基于网络分析的五味子醇沉工艺优化研究

Network analysis of ethanol precipitation process for Schisandrae Chinensis Fructus

该文采用中心复合设计(central composite design),以醇沉后的糖去除率、五味子醇甲保留率及可溶性固形物含量为指标考察醇沉乙醇加入量、乙醇浓度、冷藏温度及冷藏时间4个醇沉工艺参数对于醇沉效果的影响。运用贝叶斯网络分析发现乙醇加入量和乙醇浓度为2个重要的醇沉工艺参数,然后运用遗传算法优化的BP人工神经网络建立2输入4输出的网络模型,所得训练集回归模型R^2=0.983 8,MSE=0.001 1,验证集回归模型R^2=0.975 9,MSE=0.001 8,模型拟合精度和预测效果均比较理想。研究表明该方法可有效地用于五味子醇沉过程关键工艺参数辨析与过程建模。

A set of central composite design experiments were designed by using four factors which were ethanol amount, ethanol concentration, refrigeration temperature and refrigeration time. The relation between these factors with the target variables of the retention rate of schizandrol A, the soluble solids content , the removal rate of fructose and the removal rate of glucose were analyzed with Bayesian networks, and ethanol amount and ethanol concentration were found as the critical process parameters. Then a network model was built with 2 inputs and 4 outputs using back propagation artificial neural networks which was optimized by genetic algorithms. The R2 and MSE from the training set were 0. 983 8 and 0. 001 1. The R2 and MSE from the test set were 0. 975 9 and 0.001 8. The results showed that network analysis method could be used for modeling of Scbisandrae Chinensis Fructus ethanol precipitation process and identify critical operating parameters.