Abstract
This paper proposes a novel Differential Evolution based algorithmic structure for solving continuous global optimization problems. The proposed structure makes use of the recently introduced concept of compact Differential Evolution as a search unit. Several compact units evolve simultaneously and interact in order to solve the optimization problem. In other words, the compact units are supposed to explore the decision space from diverse perspectives. The search work performed by the compact units is coordinated by a global supervision unit which processes by means of a global search the achievements obtained by the various compact units. More specifically, each compact unit performs a step of compact Differential Evolution and then feeds the achieved results to a global optimizer which recombines during one generation the candidate solutions and returns the improved genotypes to the corresponding compact units. In this implementation we selected as a global supervision unit a Differential Evolution algorithm with self-adaptive control parameters previously proposed in literature. The concept of global supervision, here introduced, appears to be very promising as it allows the improvement and development of the results locally obtained by each compact unit, thus preventing premature convergence of each unit and promoting a successful continuation of the search. Numerical results show that the resulting algorithm considered in this study displays a promising performance for a set of challenging test problems and is competitive with the-state-of-the-art Differential Evolution based algorithms.