A dual-task-assisted self-organizing map evolutionary algorithm for constraining multimodal multi-objective optimization
Qianlin Ye, Zongda Wu, Keli Hu, Huawen Liu, Guoqing Li, Wanliang Wang · Swarm and Evolutionary Computation · 2026
• A dual-task-assisted framework is designed for constrained multimodal problems. • A self-organizing map network is constructed to learn the topology of the true PF. • The dynamic environment selection mechanism is innovatively proposed. • The adaptive pruning strategy is designed to balance the two spaces. This paper proposes a novel dual-task-assisted self-organizing multi-objective evolutionary algorithm (DSCMMOEA) for constrained multimodal multi-objective optimization problems (CMMOPs). The algorithm establishes a collaborative dual-task framework. The main task employs a self-organizing map (SOM) network to learn the topological structures of the Pareto front in both decision and objective spaces. This enables efficient exploration of discrete feasible regions. The auxiliary task, disregarding constraints, focuses on mining potentially high-quality infeasible solutions to guide the population across infeasible regions. A dynamic offspring sharing mechanism facilitates knowledge transfer between these tasks, enhancing global search capability. Furthermore, this paper designs a dynamic environment selection mechanism that adaptively adjusts weight vectors to balance diversity and convergence. An adaptive pruning strategy is also introduced to coordinate diversity in both decision and objective spaces by sequentially eliminating solutions using dual-space crowding distances. Experimental results on the CMMOP and CMOP benchmark suites, along with real-world engineering applications, confirm that DSCMMOEA provides an effective and innovative solution for complex CMMOPs.