1次元トーラス網状離島モデルに基づく遺伝的局所探索における島間距離戦略の検討

抄録

In a genetic local search (GLS) which is a hybrid technique of a genetic algorithm (GA) and a local search (LS), the undesirable phenomenon of premature convergence can often occur. Premature convergence is the phenomenon where the diversity of a group is lost. This phenomenon occurs because identical individuals are increased rapidly in the group while searching. Therefore, crossover loses its function. Once the premature convergence occurs, the search by the GLS becomes meaningless. Therefore, it is important to avoid the premature convergence and maintain the diversity. We made the parallel GLS to improve its searching ability and, in this paper, we propose a method named “Islands' Distance Strategy" to improve the searching ability of the GLS by introducing the concept of distance between the islands on one-dimensional torus type island model and by limiting the islands' range to which migrants can migrate. The problems used in these experiments are traveling salesman problems (TSPs) in 48 cities arranged in double-concentric-circle. The experimental results show that the proposed method achieved 100.0% as the attainment rate to the optimal solution, where the number of islands (subpopulations) N_sp=20, 25 and the islands' range to which migrants can migrate r=1, 2. Furthermore we demonstrated the existence of an appropriate islands' range to which migrants can migrate, and confirmed that the appropriate islands' range r=2 in these experimental conditions.