Estudo sobre métodos evolutivos multiobjetivos voltados para robustez e diversidade no espaço de decisão

Abstract

Multiobjective optimization evolutionary algorithms (MOEAs) are usually evaluated by their ability to obtain good approximations of the Pareto-optimal front with an ideally uniform spread of samples in the space of objectives. However, by discarding information about thespace of decision variables, these computational tools return solution sets that do not consider the sensitivity of points to perturbations in their variables, or that do not contain possible alternative designs leading to similar performance values. This work presents an alternative methodof selection which employs a measure of solution density in the space of decision variables in addition to the traditional ones employed in the space of objectives during the selection procedure of the algorithm. Through an experimental evaluation, it is verified that the inclusionof this approach leads the algorithms to present a greater capacity to generate a representative sampling of the Pareto-optimal set. The proposed approach makes it possible to gather complementaryinformation regarding the sensitivity of solutions belonging to different regions of the search space, providing potentially useful information for the decision maker to select which particular solution may end up being implemented.