1-215 遺伝的ネットワークプログラミングを利用したマルチエージェントのグループ化

概要

論文の詳細を見る
We apply the genetic network programming (GNP) to learning agents' decision making, where all agents don't have same ability. In our target problem, the goals are to find the optimal group structure such that agents with same ability belong to the same group, and to acquire rules for agents' decision making. To achieve the goals, we combine the automatically defined groups (ADG) with GNP, where ADG has been proposed for genetic programming techniques to acquire both the optimal group structure and its behavior. Next, we propose a new crossover method to improve the performance of GNP. Our crossover method is performed according to a series of action, while the original GNP may not take care of network structure. We have implemented the proposed method named ADGGNP+ and applied the method to a load transportation problem. Experimental results show that ADGGNP+ finds better solution than the original GNP.
社団法人日本機械学会の論文
2002-11-14