Scholarlink

With the acceleration of urbanization, traffic congestion has become increasingly severe. This study focuses on the development of intelligent adaptive traffic signal control models, aiming to enhance road capacity, reduce traffic congestion and carbon emissions, and improve traffic safety through big data analytics, machine learning. We propose a model that combines Deep Q-Networks (DQN) with an attention-enhanced state representation mechanism to generate signal timing strategies. This model dynamically adjusts the timing of traffic lights, considering multiple factors such as vehicle flow and pedestrian waiting times, and optimizes overall road accessibility through a reward mechanism. Additionally, we employ the distributed reinforcement learning framework IMPALA and the pruning tool AutoSlim to enhance computational efficiency and reduce deployment costs, enabling the model to run lightweight on edge computing device. Finally, the simulation results on the SUMO traffic simulation platform verify the feasibility and superiority of the intelligent adaptive traffic signal control model, which provides new insights and technical support for improving urban traffic management.

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

Research on Adaptive Traffic Signal Control Based on Multi-object Tracking and Deep Reinforcement Learning

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Su Hu,
  • Junyu Zeng,
  • Fangyi Liang,
  • Peixi Chen,
  • Xin Zhong,
  • Ximing Lin

摘要

With the acceleration of urbanization, traffic congestion has become increasingly severe. This study focuses on the development of intelligent adaptive traffic signal control models, aiming to enhance road capacity, reduce traffic congestion and carbon emissions, and improve traffic safety through big data analytics, machine learning. We propose a model that combines Deep Q-Networks (DQN) with an attention-enhanced state representation mechanism to generate signal timing strategies. This model dynamically adjusts the timing of traffic lights, considering multiple factors such as vehicle flow and pedestrian waiting times, and optimizes overall road accessibility through a reward mechanism. Additionally, we employ the distributed reinforcement learning framework IMPALA and the pruning tool AutoSlim to enhance computational efficiency and reduce deployment costs, enabling the model to run lightweight on edge computing device. Finally, the simulation results on the SUMO traffic simulation platform verify the feasibility and superiority of the intelligent adaptive traffic signal control model, which provides new insights and technical support for improving urban traffic management.