【作者】 陈喜群；

【导师】 史其信； 李力；

【作者基本信息】 清华大学 ， 土木工程， 2012， 博士

【摘要】 随机交通流模型是智能交通系统、交通工程设计、交通管理与控制等领域的应用基础，对丰富现代交通流理论体系具有重要意义。道路交通流具有复杂性、动态性和随机性特征，新一代智能交通系统对交通流理论提出更高要求，静态的、确定的交通流理论已不能满足动态随机建模需求。本文应用多元异构数据，建立基于车辆轨迹信息的随机交通流模型，揭示交通流复杂动态性和随机演化性的内在机理，克服传统模型在确定函数上增加随机扰动项的不足。论文的主要研究内容和研究成果表现在以下方面：（1）数据挖掘：考虑道路交通流系统内部的复杂动态性和随机演化性，对交通流特性进行了定量研究和定性分析，为构建随机交通流模型奠定了基础。选取欧拉型（伯克利I80-W和北京G6-S高速公路感应线圈）和拉格朗日型（NGSIM车辆轨迹）实测数据，描述了拥堵车队扰动特性和交通流随机振荡时频特性，提出了车头时距/车头间距/瞬时速度的对数正态型经验概率分布模型。（2）微观关联：基于实测车头时距/瞬时速度、车头间距/瞬时速度联合概率分布，建立马尔可夫微观车辆跟驰模型，研究概率分布与微观模型内在关联，利用转移概率矩阵描述驾驶员随机驾驶行为，更真实地反映交通流动态演化特征。（3）宏观关联：将车头时距/车头间距/瞬时速度经验分布与流量/密度散布特性相结合，建立随机基本图模型，探讨概率分布与宏观模型内在关联，提出随机Newell条件，研究交通流半稳态区域的分布特性和概率边界计算方法。有助于指导交通系统设计、管理和控制。（4）匝道瓶颈建模：考虑道路通行能力随机性和时变性，建立基于车头时距/间距分布的交通流崩溃概率模型，研究连续流设施瓶颈区通行能力随机特性和交通流崩溃机理，分析瓶颈拥堵形成、传播、消散规律。有助于采取主动交通管理措施、形成优化控制策略预防常发性拥堵，提高道路交通可靠性。本文依托拉格朗日型和欧拉型检测数据，围绕车头时距/间距联合概率分布，提出一系列随机交通流模型，从微观和宏观两个层面揭示交通流随机特性内在影响因素和形成机理，可辅助构建城市主动交通管理系统。更多还原

【Abstract】 Stochastic traffic flow modelling is the application foundation of Intelligent Trans-portation Systems (ITS), traffic engineering, traffic management and control and so on.It is of a certain significance to enrich the modern traffic flow theory. Road traffic flow iswith complex, dynamic and stochastic characteristics. The new generation of ITS bringsforward higher requirements to the traditional static and deterministic theory, so that itno longer satisfies the demands of dynamic and stochastic modelling. This dissertationinvestigates multiple heterogeneous data to establish stochastic traffic flow models basedon vehicle trajectory information, reveals the underlying mechanism of complexity andstochastic evolutions, and overcomes the deficiency of traditional models that added ran-dom disturbance terms to determinant functions. The main contents and results are asfollows:(1) Data mining: we first analyze the characteristics of empirical traffic measure-ments to reveal the underlying mechanism of complexity, dynamic and stochastic evo-lutions quantitatively and qualitatively. This work lays the foundation to the stochastictraffic flow modelling. By using Eulerian measurements (e.g. inductive loop data offreeways I80-W in Berkeley and G6-S in Beijing) and Lagrangian measurements (e.g.vehicular trajectories of NGSIM dataset Highway101), we study shifted lognormal dis-tributions of headway/spacing/velocity, disturbances of congested platoons (jam queues)and time-frequency properties of traffic oscillations.(2) Microscopic connection: based on the joint probability distributions of head-way/velocity and spacing/velocity, we propose a Markov model for road traffic by incor-porating the connection between empirical distributions and microscopic car-followingmodels. Applying the transition probability matrix to describe random choices of drivers,the results show that the stochastic model more veritably reflects the dynamic evolutioncharacteristics of traffic flow.(3) Macroscopic connection: we propose a stochastic FD model through the con-nection between headway/spacing distributions and the macroscopic fundamental dia-gram (FD). We also study the stochastic Newell condition, wide scattering features andprobabilistic boundaries in flow-density plot, which compensate for the lack of analyticalderivation of the meta-stable2D region. (4) On-ramp/off-ramp bottleneck modeling: we propose a traffic flow breakdownprobability model based on headway/spacing distributions. It assumes the stochastic anddynamic road capacity. We study the capacity of highway on-ramp bottlenecks and themechanism of traffic breakdown phenomena, analyze the formation, propagation and dis-sipation of bottleneck congestions, and use the spatial-temporal queueing model to derivephase transition conditions of5traffic jam patterns. This work is beneficial to take mea-sures in active traffic management (ATM), obtain optimal control strategies to preventrecurrent congestions and improve reliability.This dissertation relies on Lagrangian and Eulerian measurements to study jointdistributions of headway/spacing/velocity, proposes a series of stochastic traffic flowmodels to reveal the potential impact factors and the formation mechanism of stochasticcharacteristics. This study assists in establishing urban ATM systems.更多还原