Scholarlink

<p>Traffic congestion remains a significant challenge for urban mobility, leading to increased travel times, fuel consumption, and air pollution. Optimizing traffic signal timings plays a crucial role in mitigating congestion and improving traffic flow efficiency. In this paper, we propose a deep reinforcement learning (DRL) algorithm for traffic flow control in large-scale networks. Our approach employs a Deep Q-network (DQN) to dynamically adapt traffic signal timings, minimizing delays and maximizing overall throughput of vehicles. We compare the performance of our DRL algorithm with traditional traffic signal control methods, such as Fixed-Time Control (FTC), Adaptive Traffic Signal Control (ATSC), and Model Predictive Control (MPC) across various traffic scenarios and network sizes for average delay, throughput, and queue length. Experimental results demonstrate that the DRL algorithm outperforms these traditional methods, showcasing improved adaptability to dynamic traffic conditions, robust performance under varying demand patterns, and scalability to large networks. The findings highlight the potential of deep reinforcement learning for traffic flow control and suggest further research into addressing limitations and exploring extensions for real-world traffic management scenarios.</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

Deep Reinforcement Learning for Traffic Flow Control in Large-Scale Networks

  • 期刊论文

  • 发表日期：

    2025年10月6日

查看全文
  • Anand Gokul,
  • Sakthi Ganesh Mahalingam

摘要

Traffic congestion remains a significant challenge for urban mobility, leading to increased travel times, fuel consumption, and air pollution. Optimizing traffic signal timings plays a crucial role in mitigating congestion and improving traffic flow efficiency. In this paper, we propose a deep reinforcement learning (DRL) algorithm for traffic flow control in large-scale networks. Our approach employs a Deep Q-network (DQN) to dynamically adapt traffic signal timings, minimizing delays and maximizing overall throughput of vehicles. We compare the performance of our DRL algorithm with traditional traffic signal control methods, such as Fixed-Time Control (FTC), Adaptive Traffic Signal Control (ATSC), and Model Predictive Control (MPC) across various traffic scenarios and network sizes for average delay, throughput, and queue length. Experimental results demonstrate that the DRL algorithm outperforms these traditional methods, showcasing improved adaptability to dynamic traffic conditions, robust performance under varying demand patterns, and scalability to large networks. The findings highlight the potential of deep reinforcement learning for traffic flow control and suggest further research into addressing limitations and exploring extensions for real-world traffic management scenarios.