A Study in function optimization with the breeder genetic algorithm

Abstract

Optimization is concerned with the finding of global optima (hence the name) of problems that can be cast in the form of a function of several variables and constraints thereof. Among the searching methods, {em Evolutionary Algorithms} have been shown to be adaptable and general tools that have often outperformed traditional {em ad hoc} methods. The {em Breeder Genetic Algorithm} (BGA) combines a direct representation with a nice conceptual simplicity. This work contains a general description of the algorithm and a detailed study on a collection of function optimization tasks. The results show that the BGA is a powerful and reliable searching algorithm. The main discussion concerns the choice of genetic operators and their parameters, among which the family of Extended Intermediate Recombination (EIR) is shown to stand out. In addition, a simple method to dynamically adjust the operator is outlined and found to greatly improve on the already excellent overall performance of the algorithm.