城市交通拥堵预测研究:以西安市为例  

Urban Traffic Congestion Prediction Study:Case Study of Xi’an City

作　　者：颜建强 张霖 高原[2] 李银香 YAN Jianqiang;ZHANG Lin;GAO Yuan;LI Yinxiang(School of Information Science and Technology,Northwest University,Xi’an 710127,China;School of Economics and Management,Northwest University,Xi’an 710127,China)

机构地区：[1]西北大学信息科学与技术学院,西安710127 [2]西北大学经济管理学院,西安710127

出　　处：《计算机工程与应用》2025年第1期330-340,共11页Computer Engineering and Applications

摘　　要：随着经济的发展和生活水平的提升,出行旅游逐渐成为流行的休闲方式,尤其在国庆等假日期间。然而,集中的出行需求引起的城市交通拥堵问题,不仅影响出行效率和体验,也加大了城市运行压力。该研究以西安市为例,深入分析假日期间的交通拥堵模式和特征,提出了一种基于深度学习拥堵预测模型ConstFormer。该模型融合了图卷积网络(graph convolutional network)和Transformer架构,利用图卷积网络探索空间邻接关系,以及Transformer的自注意力机制来挖掘长时空依赖关系,从而有效预测未来的交通状况。对比实验表明,ConstFormer模型对比基线模型均取得了最佳的性能,能够为人们规避高峰期和合理安排出行提供帮助。With the development of the economy and the improvement of living standards,traveling and tourism have gradually become a popular way of leisure,especially during the National Day and other holidays.However,urban traffic congestion caused by concentrated travel demand not only affects travel efficiency and experience,but also increases the pressure of urban operation.This study takes Xi’an as an example to deeply analyze the traffic congestion patterns and characteristics during the holiday period,and proposes a deep learning-based congestion prediction model,ConstFormer.The model integrates the graph convolutional network(GCN)and Transformer architectures,which explores spatial adjacencies using the GCN,and the Transformer’s self-attention mechanism to mine long spatial and temporal dependencies to effectively predict future traffic conditions.Comparison experiments show that the ConstFormer model achieves the best performance over the baseline model,which can help people avoid peak periods and rationalize their travel arrangements.

关 键 词：拥堵预测 数据分析 城市交通 图卷积网络 TRANSFORMER 

分 类 号：TP391[自动化与计算机技术—计算机应用技术] U495[自动化与计算机技术—计算机科学与技术]