EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

Topology-Aware Graph Reinforcement Learning Method for Supplier Switching and Order Allocation Under Disruption Risks

Xiuhong Shi, Jide Nie, Tao Sun

IEEE Access · 2026 · 14 · 165555-165574

DOI: 10.1109/access.2026.3746961 ↗

Kurz gefasst

Fachartikel in der Zeitschrift IEEE Access, 2026. Verfasst von Xiuhong Shi und 2 weiteren Autor:innen. Die Arbeit nennt 36 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: Research Office of Wuhan Institute of Shipbuilding Technology.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Global supply chains, while continually improving efficiency by sourcing across regions and adopting lean production, are becoming more susceptible to regional disasters, port congestion, and upstream disruptions. In a system where a single manufacturer deals with multiple suppliers, sourcing decisions must consider not only the costs, capacities, and reliability of individual suppliers but also the risk dependencies stemming from shared regions, logistics channels, and upstream sources. This study introduces a method based on graph reinforcement learning that is aware of the supply chain structure to facilitate supplier switching and order allocation while considering disruption risks. This method involves representing manufacturers, potential suppliers, supply links, and risk dependencies in a dynamic supply chain graph. It utilizes a graph attention network that is aware of edge types to extract topological embeddings and learn order allocation preferences within a MAPPO framework. Moreover, a feasibility projection mechanism is devised to convert the learned actions into actual order quantities that meet demand, inventory, and capacity requirements. The results show that the topology-aware approach reduces total costs by approximately 5.22%–10.18% and enhances order fulfillment rates by approximately 4.94%–7.62% compared with models lacking graph structures. Ablation studies focusing on risk links and scenarios without projections confirm the effectiveness of identifying shared exposures through risk dependency links and the significance of feasibility projections in preventing constraint violations and demand deficits. The study concludes that incorporating shared risk exposures as structured graphs in operational order allocation is crucial, and feasibility projection is vital for ensuring successful order execution in supply chain decision-making based on learning principles.

Förderung

  • Research Office of Wuhan Institute of Shipbuilding Technology

Zitieren

APA

Shi, X., Nie, J., & Sun, T. (2026). Topology-Aware Graph Reinforcement Learning Method for Supplier Switching and Order Allocation Under Disruption Risks. IEEE Access, 14, 165555-165574. https://doi.org/10.1109/access.2026.3746961

Vancouver

Shi X, Nie J, Sun T. Topology-Aware Graph Reinforcement Learning Method for Supplier Switching and Order Allocation Under Disruption Risks. IEEE Access. 2026;14:165555-165574. doi:10.1109/access.2026.3746961

BibTeX
@article{Shi2026,
  author = {Shi, Xiuhong and Nie, Jide and Sun, Tao},
  title = {Topology-Aware Graph Reinforcement Learning Method for Supplier Switching and Order Allocation Under Disruption Risks},
  journal = {IEEE Access},
  year = {2026},
  volume = {14},
  pages = {165555--165574},
  doi = {10.1109/access.2026.3746961}
}
Kostenlos starten →