EUVIMEDEUVIMED entdecken
Open Access · CC BY-NC-ND 4.0

Dynamic 3-D Bin Packing Collaborative Optimization Method Based on Hierarchical Reinforcement Learning Driven by Digital Twins

Jiquan Zhang, Jiyuan Chen, Wenxian Guo, Yongxin Jing, Junwei Yang

IEEE Access · 2026 · 14 · 120018-120041

DOI: 10.1109/access.2026.3720551 ↗

Kurz gefasst

Fachartikel in der Zeitschrift IEEE Access, 2026. Verfasst von Jiquan Zhang und 4 weiteren Autor:innen. Die Arbeit nennt 34 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY-NC-ND 4.0). Gefördert von: Management Innovation Project of Southern Power Grid Supply Chain Group Company Ltd..

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Modern logistics systems are confronted with the challenge of the collaborative optimization of container loading efficiency and stability under dynamic environments and multi-dimensional constraints. Traditional heuristic algorithms suffer from bottlenecks such as low space utilization and poor dynamic adaptability. To address the 3D bin packing problem, this study proposes a collaborative optimization method driven by digital twins and hierarchical reinforcement learning. By constructing a five-layer digital twin architecture, real-time mapping between the physical and virtual Spaces is achieved, and combined with the hierarchical reinforcement learning framework (Manager-Worker network), the task decomposition and action execution separation of packing decisions are realized. Firstly, the construction content and methods of the digital twin during the packing process were proposed. The basic digital twin of the distribution process, the digital twin model of bin packing planning, and the digital twin model of packing process optimization were designed and constructed. Then a heuristic rule fusion strategy was designed to improve the original bin packing configuration tree model. The problem was stratified through the idea of hierarchical reinforcement learning, and the graph attention classification model was introduced to determine the optimal spatial point expansion scheme in the current situation, thereby providing more strategies in the 3D bin packing process. Experiments show that the hierarchical design heuristic fusion strategy performs well on multiple datasets and can collaboratively optimize the 3D bin packing process. Experiments on three synthetic datasets with varying rotation constraints and density conditions demonstrate that the proposed method achieves superior packing performance, optimizing space utilization, packing stability, and the number of packed items simultaneously. Specifically, compared to the state-of-the-art configuration tree method with maximum subspace expansion (PCT&EMS), the proposed model improves space utilization by 1.3 percentage points in discrete scenarios and maintains higher packing efficiency in continuous action spaces, demonstrating its effectiveness in collaborative optimization of the 3D bin packing process.

Förderung

  • Management Innovation Project of Southern Power Grid Supply Chain Group Company Ltd.

Zitieren

APA

Zhang, J., Chen, J., Guo, W., Jing, Y., & Yang, J. (2026). Dynamic 3-D Bin Packing Collaborative Optimization Method Based on Hierarchical Reinforcement Learning Driven by Digital Twins. IEEE Access, 14, 120018-120041. https://doi.org/10.1109/access.2026.3720551

Vancouver

Zhang J, Chen J, Guo W, Jing Y, Yang J. Dynamic 3-D Bin Packing Collaborative Optimization Method Based on Hierarchical Reinforcement Learning Driven by Digital Twins. IEEE Access. 2026;14:120018-120041. doi:10.1109/access.2026.3720551

BibTeX
@article{Zhang2026,
  author = {Zhang, Jiquan and Chen, Jiyuan and Guo, Wenxian and Jing, Yongxin and Yang, Junwei},
  title = {Dynamic 3-D Bin Packing Collaborative Optimization Method Based on Hierarchical Reinforcement Learning Driven by Digital Twins},
  journal = {IEEE Access},
  year = {2026},
  volume = {14},
  pages = {120018--120041},
  doi = {10.1109/access.2026.3720551}
}
Kostenlos starten →