首页|基于改进时空图卷积网络的道路行程时间预测模型

基于改进时空图卷积网络的道路行程时间预测模型

Road travel time prediction model based on improved spatiotemporal graph convolutional network

扫码查看
原文链接
  • 国家科技期刊平台
  • NETL
  • NSTL
  • 万方数据
为提高道路网行程时间预测精度,综合考虑行程时间的空间依赖性、时间依赖性和天气因素影响,提出了基于属性增强和注意力机制的时空图卷积网络模型.首先,构建属性增强单元,将行程时间和天气信息融合;然后,利用图卷积网络捕获空间依赖性,利用门控递归单元捕获时间依赖性,并利用注意力机制增强模型对特征的学习;最后,利用该模型在真实数据集上对未来15、30、45和60 min的行程时间进行预测.结果表明:预测结果的均方根误差(RMSE)分别为0.045 3、0.045 6、0.045 7和0.046 8,与其他模型相比表现更优;考虑了时间、空间和天气因素后,相较于不考虑天气的情况,预测误差降低了约10.3％;相较于不考虑空间依赖性的情况,降低了约24.2％,表明所提模型能更好表达时空依赖性和外部条件影响.
To enhance the accuracy of travel time prediction in road networks,a spatiotemporal graph convolu-tional network model based on attribute enhancement and attention mechanisms was proposed considering spa-tial dependencies,temporal dependencies,and weather impact of travel time.First,an attribute enhancement unit was constructed to integrate travel time and weather information.Subsequently,spatial dependencies were captured using a graph convolutional network,and temporal dependencies were captured using gate-recurrent units.An attention mechanism was employed to enhance the model's learning of features.Finally,the model was utilized to predict future travel times at 15,30,45,and 60 min intervals on a real dataset.The results show that the root mean square error(RMSE)of the prediction results are 0.045 3,0.045 6,0.045 7,and 0.046 8,respectively,which are better than other models.When temporal,spatial,and weather factors are considered,a reduction of about 10.3％in prediction error is observed compared with scenarios without weather consideration.Similarly,a reduction of about 24.2％in prediction error is observed compared with scenarios without accounting for spatial dependency.It shows that the model can better describe the spatiotem-poral dependence and the influence of external conditions.

transportation engineeringtravel time predictiongraph convolutional networksspatiotemporal dependenceweather factors

王忠宇、李盼归、杨航、吴兵

展开 >

上海海事大学交通运输学院,上海 201306

宁波大学海运学院,宁波 315211

同济大学道路与交通工程教育部重点实验室,上海 201804

交通工程 行程时间预测 图卷积网络 时空依赖 天气因素

国家自然科学基金资助项目国家自然科学基金资助项目上海市"科技创新行动计划"软科学研究领域重点资助项目宁波市国际科技合作资助项目

5217233152272334186921112002023H020

2024

10.3969/j.issn.1001-0505.2024.04.027

东南大学学报(自然科学版)
东南大学

东南大学学报(自然科学版)

CSTPCD北大核心
影响因子：0.989
ISSN：1001-0505
年,卷(期)：2024.54(4)
  • 12