Scholarlink

The convergence of 5G, AI, and IoT technologies is enabling the transition of Unmanned Aerial Vehicle (UAV) systems from single-unit operations to intelligent swarm collaboration. Addressing the challenges of dynamic topologies and resource constraints in these networks, this paper proposes a spatiotemporal joint optimization framework. Our cross-layer design integrates an enhanced time synchronization algorithm with an intelligent routing mechanism to improve network performance. First, to mitigate cumulative clock errors, our synchronization protocol combines multi-path broadcasting with bidirectional exchange, using Maximum Likelihood Estimation (MLE) to transform the exponential accumulation of multi-hop errors into linear growth. Second, a Dueling Double Deep Q-Network (D3QN) routing algorithm uses a mobility-aware state representation and a multi-objective reward function to optimize for energy efficiency, link stability, and low latency. We fuse these components into a unified model using a Graph Neural Network (GNN) that enables closed-loop interaction via a bidirectional feedback mechanism. Experimental results demonstrate our architecture excels in suppressing cumulative errors, enhancing network throughput, and reducing packet loss, validating the effectiveness of the spatiotemporal joint optimization design.

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

SJ-GNN: A Graph Neural Network-Based Spatio-Temporal Joint Optimization Routing Algorithm

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Changwei Liu,
  • Jie Li,
  • Yuhe Zhang,
  • Yinrui Yu,
  • Jiaxin Lu,
  • Hongjun Ma

摘要

The convergence of 5G, AI, and IoT technologies is enabling the transition of Unmanned Aerial Vehicle (UAV) systems from single-unit operations to intelligent swarm collaboration. Addressing the challenges of dynamic topologies and resource constraints in these networks, this paper proposes a spatiotemporal joint optimization framework. Our cross-layer design integrates an enhanced time synchronization algorithm with an intelligent routing mechanism to improve network performance. First, to mitigate cumulative clock errors, our synchronization protocol combines multi-path broadcasting with bidirectional exchange, using Maximum Likelihood Estimation (MLE) to transform the exponential accumulation of multi-hop errors into linear growth. Second, a Dueling Double Deep Q-Network (D3QN) routing algorithm uses a mobility-aware state representation and a multi-objective reward function to optimize for energy efficiency, link stability, and low latency. We fuse these components into a unified model using a Graph Neural Network (GNN) that enables closed-loop interaction via a bidirectional feedback mechanism. Experimental results demonstrate our architecture excels in suppressing cumulative errors, enhancing network throughput, and reducing packet loss, validating the effectiveness of the spatiotemporal joint optimization design.