城市排水管网流量预测多视图时空图神经网络模型  

Multi-view spatio-temporal graph convolutional networks model for urban drainage networks flow prediction

作　　者：涂伟[1,2,3] 池向沅 赵天鸿 杨剑 朱世平 陈德莉 TU Wei;CHI Xiangyuan;ZHAO Tianhong;YANG Jian;ZHU Shiping;CHEN Deli(Guangdong Key Laboratory of Urban Informatics,Shenzhen University,Shenzhen 518060,China;Key Laboratory for Geo-Environmental Monitoring of Great Bay Area,MNR,Shenzhen University,Shenzhen 518060,China;School of Architecture and Urban Planning,Shenzhen University,Shenzhen 518060,China;College of Big Data and Internet,Shenzhen Technology University,Shenzhen 518057,China;School of Geospatial Information,Information Engineering University,Zhengzhou 450052,China;Wuxi Hangzheng Science&Technology Co.,Ltd.,Wuxi 214135,China)

机构地区：[1]深圳大学广东省城市空间信息工程重点实验室,广东深圳518060 [2]深圳大学自然资源部大湾区地理环境监测重点实验室,广东深圳518060 [3]深圳大学建筑与城市规划学院,广东深圳518060 [4]深圳技术大学大数据与互联网学院,广东深圳518057 [5]信息工程大学地理空间信息学院,河南郑州450052 [6]无锡航征科技有限公司,江苏无锡214135

出　　处：《测绘学报》2025年第2期334-344,共11页Acta Geodaetica et Cartographica Sinica

基　　金：广东省教育厅创新团队项目(2024KCXTD013);深圳市基础研究重点项目(JCYJ20220818100200001);国家自然科学基金(42401553);KartoBit Research Network(KRN2202GK)。

摘　　要：城市排水管网的流量是其运行效率和安全的关键指标,准确的流量预测对排水管网运行风险预警、优化其运行效率及规划布局至关重要。水流量不仅受到其自身动力学特性的影响,还与管网的空间结构紧密相关,但传统水流量预测方法较少关注水流在管道之间复杂多维的空间依赖关系。针对这一问题,本文提出了一种基于多视图的时空图网络模型,该模型综合考虑了排水管网的空间邻近性和节点间的属性相似性。分别构建最近邻拓扑视图与流量相似性属性视图,使用时空图卷积网络挖掘流量特征的内在时空依赖,利用注意力机制对多个视图的时空依赖特征进行融合以获得流量预测值。利用某市排水管网历史水流监测数据进行试验,结果表明本文提出的多视图时空图神经网络模型取得了较好的预测性能,多视图对比试验验证了不同视图在模型中起到的贡献。Flow of urban drainage networks are critical indicators of their operational efficiency and safety.Accurately forecasting these parameters is crucial for risk mitigation,performance enhancement,and layout planning of the networks.Traditional flow forecasting methods typically overlook the complex multidimensional spatial dependencies between the flows in pipelines.This paper proposes a multi-view spatio-temporal graph convolutional networks model that considers both the spatial proximity and attribute similarity of network nodes.It constructs the nearest neighbor-based graph and the flow similarity-based graph,utilizes spatio-temporal graph convolutional networks to uncover intrinsic dependencies,and applies an attention mechanism to merge features from multiple views for enhanced flow predictions.Experiments with historical flow data from an urban drainage network confirm the superior predictive capabilities of our model,with ablation studies validating the contributions of different views.

关 键 词：管网流量预测 多视图 时空图网络 图深度学习 

分 类 号：P208[天文地球—地图制图学与地理信息工程]