EUVIMEDEUVIMED entdecken
Crossref

Graph Neural Network-Enhanced Multi-Objective Scheduling and Offloading for Hybrid Applications in Edge-Cloud Networks

Yang Sun, Jitao Yi, Yuwei Bian, Meng Li, Pengbo Si, Fei Richard Yu

IEEE Transactions on Cognitive Communications and Networking · 2026 · 12 · 7731-7744

DOI: 10.1109/tccn.2026.3685429 ↗

Kurz gefasst

Fachartikel in der Zeitschrift IEEE Transactions on Cognitive Communications and Networking, 2026. Verfasst von Yang Sun und 5 weiteren Autor:innen. Die Arbeit wurde bisher 1-mal zitiert (Crossref) und nennt 47 Literaturangaben. Gefördert von: Beijing Municipal Natural Science Foundation, National Natural Science Foundation of China.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

The rapid proliferation of the Internet of Things (IoT) has driven the emergence of intelligent applications characterized by complex task dependencies and heterogeneous quality of service (QoS) requirements. Efficient coordination of edge-cloud resources to satisfy diverse QoS demands of hybrid applications remains a challenging problem. In this paper, we formulate the task scheduling and offloading problem in edge-cloud networks as a multi-constraint, multi-objective combinatorial optimization problem through a two-dimensional decision model that jointly determines task prioritization and server selection. To support adaptive decision-making in dynamic environments, the problem is further modeled as a multi-objective Markov decision process (MOMDP). Based on this formulation, we propose a graph neural network-enhanced multi-objective reinforcement learning (GEMORL) framework that integrates a graph neural network (GNN) with multi-objective reinforcement learning (MORL). The GNN hierarchically extracts task-level, application-level, and global-level representations from task-related and latency-related attributes of applications, providing expressive state embeddings for policy learning. A multi-objective policy gradient (MOPG) algorithm is employed to train an independent policy network for each objective, and a deterministic policy is derived via a scalarization method. Extensive simulations demonstrate that GEMORL achieves superior trade-offs among conflicting objectives and consistently outperforms state-of-the-art baseline schemes across diverse network scenarios.

Förderung

  • Beijing Municipal Natural Science Foundation
  • National Natural Science Foundation of China

Zitieren

APA

Sun, Y., Yi, J., Bian, Y., Li, M., Si, P., & Yu, F. R. (2026). Graph Neural Network-Enhanced Multi-Objective Scheduling and Offloading for Hybrid Applications in Edge-Cloud Networks. IEEE Transactions on Cognitive Communications and Networking, 12, 7731-7744. https://doi.org/10.1109/tccn.2026.3685429

Vancouver

Sun Y, Yi J, Bian Y, Li M, Si P, Yu FR. Graph Neural Network-Enhanced Multi-Objective Scheduling and Offloading for Hybrid Applications in Edge-Cloud Networks. IEEE Transactions on Cognitive Communications and Networking. 2026;12:7731-7744. doi:10.1109/tccn.2026.3685429

BibTeX
@article{Sun2026,
  author = {Sun, Yang and Yi, Jitao and Bian, Yuwei and Li, Meng and Si, Pengbo and Yu, Fei Richard},
  title = {Graph Neural Network-Enhanced Multi-Objective Scheduling and Offloading for Hybrid Applications in Edge-Cloud Networks},
  journal = {IEEE Transactions on Cognitive Communications and Networking},
  year = {2026},
  volume = {12},
  pages = {7731--7744},
  doi = {10.1109/tccn.2026.3685429}
}
Kostenlos starten →