A study of topology in insular Parallel Genetic Algorithms

In this paper we study how the connectivity affects the performance of insular parallel genetic algorithms (PGAs). Seven topologies PGAs were proposed, with growing number of connections. We used three instances of the well-known traveling salesman problem as benchmark. Each island of the PGA had different parameters and we established a fixed migration policy for all islands. Experiments were done and average results were reported. The effect of coevolution in PGAs was evidenced. The convergence time increased with the number of connections of the topology. The quality of solutions also increased in the same way. Although topologies with large connectivity increases the overall processing time, they take benefits to the quality of solutions found.