A stochastic model of traffic flow: Theoretical foundations

In a variety of applications of traffic flow, including traffic simulation, real-time estimation and prediction, one requires a probabilistic model of traffic flow. The usual approach to constructing such models involves the addition of random noise terms to deterministic equations, which could lead to negative traffic densities and mean dynamics that are inconsistent with the original deterministic dynamics. This paper offers a new stochastic model of traffic flow that addresses these issues. The source of randomness in the proposed model is the uncertainty inherent in driver gap choice, which is represented by random state dependent vehicle time headways. A wide range of time headway distributions is allowed. From the random time headways, counting processes are defined, which represent cumulative flows across cell boundaries in a discrete space and continuous time conservation framework. We show that our construction implicitly ensures non-negativity of traffic densities and that the fluid limit of the stochastic model is consistent with cell transmission model (CTM) based deterministic dynamics.     

Heteroscedastic Gaussian processes for uncertainty modeling in large-scale crowdsourced traffic data

Accurately modeling traffic speeds is a fundamental part of efficient intelligent transportation systems. Nowadays, with the widespread deployment of GPS-enabled devices, it has become possible to crowdsource the collection of speed information to road users (e.g. through mobile applications or dedicated in-vehicle devices). Despite its rather wide spatial coverage, crowdsourced speed data also brings very important challenges, such as the highly variable measurement noise in the data due to a variety of driving behaviors and sample sizes. When not properly accounted for, this noise can severely compromise any application that relies on accurate traffic data. In this article, we propose the use of heteroscedastic Gaussian processes (HGP) to model the time-varying uncertainty in large-scale crowdsourced traffic data. Furthermore, we develop a HGP conditioned on sample size and traffic regime (SSRC-HGP), which makes use of sample size information (probe vehicles per minute) as well as previous observed speeds, in order to more accurately model the uncertainty in observed speeds. Using 6 months of crowdsourced traffic data from Copenhagen, we empirically show that the proposed heteroscedastic models produce significantly better predictive distributions when compared to current state-of-the-art methods for both speed imputation and short-term forecasting tasks.     

经验过程的一般加权大数定律

研究了经验过程中加权系数具有某些较弱性质的加权和收敛问题.利用经验过程中已有的几个概率不等式与一般加权和的对称化不等式,得到了经验过程中的独立同分布随机元序列的这类加权和的强、弱大数定律成立的充分条件(E‖f(X0)‖1/αG<∞).同时,对经验过程中的Cesaro大数定律和欧拉弱大数定律进行了推广.     

Assessing the impact of reduced visibility on traffic crash risk using microscopic data and surrogate safety measures

Due to the difficulty of obtaining accurate real-time visibility and vehicle based traffic data at the same time, there are only few research studies that addressed the impact of reduced visibility on traffic crash risk. This research was conducted based on a new visibility detection system by mounting visibility sensor arrays combined with adaptive learning modules to provide more accurate visibility detections. The vehicle-based detector, Wavetronix SmartSensor HD, was installed at the same place to collect traffic data. Reduced visibility due to fog were selected and analyzed by comparing them with clear cases to identify the differences based on several surrogate measures of safety under different visibility classes. Moreover, vehicles were divided into different types and the vehicles in different lanes were compared in order to identify whether the impact of reduced visibility due to fog on traffic crash risk varies depending on vehicle types and lanes. Log-Inverse Gaussian regression modeling was then applied to explore the relationship between time to collision and visibility together with other traffic parameters. Based on the accurate visibility and traffic data collected by the new visibility and traffic detection system, it was concluded that reduced visibility would significantly increase the traffic crash risk especially rear-end crashes and the impact on crash risk was different for different vehicle types and for different lanes. The results would be helpful to understand the change in traffic crash risk and crash contributing factors under fog conditions. We suggest implementing the algorithms in real-time and augmenting it with ITS measures such as VSL and DMS to reduce crash risk.     

Short-term Hourly Traffic Forecasts using Hong Kong Annual Traffic Census

The need for acquiring the current-year traffic data is a problem for transport planners since such data may not be available for on-going transport studies. A method is proposed in this paper to predict hourly traffic flows up to and into the near future, using historical data collected from the Hong Kong Annual Traffic Census (ATC). Two parametric and two non-parametric models have been employed and evaluated in this study. The results show that the non-parametric models (Non-Parametric Regression (NPR) and Gaussian Maximum Likelihood (GML)) were more promising for predicting hourly traffic flows at the selected ATC station. Further analysis encompassing 87 ATC stations revealed that the NPR is likely to react to unexpected changes more effectively than the GML method, while the GML model performs better under steady traffic flows. Taking into consideration the dynamic nature of the common traffic patterns in Hong Kong and the advantages/disadvantages of the various models, the NPR model is recommended for predicting the hourly traffic flows in that region.     

