朱晓波

入学时间：2012级

答辩时间：2018年

论文题目：基于数据的通行能力突变条件下城市道路交通状态演化分析方法研究

中文摘要

摘要

城市道路交通系统作为一个具有高度随机性、不确定性及动态性等特征的复杂系统，其交通状态十分容易受到突发交通事件影响。突发交通事件通常指导致道路通行能力临时降低的意外事件，包括交通事故、车辆抛锚、道路施工乃至恶劣天气、自然灾害等。当突发交通事件发生于城市道路某一位置时，交通流会在不同交通状态间转移。城市道路交通状态是交通流在一定的交通需求、供给、控制管理及信息服务等条件下所表现出的畅通、拥挤或阻塞的状态。交通状态演化过程和规律体现为路径级出行活动在路网中随时空变化的情况，包括突发交通事件导致交通拥堵的形成和消散、路径及路网交通流的变换与转移等。快速、科学分析突发事件发生后城市道路交通状态演化规律对城市路网级交通管控、应急管理决策、出行信息服务等具有重要意义。

本文以路段行程时间作为城市路网交通状态演化分析的核心参数。论文首先对突发交通事件、通行能力突变相关理论与基础问题展开研究，在此基础上，运用实验交通工程学的方法调查、统计突发交通事件发生处交通流数据，分析突发交通事件对道路交通流、道路通行能力的影响，并用解析方法对通行能力突变点处交通流演化规律和过程进行理论分析；在路网交通状态分析方面，基于路网实际交通需求数据，运用仿真实验建立突发事件场景，将仿真实验与理论解析相结合，提出基于交通等时线的方法以分析通行能力突变条件下路网层面交通状态演化过程；最后，从出行者路径变化的角度，分析通行能力突变对出行路径选择的影响，分别运用宏观基本图和交通等时线两种方法探索了通行能力突变条件下路网交通状态与交通分布、出行路径间的关系，并比较了两种方法的效果。

本文从突发交通事件定义、类型和特点出发，分析了不同突发交通事件对城市道路交通流和通行能力影响。在此基础上，划分了通行能力突变类型并归纳总结了各类型特征，建立相应突变场景，为突变条件下城市道路交通状态分析奠定基础。

在通行能力突变点处，以往研究调查统计了突发事件下交通流参数并与常态下交通流特征进行比较，但较少定量地确定突发事件对道路通行能力影响程度，多取近似值。本研究基于实测数据统计分析了受突发交通事件影响的、城市道路不同车道的交通流特征，并定量地分析了突发事件对不同车道通行能力影响程度，为同类研究提供一定借鉴。在此基础上，根据路段行程时间和出行者可接受的延误阈值关系，从理论上分析了通行能力突变点处排队长度、路段行程时间的演化过程和规律。

在路网层面，论文提出了基于交通等时线的通行能力突变条件下城市路网交通流演化分析方法。通过分析路段行程时间在路网上的空间分布特征，将以往路段级的交通流分析上升到路网级的宏观角度。在总结等时线定义、基本参数的基础上，定义了等时线密度、等时线角、平均交通扩散速度及平均空间扩散速度四个指标以定量地分析突发交通事件发生后路网交通流演化特征，并提出了分析流程。通过仿真实验对该方法的实用性和有效性进行检验。结果表明，基于交通等时线的方法定性与定量相结合地分析突变条件下路网交通流演化过程，等时线图直观地反映了突发交通事件对路网交通流的影响，可以快速、准确地判断事件发生方向与位置，指标则定量地分析交通流变化特征和程度、出行者绕行方向和趋势等。该方法为城市交通管控、应急管理等提供科学依据和决策支持，使路网交通管理控制措施更合理、更具有针对性。

本文进一步分析了通行能力突变条件下路网交通状态与交通分布、出行路径选择的关系，以探索突变条件下路网交通状态演化机理。基于以往研究，借鉴了突发事件下道路通行能力计算方法和路阻函数，并根据用户动态性，建立动态交通网络，运用遗传算法建立动态路径选择模型。在此基础上，通过仿真实验，分别运用宏观基本图和交通等时线探索了通行能力突变条件下路网交通状态与交通分布的关系，并对比了两种方法的效果和适用性。宏观基本图从路网整体角度分析了交通状态和交通分布之间的关联性，当路网中交通需求处于一定范围时，交通分布越均匀，交通状态越好；然而，此方法无法分析路网内部交通流空间分布情况以及突变条件下交通分布变化和交通流转移规律。基于交通等时线的方法可以定量地分析突变条件下每一时刻路网交通流空间分布、不同区域交通状态及区域间状态优劣对比，从而进一步分析出行路径变化、路网交通流转移趋势以及交通状态演化规律和过程。因此，基于交通等时线的分析方法更好地揭示了通行能力突变条件下路网交通状态演化规律和机理。

