Simulated Annealing-based Evolutionary Algorithm for Constrained Multimodal Multiobjective Optimization
Juan Zou, Yu Li, Qi Deng, Tianbin Xie, Shengxiang Yang, Jinhua Zheng · Proceedings of the Genetic and Evolutionary Computation Conference Companion · 2025
Constrained multimodal multiobjective problems (CMMOPs) are challenging due to multiple objectives, modalities, and constraints. State-of-the-art algorithms often use the constrained dominance principle (CDP) and multimodal mechanisms to balance feasibility and diversity. However, relying solely on CDP may overemphasize feasible solutions, neglect infeasible ones, and affect diversity and global search performance. To address this, we propose a simulated annealing-based evolutionary algorithm, called CMMOSA. It integrates the constrained dominance principle based on simulated annealing (CDP-SA) with an archival mechanism. In the high-temperature stage, the algorithm accepts larger constraint-violating solutions, preventing premature convergence to local optima while maintaining feasible solutions in the archive. As the temperature decreases, the algorithm converges towards feasible solutions, exploring multiple constrained Pareto optimal solutions using global information from the archive. Experimental results show that CM-MOSA enhances global search capability in CMMOPs, overcomes local search limitations, and significantly improves solution quality and optimization efficiency.