Evaluation of differential evolution with interaction network on a real-parameter optimization benchmark
Pavel Krömer, Petr Gajdoš · 2016
Differential evolution (DE) is a popular member of the wide family of population-based evolutionary optimization methods. These general purpose methods solve arbitrary problems by iteratively evolving a pool (population) of candidate solutions. Candidate solutions are during the artificial evolution updated and modified so that they efficiently explore solution space of the solved problem. Population-based metaheuristics focus on finding local or global optima with respect to selected optimization criteria (objective function). The iterative updates of candidate solutions usually involve some sort of interaction and information exchange. This behaviour has been recently cast as a temporal interaction network to allow a network-centric representation and research of artificial evolution. In this paper, we use a network-based model of the interactions in DE to improve the underlying algorithm. A simple extension of a traditional DE utilizing the properties of its interaction network is proposed in order to study the usefulness of this concept. The extended algorithm is evaluated on the CEC 2016 real-parameter optimization benchmark.