朱际宸

入学时间：2022级

答辩时间：2026年

论文题目：面向预约通行的城市道路交通协同控制方法

中文摘要

摘要

道路交通控制是通过交通信号方案调控，辅以交通流组织和通行条件优化，按特定目标最佳配置通行权和供需关系，是改善交通阻塞、提升城市道路综合效能的重要手段之一。然而，由于个体出行具有高度不确定性，传统交通控制方法尚未能精细化协调多模式异质交通群体。在智能网联技术不断发展的背景下，城市道路交通亟须面向精细个体行为和主动需求演化的新一代交通控制系统。与此同时，预约通行正作为一种新兴的需求管理手段被逐步引入城市道路交通系统。通过在出行前对车辆的时刻表与路径进行规划，预约通行能够将原本无序随机的即时出行需求转化为具有明确时空约束的计划性需求。广义的预约通行已在多类典型场景中得到应用与探索，例如按照固定时刻表运行的城市公交系统、具有强时效约束的应急救援车辆，以及提供出行计划的出行预定服务等。

预约通行与交通控制之间天然存在紧密关联。一方面，预约通行能够削弱自发出行需求的随机波动性，从而使控制系统具备前瞻性和主动性；另一方面，预约通行车辆需要通过交通控制系统最佳地分配通行权来兑现服务承诺，以保障预约通行车辆按时到达。本文即在预约通行车辆（Scheduled Vehicle, SV）与常规车辆（Regular Vehicle, RV）混行的条件下，面向已经规划好的预约通行服务，研究交通协同控制机制及其方法，以充分利用高精度预约信息开展更精准、主动的交通控制，并有效保障预约通行的服务可靠性。

首先，针对当前研究主要集中在预约服务的事前安排阶段、导致面向预约通行的交通控制机制架构缺失的问题，本文首先分析典型预约通行场景对交通控制提出的功能需求，建立包含需求感知、状态预测、综合决策与伴随控制等环节的控制逻辑框架，并设计了支持预约信息接入与控制决策协同的系统架构，为后续章节的研究奠定了技术路径。

其次，针对部分观测的SV信息仍难以充分支撑主动控制的问题，提出基于预约通行信息的状态预测与主动控制方法，完成了面向预约通行的交通控制机制中“感知预测”部分的工作。具体而言，首先提出了预约通行环境下转向流量随机性横纵向解耦框架，纵向上采用车队散布模型描述行驶速度的随机性，横向上基于部分观测的SV路径信息采用基于深度核学习的高斯过程来预测转向比例的随机性，提升转向流量的预测精度；随后，提出了预约通行环境下信号主动控制异步框架，每个交叉口独立设计优化时域与信息更新周期，使控制决策与局部交通演化过程相匹配。实验结果表明，所提出的转向流量预测模型能够有效刻画未来到达流量的随机波动特征，显著提升预测精度；在此基础上，信号主动控制异步框架能够充分释放该预测模型的潜力，实现更高效的信号主动控制。

随后，针对预约通行个体车辆服务目标与系统整体效率最优目标的矛盾，提出了一种交叉口预约通行车辆轨迹协同决策优化方法，完成了面向预约通行的交通控制机制中“综合决策”部分的工作。具体而言，首先分析了个体服务可靠性与系统整体效率的矛盾，构建了以空间位置为建模域的协同决策优化模型，该模型以SV的时刻表延误与RV的平均延误为联合优化目标，实现了车辆轨迹规划与信号控制的统一建模；随后，设计了并行自主决策算法，每辆SV根据自身时刻表自主调整轨迹，信号控制器根据全局延误最小化优化信号方案，各控制单元通过一致性变量传递实现决策结果的收敛。仿真实验证明，协同决策方法能够实现个体服务优先与系统整体效率之间的协同，所提出的并行自主决策算法能够保持高效的求解能力。

最后，针对路网中预约通行车辆的优先需求相互冲突、交通控制缺乏有效协调机制的问题，提出了保障预约通行服务可靠性的伴随式信号控制方法，完成了预约通行交通控制机制中“伴随控制”环节的工作。具体而言，首先解析了预约通行车辆的服务性质：往往只需要保证其在终点处准时到达即可，因此信号控制可提供“伴随式”优先，即部分刚起程的SV可允许适度延误，因为其在后续路段仍有获得优先的机会。基于此思想，将伴随式信号控制建模为部分可观测马尔可夫决策过程，通过构造基于时刻表延误与剩余路程的动态优先级机制；结合端到端的多智能体强化学习框架，实现了状态编码、策略学习和价值评估的统一训练。通过仿真实验，证明了伴随式控制能够在服务可靠性与系统效率之间取得良好协同，不仅显著提升了SV的准点性，还能够进一步降低RV延误。

综上，本文针对面向预约通行的交通控制问题，首先解析了预约通行环境下交通控制机制的革新与重构，在此基础上，进一步构建了基于预约通行信息的状态预测与主动控制方法、交叉口预约通行车辆轨迹协同决策优化方法和保障预约通行服务可靠性的伴随式信号控制方法。上述工作共同完善了面向预约通行的交通控制机制体系，推动新一代交通控制系统向精细化预测、主动化控制与智能化决策的方向演进。

