Scholarlink

<p>The rapid growth of urban populations has led to increased traffic congestion, posing significant challenges to safety, efficiency, and environmental sustainability. To address these issues, this research presents an artificial intelligence (AI)-driven framework that powers a state-of-the-art Internet of Things (IoT)-enabled distributed communication network for smart city traffic management. The proposed system integrates edge IoT sensors, real-time traffic data analytics, and intelligent decision-making modules to monitor and optimize traffic flows dynamically. A hybrid AI model—combining vision transformer (ViT) for advanced visual traffic detection and temporal convolutional network (TCN) for sequential traffic pattern prediction—is deployed across distributed edge nodes to enable localized, low-latency decision-making. The training process is optimized using the Ranger optimizer, a novel fusion of Lookahead and Rectified Adam (RAdam), ensuring faster convergence and improved model generalization in real-time conditions. The system is evaluated on the Dublin Traffic Sensor Dataset, which contains rich real-time vehicular and environmental sensor data from a smart city network. Experimental results show that the proposed framework significantly enhances traffic flow efficiency, reduces congestion and latency, and outperforms traditional centralized traffic control methods in terms of responsiveness and scalability. Deployment of the hybrid ViT-TCN model led to a 40.4% reduction in average vehicle delay time, a 42.2% decrease in average stop time, and a 29.5% drop in peak hour traffic density, alongside a 32.7% improvement in intersection throughput, reflecting enhanced traffic flow coordination. In terms of real-time responsiveness, the distributed edge deployment achieved a mean inference time of 47.6&#xa0;ms and a communication latency of 17.3&#xa0;ms, resulting in a total decision latency of 64.9&#xa0;ms, significantly outperforming centralized (126.9&#xa0;ms) and cloud-based systems (209.9&#xa0;ms).</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

AI-driven distributed IoT communication architecture for smart city traffic optimization

  • 期刊论文

  • 发表日期：

    2025年5月29日

查看全文
  • Alaa A. Qaffas

摘要

The rapid growth of urban populations has led to increased traffic congestion, posing significant challenges to safety, efficiency, and environmental sustainability. To address these issues, this research presents an artificial intelligence (AI)-driven framework that powers a state-of-the-art Internet of Things (IoT)-enabled distributed communication network for smart city traffic management. The proposed system integrates edge IoT sensors, real-time traffic data analytics, and intelligent decision-making modules to monitor and optimize traffic flows dynamically. A hybrid AI model—combining vision transformer (ViT) for advanced visual traffic detection and temporal convolutional network (TCN) for sequential traffic pattern prediction—is deployed across distributed edge nodes to enable localized, low-latency decision-making. The training process is optimized using the Ranger optimizer, a novel fusion of Lookahead and Rectified Adam (RAdam), ensuring faster convergence and improved model generalization in real-time conditions. The system is evaluated on the Dublin Traffic Sensor Dataset, which contains rich real-time vehicular and environmental sensor data from a smart city network. Experimental results show that the proposed framework significantly enhances traffic flow efficiency, reduces congestion and latency, and outperforms traditional centralized traffic control methods in terms of responsiveness and scalability. Deployment of the hybrid ViT-TCN model led to a 40.4% reduction in average vehicle delay time, a 42.2% decrease in average stop time, and a 29.5% drop in peak hour traffic density, alongside a 32.7% improvement in intersection throughput, reflecting enhanced traffic flow coordination. In terms of real-time responsiveness, the distributed edge deployment achieved a mean inference time of 47.6 ms and a communication latency of 17.3 ms, resulting in a total decision latency of 64.9 ms, significantly outperforming centralized (126.9 ms) and cloud-based systems (209.9 ms).