章程

入学时间：2016级

答辩时间：2021年

论文题目：基于数据的城市快速路交通主动协同控制方法研究

中文摘要

摘要

为了破解日趋严重的快速路交通拥堵问题，快速路交通控制是重要手段，自20世纪60年代以来出现了许多理论探索及实践。但是，经历了半个多世纪的发展，现实运行中仍然存在控制失效的问题。这是由于快速路交通系统是个复杂巨系统，难以通过简单模型精准还原，而早先的控制方法通常将快速路交通系统过度简化，引入了过强的模型假设，使得控制方法难以及时准确地感知交通状态、难以直接建立系统内的复杂关联、并且由于复杂的模型推导难以快速求解。这些缺陷造成了理论与实际的脱离，导致控制方法的时效性与系统性不足。因此，控制方法亟需实现理论与现实的耦合，提升控制的科学性与实用性。

随着交通系统感知与交互能力的显著提升，精准、可靠、结构化的数据获取成为可能，可以用于精准感知系统实时状态，以及直接建立系统内的复杂关联。不仅如此，还可通过他组织与自组织的结合，实现系统最优控制。在此背景下，本研究提出了“基于数据的城市快速路交通主动协同控制方法”。其中，“主动”即“防患于未然”，指的是提前采取控制，预防拥堵发生；或是当局部拥堵出现时，尽快启动控制，及时消散拥堵。“协同”和“协调”不同，“协调”只是考虑了系统各部分之间的合作，而“协同”指的是系统各部分围绕共同目标，通过合作来实现，充分考虑系统内部的高度关联以及相互作用，旨在高效利用有限的道路资源，提升整体运行水平。“基于数据”则是试图利用数据的时效性、客观性、精准性、信息丰富、便于大规模计算及应用等诸多优点，在一定程度上替代繁琐的模型推导，放宽人为设置的模型假设，为控制问题提供新的解决思路，推动控制理论的进一步发展。要注意的是，现实中的数据条件虽然较以往丰富许多，但普遍存在多源异构、数据质量、隐私顾虑、数据共享等问题，难以直接用于控制。同时，许多数据的最初用途并非用于控制，如何充分利用这些数据，转化为适用于控制的精准、可靠、结构化的数据，也是需要解决的问题。

本文首先回顾了快速路交通控制的发展历程，从控制模型、优化目标、约束条件、数据需求等方面，系统梳理了现有的控制方法，主要存在以下两方面不足。一方面，状态变量的获取及时空关联的建立大量依赖模型推导及假设，与现实存在偏差，容易造成控制失效。另一方面，受限于当时的感知、交互与计算能力，一些学术思想难以实现，如大规模控制、实时控制、多约束多目标的优化控制等，无奈通过简化问题来实现求解，限制了控制方法的理论拓展及实际应用。本研究提出了一种新型的快速路主线路段及出入口匝道一体化实时控制理论方法，新的数据及技术环境从以下三个方面为控制理论提供支持：（1）基于实时更新的交通状态数据，提高控制的时效性；（2）基于行程时间与OD数据，研究快速路系统内的时空关联性，提高控制的系统性；（3）基于实时轨迹数据，对潜在拥堵进行精准预测与溯源，实现主动协同控制。

然后，基于主动协同控制的原理，研究确定了主动协同控制的数据需求及获取方法。其中，基本数据需求包括交通状态、交通需求以及道路基础信息，可通过环形线圈数据、浮动车GPS数据及路网地图来获取。为了研究快速路上下游的时空关联，行程时间及OD是重要的数据需求，可基于视频数据与手机导航数据，提高两者的估计精度。实时轨迹数据将进一步支撑主动协同控制，随着交通网联化、共享化、交通管控与服务一体化，这一数据需求有望在未来得到满足。

为了研究快速路上下游的时间关联性，为主动协同控制中的行程时间预测提供重要依据，本文基于广泛分布的视频数据，通过车辆重识别技术，采集两个视频监控点位之间的历史行程时间样本。研究提出了基于车道的车辆重识别方法，显著提升了车辆匹配的正确率，进而提高了行程时间估计精度。估计内容除了平均值还包括标准差与分布类型，不仅可得到基于路段的行程时间估计，还可得到基于车道的行程时间估计。提出的方法不依赖车牌信息，不涉及隐私问题，对于低清晰度视频以及拥堵场景同样适用。

