Convergence-Driven Adaptive Many-Objective Particle Swarm Optimization

Abstract

In recent years, the prevalence of Many-Objective Optimization Problems (MaOPs) in practical applications has been increasing. However, traditional multi-objective optimization algorithms, such as Multiple Objective Particle Swarm Optimization (MOPSO), often face challenges of dimensionality and selection pressure when handling MaOPs. To overcome these challenges, this study proposes a Convergence-Driven Adaptive Many-Objective Particle Swarm Optimization (CDA-MOPSO) algorithm. This algorithm introduces a convergence metric to assess the convergence status and solution distribution quality of the particle swarm during iterations. Based on this metric, Convergence-Aware Learning Factor Adjustment (CALFA), Convergence-Oriented Dimension Variation Strategy (CODVS), and Convergence-Driven Archive Maintenance (CDAM) operations are proposed. Additionally, evolutionary search is further conducted on the external archive to enhance algorithm performance. To validate the performance of the CDA-MOPSO algorithm, extensive experiments are conducted using standard test problems such as DTLZ and WFG. Experimental results demonstrate that the CDA-MOPSO algorithm exhibits superior convergence and solution distribution characteristics across multiple standard test functions, particularly in handling many-objective optimization problems, outperforming traditional multi-objective algorithms significantly. In conclusion, the CDA-MOPSO algorithm provides a novel solution for many-objective optimization problems, offering strong convergence capability and solution diversity, with broad prospects for practical applications.