Speeding up a multiobjective genetic algorithm with constraints through artificial neuronal networks

Abstract Evolutionary algorithms have been recognized to be well suited for multiobjective optimization [1]; their principal disadvantage is the large number of evaluations of objective function required [2], which is accentuated when those are computationally expensive. In this work, we propose the use of artificial neuronal networks, ANN, to speed up a multiobjective genetic algorithm with constraints, with base on the work of Gaspar-Cunha [3]. The neuronal network generates an approximated function of the original objective function, which are switched during the optimization; so, we reduce the evaluations of the original objective function and the computational time. The use of approximated functions created by the ANN allows reaching the optimal zone rapidly. Results show a significant reduction in the number of evaluations of the objective function, as well as in computational time, required to reaching the Pareto front.