Abstract:Comparing the effect of amino acid nitrogen production from coix seed wine through the backpropagation （BP）neural network and response surface methodology，a more suitable model was selected in optimizing the content of acid nitrogen production.Acid nitrogen production was used as a survey indicator，with the ratio of coix seed and glutinous rice，inoculation amount，temperature，and water-to-material ratio used as influencing factors.A central composite design （CCD）was adopted for the fermentation conditions of the brewery，and the results were analyzed using the response surface methodology and back-propagation neural network.The results showed that the correlation（R），R-squared（R2），root mean square error（RMSE）and absolute average relative error（AARE） correlation coefficients of the back-propagation neural network model，and the response surface methodology were 0.994 94，0.989 91，13.206 39，0.008 84 and 0.984 77，0.969 77，22.759 38 and 0.031 15，respectively.At the same time，the optimal fermentation conditions generated by the back-propagation neural network were obtained as follows：coix seed ∶glutinous rice 2.8∶1 （mass ratio），inoculum 4.8%，temperature 28.2℃，water material ratio 2.2∶1 （mL/g）.Moreover，the actual and predicted values of the back-propagation neural network under optimal conditions were almost the same.Overall，compared with the response surface method，the back-propagation neural network had better fitting ability and optimization for amino acid nitrogen production from coix seed wine.