Scholarlink

<p>Urban traffic congestion remains a critical challenge for modern cities, driving the advancement of Intelligent Transportation Systems (ITS) aimed at optimizing traffic flow through efficient Traffic Signal Control (TSC). Deep Reinforcement Learning (DRL) has shown significant potential in enhancing TSC performance; however, existing approaches often suffer from unstable learning due to suboptimal reward designs and state representations. This study proposes a Double Deep Q-Network-based Traffic Signal Control Agent (DDQNTSCA) integrated with Prioritized Experience Replay (PER) to improve learning stability and convergence. The proposed framework incorporates a compact Combined State Index (CSI) for efficient state representation, a congestion-sensitive reward formulation, a hybrid action mechanism that jointly optimizes phase selection and signal duration, and a neighbor-aware decentralized coordination strategy. Simulation results in the SUMO environment demonstrate that DDQNTSCA achieves faster convergence, enhanced adaptability, and significant reductions in average travel time, queue length, and cumulative delay compared to existing DRL-based TSC methods. These results highlight the effectiveness and potential applicability of the proposed approach for real-world traffic management systems.</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

Optimized adaptive traffic management using double DQN and prioritized experience replay in a multi-agent traffic control setup

  • 期刊论文

  • 发表日期：

    2026年7月16日

查看全文
  • Anurag Agrahari,
  • Mogal Aftab Baig,
  • Meera M. Dhabu,
  • Ashish Tiwari

摘要

Urban traffic congestion remains a critical challenge for modern cities, driving the advancement of Intelligent Transportation Systems (ITS) aimed at optimizing traffic flow through efficient Traffic Signal Control (TSC). Deep Reinforcement Learning (DRL) has shown significant potential in enhancing TSC performance; however, existing approaches often suffer from unstable learning due to suboptimal reward designs and state representations. This study proposes a Double Deep Q-Network-based Traffic Signal Control Agent (DDQNTSCA) integrated with Prioritized Experience Replay (PER) to improve learning stability and convergence. The proposed framework incorporates a compact Combined State Index (CSI) for efficient state representation, a congestion-sensitive reward formulation, a hybrid action mechanism that jointly optimizes phase selection and signal duration, and a neighbor-aware decentralized coordination strategy. Simulation results in the SUMO environment demonstrate that DDQNTSCA achieves faster convergence, enhanced adaptability, and significant reductions in average travel time, queue length, and cumulative delay compared to existing DRL-based TSC methods. These results highlight the effectiveness and potential applicability of the proposed approach for real-world traffic management systems.