A reference vector-guided many-objective evolutionary algorithm based on an adaptive adjustment strategy
Baihao Qiao, Rongrong Bai, Ziru Feng, Boyang Qu, Li Yan, Haimin Hu · Applied Soft Computing · 2026
In recent years, numerous many-objective evolutionary algorithms (MaOEAs) have been developed to solve objective conflicts. However, as the number of objectives scales, effectively managing the inherent trade-off between diversity preservation and convergence efficiency in MaOEAs remains a persistent challenge. Therefore, we propose a reference vector-guided many-objective evolutionary algorithm based on an adaptive adjustment strategy (RVEAADS) to dynamically regulate exploration and exploitation trade-offs. In RVEAADS, the population is divided into two subpopulations, and a pre-defined proportion of individuals that are superior in performance are archived to ensure that the solution set converges to the Pareto optimal frontier. Subsequently, the inferior individuals use the reference vector to guide the selection of elite individuals, and implement an adaptive adjustment combining the diversity maintenance strategy and the convergence promotion strategy for the population after the guided selection, avoiding the problem of insufficient solutions and further balancing the convergence and diversity of the population. In addition, the diversity maintenance strategy can retrieve excellent solutions in the partial space twice to fill the insufficient number of solutions, and the promote convergence strategy can avoid the loss of solutions with significant convergence performance. Finally, the proposed RVEAADS is compared with state-of-the-art algorithms on 14 test problems, and the experimental results indicate the competitiveness of the proposed algorithm.