Obtaining Multiple Distinct Solutions with Genetic Algorithm Niching Methods

This paper proposes a new technique for improving the number of usefully distinct solutions produced by a Genetic Algorithm (GA) when applied to multimodal problems. The tribes method builds on the spatial selection methods proposed by Collins and Jefferson [1]. We compare the technique with two well-known niching methods (crowding and sharing), spatial selection alone, and a simple control GA method, in the domain of simple timetabling problems. We demonstrate that the tribes technique can greatly improve the efficiency with which a GA can obtain multiple distinct solutions to a problem.