Transformer-enhanced Artificial Bee Colony with self-attention-based relational selection
Applied Soft Computing, cilt.203, 2026 (SCI-Expanded, Scopus)
- Yayın Türü: Makale / Tam Makale
- Cilt numarası: 203
- Basım Tarihi: 2026
- Doi Numarası: 10.1016/j.asoc.2026.116233
- Dergi Adı: Applied Soft Computing
- Derginin Tarandığı İndeksler: Science Citation Index Expanded (SCI-EXPANDED), Scopus, Applied Science & Technology Source, Compendex, INSPEC
- Anahtar Kelimeler: Artificial Bee Colony, Evolutionary optimization, Relational modeling, Self-attention mechanism, Swarm intelligence
- Sivas Cumhuriyet Üniversitesi Adresli: Evet
Özet
Standard Artificial Bee Colony (ABC) and many fitness-driven ABC variants mainly rely on scalar fitness values, local perturbations, and trial-counter-based scout decisions, which may underutilize explicit population-wide relational information. This study introduces the Transformer-enhanced Artificial Bee Colony algorithm (T-ABC), a novel ABC variant that integrates multi-head self-attention mechanisms into the onlooker and scout phases to improve population-level search behavior. The method uses attention-derived relational scores for food-source selection, attention-guided scout reinitialization, adaptive stochastic perturbation of projection matrices, and single-dimension best-guided perturbation. T-ABC was evaluated on the CEC2022 benchmark suite at D=2, 10, and 20, frequency-constrained truss optimization problems, and the three-dimensional AB off-lattice protein-folding energy model. On CEC2022 at D=20, T-ABC achieved the best mean result on 7 of the 12 functions and yielded 123 wins, 5 ties, and 40 losses against 14 competing algorithms across 168 pairwise comparisons according to the Wilcoxon rank-sum test. In the engineering and bioinformatics experiments, T-ABC achieved the best mean performance on 3 of the 5 truss optimization problems and 8 of the 11 protein-folding instances, including high-dimensional cases in both frequency-constrained truss optimization and off-lattice protein folding. These results indicate that attention-derived relational scoring can improve ABC search behavior under the tested settings. Nevertheless, the additional computational cost introduced by the attention mechanism and the scalability of T-ABC to substantially larger populations remain important limitations. The source code is available at https://github.com/Yeness/T-ABC-Transformer-based-Artificial-Bee-Colony-Algorithm.git.