最后，对全文进行了总结，指出了论文的创新点，对有待于进一步研究的问题进行了展望。

关键词：通行能力突变，路网交通状态演化，等时线分析方法，行程时间

英文摘要

ABSTRACT

Since urban road traffic system is a complex system with high randomness, uncertainty and dynamics, network traffic flow is easily to be influenced by traffic emergency conditions under which segment capacity declines temporarily because of traffic events, including traffic accidents, vehicle breakdowns, road maintenance, bad weather and even natural disaster. After a traffic event happens on a road, traffic flow would evolve in different states. Urban network traffic flow states, including free flow state, congestion state and gridlock state, depend on many factors, such as traffic demand, supply, management, traffic control and information. The evolution of network traffic flow represents the space-time changing characteristics of path-level trips, including the formation and dissipation of congestion under traffic emergency conditions, network traffic states transfer, etc. It is significant for urban network management and control, emergency management and traffic information services to analyze the evolution of network traffic flow rapidly and accurately under traffic emergency conditions.

In this study, segment travel time is applied as the main parameter which is used to analyze the evolution of urban network traffic flow. First of all, some related theories and basic issues about traffic emergency conditions and segment capacity mutation are studies. On this basis traffic data under a traffic emergency condition are collected to analyze the effects of the traffic event on segment capacity and the evolution of network traffic flow based on experimental traffic theory. Then, mathematical analysis is applied to analyze the process of traffic flow evolution at the location where the traffic event happens. For the analysis of network traffic states under traffic emergency conditions, based on real traffic demand data, simulation experiments are applied. By combining simulation experiments and mathematical analysis, a traffic-isochrone map based method is proposed to analyze the transfer and changing characteristics of urban network traffic states under traffic emergency conditions. Finally, the effects of segment capacity mutation on network flow are analyzed from the travel path perspective. Macroscopic Fundamental Diagram and the proposed traffic-isochrone map are respectively applied to explore the relationships between network traffic states, traffic distributions and travel paths under traffic emergency condtions. The performance of these two methods are also analyzed.

The concepts of traffic events which cause traffic emergency conditions are illustrated in this study. The effects of different traffic events on segment capacity and network flow are analyzed theoretically based on previous studies. The types and characteristics of segment capacity mutation are concluded and the corresponding scenes of traffic emergency conditions are established, which is the foundation of analysis of network traffic flow evolution.

At the location where a traffic event happens, statistical data are used to analyze the characteristics of traffic flow under traffic emergency conditions and compare with normal conditions in previous studies, while few of them studied the effects of traffic events on segment capacity quantificationally. In this study, based on collected data, characteristics of traffic flow on different lanes are analyzed under traffic emergency conditions. Then, the change degrees of segment capacity for different lanes are analyzed quantificationally, which could provide reference to future studies. By comparing segment travel time with the threshold value of acceptable travel delay, the evolutions of queue length and travel time at the location where a traffic event happens are analyzed theoretically.

From the network-wide perspective, to analyze the evolution of network traffic flow under traffic emergency conditions, an approach based on traffic isochrones map which represents spatial distribution of travel time in an urban road network is proposed in this study. Based on introductions to definition and essential parameters of traffic isochrones map, four indexes are proposed to quantitatively analyze characteristics and evolution of network traffic flow under traffic emergency conditions and thereupon an analysis process is presented. Simulations are used to analyze the performance of the proposed approach. Results demonstrate that this traffic isochrones map based approach analyzes the evolution of network traffic flow under emergency conditions from both qualitative and quantitative aspects. The effects of traffic events on network flow are shown qualitatively and the locations of events are found quickly based on the changes in the traffic isochrones map, while the change degree and characteristics of network traffic flow are analyzed quantitatively using proposed travel time related indexes. In summary, the proposed approach provides a macroscopic view for the evolution of network traffic flow under traffic emergency conditions and is significant to emergency response and urban traffic management and control.

