Filtering-GA: the evolutionary TSB landscape modifier

Genetic algorithms (GAs) are developed to model mechanisms of natural evolution, and are known as robust search procedures. It has been reported, however, that the canonical GA cannot discover optima on some types of problems called GA-hard problems easily. In this paper, to overcome defects of the canonical GA without disturbing its characteristics, we propose the filtering-GA which can modify the fitness landscape in an adaptive way. On the modified landscape, it will discover many local or global optima sequentially. Since this approach does not depend on representation, the filtering-GA with genetic operators developed to solve specific problems must also work well. To make sure of this proposition, we apply the filtering-GA with edge-recombination to traveling salesman problems (TSP) of which configuration of cities forms double concentric circles. In the computer simulation, comparing the discovered results on various configurations, the characteristics of this approach become clear, and we discuss the efficiency and problems of the filtering-GA through observing its search process on these configurations