An empirical study of genetic algorithm parameter values in a network optimization problem.

摘要

The effect of different values of the genetic algorithm parameters determinant encoding, exchange mutation, uniform crossover, and population size on solution quality and computation time was studied in the context of the genetic algorithm optimization of a computer network represented by a degree-constrained minimum spanning tree. The impact of a gender differentiated population, in which only individuals of opposite gender could mate, was studied as well. Results showed that the mutation probability and population size were significant factors in the performance of the genetic algorithm in terms of solution quality and computation time. The genetic algorithm runs with higher mutation probabilities, higher crossover probabilities, and larger population sizes produced the lowest cost solutions, but also produced the largest computation times. The runs with the gender differentiated population produced solutions of similar quality to those produced by runs without the gender differentiation and with similar computation times.