Optimización evolutiva de los parámetros de control de un algoritmo genético

Abstract

The behavior of a Genetic Algorithm is determined by the control parameter settings, such as selection probability, crossover probability, mutation probability and population size. Nevertheless, there are no standard rules for choosing appropriate values for these parameters. This decision is usually taken in terms of the most common values or experimental formulas given in literature, or by means of trial an error methods. Furthermore, any authors give arguments that any static set of parameters, having the values fixed during an Genetic Algorithm, seems to be inappropriate; a Genetic Algorithm is an intrinsically dynamic, adaptive process. This Ph. D. Thesis propose an approach based on a meta-level GA combined with an adaptation strategy of the GA control parameters to find and adjust the best control parameters to improve the GA performance. In order to validate the approach, it have been applied to three Genetics Algorithms which use different coding schemes to solve different types of optimization problems, such as: 1. a binary-coded Genetic Algorithm. The objective of this algorithm is to minimize six frequently used test functions. 2. a hybrid-coded (binary and real) Genetic Algorithm, which belong to a Genetic Fuzzy Rule-Based System based on the Pittsburgh approach. 3. a real-coded Genetic Algorithm, which is integrated with a network simulator to verify the network protocol performance. In this case, the algorithm aims at generating the worst case traffic for the protocol under analysis. Finally, different comparisons are performed, aiming to assess the acceptable optimization power of the proposed system. The results have been compare with the ones obtained for other methods for changing the values of parameters. Moreover, a statiscal analysis have been done to ascertain if differences are significant between the proposed system and the other algorithms.