为了研究快速路上下游的空间关联性，为主动协同控制提供精准可靠的数据输入，本文基于真实的历史轨迹数据，提取历史OD样本及路径选择比例，结合断面流量数据进行OD估计。研究分别基于线圈数据和轨迹数据，进行OD估计并用于入口匝道控制。实验结果表明，基于轨迹数据的方法能更有效地缓解甚至消除部分瓶颈拥堵。随后，通过简单随机抽样，逐步降低轨迹数据的渗透率，分析渗透率对控制效果的影响，由此确定轨迹数据的最小渗透率。

之后，本文提出了一种新型的快速路主线路段及出入口匝道一体化实时控制理论。研究假设所有车辆的轨迹数据可以实时获取，利用实时OD替代原先控制理论中的预设OD或主线驶出匝道的预设交通量比值，实现对潜在拥堵的精准预测与溯源。具体来说，在车辆进入主线前，精准预测其可能造成的主线拥堵及出口匝道溢流，据此提出主动协同的入口匝道控制方法。控制以总耗费时间最少为优化目标，综合考虑主线通行能力约束、主线行驶速度约束、入口匝道排队约束、出口匝道排队约束、入口匝道流量约束，并提出了问题无解时的约束松弛策略。实验对比了不同控制参数以及不同控制输入对控制效果的影响，由此对数据质量提出要求，为数据采集提供理论支撑。

本研究具有前沿的学术价值和实用价值。与目前最先进的控制方法相比，提出的方法能更及时地启动控制，能对距离拥堵路段更上游的入口匝道采取控制，能更有效地缓解甚至消除拥堵，体现出主动性及协同性的增强。提出的控制理论可为快速路网控制、车路协同控制以及大规模应用奠定基础。并且，本研究探索了控制方法的应用，基于现实数据条件，提高了行程时间与OD估计精度，为主动协同控制提供了更精准可靠的数据输入，在实时轨迹数据难以获取时，一定程度上可作为替代，从而提升了控制方法的实用价值。

关键词：快速路交通控制，主动控制，协同控制，入口匝道控制，轨迹数据

英文摘要

ABSTRACT

Expressway traffic control is an important measure to solve the increasingly serious problem of expressway traffic congestion. There have been many theoretical explorations and practices since the 1960s. Nevertheless, after more than half a century of development, the problem of control failure still exists in reality. The expressway traffic system is a complex giant system, so it cannot be accurately described by simple models. Previous control methods oversimplified the expressway traffic system and introduced too strong model assumptions, making it difficult to perceive traffic states timely and accurately, difficult to directly establish complex relationships within the traffic system, and difficult to quickly solve the control problem due to complex model derivation. These defects made the control theory separated from the reality, which limited the timeliness and systemicity of control methods. Therefore, it is urgent to integrate theory and reality and to improve the scientific and practical nature of control methods.

The significant improvement of traffic system perception and interaction capabilities enables accurate, reliable and structured data acquisition. These data can be utilized to accurately perceive real-time traffic states and directly establish the complex relationships within the traffic system. Moreover, system optimal control can be realized with the combination of hetero-organization and self-organization. In this context, this paper proposes “Data-based Proactive and Integrated Control of Urban Expressway Traffic”. “Proactive” means “taking precautions”, which refers to taking control in advance to prevent congestion from occurring or taking control to address congestion as soon as possible when it occurs. “Integrated” is different from “coordinated”. “Coordinated” only considers the cooperation between the various parts of the system, while "integrated" indicates that all parts of the system cooperate with each other to realize the same goal. The control methods should fully consider the high correlation and interaction within the system and improve efficiency with limited road resources. “Data-based” methods try to take advantages of data, such as timeliness, objectivity, accuracy, richness of information, feasibility for large-scale calculation and application, etc. To some extents, data can replace complex model derivation, relax model assumptions, provide new ideas of control, and promote further development of control theories. It should be noted that though data become much richer nowadays, they can seldom be directly utilized for traffic control due to the following problems, such as multi-source heterogeneity, poor data quality, privacy concerns, data sharing problems, etc. Also, many data are not originally collected for traffic control. How to make good use of these data? How to transform them into accurate, reliable, and structured data for traffic control? These are also the problems that needs to be solved.

