Mixed strategy in genetic algorithms: Domains reduction and Multirecombination

Abstract

The strategy of domain's reduction and the strategy of multirecombination are based on previous works 1,2 and have achieved some promising results. Like continuation has been proposed to combine both strategies for problems of real functions optimization. The strategy of domains's reduction is based on taking advantage the information that provide the better objective functions of a statistics sample to go reducing the domain. While the domain is reduced we go ourselves approximating to the optimum. The strategy of multirecombination is based on taking advantage the non-linear interactions between chromosomes. The non-linear interactions are produced when are accomplished several recombinations (crossovers and mutations) in populations, without selection processes; arrived to a point we select the better strings opening step to a new population. The multirecombination can accelerate or stop the search of the optimum. Both strategies' combination consists of launching a statistics sample that use multirecombination instead of the usual recombination of Simple Genetic Algorithm, SGA, with the idea that the mutirecombination in some launchings carry us toward proximities of the optimum, and the domain reduction permit us to locate with the wished precision. In this work is presented a study of the mixed strategy. In a first section are studied problems with connected domains; after, problems with not connected domains. The incorporation of this last is due to the existence of practical problems with constraints that produce this type of domains. Customary academics test functions and comparisons with other evolutionary algorithms are showed.