Initial approaches to the application of islands-based parallel EDAs in continuous domains

Evolutionary algorithms in general and estimation of distribution algorithms in particular have been widely applied to solve combinatorial optimization problems, that is, problems in which the variables involved are discrete. However, there are also modifications or extension of these algorithms to deal with numerical optimization problems, i.e., continuous variables. In this paper we focus on dealing with numerical optimization problems by using island-based estimation of distribution algorithms. As far as we know, this is the first time parallel EDAs are used in continuous domains. Our proposal includes the use of two different topologies (ring and star) and two different types of information sharing between islands: individuals and models. In the case of model migration we specify the way in which models are combined and also study whether it is better to use or not adaptive combination. The proposed algorithms are tested over sixteen problem instances, and from the analysis of the results we can conclude that model migration using adaptive combination is, in general, the outstanding approach.