ESOEA: Ensemble of single objective evolutionary algorithms for many-objective optimization

Inspired by the success of decomposition based evolutionary algorithms and the necessary search for a versatile many-objective optimization algorithm which is adaptive to several kinds of characteristics of the search space, the proposed work presents an adaptive framework which addresses many-objective optimization problems by using an ensemble of single objective evolutionary algorithms (ESOEA). It adopts a reference-direction based approach to decompose the population, followed by scalarization to transform the many-objective problem into several single objective sub-problems which further enhances the selection pressure. Additionally, with a feedback strategy, ESOEA explores the directions along difficult regions and thus, improving the search capabilities along those directions. For experimental validation, ESOEA is integrated with an adaptive Differential Evolution and experimented on several benchmark problems from the DTLZ, WFG, IMB and CEC 2009 competition test suites. To assess the efficacy of ESOEA, the performance is noted in terms of convergence metric, inverted generational distance, and hypervolume indicator, and is compared with numerous other multi- and/or many-objective evolutionary algorithms. For a few test cases, the resulting Pareto-fronts are also visualized which help in the further analysis of the results and in establishing the robustness of ESOEA.