Abstract
A searching method for the blending ratio of regular coffee with a designed flavor and taste has been developed. A fuzzy neural network (FNN) model was at first constructed for estimation of the sensory scores of various blending ratios. Using the model, the following four methods were attempted in order to estimate the blending ratio with the designed sensory score: 1) an exhaustive search, 2) a method using a genetic algorithm (GA), 3) a method using the GA combined with a reliability index (RIGA), and 4) a reverse FNN model. Among the results of these methods, the RIGA showed a good performance not only on estimation of the known blending ratio but also on the actual blending test in the present study.
