Scholarlink

<p>Rapid urbanization has significantly increased traffic congestion, adversely impacting travel efficiency, fuel consumption, and environmental sustainability. To address this challenge, this paper proposes RATM-FL, a multi-layer RSU-assisted traffic management framework based on federated learning. In the proposed system, roadside units (RSUs) locally collect and process real-time traffic data to predict congestion within their coverage areas, while a cloud server aggregate selected RSU model updates to construct a global traffic prediction model. Since RSU deployment is constrained by cost, the cloud server manages regions not covered by RSUs. Unlike conventional federated learning, RATM-FL employs a traffic-aware RSU selection strategy, formulated as an integer linear programming problem, which dynamically selects influential RSUs based on vehicle density and energy consumption. This approach reduces communication overhead and energy usage without degrading system performance. Extensive simulations conducted using the SUMO traffic simulator demonstrate that RATM-FL outperforms existing traffic management schemes by significantly reducing travel time, fuel consumption, and CO<sub>2</sub> emissions.</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

RATM-FL: RSU assisted traffic management using federated learning for intelligent transportation systems

  • 期刊论文

  • 发表日期：

    2026年4月25日

查看全文
  • Sreya Ghosh,
  • Iti Saha Misra,
  • Tamal Chakraborty

摘要

Rapid urbanization has significantly increased traffic congestion, adversely impacting travel efficiency, fuel consumption, and environmental sustainability. To address this challenge, this paper proposes RATM-FL, a multi-layer RSU-assisted traffic management framework based on federated learning. In the proposed system, roadside units (RSUs) locally collect and process real-time traffic data to predict congestion within their coverage areas, while a cloud server aggregate selected RSU model updates to construct a global traffic prediction model. Since RSU deployment is constrained by cost, the cloud server manages regions not covered by RSUs. Unlike conventional federated learning, RATM-FL employs a traffic-aware RSU selection strategy, formulated as an integer linear programming problem, which dynamically selects influential RSUs based on vehicle density and energy consumption. This approach reduces communication overhead and energy usage without degrading system performance. Extensive simulations conducted using the SUMO traffic simulator demonstrate that RATM-FL outperforms existing traffic management schemes by significantly reducing travel time, fuel consumption, and CO2 emissions.