EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

A Deep Reinforcement Learning Framework for Centralized Dynamic Routing Optimization via Integrated Link Weight Prediction and Path Selection

Erika A. Katonova, Juraj Jurcisin, Rastislav Petija, Eugen Slapak, Peter Fecilak

IEEE Access · 2026 · 14 · 86605-86632

DOI: 10.1109/access.2026.3700962 ↗

Kurz gefasst

Fachartikel in der Zeitschrift IEEE Access, 2026. Verfasst von Erika A. Katonova und 4 weiteren Autor:innen. Die Arbeit nennt 51 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: Research and Development Agency of the Ministry of Education, Research, Development and Youth of Slovakia.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Traditional routing protocols such as OSPF rely on static metrics, failing to account for real-time link utilization. This static approach often directs traffic to congested bottlenecks while redundant paths remain idle, leading to increased latency, packet loss, and throughput degradation. To address this, we propose a centralized dynamic routing system powered by Deep Reinforcement Learning (DRL). We introduce three distinct routing schemes: 1) a weight-prediction scheme, where a DRL agent predicts dynamic edge weights based on network states; 2) a path-selection scheme, where a DRL agent directly selects the optimal route from a set of candidate paths; and 3) a hybrid scheme that combines both approaches, utilizing one agent to predict link weights and a second agent to select the final path from the resulting set. The results of the conducted experiments and simulation demonstrate that the proposed hybrid method inherently ensures Quality of Service (QoS) without manual configuration, reduces retransmissions by reducing packet loss, and maximizes total network capacity by actively using redundant links to distribute load.

Förderung

  • Research and Development Agency of the Ministry of Education, Research, Development and Youth of Slovakia

Zitieren

APA

Katonova, E. A., Jurcisin, J., Petija, R., Slapak, E., & Fecilak, P. (2026). A Deep Reinforcement Learning Framework for Centralized Dynamic Routing Optimization via Integrated Link Weight Prediction and Path Selection. IEEE Access, 14, 86605-86632. https://doi.org/10.1109/access.2026.3700962

Vancouver

Katonova EA, Jurcisin J, Petija R, Slapak E, Fecilak P. A Deep Reinforcement Learning Framework for Centralized Dynamic Routing Optimization via Integrated Link Weight Prediction and Path Selection. IEEE Access. 2026;14:86605-86632. doi:10.1109/access.2026.3700962

BibTeX
@article{Katonova2026,
  author = {Katonova, Erika A. and Jurcisin, Juraj and Petija, Rastislav and Slapak, Eugen and Fecilak, Peter},
  title = {A Deep Reinforcement Learning Framework for Centralized Dynamic Routing Optimization via Integrated Link Weight Prediction and Path Selection},
  journal = {IEEE Access},
  year = {2026},
  volume = {14},
  pages = {86605--86632},
  doi = {10.1109/access.2026.3700962}
}
Kostenlos starten →