A study of alternative selection mechanisms for multiple cossover per couple in genetic algorithms

• Autores: Esquivel, Susana Cecilia; Leiva, Héctor Ariel; Gallard, Raúl Hector
• Año de publicación: 1998
• Idioma: inglés
• Tipo de recurso: documento de conferencia
• Materia: Ciencias Informáticas; Informática; Biology and genetics; Algorithms; Combinatorial algorithms; Selection process; genetic algorithms; selections mechanism; crossover; function optimization
• Institución: Universidad Nacional de La Plata

Multiple crossover per couple (MCPC) is a newly introduced crossover method which in contrast with the single crossover per couple approach (SCPC), permits more than one crossover operation for each mating pair. MCPC was applied to optimise classic testing functions and some harder (non-linear, non-separable) functions. The goodness of this approach prevailed under all tests and revealed that, when MCPC is applied with 2, 3 and 4 crossovers per couple, results as good as under SCPC can be expected with an additional benefit in processing time. This performance was obtained through the ability showed by MCPC of exploiting the recombination of good, formerly found solutions. But on the other hand, those experiments also showed that, in some cases, the method increased the risk of premature convergence due to a loss of genetic diversity. This paper gives an insight of the convenience of binding the choice of a selection mechanism to the genetic operators used. Focussing on this problem experiments with MCPC under proportional, and ranking selection methods were performed. In the case of ranking, an adaptive approach was carried out to adjust selective pressure. Descriptions of the alternative selection mechanisms used, experiments and some of the results obtained under each method are shown.
Sistemas Inteligentes
Red de Universidades con Carreras en Informática (RedUNCI)