朱怀中

入学时间：2013级

答辩时间：2019年

论文题目：基于数据的公交服务特征提取方法研究

中文摘要

摘要

公交服务（Bus Service）可以理解为：政府、公交企业以及相关单位组织车辆、路线、人力等要素，用以满足乘客出行需求的有偿/无偿活动，本研究试图从客观层面对其质量进行评价，确定主要的服务影响因素，从而对公交服务质量进行优劣分析，并提出指向明确、有针对性、效果显著的改进方案。基于以往的研究，公交服务质量评价可划分为可用性（基本功能）、舒适性（性能）以及便捷性（使用体验）三个评价维度，每个维度均有对应的“服务特征”，显然应具备高辨识度，应能够准确反映该评价维度的现象和规律。尽管越来越多的城市管理者及公交企业意识到公交服务质量的重要性，但鲜有城市提出完整科学的服务质量提升计划和体系，且服务改善流于表面，无法有效破解服务质量低下的顽疾。与此同时，国内外学者在公交服务质量及服务特征等方面也尚未形成系统的、定量的公交服务特征提取方法。

基于上述考虑，本研究围绕公交服务特征，以数据驱动的方式，系统探讨了公交服务特征的概念及影响因素，并以定量方式完整地提出了公交服务特征提取方法：以公交车站、公交线路、公交系统为分析层次，从可用性、舒适性、便捷性分析维度，基于各个公交服务特征的内涵，包括物理意义与实用意义，提出服务特征提取的思想、方法与算法，并开发了应用系统，且进行了实证。此外，本研究还探讨了公交服务特征改善措施，明晰未来服务质量提升的发展方向及具体思路

划分公交服务特征分析层次，解析公交服务特征提取维度，不仅能够解释乘客对公交服务的认知与强调乘客的实际感知，还能反映出行决策行为的内在机理。公交服务可用性的影响因素包括服务覆盖范围、运行时刻表、运输能力、服务信息；公交服务舒适性的影响因素包括载客量、外观和舒适度、出行时间、安全性、成本；公交服务便捷性的影响因素包括可靠性、出行时间、系统智能化、人性化水平。理解影响因素有利于建立对公交服务特征体系的深刻认识。公交服务特征体系精简了以往不少城市庞大的评价指标分类，突出乘客使用公交方式的切身感受，三个分析层次和提取维度化繁为简，既考虑到不同城市之间的发展差异，又在一定程度上解决各城市指标体系自成一派的问题。

公交车站是城市居民使用公交方式出行的基本载体，是检验公交服务质量的第一试金石，其可表征的服务特征包括车头时距、承载系数与站立乘客面积，以及服务半径。公交线路与城市发展相适应，线路设置应满足出行基本需求，其服务特征可以用服务时间、行车速度与行车加速度，以及准点率与车头时距变异系数表征。针对单条公交线路的服务特征提取是评价公交服务质量的关键，因其直接反映了公交在途的“综合实力”。公交系统是城市公交发展的“统一体”，其服务质量提升关系全局，高效运作的公交系统能够保障公交都市顺利建成，公交系统服务特征包括服务覆盖率、行程时间，以及换乘系数。本研究围绕公交车站、公交线路、公交系统的服务特征提出相应提取方法。

本研究在各提取方法基础上，开发了相应的应用系统，以实证方式验证了公交服务特征提取算法。通过将实时、大批量的数据整合到系统平台，可以深入解析公交基础数据、环境数据及其他异构数据，为公交服务质量评估与优化提供标准化数据支撑，同时也有助于对城市公交出行方式进行深入挖掘，以期提高公交系统的整体功效。

本研究的创新性在于率先系统性阐述了公交服务特征提取的完整概念及影响因素，填补了国内公交服务质量评价在这一领域研究的空白。基于移动互联网、大数据等新技术提供的全息数据新环境，提出了适用于新型数据和技术环境下的公交服务特征的提取方法，针对各公交服务特征提出的相应模型和算法具有较强的适应性和

可迁移性，能够推动公交服务质量的快速提升。当然，本研究还存在一些值得思考的地方，例如使用的共享单车数据仅涉及少数运营主体，在图像、语音等人工智能数据利用方面尚显不足。不过，随着智能公共交通快速发展，建设与移动互联网深度融合的智能公交系统恰逢其时，这也为本研究未来的深入探索指明了方向。

基于现代科技手段的公交服务特征提取，能够保证数据的科学、准确和有效。运用最新计算方法、手段和模型，公交企业和政府管理部门可以实时、高效地对公交服务质量进行评估与分析，实现公交服务质量的不断提升。本研究能够为国内公交服务质量评价和公交服务特征提取提供重要参考依据，相关研究方法和结论对城市公交系统理论与技术具有关键指导意义。

