Gustav Sennton

Comparing different Genetic Algorithms through solving Steiner Networks

Abstract

The purpose of this paper is to inspect the behavior of different genetic algorithms. Some different characteristics, of selection and reproduction of individuals, are implemented in a genetic algorithm and the results are then compared to see if some characteristics are more important than others. More specifically the reproduction methods mutation and crossover are compared and the selection methods elitism selection and biased random selection are compared. The problem, which the genetic algorithms are tested on, is called the Steiner network problem. The results indicates that using elitism selection together with mutations is the best method when solving trivial problems. For more difficult problems mixing mutations and crossovers and using biased random selection seems to be the best alternative though this result is not as certain as the fast convergence of elitism-mutation algorithms for trivial problems.