A Non-uniform Clustering Based Evolutionary Algorithm for Solving Large-Scale Sparse Multi-objective Optimization Problems

Evolutionary algorithms have shown their effectiveness in solving sparse multi-objective optimization problems (SMOPs). However, for most of the existing multi-objective optimization algorithms (MOEAs) for solving SMOPs, their search granularity keeps the same for all the decision variables, which leads to significant performance deterioration when dealing with SMOPs in high-dimensional decision spaces. To tackle the issue, in this paper, a non-uniform clustering based evolutionary algorithm, termed NUCEA, is proposed for solving large-scale SMOPs. The proposed algorithm divides the decision variables into multiple groups with varying sizes, so as to reduce the search space with different granularity. These clustering outcomes inspire the development of new genetic operators, which have been proven to efficiently perform dimensionality reduction when approximating sparse Pareto optimal solutions. Experimental results on both benchmark and real-world SMOPs have shown that the proposed algorithm has significant advantages in comparison with the state-of-the-art evolutionary algorithms.