摘要
【目标】精准的交通流量预测是实现高速公路主动管控的基础,但现有方法因忽略交通流时空相关性及天气等外部特征影响,导致预测精度和稳定性不足。为充分挖掘交通流量与外部特征的内在联系以及交通流复杂的时空关联性,提出一种基于特征融合的时空图卷积网络(FSTGCN)交通流量预测模型。【方法】通过FSTGCN构建一种融合交通流量与外部特征的特征卷积网络,并将其应用到特征模块中来捕捉交通流量与外部特征的关联性,同时使用基于图卷积网络的空间模块来挖掘交通流量的空间关联性,最后将特征模块和空间模块的输出连接起来输入到基于门控循环单元的时间模块中来学习交通流量中的时间相关性。【结果】基于山东省高速公路实测数据的试验表明,FSTGCN在长期预测任务中精度显著优于主流基准模型,且在短期和长期预测中均表现出更优的稳定性。消融试验验证了特征模块、时间模块和空间模块对预测性能的积极贡献,极端误差对比表明FSTGCN模型具备较强的鲁棒性,同时训练效率验证了其在实际部署中的可行性。【结论】FSTGCN通过融合多源外部特征与时空关联性建模,有效解决了高速公路交通流量预测的精度与稳定性问题。
[Objective]Accurate traffic volume prediction is the basis for achieving expressway active control.However existing methods suffer from insufficient prediction accuracy and stability due to neglecting spatio-temporal correlations in traffic volume and external factors,e.g.,weather conditions.To fully explore the intrinsic connection between traffic volume and external features as well as the complex spatial-temporal correlation of traffic volume itself,a novel feature-fused spatio-temporal graph convolutional networks(FSTGCN)was proposed for traffic volume prediction.[Method]The feature convolutional networks,integrating traffic volume and external features,were established with FSTGCN.It was used to capture correlations between traffic volume and external features in feature modules.The graph convolutional networks based spatial modules were employed to extract spatial correlation of traffic volume.Finally,the outputs from feature module and spatial module were connected and input into Gated Recurrent Unit based temporal modules for spatio-temporal correlation learning.[Result]Experimental results with real data from expressways in Shandong province indicate that FSTGCN significantly outperforms mainstream benchmarks in long-term prediction tasks,while maintaining superior stability across both short-and long-term horizons.Ablation tests confirm the positive contributions of feature module,spatial module and temporal module.The extreme error comparison shows that FSTGCN model has strong robustness.Simultaneously,the training efficiency analysis further validates its practicality for real deployment.[Conclusion]The FSTGCN effectively addresses accuracy and stability challenges in expressway traffic volume prediction through multisource feature fusion and spatio-temporal correlation modeling.
作者
唐进君
段一鑫
商淑杰
王骋程
陈群
TANG Jinjun;DUAN Yixin;SHANG Shujie;WANG Chengcheng;CHEN Qun(School of Traffic&Transportation Engineering,Central South University,Changsha,Hunan 410075,China;Shandong Hi-speed Infrastructure Construction Co.,Ltd.,Jinan,Shandong 250000,China;Shandong Provincial Communications Planning and Design Institute Group Co.,Ltd.,Jinan,Shandong 250000,China)
出处
《公路交通科技》
北大核心
2025年第6期22-31,共10页
Journal of Highway and Transportation Research and Development
基金
国家自然科学基金项目(52172310)
山东省交通运输厅科技计划项目(2021B68)
山东省自然科学基金青年基金(ZR202103040494)。
关键词
智能交通
交通流量预测
图卷积网络
ETC门架数据
特征融合
intelligent transport
traffic volume prediction
graph convolutional networks
ETC gantry system data
feature fusion
