Two-Archive Based Evolutionary Algorithm Using Adaptive Reference Direction and Decomposition for Many-Objective Optimization

Abstract

Many real world problems can be formulated as many objective optimization problems (MaOPs) which can not be solved easily. Although a lot of many-objective evolutionary algorithms(MOEAs) have been proposed, balancing the diversity and convergence is still an unsolved issue. In this paper, a two-archive based evolutionary algorithm based on adaptive reference point and decomposition method is proposed. Firstly, we use binary indicator to update convergence archive(CA). Then, we use the updated CA to generate adaptive reference points. Furthermore, we update diversity archive (DA) with modified penalty-based boundary intersection approach. Finally, the proposed algorithm has been tested on DTLZ1-DTLZ4 and WFG1-WFG9 benchmark problems with 10-22 objectives, and is compared with three state-of-art algorithms. The experimental results indicate that the proposed algorithm has great advantage to handle many-objective optimization.