Scholarlink

<p>Urban traffic congestion has emerged as a significant challenge impairing travel efficiency and transportation resource utilization. Although vehicle route guidance plays a crucial role in mitigating urban traffic congestion, traditional methods fail to accurately assess the spatiotemporal traffic patterns and adapt swiftly to dynamic traffic conditions and vehicular interactions. To address the issues, this paper proposes a Multi-agent Deep Reinforcement Learning framework based on cloud-edge computing for urban vehicle route guidance. The framework deploys intersection-based agents to collect realtime traffic data, facilitate global deep reinforcement learning, and offer intelligent and interactive route guidance. Moreover, an accurately spatiotemporal traffic state information mining method is proposed by integrating Graph Convolutional Network (GCN) and Gated Recurrent Unit network (GRU) with an attention mechanism. Subsequently, a modeling approach for inter-agent spatial relationship based on Graph Attention Network (GAT) is presented to facilitate the transmission and fusion of traffic characteristics on cloud. Finally, the Spatio-Temporal Mean-Field multi-agent Dueling Q-Routing (ST-MF-DQR) algorithm is proposed to optimize vehicular spatiotemporal coordination. The experimental evaluations on both synthetic and real traffic networks demonstrate that our method minimizes average vehicle waiting and travel times, significantly improving urban traffic mobility.</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

Multi-agent deep reinforcement learning based on cloud-edge computing for urban vehicle route guidance

  • 期刊论文

  • 发表日期：

    2025年11月11日

查看全文
  • Zhuhua Liao,
  • Junjian Gao,
  • Aiping Yi,
  • Yijiang Zhao,
  • Yue Tang

摘要

Urban traffic congestion has emerged as a significant challenge impairing travel efficiency and transportation resource utilization. Although vehicle route guidance plays a crucial role in mitigating urban traffic congestion, traditional methods fail to accurately assess the spatiotemporal traffic patterns and adapt swiftly to dynamic traffic conditions and vehicular interactions. To address the issues, this paper proposes a Multi-agent Deep Reinforcement Learning framework based on cloud-edge computing for urban vehicle route guidance. The framework deploys intersection-based agents to collect realtime traffic data, facilitate global deep reinforcement learning, and offer intelligent and interactive route guidance. Moreover, an accurately spatiotemporal traffic state information mining method is proposed by integrating Graph Convolutional Network (GCN) and Gated Recurrent Unit network (GRU) with an attention mechanism. Subsequently, a modeling approach for inter-agent spatial relationship based on Graph Attention Network (GAT) is presented to facilitate the transmission and fusion of traffic characteristics on cloud. Finally, the Spatio-Temporal Mean-Field multi-agent Dueling Q-Routing (ST-MF-DQR) algorithm is proposed to optimize vehicular spatiotemporal coordination. The experimental evaluations on both synthetic and real traffic networks demonstrate that our method minimizes average vehicle waiting and travel times, significantly improving urban traffic mobility.