A two-mode offspring generation selection mechanism with co-evolution for sparse large-scale multiobjective optimization

Sparse large-scale multiobjective optimization problems (LSMOPs) have a wide range of practical applications. In recent years, numerous multiobjective evolutionary algorithms (MOEAs) have been developed to address the complexities of these problems. However, many existing MOEAs designed to solve sparse LSMOPs typically rely on fixed, experience-based vectors to guide offspring generation, which often makes it challenging to determine the optimal guiding vectors for different population states, leading to premature convergence and loss of population diversity. To some extent, this leads to a subjective selection of the vector used for offspring generation. To address this issue, we propose a two-mode offspring generation selection mechanism (TOGSM) that incorporates diversified sparse knowledge into the offspring generation process. The switching between these two modes is based on a designed offspring performance indicator. We also divide the population into two subpopulations by employing techniques of Pareto dominance relationship and fitness values. In each generation, the loser subpopulation generates offspring solutions during the reproduction process, under the guidance of the winner subpopulation. Experimental results confirm that TOGSM incorporating two-mode mechanism and co-evolution strategy can generate higher quality Pareto optimal solutions with faster convergence speed than the state-of-the-art (SOTA) comparative algorithms.