This paper firstly reviews the development of expressway traffic control. Existing methods in literature are comprehensively summarized regarding to control models, optimization objectives, constraints, data requirements, etc. There are two aspects of limits in the existing methods. On one hand, model derivation and assumptions were heavily relied on to acquire traffic states and establish spatial-temporal relationships, which caused deviation from real-world traffic systems and led to control failures. On the other hand, limited by the perception, interaction and computing capabilities at that time, a lot of ideas of traffic control could not be realized, such as large-scale control, real-time control, multi-constraint and multi-objective optimization control, etc. Control problems had to be simplified for being solved, which limited the theoretical extension and practical application of control methods. This study proposes a new real-time and integrated control theory for expressway traffic on mainline sections, on-ramps and off-ramps. New data and new technologies provide support for control theory from the following three aspects: (1) based on real-time traffic state measurement, improve the timeliness of control methods; (2) based on travel time and origin-destination (OD) data, investigate the spatial-temporal relationships in the expressway traffic system and improve the systemicity of control methods; (3) based on real-time trajectory data, accurately predict and back-trace potential congestion for proactive and integrated control.

Then, based on the principle of proactive and integrated control, this study determines the data requirements and acquisition methods. Basic data requirements include traffic states, traffic demands, and road information. They can be respectively obtained by loop detector data, floating car GPS data, and road maps. To investigate the spatial-temporal relationships between the upstream and downstream traffic, travel time and OD are important data requirements. Their estimation accuracy can be enhanced by video data and mobile phone navigation data. Real-time trajectory data will further support proactive and integrated control. With the development of connected vehicles, shared mobility, and the integration of transportation management, control and services, the data requirement of real-time trajectory is expected to be met in the future.

In order to investigate the temporal relationships between the upstream and downstream traffic and provide an important basis for travel time prediction in proactive and integrated control, this paper utilizes widely distributed video data and vehicle re-identification technology to collect historical travel travel samples between two video surveillance points. A lane-based vehicle re-identification is proposed to significantly improve the correctness of vehicle matching and the accuracy of travel time estimation. The travel time estimation includes not only mean but also standard deviation and distribution type. Link-based and lane-based travel time estimations are obtained. The proposed method does not rely on license plate information, so it does not involve privacy issues. The proposed method can also be applied to low-resolution videos and congested scenarios.

In order to investigate the spatial relationships between the upstream and downstream traffic and provide accurate and reliable inputs for proactive and integrated control, this paper extracts historical OD samples from real trajectory data. OD estimation is obtained via the route choice probabilities from trajectory data and traffic counts from loop detector data. Two methods, loop detector data-based and trajectory data-based methods, are respectively implemented for OD estimation and ramp metering control. Experimental results show that the trajectory data-based method can alleviate congestion more effectively or even eliminate congestion. The penetration rate of the trajectory data is then reduced by simple random sampling, to analyze control effects by various penetration rates. The minimum penetration rate of the trajectory data is determined.

After that, this paper proposes a new real-time and integrated control theory for expressway traffic on mainline sections, on-ramps and off-ramps. The study assumes that trajectory data of all vehicles can be obtained in real time, and then makes use of real-time OD to replace the preset OD or the preset ratio of off-ramp traffic volume in the previous control theories. Accurate prediction of potential congestion and back-tracing are realized for proactive and integrated control. Specifically, before a vehicle enters the mainline, the proposed method accurately predicts mainline congestion and off-ramp queue spillback caused by this vehicle. Based on the accurate prediction, the optimal inflow is determined to minimize total time spent. The optimization problem considers constraints of mainline capacity, mainline speed, on-ramp inflow, on- and off-ramp queues. Constraint relaxation strategy is proposed when the optimization problem has no solution. Experiments compare the control effects by control parameters and by control inputs. The experimental results put forward requirements on data quality and provides theoretical support for data collection.

This study has cutting-edge academic value and practical value. Compared with the state-of-the-art methods, the proposed method can start the control earlier, control inflow at more upstream on-ramps, and thus alleviate or eliminate congestion more effectively. Experimental results demonstrate the improvement of the proactive and integrated control. The proposed control theory lays foundation for expressway network control, connected vehicle control, and large-scale applications. Moreover, this study explores the application of control methods. Based on typical data sources in the real world, travel time and OD estimations are improved to provide more accurate and reliable inputs for proactive and integrated control. When real-time trajectory data are not available, travel time and OD estimations can be utilized as alternative control inputs, and hence the practical value of the control method is improved.

