An Evolution Algorithm with Objective-Wise Variable Analysis for Sparse Large-Scale Multi-Objective Optimization

Sparse large-scale multiobjective optimization problems are widely encountered in scientific research and engineering applications. However, the enormous search space and the intrinsic sparsity of the Pareto-optimal solutions substantially increase the difficulty of the optimization process. Different from existing approaches that directly rely on all objective values, this paper proposes an objective-wise variable analysis method. Specifically, the analysis method first evaluates the sensitivity of each objective to all decision variables, and then comprehensively aggregates the sensitivity information across multiple objectives to estimate the overall importance of decision variables. As a result, critical variables can be accurately identified to generate initial population, enabling the population to focus on the critical decision subspace efficiently. Furthermore, a non-coordinated mask-real evolution strategy is proposed to fully exploit the potential of mask vector and reduce the interference of real vectors. In this strategy, mask vector and real component evolve at separate update frequencies. This non-coordinate updating mechanism enables the mask structure to better match the corresponding real number vector. Experimental results demonstrate that the proposed algorithm significantly outperforms several state-of-the-art algorithms on the SMOP benchmark suite and exhibits promising potential in four practical real-world applications.