Local Adaptive Operator Selection in Adaptive Non-Dominated Tournament Genetic Algorithm (A-NTGA) in Solving Multi-Objective NP-Hard Problems With Constraints
Michał Antkiewicz, Paweł B. Myszkowski · IEEE Access · 2026
The incorporation of multiple specialized operators can significantly enhance the Multi-Objective Evolutionary Algorithms in handling constraints. However, selection or manual setup of operator proportions is challenging. The Adaptive Operator Selection is applied to promote the selection of specific operators based on their effectiveness in the optimization process. Most existing methods rely on global adaptation or state-based models, which require additional learning parameters and do not capture local heterogeneity. It overlooks the fact that constraints and, therefore, an operator’s effectiveness can vary across different regions of the solution landscape. This paper introduces A-NTGA, an extension of the B-NTGA framework, which implements stateless Local Adaptive Operator Selection with individual-level memory, allowing operators to adapt to local solution characteristics across the Pareto front. A memory decay mechanism enables adaptation to different stages of the evolutionary process. To verify the generalization, experiments were conducted on a diverse benchmark suite of 3 multi-objective real-world combinatorial constrained NP-hard problems: Multi-Skill Resource Constrained Project Scheduling Problem, Traveling Thief Problem, and Multi-Stage Resource Allocation. The proposed method outperforms all others examined, while the applied adaptation mechanism adjusts the operator selection both in the optimization time and the objective space.