关键词：公交，服务质量，服务特征，数据驱动，服务特征提取

英文摘要

ABSTRACT

俞春辉

入学时间：2013级

答辩时间：2018年

论文题目：基于数据的城市道路交通网络主动控制基础问题研究

中文摘要

摘要

城市道路交通控制理论的发展经历了从单交叉口控制到多交叉口协调控制，从固定配时控制到自适应控制，从离线优化控制到在线优化控制。广泛采用的城市道路交通网络控制系统包括集中式控制系统（如TRANSYT、SCOOT）和分层式控制系统（如SCATS、UTOPIA、RHODES、MOTION）。这些控制系统的控制原理基本可归纳为，基于集计交通流数据，即历史统计数据或固定检测器数据，从机理解析的角度出发，简化实际问题，基于当前时刻或预测未来时刻的交通需求，建立数学解析模型，以最小化车辆延误、停车次数等为优化目标，采用最优化理论优化相位相序、信号周期、绿信比、相位差等交通控制参数。同时，受现有技术制约，交通控制的主要手段为交通信号灯控制，即基于集计交通流的控制模式。由于交通系统的高度复杂性以及出行车辆的个体差异性，基于集计交通流数据的交通控制方法已在日益严重的交通拥堵中显现其局限性。

无线通信技术和自动驾驶车技术的发展促进了网联车（Connected Vehicle（CV））和网联自动驾驶车（Connected and Automated Vehicle（CAV））在交通控制领域的应用，为交通控制提供了新的数据源和新的控制模式。相比于传统的固定检测器数据，CV和CAV数据包含更为详细的车辆轨迹信息，例如实时车辆位置和车辆速度。固定检测器的数据是断面的、集计的，仅在时间维度上是连续的。而CV和CAV的轨迹数据是非集计的，在时间维度和空间维度上均具有连续性。因此，更丰富的数据源为进一步提高交通网络的控制效果提供了可能。进一步，由于CAV具有较强的可控性，使得车辆轨迹的精准控制成为可能。交通控制模式也将从单一的时间维度转变为时间空间两个维度、从针对交通流的集计层面转变为针对车辆个体的非集计层面。本文基于CAV的轨迹数据和轨迹控制技术，提出了基于CAV的交通网络主动控制理论和方法。

首先，针对八相位单交叉口，建立统一框架下的交通信号配时和车辆轨迹的集成优化模型。以车辆总延误最小为目标，优化信号配时的周期长度、相序、绿信比以及车辆驶离交叉口的时刻。根据车辆的最优驶离时刻，以油耗/排放最小为目标建立最优控制模型，规划车辆的最优轨迹，即车辆在交叉口进口道的跟车行为和换道行为，保证车辆以期望速度不停车通过交叉口。由于交通流量的不确定性，利用滚动规划时间窗实现交通信号配时和车辆轨迹的动态优化。通过算例证明，在不同交通需求下，集成优化模型较车辆感应控制均具有更显著的控制效果。

其次，针对“signal free”单交叉口，以排队论为基础，从理论上分析基于规则的CAV控制策略的控制效果。以解析形式分析“first come, first service”（FCFS）策略和BATCH策略下的交叉口通行能力和车辆平均延误，并与传统的固定信号配时做比较。在理论上证明：在低流量下，FCFS控制策略较固定信号配时具有更好的控制效果；在接近饱和的高流量下，FCFS控制策略的平均车辆延误更大；从交叉口通行能力的角度看，固定信号配时和BATCH策略是等价的。同时，本文提出基于优化的CAV控制模型，优化车辆在交叉口的放行次序，基于车辆个体轨迹保证车辆在交叉口内部的安全性，提高交叉口通行能力，降低车辆延误。并通过算例验证理论分析的结论，以及证明基于优化的CAV控制模型的控制效果。

再次，将基于CAV的“signal free”交叉口控制模型扩展到交通网络层面。在车辆路径确定的条件下，以系统总延误最小为目标，基于离散时间构建混合整数线性规划（MILP）模型，协调优化车辆轨迹。微观上，模型考虑了车辆轨迹在路段和交叉口内部的交互，包括跟车行为、换道行为和碰撞避免。模型可应用于所有转向车辆（左转、直行和右转），并且交叉口进口道不需要划分车道功能。只要交叉口几何条件允许，任何转向车辆可利用任何一条进口道。应用滚动规划时间窗动态优化车辆轨迹。为平衡模型可行性和计算负荷，设计了自适应算法动态调整规划时间窗长度。

最后，以确定路径下的车辆轨迹优化模型为基础，建立基于CAV控制下的交通网络均衡模型，在同一框架下将宏观的车辆路径选择和微观的车辆轨迹规划相结合。在系统最优原则下，以系统总延误最小为优化目标建立双层规模型。在用户最优原则下，基于利己原则建立车辆个体的路径选择和轨迹优化模型，并提出基于CAV控制下用户均衡条件。针对系统最优模型和用户最优模型，分别设计遗传算法和启发式迭代算法进行求解。

本文从CAV技术的发展和交通控制系统发展的趋势出发，提出基于CAV的城市道路交通网络主动控制机制，从单交叉口层面到交通网络层面提出了CAV控制理论和方法，拓展了交通控制的内涵和外延，对后续相关研究有重要的借鉴意义。

