It has long been recognised that the choice of recombination and mutation operators and the rates at which they are applied to a Genetic Algorithm will have a significant effect on the outcome of the evolutionary search, with sub-optimal values often leading to poor performance. In this paper an evolutionary algorithm (APES) is presented within which both the units of heredity and the probability that those units will subject to mutation are learnt via self adaptation of the genetic material. Using Kaufmann's NK model, this algorithm is compared to a number of combinations of frequently used crossover operators with "standard" mutation rates. The results demonstrate competitive times to find maxima on simple problems, and (on the most complex problems) results which are significantly better than the majority of other algorithms tested. This algorithm represents a robust adaptive search method which is not dependant on expert knowledge of genetic algorithm theory or practice in order to perform effectively.
展开▼
