[Network analysis of ethanol precipitation process for Schisandrae chinensis fructus]

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 Schisandrae Chinensis Fructus ethanol precipitation process and identify critical operating parameters.