Parallel Genetic Algorithm with Strategy Parameters Encoded as Chromosomes

A.F. Tulai and F. Oppacher (Canada)


parameter adaptation, parallel geneticalgorithms, coevolution


In this paper, we address the issue of pre-determining the probabilities for various genetic operators. We show that a parallel genetic algorithm (PGA) that has randomly initialized crossover, mutation and recombination probabilities included as chromosomes in every individual genome in the population could be successfully used in function optimization problems. When used, as part of a cooperative coevolution model, for the difficult task of evolving cascade neural networks to solve the two-spiral classification problem, the algorithm matches the very good performance of an earlier version of the same algorithm that does not let the strategy parameters evolve freely.

Important Links:

Go Back