An evolutionary algorithm with dynamic grouping-based reproduction for sparse multi-objective optimization

Sparse large-scale multi-objective optimization problems (SLMOPs), in which most decision variables in the Pareto-optimal solution are zero, have attracted significant attention in recent years. However, most existing algorithms struggle to accurately identify the non-zero decision variables of the Pareto-optimal solution. To address this issue, an evolutionary algorithm with dynamic grouping-based reproduction (DGREA) is proposed. The proposed DGREA first employs a probabilistic empirical score selection mechanism to evaluate variable importance. Then, decision variables are dynamically grouped by a dynamic grouping method. Finally, sparse solutions are precisely generated through a dynamic group reproduction operator. The proposed algorithm is evaluated on a benchmark suite and compared against five state-of-the-art algorithms. Experimental results demonstrate that DGREA outperforms the competing algorithms in terms of convergence and diversity, particularly in high-dimensional decision spaces. Furthermore, DGREA effectively identifies the non-zero decision variables in Pareto-optimal solutions.