A sparse large-scale multi-objective evolutionary algorithm based on sparsity detection

Sparse large-scale multi-objective optimization problems (LSMOPs), in which most decision variables of the Pareto-optimal solutions are zero, have become increasingly prevalent in real-world applications. An effective approach for addressing the sparse LSMOPs is the two-layer encoding scheme. However, a significant challenge when adapting the two-layer encoding scheme remains in accurately identifying the sparse distribution of Pareto-optimal solutions. Therefore, this paper proposes a sparse large-scale multi-objective evolutionary algorithm based on sparsity detection. The proposed algorithm uses the two-layer encoding scheme with a specialized focus on finding the positions of sparse non-zero variables by optimizing the binary vector. In the algorithm, an initialization strategy based on sparsity detection is proposed. This strategy employs a sparsity detection method to identify non-zero variables at the beginning of the algorithm. Moreover, a two- stage search strategy is designed to enhance the ability to search the sparse Pareto-optimal solutions. In this strategy, two types of knowledge are adaptively selected to guide the genetic operators of the binary vector. Experimental results from the benchmark suite and real-world applications demonstrate that the proposed algorithm outperforms the existing state-of-the-art algorithms in solving sparse LSMOPs.