An adaptive metaheuristic for unconstrained multimodal numerical optimization

Abstract

The purpose of this paper is to show an adaptive metaheuristic based on GA, DE, and PSO. The choice of which one will be used is made based on a probability that is uniform at the beginning of the execution, and it is updated as the algorithm evolves. That algorithm producing better results tend to present higher probabilities of being selected. The metaheuristic has been tested in four multimodal benchmark functions for 1000, 2000, and 3000 iterations, managing to reach better results than the canonical GA, DE, and PSO. A comparison between our adaptive metaheuristic and an adaptive GA has shown that our approach presents better outcomes, which was proved by a t-test, as well.