Key Words:Expressway traffic control, proactive control, integrated control, ramp metering control, trajectory data

江泽浩

入学时间：2016级

答辩时间：2020年

论文题目：基于数据的城市道路信号控制交叉口健康诊断理论研究

中文摘要

摘要

我国城市道路交通设施的效能尚不足发达国家的60%，而城市交通系统的发展趋势已由“增量建设”转变为“存量提升”。城市“交通病”的日益严重和交通数据的飞速积累，使得“基于数据的城市道路健康诊断”具有了充分性与必要性。交叉口是城市道路网络中的瓶颈，其健康与否直接影响到城市路网的运行情况。因此，“基于数据的城市道路信号控制交叉口健康诊断”的提出，对于提升交叉口的安全、效率与秩序，在传统的“规划→设计→建设→管理→服务”单向链条增加反馈形成闭环，具有重要意义。

首先，本文在借鉴“临床健康诊断”、“桥梁健康诊断”的基础之上，提出了“交叉口健康诊断系统”的概念，并对系统功能和系统框架进行了设计。系统功能方面，本系统主要包含“诊断指标的选取”、“诊断标准的确定”、“健康诊断的实施”及“致病原因的分析”四大功能。系统框架方面，本系统主要包括“问题库”、“致因库”及“诊断库”三大部分。其中，“问题库”对应的是交叉口运行表现出的不良健康状况，即：安全病、效率病与秩序病；“致因库”表征的是交叉口的系统要素；“诊断库”的定位是寻求问题库与致因库之间的关系，即在交叉口健康状况与系统要素之间建立联系。

其次，针对“健康诊断阈值确定”，本文在传统的“确定论”方法的基础之上，提出了一种“概率论”的基于可靠性的方法论，并以信号相位切换为例对这种方法进行了研究。利用高精度驾驶模拟器对“感知-反应时间（PRT）”与“可接受的安全减速度（SAA）”进行参数标定，并基于驾驶员在绿灯闪烁和绿灯倒计时两种模式下的微观驾驶行为给出了信号相位切换的安全建议，最后计算了考虑安全可靠度的黄灯时长。本部分着重强调，交叉口健康诊断的阈值不应是一个确定值，而应该是由交叉口系统要素、交通参与者个体信息等情况共同确定的。

再次，本文对交叉口安全诊断进行了详细的论述。在简述我国道路交通安全形势的基础上，指出“基于事故”和“基于冲突”的安全诊断方法存在相关问题，并基于此提出一种“基于冲突”的交叉口安全诊断方法。通过在宿迁与银川两个城市设计的一组实证实验，证明了交通违法与交通事故之间存在显著的相关性，从而交通违法可以作为交通事故的替代指标以评价交叉口的安全性。在此基础上，构建了基于交通违法的交通安全指数，并基于此指数对交叉口安全性进行诊断。

再次，本文对交叉口运行效率也做了诊断。在总结传统的交叉口通行效率评价指标的基础之上，指出这些指标的基础仍然是通行能力。通过对传统通行能力概念的反思，指出仍然存在“对空间资源利用率考虑不足”、“较少考虑非机动车对通行能力的影响”及“没有反映出交叉口内部冲突约束”等的缺陷。针对上述问题，提出了一种考虑土地利用的交叉口效率排序方法，并重点探究了提升电动自行车在交叉口秩序与效率的方法，并以大型环岛为例讨论了考虑交叉口内部约束的通行能力计算方法。

最后，对本研究之不足及进一步的研究方向进行了简要的讨论。

综上所述，本文提出了“交叉口健康诊断系统”的概念，并设计了系统的功能与框架；在传统的“确定论”方法的基础之上，提出了一种“概率论”的基于可靠性的健康诊断阈值确定方法；在传统的“基于事故”与“基于冲突”的安全诊断方法的基础之上，提出了一种“基于违法”的交叉口安全诊断方法；在反思传统的通行能力计算方法的基础之上，提出了交叉口通行能力的计算与评价应考虑土地利用效率、交叉口内部要素的约束及应该重点考虑电动自行车的影响。

