【作者】 张燕；

【导师】 赵德； 王树盛；

【作者基本信息】 东南大学 ， 交通运输， 2025， 硕士

【摘要】 当前,我国道路交通供给的增长速度远比不上居民出行需求的增长速度,交通供需不平衡之间的矛盾日益突出。为有效解决交通拥堵问题,构建准确实时的城市路网流量监测体系至关重要,其可全面感知交通网络运行状态,为精细化管控提供数据支撑。然而,路网流量与交通检测器布设的数量、位置、采集的数据质量密切相关。基于此,本文从检测器最优布设、稀疏布设和密集布设三个场景出发,分别探讨适用于不同布设场景的城市路网流量推断方法,旨在为城市路网流量监测提供一套完整的解决方案。结合国内外研究进展,当前检测器布设方案未充分考虑多维路段属性和初始检测器、流量估计模型在低覆盖率时精度不足、流量修正模型计算复杂度高。因此,本文依托交通检测器采集的数据,融合三种检测器布设场景,提出了检测器布设方法、路网流量推算方法、路网流量估计方法和路网流量修正方法,并选取河北省雄安新区路网进行实验。本文研究工作包括以下几方面:(1)面向路网流量推算的检测器最优布设方法。针对检测器最优布设场景,推导并证明了路网流量推算理论,得到路网流量推算公式,检测器最优布设判断条件等。设计了多指标综合评分法进行路段重要性排序,考虑路段重要性和初始检测器,提出了基于关键路段的检测器布设模型和存在初始检测器的检测器布设模型。实验验证了考虑路段重要性的必要性,揭示了初始检测器比例对整体布设的影响。(2)基于检测器稀疏布设的路网流量估计模型。针对检测器稀疏布设场景,改进现有路网流量估计模型,提出了一种基于交通信息的全局图注意力网络模型,简称TIG-GAT模型。TIG-GAT模型基于交通基本图和模糊聚类,设计交通信息模块和全局信息模块,补充交通领域先验知识和路段交通流特性相关性,并结合图注意力网络和图嵌入,挖掘路段拓扑结构相关性,最后,融合路段动态运行特征和路段静态特征构建路段多视图特征,以提升模型流量估计性能。对比实验表明,TIG-GAT模型较经典估计模型具有显著优势;消融实验表明,路段多视图特征对模型提升均有效。(3)基于检测器密集布设的路网流量修正模型。针对检测器密集布设场景,结合检测器误差模型与路网流量推算理论,提出了一种基于计算图的路网流量修正模型,简称CG-TFCM模型。CG-TFCM模型将检测器系统误差比率和随机误差比率建模为可学习的参数,并利用路网流量推算的唯一性构建路网流量不一致损失函数,通过最小化损失函数实现路网观测流量修正。多场景实验验证了CG-TFCM模型对流量修正的有效性。(4)检测器布设专项分析模块开发与案例应用。基于本文研究成果,依托“交运之星—Tran Star”交通版软件,完成检测器布设专项分析模块的开发,对软件交通运行分析模块进行了扩展。通过案例分析验证了模块功能的有效性及其工程实践价值。更多还原

【Abstract】 Currently,the growth rate of road traffic supply in China significantly lags behind the increasing demand for resident travel,leading to an increasingly prominent imbalance between transportation supply and demand.To effectively address traffic congestion issues,establishing an accurate and real-time urban road network traffic monitoring system is crucial,as it enables comprehensive perception of traffic network operating status and provides data support for refined management and control.However,the road network traffic flow is closely related to the number,location,and data quality of traffic detectors.Based on this,this work explores urban road network traffic flow inference methods under three detector deployment scenarios,optimal,sparse,and dense,aiming to provide a complete solution for urban road network traffic monitoring.Drawing on domestic and international literature,current detector deployment schemes fail to adequately consider multi-dimensional road segment attributes.and initial detector configurations,traffic estimation models exhibit insufficient accuracy under low-coverage conditions,and traffic correction models suffer from high computational complexity.Therefore,based on the data collected by detectors,this work integrates multiple detector deployment scenarios,proposing detector deployment methods,road network traffic flow derivation methods,road network traffic flow estimation methods,and road network traffic flow correction methods.The road network in Xiong’an New Area,Hebei Province,is selected for experiments.The main work of this dissertation includes the following aspects:(1)Optimal detector deployment method for road network traffic flow derivation.For the optimal detector deployment scenario,it deduces and proves the theory of road network traffic flow derivation,and obtain the road network traffic flow derivation formula,the judgment conditions for optimal detector deployment,etc.a.Then,A multi-index comprehensive scoring method is designed to rank the importance of road.Considering the importance of roads and initial detectors,it offers a detector deployment model based on key roads and a detector deployment model with initial detectors.Experiments verify the necessity of considering multidimensional road attributes and reveal the impact of the proportion of initial detectors on the overall detector deployment.(2)Road network traffic flow estimation model based on sparse detector deployment.For the sparse detector deployment scenario,it improves current road network traffic flow estimation models,proposing a traffic information global graph attention network,abbreviated as TIG-GAT model.The TIG-GAT model is based on the traffic fundamental diagram and fuzzy clustering,designs the traffic information module and global information module,supplements traffic domain prior knowledge and the correlation of road traffic flow,and combines the graph attention network and graph embedding to mine the correlation of road topology structure.Finally,the road traffic operation features and road static features are fused to construct the multi-view features of roads to improve the traffic flow estimation performance.Comparative experiments show that the TIG-GAT model has significant advantages over the classical estimation models,and ablation experiments show that the multi-view features of road sections are effective for model improvement.(3)Road network traffic flow correction model based on dense detector deployment.For the dense detector deployment scenario,it combines the detector error model and the road network traffic flow inference theory,proposing a computational graph-based traffic flow correction model,abbreviated as CG-TFCM model.The CG-TFCM model takes the system error ratioμand random error ratioσof the detector system as learnable parameters,and constructs a road network traffic flow inconsistency loss function based on the uniqueness of road network traffic flow inference.The road network observed traffic flow is corrected by minimizing the loss function.Multi-scenario experiments verify the effectiveness of the CG-TFCM model for traffic flow correction.(4)Development of detector deployment special analysis module and case application.Based on the research results of this work,the"Transportation Star-Tran Star"transportation version software is used to complete the development of the detector deployment special analysis module,which expands the traffic operation analysis module of the software.Through case analysis,the effectiveness of the module function and its engineering practical value are verified.更多还原