OPTIMIZING CONTROL OF SENSORY EVALUATION IN THE SAKE MASHING PROCESS BY DECENTRALIZED LEARNING OF FUZZY INFERENCES USING A GENETIC ALGORITHM

Optimal control of sensory evaluation estimated from 13 component conc entrations on the basis of Dempster-Shafer's measure (DS) was attempte d in the fermentation process for mashing Ginjyo-shu (sake). The contr ol system consisted of fuzzy simulators generated by a genetic algorit hm (GA) and an optimization procedure based on another GA. The fuzzy s imulators simulated the dynamics of the ethanol production rate and se nsory evaluation. Decentralized learning of fuzzy rules was also intro duced. The fermentation period was divided into 4 phases, with a set o f fuzzy rules corresponding to each phase. In order to construct an ad aptive system based on the fuzzy simulators, only the set of rules cor responding to the current phase was adaptively identified, with the re sult that the fuzzy rules adapted to fluctuations in the relationship between the temperature and the ethanol production rate. By optimizing the control in this way, the optimal quality sake was successfully ob tained.