In this study, the relationship between network traffic states, traffic distributions and travel paths under traffic emergency conditions are further anazlyed to explore the

张楠

入学时间：2012级

答辩时间：2018年

论文题目：信号控制交通流再现的数值模拟方法

中文摘要

摘要

基于交通流观测数据，对动态交通流演化的复杂现象进行建模和分析是交通流研究的基础问题，对于交通预测，交通控制和交通管理具有理论意义和实用价值。但是，能够准确描述现实交通流的动态变化特征并不是容易解决的问题。首先，虽然观测方法不断地进步，现阶段能够获取多样性的观测数据，但是由于交通流的空间和时间分布特性，直接利用观测方法获取全面的交通流观测样本仍然是巨大挑战。另一方面，受到观测条件和数据样本的限制，传统的交通流模型难于应用于对现实交通流演化特征的分析。另外，不同于连续交通流情况，由于存在信号控制的外在因素，信号控制交通流的演变过程不仅受到交通流自身相互作用的影响同时受到信号控制对交通流的影响。所以，在相同的观测条件下，信号控制交通流的建模分析会比连续流情况更为复杂。所以，本研究选择信号控制条件下的交通流作为研究对象。

本研究提出以实验交通工程学为理论依据，应用实验方法应对以上交通流建模和分析遇到的关键问题和难点问题。本研究提出的实验方法包括：信号控制交通流的随机过程模型和结合观测数据的交通流随机模拟实验算法两部分。随后，利用模拟和实测数据对本研究提出的模型和算法进行了测试和实验验证。

类似于连续交通流的情况，考虑交通流内在作用机制的普遍性，将交通密度、交通流量和交通流波动速度作为信号控制交通流的特征变量来描述信号控制交通流本身随时间演变的动态特征。其次，由于信号控制因素的特殊性，通过划分信号控制交通流的状态模式来描述交通流在信号控制过程中的动态特征。充分考虑以上因素，本研究应用随机过程模型描述信号控制交通流动态过程中的不确定性和非线性性，该模型可以被看作是状态转换动态线性系统（Switching Linear Dynamic Systems，SLDS）。在该模型中，不仅考虑到交通流状态模式的随机动态过程，而且不同于连续交通流密度-流量的三角形关系模型，为了更好的描述现实交通流密度-流量关系的不确定性特征，本研究提出混合状态特征下的随机密度-流量关系模型（Stochastic Fundamental Diagram，SFD）。该SFD模型利用高斯分布作为实验分布近似描述现实信号控制交通流波动速度的不确定性。随后，从理论上论证了本研究提出的SFD模型可以近似的表示现实交通流的波动速度，更进一步应用现实数据对该结论进行了实证分析。

在交通流模型的应用过程中，需要考虑现实动态交通流分析时的数据观测条件制约。本研究选用信号控制路段上下游到达和消散的交通流流量作为观测数据。该数据是全时段的观测数据，但是受空间条件限制，只能作为局部交通流观测数据。另外对于信号控制交通流，该数据存在着“数据缺失问题”。应对以上问题，在该观测数据的条件下，本研究详细设计了信号控制交通流SLDS模型的贝叶斯框架。通过贝叶斯框架的学习能够解决局部观测和观测缺失的问题，利用观测数据推断出现实信号控制交通流的SFD模型，随时间变化的交通流特征参数和交通流状态模式序列。本研究采用马尔科夫-蒙特卡罗（Markov Chain Monte Carlo，MCMC）的随机数值模拟方法作为该贝叶斯框架的实验计算方法。

最后，选用NGSIM（Next Generation Simulation, NGSIM）数据和SLDS模型生成的模拟数据对信号控制交流的随机模型和贝叶斯框架的学习算法进行了测试和验证。测试结果表明：利用本研究提出的模型和相应的算法，将设计的信号控制路段上下游交通流观测数据作为训练的输入数据，计算算法可以收敛并且计算结果到达一定的精度要求。验证结果表明：利用本论文提出的模型和算法，选用信号控制路段上下游实际检测的到达和消散的交通流量观测数据作为训练的输入数据，可以推断出与现实交通流近似的实验交通流。在本论文中提出的模型和方法可以实现信号控制交通流的再现，作为信号控制，交通流预测以及交通管理的依据。

