Ensemble of resource allocation strategies in decision and objective spaces for multiobjective optimization

Abstract

A well exploitation of computational resource is essential when solving high-dimensional multiobjective problems (MOPs). Since many recent strategies presented for decision space and objective space are in consideration of searching efficiency, the collaboration between the two spaces is a promising approach for high-dimensional optimization. In this article, resource allocation strategies for both decision and objective spaces are readjusted and cooperated for complex MOPs. A metric-based variable partition strategy is introduced and a simple reference adaptation strategy is adopted to specify the searching orientations in the decision space and the objective space respectively. Subsequently, based on the above-mentioned basic techniques, three different evolutionary strategies are further designed to strengthen the directional convergence and preserve the diversity of target regions in a collaborative fashion. Several benchmark instances and a practical optimization problem are adopted in the experimental study. The effectiveness and rationality of the proposed resource allocation approach have been demonstrated by the experimental results.