城市交通路网动态短时推理与精准预测研究  

Short-term traffic state reasoning and precise prediction in urban networks

作　　者：覃缘琪 季青原 葛俊 戴星原 陈圆圆[3] 王晓[3,4] QIN Yuanqi;JI Qingyuan;GE Jun;DAI Xingyuan;CHEN Yuanyuan;WANG Xiao(Zhejiang Lab,Hangzhou 310000,China;Enjoyor Technology Co.,Ltd.,Hangzhou 310030,China;The State Key Laboratory for Management and Control of Complex Systems,Institute of Automation,Chinese Academy of Sciences,Beijing 100190,China;Qingdao Academy of Intelligent Industries,Qingdao 266000,China)

机构地区：[1]之江实验室,浙江杭州310000 [2]银江技术股份有限公司,浙江杭州310030 [3]中国科学院自动化研究所复杂系统管理与控制国家重点实验室,北京100190 [4]青岛智能产业技术研究院,山东青岛266000

出　　处：《智能科学与技术学报》2022年第3期380-395,共16页Chinese Journal of Intelligent Science and Technology

基　　金：国家重点研发计划基金资助项目(No.SQ2019YFE012476)。

摘　　要：城市路网结构对于交通拥堵的形成及其在时空上的传播过程具有重要的影响。然而,在基于传统交通模型或深度学习模型的研究中,交通模式的生成与传播往往并未考虑路网特征,且只能通过交通特征指标进行间接刻画,使得难以动态描述其在时间及空间维度上的特征,导致交通模式传播预测精准度不高且缺乏针对性。为了解决上述问题,提出了一种新型的基于交通模式推理的交通预测框架TP2。该框架将拥堵传播模式建模为一个随时间变化的时序知识图谱,并使用一种包含了全新聚合函数RGraAN的推理框架进行时序推理,以捕捉交通拥堵的动态时变传播模式,将路段和与其存在交通模式关联的路段进行组合,并构建时空关联路网子区域,然后基于图神经网络的交通短时预测模块充分挖掘子区内的交通流时空相关性,并预测子区内各个路段的未来速度变化情况。与现有方法相比,TP2预测精度相比A3T-GCN模型有1%~2%的提升。The structure of urban traffic network has a significant impact on the formation and spatio-temporal pattern propagation of traffic congestions.However,in studies based on traditional traffic models or deep learning models,the generation of traffic mode can only be described indirectly by traffic indicators,without considering the traffic network feature.This makes it very difficult to accurately describe the propagation dynamics both in temporal and spatial dimen-sions and lacks specificity.To tackle the above-mentioned problems,a novel traffic state prediction approach based on traffic pattern reasoning(TP2)framework was proposed.The framework modeled congestion propagation as a dynami-cally evolving temporal knowledge graph(TKG),and applied an inferencing framework(TPP-TKG)that was based on a novel aggregator called RGraAN.TPP-TKG captured the spatial-temporal propagation pattern of traffic congestion,and combined related road links to a given link,and constructed correlated sub region of the traffic network.Then a traffic state predicting based on graph neural network was employed to predict short-term speed evolution of road links in this sub region.Comparing to the state-of-the-art benchmark models,TP2 achieves 1%~2%higher accuracy.

关 键 词：路网结构 传播模式 交通状态推理与预测 时序知识图谱 图神经网络 

分 类 号：U495[交通运输工程—交通运输规划与管理]