A Directional Vector-Guided Large-Scale MultiObjective Optimization Algorithm

For large-scale multiobjective evolutionary algorithms (LSMOEAs), maintaining algorithmic performance to generate high-quality solutions while continuously increasing the dimensionality of decision variables remains a major challenge. Directional vectors systematically guide search directions, narrow the effective search space, focus on more promising regions, and avoid ineffective blind exploration. This study proposes a directional vector-guided large-scale multiobjective optimization algorithm (DV-LSMOEA). The algorithm obtains representative solutions through dynamic step size sampling to assist offspring reproduction. The proposed algorithm constructs four distinct types of directional vectors using two groups of individuals closer to the ideal points and dynamically adjust the sampling step size throughout the optimization process, ensuring that representative solutions involved in reproduction exhibit stronger advantages. Comprehensive experiments show that the proposed algorithm outperforms five other state-of-the-art LSMOEAs for nine large-scale multiobjective benchmark problems LSMOP1-LSMOP9 with up to three objectives and 1000 decision variables.