Migration Guided by a Performance Index in Heterogeneous Island Models

Evolutionary Algorithms are efficient alternatives to solve complex optimization problems. The high computational cost of these algorithms commonly motivates their implementation to run in parallel computational environments. Island Model enables the parallel implementation of Evolutionary Algorithms relatively easily to incorporate the migration operation into the evolutionary process. The inclusion of new solutions in a population, previously evolved in another population, can contribute positively to the problem’s solution quality. In this work, a performance index was added to the Island Model, aiming to indicate how efficiently each island’s population is in solving the problem according to its algorithm. Islands with higher performance indexes receive more individuals in migrations. In this way, these algorithms become more active in the evolutionary process. The experiments demonstrated that the new model solutions were as good as the solutions from each problem’s best algorithm. We also noticed that even if we remove the most efficient algorithm from the model, it still adapts and provides efficient solutions.