Modelling gas with self adaptive mutation rates
James E. Smith · UWE Research Repository (UWE Bristol) · 2001
This paper introduces a dynamical systems model of a generational Genetic Algorithm with Self-Adaptation of mutation rates.This model is used to predict the mean fitness of an evolving population as a function of time.The accuracy of these predictions are then tested by running a series of experiments using Genetic Algorithms with different population sizes.It is shown that although there is a threshold below which the "real" populations do not closely follow the predictions, the model is still able to give us useful information about the behaviour of the "real" GAs, since the populations tend to get "stuck" at points close to certain eigenvectors of the infinite population model.Arguments are given which allow the prediction of which eigenvectors will be important.The dynamics of the population evolving on a non-stationary environment are then considered, and some conclusions drawn about the nature of environmental change to which the algorithm will be able to respond.