关键词：交通控制，预约通行，出行服务，主动控制，协同优化

英文摘要

ABSTRACT

Traffic signal control regulates traffic signal plans and, together with traffic flow organization and operational condition optimization, allocates roadway right-of-way and balances supply–demand relationships to achieve specific objectives. It is one of the most important technical approaches for alleviating congestion and improving the overall efficiency of urban road networks. However, due to the high uncertainty and uncontrollability inherent in individual, spontaneous travel behavior, conventional traffic control methods—largely relying on aggregated information—have limited capability in finely coordinating heterogeneous traffic participants. With the rapid development of intelligent and connected vehicle technologies, urban traffic systems urgently require a new generation of traffic control frameworks that explicitly consider fine-grained individual behavior and proactive demand evolution. Meanwhile, urban scheduled travel reservation is being progressively introduced as an emerging demand management strategy. By planning vehicle departure times and routes in advance, urban scheduled travel reservation transforms originally random travel demand into planned demand with explicit spatiotemporal constraints. At present, such reservation mechanisms have been explored in several scenarios, including urban public transit operating under fixed timetables, emergency vehicles with strict time-critical requirements, and travel services that provide pre-planned itineraries.

A natural coupling exists between urban travel reservation and traffic control. On the one hand, reservation systems reduce random fluctuations in spontaneous travel demand, enabling the control system to become more proactive and forward-looking. On the other hand, Scheduled Vehicles (SVs) must rely on traffic control to allocate right of way in an optimal manner so that their promised arrival times can be fulfilled. This study investigates the reservation-oriented coordinated traffic control mechanism under mixed traffic conditions with SVs and Regular Vehicles (RVs), where the scheduled travel reservation services are already planned. It aims to leverage high-precision reservation information for more accurate and proactive control and to ensure reliable service for reserved travel.

First, to address the limitation that existing studies mainly focus on the pre-trip scheduling stage of reservation services and lack a dedicated traffic control mechanism for reservation-based demand, this study analyzes the functional requirements imposed by typical reservation scenarios on traffic control systems. A closed-loop control logic framework is proposed, comprising demand sensing, state prediction, integrated decision-making, and companion control modules. Based on this framework, a system architecture is designed to support the integration of reservation information and its coordination with signal control decisions, thereby providing a structural foundation for the subsequent model development and algorithm design.

Second, to overcome the challenge that partially observed SV information is still insufficient for proactive control, this study develops a reservation-based state prediction and proactive signal control method, corresponding to the “sensing and prediction”module of the mechanism. A longitudinal–lateral decoupled framework is proposed to model the stochasticity of turning flow evolution: longitudinal randomness is captured using a platoon dispersion model, while lateral randomness is estimated using a deep kernel learning-based Gaussian process model that predicts unobservable RV turning proportions from partial SV route information. Building on the improved prediction accuracy, an asynchronous proactive signal control framework is further developed, where each intersection designs its own optimization horizon and update cycle to match local traffic evolution prediction results. Experiments demonstrate that the proposed prediction model effectively captures stochastic variations in future arrivals and significantly improves accuracy, and the asynchronous control framework further enhances efficiency by fully exploiting the predictive capability.

Third, to address the conflict between individual service reliability for SVs and overall system efficiency, this study proposes a trajectory-level coordinated decision-making method for SVs at signalized intersections, completing the “integrated decision-making” component of the mechanism. A spatial-domain formulation is developed to jointly reduce SV schedule delay and RV average delay, achieving unified modeling of SV trajectory planning and signal control. A parallel decision-making algorithm is then designed, where each SV adjusts its trajectory based on its own schedule, while the signal controller optimizes signal plans to minimize global delay. Consensus is achieved through the exchange of coordination variables among control units. Simulation results confirm that the method effectively balances individual service and system efficiency while maintaining high computational efficiency.

Finally, to address conflicts among multiple SVs’ priority requirements and the lack of effective coordination in existing control systems, this study develops a companion signal control method that ensures reservation service reliability, completing the “companion control” module of the mechanism. The analysis shows that general SVs only need to ensure on-time arrival at their final destination, meaning that some SVs at the early stage of their journey may tolerate moderate delay if they still have opportunities to regain priority later. Based on this insight, the companion control problem is formulated as a partially observable Markov decision process with dynamic priorities defined by schedule delay and remaining travel distance. An end-to-end multi-agent reinforcement learning framework is employed for state encoding, policy learning, and value estimation. Experiments show that the proposed method achieves a desirable trade-off between service reliability and system efficiency, significantly improving SV punctuality while reducing RV delay.

In summary, this study systematically examines reservation-oriented traffic control by reconstructing the control mechanism for urban scheduled travel reservation and proposing three key methodological advances: a reservation-based state prediction and active control method, a trajectory-level coordinated decision-making method, and a companion signal control method. Together, these contributions enrich the mechanism framework for reservation-oriented traffic control and promote the evolution of next-generation traffic control systems toward refined prediction, proactive control, and intelligent decision-making.

Key Words: Traffic Control, Scheduled Travel Reservation, Travel Service, Proactive Control, Coordinated Control