Increasing Diversity in Genetic Algorithms
Tim Watson, Peter Messer · 2001
Providing a genetic algorithm (GA) with the ability to control population diversity has been shown to be advantageous in both static and dynamic environments. Previous work has demonstrated that if the mutation rate of individuals is under genetic control then the optimal mutation rate rises in proportion to the speed of environmental change. This paper attempts to show that such an ‘automute’ GA outperforms a standard GA at keeping track of the fitness optimum in a simple, fast-changing environment. The paper also introduces an apparently equally effective method of controlling population diversity, based on the Hamming distance between pairs of individuals. It is argued that this ‘autoham’ GA is more suited to co-operative evolutionary systems since it does not rely on an increase in mutational ‘noise’ to provide an increase in diversity.