Improving evolutionary algorithms performance by extending incest prevention

Provision of population diversity is one of the main goals to avoid premature convergence in Evolutionary Algorithms (EAs). In this way the risk of being trapped in local optima is minimised. Eshelman and Shaffer [4] attempted to maintain population diversity by using diverse strategies focusing on mating, recombination and replacement. One of their approaches, called incest prevention, avoided mating of pairs showing similarities based on the parent’s hamming distance. Conventional selection mechanisms does not consider if the members of the new population have common ancestors and consequently due to a finite fixed population size, a loss of genetic diversity can frequently arise. This paper shows an extended approach of incest prevention by maintaining information about ancestors within the chromosome and modifying the selection for reproduction in order to impede mating of individuals belonging to the same “family”, for a predefined number of generations. This novel approach was tested on a set of multimodal functions. Description of experiments and analyses of improved results are also shown. Sistemas Inteligentes Red de Universidades con Carreras en Informática (RedUNCI)