A large-scale multi-objective evolutionary algorithm based on importance rankings and information feedback

For large-scale multi-objective optimization problems, the trade-off between convergence and diversity brings significant challenges for researchers. Most of the reproduction operators in the evolutionary algorithms fail to achieve a superior performance. In order to address this issue, this work proposes a large-scale multi-objective evolutionary algorithm (LSMOEA) named LMOEA-IRIF. In the LMOEA-IRIF, a novel grouping strategy and an information feedback model (IFM) are designed to evolve the population. Specifically, the decision variables are clustered into multiple convergence-related and diversity-related subgroups based on their importance rankings. The importance rankings of decision variables are quantized by the maximum Euclidean distance between individuals generated in the objective space. Then the decision variables in each subgroup are optimized in a low-dimensional decision subspace, which can effectively speed up the convergence of population. Furthermore, the IFM, which takes the information from the previous generation into consideration, is devised to generate high-quality offspring and used to enhance the diversity of population. Comprehensive experiments are performed to validate the effectiveness of the LMOEA-IRIF. The experimental results show that the proposed algorithm obtains competitive performance in 56 of 76 benchmark instances against five state-of-the-art LSMOEAs.