基于深度强化学习的车辆跟驰控制

针对自适应巡航控制系统在控制主车跟驰行驶中受前车运动状态的不确定性影响问题，在分析车辆运动特点的基础上，提出一种能够考虑前车运动随机性的跟驰控制策略。搭建驾驶人实车驾驶数据采集平台，招募驾驶人进行实车跟驰道路试验，建立驾驶人真实驾驶数据库。假设车辆未来时刻的加速度决策主要受前方目标车辆运动影响，建立基于双前车跟驰结构的主车纵向控制架构。将驾驶数据库中的驾驶数据分别视作前车和前前车运动变化历程，利用高斯过程算法建立了前车纵向加速度变化随机过程模型，实现对前方目标车运动状态分布的概率性建模。将车辆跟驰问题构建为一定奖励函数下的马尔可夫决策过程，引入深度强化学习研究主车跟驰控制问题。利用近端策略优化算法建立车辆跟驰控制策略，通过与前车运动随机过程模型进行交互式迭代学习，得到具有运动不确定性跟驰环境下的主车纵向控制策略，实现对车辆纵向控制的最优决策。最后基于真实驾驶数据，对控制策略进行测试。研究结果表明：该策略建立了车辆纵向控制与主车和双前车状态之间的映射关系，在迭代学习过程中对前车运动的随机性进行考虑，跟驰控制中不需要对前车运动进行额外的概率预测，能够以较低的计算量实现主车稳定跟随前车行驶。     

考虑驾驶人风格的跟车预警规则研究

为研究驾驶人的跟车特性及探究可适用于不同风格驾驶人的跟车预警规则,为自动驾驶车辆开发可满足不同用户驾驶需求和驾乘体验的主动安全预警系统,选取50名被试驾驶人开展实车试验,采集驾驶人跟车行为表征参数并基于雷达数据确定跟车事件提取规则。选取平均跟车时距和平均制动时距为二维向量,使用基于K-means聚类结果的高斯混合模型将驾驶人聚类为3种风格类型(冒进型、平稳型、保守型)。通过分析3组驾驶人的跟车及制动数据,将不同类型驾驶人的制动时距分位数作为跟车预警阈值,结合实际预警数据及不同制动时距分位数对应的预警正确率,对现有跟车预警规则进行调整,以适应不同类型驾驶人的驾驶需求。研究结果表明:3组驾驶人的平均跟车时距和平均制动时距差异显著,冒进型驾驶人倾向于选择较小的跟车时距和制动时距,保守型驾驶人的跟车时距和制动时距则普遍较大;3组驾驶人的实际跟车预警次数为215次,驾驶人采取制动操作而系统未予以预警的次数为329次,系统整体预警正确率为21.9%,漏警率为87.5%,通过分析信息熵等判定当前预警规则并不合理;将每类驾驶人制动时距的10%分位数作为阈值时的预警效果较好,调整后的跟车预警规则能在一定程度上适应不同的驾驶人类型。     

基于高斯过程机器学习方法的隧道围岩分类模型

针对现有围岩分类方法的局限性,基于工程实例,利用分类性能优异的高斯过程机器学习模型建立围岩类别与其主要影响因素之间的非线性映射关系,进而提出一种基于高斯过程的隧道围岩分类模型,实现不同情况下围岩分类的合理识别.将该模型应用于川藏公路二郎山隧道围岩分类,研究结果表明,隧道围岩分类的高斯过程机器学习模型是科学可行的,与人工神经网络模型、支持向量机模型相比较,该模型具有参数自适应化的优点,能方便快捷地给出合理可靠且具有概率意义的围岩分类评价结果,可对围岩分类结果的不确定性或可信度进行定量化评价.     

数字高斯白噪声的频域和时域特性分析

分析了数字高斯白噪声在频域的频谱特性和在时间域的统计特性,提出在数字处理速度一定时,随着输出噪声频域带宽的增加,输出噪声的时间统计特性不再符合高斯分布.通过增加滤波器阶数可以提高噪声的时间统计特性,但是将增加输出噪声的带内波动.通过计算机仿真验证了上述的理论分析,并给出了频域和时域特性都满足相应要求的带宽范围实验值.     

基于核函数变换的NLPCC分析在洪水分类中的应用

通过拟线性化变换和降维去噪,得出多维分类指标的低维主成分值,然后通过对每一待分类样本的低维主成分值进行聚类分析,最终得出洪水的自然分类结果.洪水分类实例结果表明,该计算方法不需要复杂的计算机专业知识和优化算法知识,原理清楚,计算简单,结果客观有效,不失为一种洪水分类评价的新途径.选择适宜的非线性变换函数是正确应用该方法的关键,同时对于能够预先给出分类指标值且数值范围较小时,需要指标标准数值判定和聚类效果判定相结合.