Deciding the Temperature Course during Sake Mashing Using a GA-FNN for Quality Control of Sake

Simulation models for Baumé and alcohol concentration from the 11th day to the end of the sake mashing were constructed using a fuzzy neural network (FNN). The models could simulate the time courses of Baumé and alcohol concentration in 17 actual sake mashings. Average errors at the ends of the mashings were 0.22 and 0.40% for Baumé and alcohol concentration, respectively. By applying a genetic algorithm (GA) with the simulation models, temperature time courses were calculated with good accuracy, and the target values for Baumé and alcohol concentration on the final day could be achieved. To make a variety of sakes with different qualities, temperature courses were calculated against 3 target values: higher (+0.3), ordinary (0.0), and lower (-0.3) final day Baumés. The calculated temperature courses were found to be similar to a Toji's (expert's) strategy for making decisions on temperature. By applying this procedure, quality control of sake can be realized.