Scholarlink

<p>High frequency vehicle-to-vehicle (V2V) communication in the Internet of Vehicles (IoV) leads to severe spectrum collisions and limited system capacity. Meanwhile, safety information transmission requires V2V communication with high transmission success rate. This paper proposes a spectrum and power allocation method based on the integration of graph neural networks (GNNs) and dueling double deep-Q network (D3QN) reinforcement learning to address the above problems. Specifically, this method first constructs a graph with V2V communication links as nodes and the interference relationships between different V2V links as edges. Then, it utilizes GNN to extract low-dimensional features of graph nodes for characterizing the interference relationships between links. Finally, by using the learned low-dimensional features combined with local observations of V2V links, the Double Deep Q-Network mitigates Q-value overestimation by separating action selection from value evaluation, and decouples state value and action advantage through a dual-branch network structure. This design optimizes spectrum allocation and power selection, improves the information transmission success rate of V2V links, and reduces interference to vehicle-to-infrastructure (V2I) links. Simulation results demonstrate that the proposed resource allocation method improves both the successful transmission rate of safety-critical messages over V2V links and the sum capacity of V2I links.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

A V2X communication resource allocation method based on graph neural networks and deep reinforcement learning

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年5月21日

查看全文
  • Wenhong Yu,
  • Xinran Yang,
  • Shuo Yu

摘要

High frequency vehicle-to-vehicle (V2V) communication in the Internet of Vehicles (IoV) leads to severe spectrum collisions and limited system capacity. Meanwhile, safety information transmission requires V2V communication with high transmission success rate. This paper proposes a spectrum and power allocation method based on the integration of graph neural networks (GNNs) and dueling double deep-Q network (D3QN) reinforcement learning to address the above problems. Specifically, this method first constructs a graph with V2V communication links as nodes and the interference relationships between different V2V links as edges. Then, it utilizes GNN to extract low-dimensional features of graph nodes for characterizing the interference relationships between links. Finally, by using the learned low-dimensional features combined with local observations of V2V links, the Double Deep Q-Network mitigates Q-value overestimation by separating action selection from value evaluation, and decouples state value and action advantage through a dual-branch network structure. This design optimizes spectrum allocation and power selection, improves the information transmission success rate of V2V links, and reduces interference to vehicle-to-infrastructure (V2I) links. Simulation results demonstrate that the proposed resource allocation method improves both the successful transmission rate of safety-critical messages over V2V links and the sum capacity of V2I links.