关键词：交通控制，主动控制，网联自动驾驶车，轨迹数据，轨迹控制

英文摘要

ABSTRACT

The theory of urban traffic network control has developed from isolated intersection control to the coordinated control of multiple intersections, from fixed-time control to adaptive control, and from offline optimization to online optimization. There are two categories of widely applied control systems for urban traffic networks: centralized control systems (e.g., TRANSYT, SCOOT) and hierarchical control systems (e.g., SCATS, UTOPIA, RHODES, MOTION). These control systems are based on aggregated data,i.e., historical statistics or infrastructure-based detector data. Control mechanisms are analyzed. Problems are simplified. On the basis of the current or predicted traffic flow conditions, analytical models are built to minimize vehicle delay, stops, etc. Optimization theories are applied to optimize signal parameters including phase sequences, cycle lengths, splits and offsets. However, constrained by existing technologies, the main approach to traffic control is traffic lights, which is based on aggregated traffic flows. Due to the high complexity of traffic systems and the difference between vehicle individuals, the conventional traffic control methods based on aggregated traffic date are facing challenges with increasing traffic demand and congestion.

Wireless communication technologies and automated driving technologies have advanced the application of connected vehicles (CVs) and connected and automated vehicles (CAVs) in traffic control, which provides a new data source and a new control approach. Compared with conventional infrastructure-based detector data, CV and CAV data have more detailed information of vehicle trajectories such as real-time vehicle positions and speeds. Infrastructure-based detector data are aggregated. They are collected at one single point, and thus they are continuous only in the temporal dimension. However, CV and CAV trajectory data are disaggregated and are continuous in the both temporal and spatial dimensions. Therefore, the richer data source provides the potential to improve the efficiency of traffic network operation. Further, the controllability of CAVs makes it possible to accurately control vehicle trajectories. The traffic control approach is expected to develop from the single temporal dimension to the spatiotemporal dimensions, and from aggregated traffic flows to disaggregated vehicle individuals. This study proposes the theory and methodologies of traffic network control based on CAV trajectory data and CAV trajectory control technologies.

Firstly, an integrated optimization model of traffic signal timings and vehicle trajectories is presented for a typical eight-phase intersection. Signal parameters, i.e., cycle lengths, phase sequences, and splits are optimized together with vehicle departure times at stop bars in a unified framework. On the basis of optimal departure times, an optimal control model is built to optimize vehicle trajectories, i.e., car-following and lane-changing behaviors in intersection approaches, with the objective of minimizing fuel consumption/emission. Vehicles are guaranteed to pass through intersections at desired speeds without stops at stop bars. Due to the uncertainty of traffic flows, a rolling planning horizon is used to dynamically optimize signal timings and vehicle trajectories. Numerical studies show that the integrated optimization model significantly outperforms vehicle-actuated control at different levels of traffic demand.

Secondly, theoretical analyses of rule-based CAV management at“signal free”intersections are conducted on the basis of queueing theory. Intersection capacity and average vehicle delay are derived analytically under“first come, first service”(FCFS) strategy and BATCH strategy. Conventional fixed-time control is used as the benchmark. It is theoretically proved that: 1) FCFS strategy outperforms fixed-time control under low demand; 2) FCFS strategy has larger average vehicle delay than fixed-time control under high demand close to the saturated one; and 3) BATCH strategy and fixed time control are equivalent in terms of intersection capacity. This study presents an optimization model for CAV management at“signal free”intersections. The service sequence of vehicles is optimized. Vehicle safety within intersection areas are guaranteed based on individual vehicle trajectories. Intersection capacity is raised and vehicle delay is reduced. Numerical studies validate the conclusions given by the theoretical analyses as well as the advantages of the proposed optimization-based CAV management model.

Thirdly, the CAV management model at“signal free”intersections is extended to an urban road network. With the assumption of determined vehicle paths, a mixed integer linear programming (MILP) model is presented to cooperatively optimize vehicle trajectories based on discrete time. In a microscopic view, the proposed model takes into consideration the interaction between vehicle trajectories on roads and within intersection areas, including car-following behaviors, lane-changing behaviors, and collision avoidance. The model considers the vehicles with all movements (i.e., left-turning, through, and right-turning). No lane markings are needed. Each vehicle can use any approach lane if the intersection geometry permits. A rolling planning time window is applied for dynamical optimization of vehicle trajectories. And an adaptive algorithm is designed to dynamically adjust the length of the planning time window for the trade-off between model feasibility and computational burden.

Finally, CAV-based equilibrium models over a traffic network are presented based on the vehicle trajectory optimization model with determined paths. Macroscopic routing and microscopic trajectory planning are combined in one framework. In the system optimal principle, a bi-level programming model is built to optimize vehicle routing and trajectories for minimum vehicle delay. In the user optimal principle, a model of routing and trajectory planning is proposed for individual vehicles assuming that vehicles are selfish. And the CAV-based user equilibrium conditions are presented accordingly. Genetic algorithms and heuristic iterative algorithms are designed to solve the system and the user optimal models.

This study looks into the development of CAV technologies and traffic control systems and then proposes the proactive control mechanism of urban traffic networks based on CAVs. CAV management from the single intersection level to the traffic network level is investigated. This study enriches the connotations and extensions of traffic control, which lays the foundation for follow-up studies.

Key words:traffic control, proactive control, connected and automated vehicle, trajectory data, trajectory control