关键词：信号控制交叉口，健康诊断系统，可靠性理论，安全诊断，效率诊断

英文摘要

ABSTRACT

The efficiency of urban-road transportation facilities in China is still less than 60% of that in developed countries, and the development trend of urban transportation systems has changed from "incremental construction" to "stock upgrading." The increasing severity of urban "traffic diseases" and the rapid accumulation of traffic data makes "data-based urban road health diagnosis" sufficient and necessary. Intersections are bottlenecks of urban-road networks, and their health directly affects the operation of urban-road networks. Therefore, the "data-based urban road signalized intersection health diagnosis" is proposed to improve the safety, efficiency and order of intersections by adding feedback to the traditional "planning→design→construction→management→service" unidirectional chain to form a closed loop.

Firstly, based on the reference to "clinical health diagnosis" and "bridge health diagnosis", this paper puts forward the concept of "intersection health diagnosis system" and designs the system functions and system framework. In terms of system functions, this system mainly includes four functions: "selection of diagnostic indicators", "determination of diagnostic criteria", "implementation of health diagnosis" and "analysis of the causes of disease". In terms of system framework, the system mainly includes three parts: "problem database", "cause database" and "diagnostic database". Among them, the "problem database" corresponds to the poor health conditions manifested in the operation of the intersection: safety diseases, efficiency diseases and order diseases; the "cause library" represents the system elements of the intersection; the location of the "diagnostic library" is to seek the relationship between the problem database and the cause database is to establish a connection between the health status of the intersection and the system elements.

Secondly, for "health diagnosis threshold determination", based on the traditional "deterministic methodology", this paper proposes a "probabilistic" reliability-based methodology, and takes signal changing as an example to analysis. Use high-precision driving simulator to calibrate the parameters of "perception-reaction time (PRT)" and "safety accepted acceleration (SAA)", and give safety suggestions for signal change based on the driver's microscopic driving behavior in the two modes of green signal flashing and green signal countdown, and finally calculated the duration of yellow light (DYL) considering safety reliability. This section emphasizes that the threshold of intersection health diagnosis should not be a certain value, but should be jointly determined by the intersection system elements and individual demographic of traffic participants.

Thirdly, this paper discusses the safety diagnosis of intersections in detail. On the basis of a brief description of the urban-road traffic safety situation in China, it points out that there are related problems in the "accident-based" and "conflict-based" safety diagnosis methods, and proposes a "traffic-low-violation-based" method to do the safety diagnosis of intersections. Through a group of empirical experiments designed in Suqian City and Yinchuan City, it proves that there is a significant correlation between traffic-law violations and traffic accidents, so that traffic-law violations can be used as a substitute for traffic accidents to evaluate the safety of intersections. On this basis, a traffic safety index based on traffic-law violations was built, and the safety of intersections was diagnosed based on this index.

Forthly, this paper also makes a diagnosis of the intersection's operating efficiency. On the basis of summarizing the conventional evaluation indicators of intersection traffic efficiency, it is pointed out that the basis of these indicators is still capacity. Through the reflection on the concept of traditional capacity, it is pointed out that there are still defects such as "under-consideration of utilization of space resources", "less consideration of the impact of non-motor vehicles on capacity" and "not reflecting the internal conflict constraints of intersections". In view of the above problems, a method for sorting the efficiency of intersections considering land use was proposed, and the method of improving the order and efficiency of electric bicycles at intersections was focused on, and the capacity of considering the internal constraints of intersections was explored by taking a large roundabout as an example calculation method.

Finally, the shortcomings of this study and further research directions were briefly discussed.

In summary, this article proposes the concept of "intersection health diagnosis system" and designs the system's function and framework; on the basis of the traditional "deterministic methodology", a probabilistic health diagnosis threshold determination method is put forward; on the basis of the traditional "accident-based" and "conflict-based" safety diagnostic methods, an "traffic-law-violation-based" safety diagnosis method for intersections is proposed; On the basis of the traditional capacity calculation method of intersections, it is proposed that the calculation and evaluation of the intersection's capacity should consider the constraints of land use and internal factors of the intersection and should focus on the impact of electric bicycles.

Keywords:signalized intersection, health diagnosis system, reliability theory, safety diagnosis, efficicency diagnosis