达到最小值0.002242,样本相关系数达到最大值0.9569);使用9个样本对模型进行测试发现,9组数据的MSE=0.003889,R=0.9856,表明该工艺模型可以准确地描述和预测不同工艺参数下南极磷虾酶解反应的结果;使用神经网络模型求解水解度的极大值和最优工艺条件,在酶添加量为4.73%,pH 6.99,温度54.0℃,时间201.0 min时,水解度最大,该条件下实验实测值为41.20%,与预测值41.36%无显著差异,该模型工艺优化结果准确。与响应面相比,BP神经网络具有避免舍去高次交互项所引起的误差,拟合模型纠偏性强,拟合结果更加精准等方面的优势。

To study the influence of process parameters on the enzymatic hydrolysis of Antarctic krill,enzymatic hydrolysis experiments under multiple factors were carried out and a process model based on BP neural network was built.In the model,after 83 epochs with 78 training samples,the mean square error(MSE)of model reached a minimum value of 0.002242,and the correlation coefficient of the model samples reached a maximum value of 0.9569,which confirmed that the accuracy of the model was optimal.MSE of 0.003889 and R of 0.9856 was obtained for nine tested data sets,which indicated that the model could accurately describe and predict the results of enzymatic hydrolysis of Antarctic krill under different process parameters.Finally,the optimal parameters were found by solving the maximum degree of hydrolysis in the model,i.e.,enzyme addition amount 4.73%,pH 6.99,enzyme hydrolysis time 201.0 min,and enzyme hydrolysis temperature 54.0℃.With the above parameters,the degree of hydrolysis determined by experiment was 41.20%.which was close to the predicted value of 41.36%.