Convergent parameters value for genetic-statistic algorithm (AGE)

Abstract

In this paper it is attempted to improve the Genetic-Statistic Algorithm by studying the relations between several search parameters (mutation "μ", chromosomés length "L", population " n " and number of genertions by iteration "t"). It has been made a study over twenty-four objective functions in eight different dimensions. When the convergence is reached, it is made a correlation and regression analysis between the parameters corresponding to this state. The possible relations among search parameters can simplify the hand tuning of the algorithm when we are searching the optimum in any problem. The twentyfour tested functions are not common test problems and were originated by a random problem generator. The use of this generator tries to complete the gaps that a test on typical test functions could leave. The most meaningful result of the correlation analysis is that there is not correlation between the degree of mutation and the others search parameters. Mutation in AGE seems to depend solely on the characteristics of the objective function. It exists a notable correlation between the size of the population and the number of generations by iteration. This result would be used to reduce the launching time of AGE in problems where it will not be necessary a high precision in the approximation of the global optimum, or when AGE is used in a first approximation stage, for thereinafter, in a second stage, to use a local optimization algorithm. Correlations among the other parameters do not seem sufficiently meaningful to be considered.