An adjoint feature-selection-based evolutionary algorithm for sparse large-scale multiobjective optimization

Abstract

Sparse large-scale multiobjective optimization problems (sparse LSMOPs) are characterized by an enormous number of decision variables, and their Pareto optimal solutions consist of a majority of decision variables with zero values. This property of sparse LSMOPs presents a great challenge in terms of how to rapidly and precisely search for Pareto optimal solutions. To deal with this issue, this paper proposes an adjoint feature-selection-based evolutionary algorithm tailored for tackling sparse LSMOPs. The proposed optimization strategy combines two distinct feature selection approaches. Specifically, the paper introduces the sequential forward selection approach to investigate independent sparse distribution, denoting it as the best sequence of decision variables for generating a high-quality initial population. Furthermore, it introduces the Relief approach to determine the relative sparse distribution, identifying crucial decisive variables with dynamic updates to guide the population in a promising evolutionary direction. Experiments are conducted on eight benchmark problems and two real-world problems, and experimental results verify that the proposed algorithm outperforms the existing state-of-the-art evolutionary algorithms for solving sparse LSMOPs.