The paper proposes a hybrid adaptive operator selection (AOS) framework that combines offline and online experiences to dynamically and adaptively select promising search operators for meta-heuristics. The framework consists of three main components:
A state-based AOS module that learns the mapping from optimization state to promising operator selection. It first learns from offline experiences of solving past problems, then continuously learns from online experiences during the current problem-solving process.
A stateless AOS module that learns only from online experiences with relatively low computational cost.
A decision policy that balances the use of the state-based and stateless AOS modules in an online manner.
The effectiveness of the proposed hybrid framework is validated through extensive experiments on 170 real-value optimization problems and 34 instances of a challenging combinatorial optimization problem (Capacitated Vehicle Routing Problem with Time Windows). The results show that the hybrid framework outperforms state-of-the-art AOS methods, including both state-based and stateless approaches. Ablation studies verify the unique contributions of each component in the framework.
To Another Language
from source content
arxiv.org
Key Insights Distilled From
by Jiyuan Pei,J... at arxiv.org 04-17-2024
https://arxiv.org/pdf/2404.10252.pdfDeeper Inquiries