Interactive Differential Evolution Using Time Information Required For User'S Selection: In A Case Of Optimizing Fragrance Composition

Interactive Evolutionary Computation (IEC) spreads its area of applications. As an example of the applications, we have already proposed an IEC that optimizes fragrance composition suited to each user's feelings by using Interactive Differential Evolution (IDE). To create the fragrance, six aroma sources were mixed, and the intensity of them were target of optimization. Paired comparison was employed as subjective evaluation method. To enhance searching ability of the IDE method, this study proposes a new IDE utilizing both of evolution level of DE's vector (solution candidate) and time information required in user's selection in the paired comparison. In other words, shorter time duration required in the selection is considered as there is larger difference of fitness level between two vectors. Winner vector tended to have chance to evolve itself. These schemes in the proposed method are expected to accelerate searching better solutions. In this study, we constructed an IDE system optimizing fragrances based on the proposed method. Furthermore, with the IDE system, we conducted smelling experiments for investigating the fundamental efficiency of the proposed method. Target of creation was relaxing fragrance. The experimental results showed that obtained fragrances in the last generation was evaluated better than fragrances in initial generation significantly (P<0.01).