Self adaptation of mutation rates in a steady state genetic algorithm

This paper investigates the use of genetically encoded mutation rates within a “steady state” genetic algorithm in order to provide a self-adapting mutation mechanism for incremental evolution. One of the outcomes of this work will be a reduction in the number of parameters required to be set by the operator, thus facilitating the transfer of evolutionary computing techniques into an industrial setting. The NK family of landscapes is used to provide a variety of different problems with known statistical features in order to examine the effects of changing various parameters on the performance of the search. A number of policies are considered for the replacement of members of the population with newly created individuals and recombination of material between parents, and a number of methods of encoding for mutation rate are investigated. Empirical comparisons (using the “best-of current-population” metric) over a range of test problems show that a genetic algorithm incorporating the best “flavour” of the adaptive mutation operator outperformed the same algorithm when using any one of a variety of “standard” fixed mutation rates suggested by other authors