Prediction Of Flavor Productions In Beer Fermentation Based On Artificial Neural Networks

This article was aimed to establish an back propagation (BP) neural network model to quantitatively control the production of key flavor productions at the range of 3 key variables:temperature, inoculum size, initial wort concentration. Selected volatiles are as followed:alcohol, n-propanol, isobutyl alcohol, isoamyl alcohol, acetic ether, isoamyl acetate, ethyl caproate, ethyl caprylate.First of all, a great number of data were obtained by head-space coupled with gas chromatography at different conditions. In the second step, BP neural network are required to analysis this large volume of information. Each data was randomly devided into three parts:training subset, test subset and predict subset. When a set of training data was given to the ANN, the network weights are adjusted backwards though the ANN neurons until the error, ie, the difference between the actual output and the expected output is minimized. Finally, the best simulation was obtained to predict higher alcohols and esters in the condition of 12℃、13°P,2×107 viable cells/mL-1.From the research above, the conclusions are as follows:five different BP neural networks were established to predict the changes of n-propanol, isobutyl alcohol, isoamyl alcohol, isoamyl acetate and acetic etherat the condition of 12℃、13°P,2×107 viable cells/mL-1. A good prediction ability was obtained for the model of n-propanol, isobutyl alcohol, isoamyl alcohol, isoamyl acetate, acetic ether, where regression correlation of 0.999、0.998、0.999、0.997、0.998 is observed between the expected output and the predicted output of each data point. However, we failed to model the changes of ethyl caproate and ethyl caprylate, due to small amout during beer fermentation.Second, the method of BP neural networks was established to predict sugar density and alcohol levels during beer fermentation. Sugar density and alcohol concentrations during beer fermentation in the followed conditions:8℃,11° P,106 viable cells/mL,8, 11° P,107 viable cells/mL,8℃,11° P,4×107 viable cells/mL,8℃,11° P,8×107 viable cells/mL. Temperature, sugar density, inoculum size was set as input, sugar density and alcohol levels during beer fermentation process were set as output, and then BP neural network was applied to build model for this process. After the model was trained, sugar density and alcohol levels was predicted under the condition of 8℃,11°P,2×107 viable cells/mL-1.Sugar density and alcohol levels was predicted under the condition of 8℃,11°P, 2×107 viable cells/mL. The predictive root mean square error of sugar density and alcohol was 2.66% and 14.60%, respectively. The results showed that the model can be applied for the prediction of sugar density and alcohol levels during the beer fermentation.