A decomposition-based constrained multi-objective optimization algorithm with dynamic resource reallocation guided by niche classification

Abstract

Decomposition-based constrained multi-objective evolutionary algorithms (CMOEAs) simplify complex optimization problems by decomposing them into multiple subproblems. These subproblems contribute unevenly to population optimization and demand varying computational resources across generations. However, most existing decomposition-based CMOEAs lack the prior knowledge for predetermining subproblem distributions. This leads to a suboptimal allocation of optimization weights and inflexible resource distribution, ultimately limiting their performance. To address this, we propose a niche classification strategy that identifies the distribution characteristics of local subproblems and categorizes them into distinct niches based on feasibility and dominance. This classification, updated each generation, provides dynamic prior knowledge, enabling adaptive allocation of optimization weights and computational resources tailored to each niche category. To operationalize this, we design a dual-population co-evolution framework based on decomposition, which dynamically redistributes resources among niches. Furthermore, we introduce an novel intergenerational fitness function to better assess the optimization potential of niches within the same category. By analyzing subpopulation changes across consecutive iterations, this function evaluates niche-level performance, thereby decoupling individual performance from fitness evaluation. Comprehensive experiments on 59 benchmark functions and a collaborative path planning task for multi-unmanned surface vehicles demonstrate that the proposed algorithm achieves competitive performance compared with seven state-of-the-art CMOEAs.