Genetic algorithms based on a granular surrogate model and fuzzy aptitude functions

Genetic Algorithms are widely used in optimization and search problems trying to find the more useful and better solutions according to one or several objectives. However, due to the high dimensionality in the space of solutions, this heuristic algorithm is computationally expensive. A viable alternative that produces acceptable results, specially in engineering problems, is based in the idea of alternative models that reflects approximate solutions, and it is called a surrogate model. In this paper, the employed surrogate model is based on granular computing, an emerged concept from fuzzy theory, that deals with grouping the solutions in the search space according to some specific similarities. We present a detail algorithm to construct such granular entities, and to find the optimal adaption process routed to not only avoid excessive fitness evaluations but also to obtain a better convergence of the algorithm. The obtained results on traditional benchmark functions show satisfactory improvements to such heuristic process.