利用现实数据的验证结果表明，本研究提出的方法在一定的观测条件下可以实现现实交通流的“再现”。另外，本研究提出方法需要更进一步在多样性的数据环境中进行验证。在研究中，信号控制交通流的外部条件仅考虑最为简单的情况，如道路条件和信号控制条件，所以需要继续改进来提升现实信号控制交通流建模和分析的适用性。本研究是对实验交通工程学理论体系中的交通流的建模和实验方法的初步尝试，而交通流的建模和实验问题仍然需要进一步的研究和探索。

关键词：实验交通工程学，随机交通流模型，随机密度-流量关系模型，交通信号控制，贝叶斯学习，马尔科夫-蒙特卡罗模拟

英文摘要

ABSTRACT

Modeling and analyzing the complexity of many dynamical phenomena oftraffic flow with the observation data is available which is is also the basis of traffic prediction, traffic control and traffic management.However, describing the dynamic characteristics of traffic flow accurately by using traffic flow model quickly becomes intractable.Although the observation methods continue to progress resulting in diversity observation data, but due tothe spatial and temporal characteristics of the traffic flow, it is still a great challenge to obtain comprehensive traffic flow observation samples directly. On the other hand, the traditional traffic flow model is difficult to apply to the variation characteristics of real traffic flow due to the limitation of observation conditions and data samples.Especially for the signaled traffic flow, because of the external factors of signal control, the evolution process of traffic flow includes the informationnotonlyfromthe traffic flow itself interaction and alsofrom thefeedback of signal control. So on the same observation conditions, the modeling analysis of signaled traffic flow is more complex than continuous flow.

In this thesis, by started modeling a stochastic process of signaled traffic flow, the stochastic simulation experiments algorithm belonged to the Experimental TrafficEngineering (ETE) approach is designed to obtain the approximate or equivalent results to the real traffic flow by combined with the observation data.

Similar to the continuous flow, the traffic density, traffic flow rate and traffic flow wave speed is used as the characteristic parameters of signaled traffic flow to describe the dynamic characteristics of the traffic flow evolving with time. Particular to the signal control factors, the dynamic process of traffic flow is described by the state mode of traffic flow. Then, the stochastic process model is established, which can be regarded as theSwitching Linear Dynamic Systems(SLDS). In this model, not only the stochastic dynamic process of the traffic flow state is considered, but also the stochastic density flow relation model is defined to describe the uncertain nature of the obsvered density-flow data, named Stochastic Fundamental Diagram (SFD), where the Fundamental Diagram (FD) of continuous flow is the the triangle model. The main purpose of the SFD is to use the Gauss distribution of traffic flow wave speed to approximat the one in reality which is not analytically tractable to be modeled by a function.The experimental method of the approximate distribution of the signaled traffic flow wave speed is theoretically demonstrated and numerical verified by real data.

In application of traffic flow model, it must be suitable for the dynamic of the actual traffic flow on the limit condition of observation. The traffic flow rate at the upstream and downstream boundaries of the road is used as observation data, which is the most common data and relatively easy to collect. Based on the stochastic traffic flow model above, the complete Bayesian framework is constructed. By given the observation data as training data with consideration of the unobservable and miss-observable problem, the Bayesian framework can deduce SFD model, traffic flow characteristic parameters sequence and traffic flow state mode sequence changed in times. Markov chain Monte Carlo (MCMC) algorithm, a random numerical simulation method, is adopted for the Bayesian framework learning.

Finally, the Next Generation Simulation (NGSIM) data are used to test and verify the stochastic model of signaled traffic flow and the Bayesian framework learning algorithm.The test results show that the training algorithm can converge and the calculation results reach a certain precision requirement.The verification results show that with the training data, we could infer the experimental approximation results with the actual traffic flow of traffic flow. The model and method proposed in this thesis could realize the‘reappearance’of signaled traffic flow, where the results could be the basis of signal control, traffic flow prediction and traffic management.

In addition, the method proposed in this thesis needs to be further validated in a diverse data environment. In the thesis, it started with the basic external conditions of signaled traffic flow, such as road and signal control, so it is necessary to continue to consider a more complex situation to improve the applicability of the method. This thesis is a preliminary attempt of modeling and experimental methods of traffic flow in the theoretical system of Experimental Traffic Engineering, which still need further research and exploration.

KeyWords:Experimental TrafficEngineering, Stochastic Traffic Flow, Stochastic Fundamental Diagram, Traffic signal control, Bayesian Learning,Markov Chain Monte Carlo