A secure GNN-MADDPG framework with combinatorial action optimization for task offloading in vehicular networks

Vehicle-to-Everything (V2X) technology is rapidly developing. However, vehicular devices operate with limited computational power and energy. These constraints pose significant challenges for secure and energy-efficient task offloading. To address these challenges, this paper proposes a novel framework that integrates a Graph Neural Network (GNN) with the Multi-Agent Deep Deterministic Policy Gradient (MADDPG) algorithm for secure task offloading and resource allocation. The framework employs a GNN (GraphSAGE) to capture the dynamic network topology and global interference, overcoming the limitations of partial observability. This spatial feature representation supports coordinated decision-making by multiple agents within the MADDPG architecture. To handle the high-dimensional and coupled action space, a combinatorial action selection strategy is proposed and QMIX value function decomposition is adopted. This “optimize-then-combine” mechanism enables efficient joint optimization of continuous resources and discrete decisions. Furthermore, a hybrid RSA-AES encryption scheme combined with frequency hopping is implemented to ensure end-to-end data security and anti-jamming capabilities. Extensive comparative experiments demonstrated that the proposed framework significantly outperformed baseline methods, including DQN and standard MADDPG, in terms of task completion rate, average latency, and energy consumption, especially in high-load scenarios. Ablation studies further validated the critical contributions of the GNN, combinatorial action design, and security mechanisms. This work provides an efficient, secure, and scalable solution for resource optimization in complex V2X environments.

• Record URL:
  • https://doi.org/10.1016/j.vehcom.2026.101000
• Record URL:
  • http://www.sciencedirect.com/science/article/pii/S2214209626000021
• Availability:
  • Find a library where document is available. Order URL: http://worldcat.org/issn/22142096
• Supplemental Notes:
  • © 2026 Elsevier Inc. All rights are reserved, including those for text and data mining, AI training, and similar technologies. Abstract reprinted with permission of Elsevier.
• Authors:
  • Zhu, Yabin
  • 0009-0004-6974-7388
  • Zhao, Xu
  • 0000-0002-2351-8019
  • Zhang, Xin
  • 0009-0000-3659-8727
• Publication Date: 2026-4

Language

• English

Media Info

• Media Type: Web
• Features: Figures; References;
• Pagination: 101000
• Serial:
  • Vehicular Communications
  • Volume: 58
  • Issue Number: 0
  • Publisher: Elsevier
  • ISSN: 2214-2096
  • Serial URL: http://www.sciencedirect.com/science/journal/22142096/1/2

Subject/Index Terms

• TRT Terms: Computer security; Energy efficiency; Neural networks; Resource allocation; Vehicle to everything communications
• Subject Areas: Data and Information Technology; Highways; Security and Emergencies;

Filing Info

• Accession Number: 01977866
• Record Type: Publication
• Files: TRIS
• Created Date: Jan 29 2026 5:01PM