康振
入学时间：2020级
答辩时间：2025年
论文题目：考虑减排的多交叉口多目标协同控制方法研究
中文摘要
摘要
在全球积极应对气候变化的背景下，中国明确提出“碳达峰”与“碳中和”的战略目标。道路交通领域作为重要的排放源，其碳排放约占全国总排放量的8%至10%，已成为碳减排工作的重点领域。交通控制因其较高的效费比优势，长期以来被视为缓解道路交通问题的核心手段。传统交通控制手段主要为基于集计交通流建模的信号控制（如单交叉口信号控制、多交叉口绿波协调等），尽管在缓堵上有所成效，但其均质化假设忽略了车辆微观排放差异，导致排放计算精度不足，难以支撑精准的减排优化。随着网联通信与自动驾驶技术的快速发展，车辆端的微观控制手段（生态驾驶、速度引导）被广泛用于道路交通缓堵减排。然而，此类方法基于车载感知执行个体决策，缺乏对交通系统的整体认知与建模，难以实现系统层面的最优。
因此，本文聚焦于城市道路多交叉口场景，以“调控车辆微观速度，实现系统减排增效”为核心学术思想，分别从交叉口上游路段控制、单交叉口控制、多交叉口控制三个层次开展研究，以“车辆与交通流协同、车辆与信号协同、多交叉口协同”为三类协同控制手段，形成支撑减排增效的协同控制方法体系。
首先，针对现有控制方法中存在的交叉口排放特征不明、系统层面优化效果不足、控制框架不完善等问题，研究构建了面向减排增效多交叉口协同控制框架。基于真实的轨迹数据集，精准识别多交叉口排放特征，系统梳理面向减排增效的控制场景与控制对象。进一步地，分别针对微观的车辆排放特征与系统层面的排放优化进行了预实验分析，明确了基于加速工况优化实现减排的车辆端微观控制方法，以及“车辆与交通流协同、车辆与信号协同、多交叉口协同”三类实现交叉口减排增效的控制需求。在此基础上，归纳总结了面向减排的城市道路多交叉口信号控制的基本需求、基本条件与基本优化目标，明确了协同控制方法研究的总体理论框架。
其次，面向交叉口上游路段控制场景，考虑现有车辆端的微观控制策略多以个体最优为目标，对系统最优的考量不足，导致车辆群体的排放激增问题，提出了“车辆与交通流协同”的控制方法。该方法构建了基于“幽灵密度”量化CAV对交通流影响的系统动态模型。幽灵密度作为统一的状态量，同时参与车辆端和路侧端的控制决策：车辆端的协调速度上限受到其所在路段幽灵密度的约束；路侧端信号控制则以幽灵密度作为信号控制优化的状态变量，动态调整信号相位。微观仿真实验评估了在不同场景下（不同的CAV渗透率、交通需求水平）与生态驾驶策略相比的性能。所提出的方法可以实现路段下游交叉口排放量减少0.51%至21.52%，通行能力增加11.83%。
再次，面向单交叉口控制场景，针对交叉口已有信号控制方法未考虑车辆层面的行进过程、信号绿灯时间未能被车辆充分利用、引发车辆频繁启停增排的问题，研究构建了“车辆与信号协同”的双层控制框架。该框架通过解耦设计车辆侧与路侧的控制目标，提出了一个双层优化模型：上层对路侧端的信号配时优化，最小化交叉口延误以提升通行效率；下层对微观车辆速度引导优化，分别提出了全局排放最小的直接优化模型，和交叉口停车次数及停车时长最小的间接优化模型，实验证明间接优化模型在牺牲2.3%~2.8%的最优解性能下，能够提升求解效率以满足实时控制的需求。微观仿真实验评估了在不同场景下（不同的受引导车辆渗透率、交通需求水平、不均衡交通需求）与已有控制方法相比的性能优势：所提出的方法可实现碳排放减少5.24%至17.60%，延误减少22.82%至52.62%。此外，针对现有车辆速度引导方法落地应用困难的问题，研究创新性地提出一种网联环境下的碳激励机制，以经济激励的方式保障车辆对速度引导的服从性。在碳激励机制下，车辆每通过一个交叉口的碳激励平均奖励在0.0015至0.016元之间，交叉口的碳减排成本为7.5至31.5元/吨。
最后，面向多交叉口控制场景，针对传统多交叉口协调控制方法难以协调不同路径上的微观车辆通行需求，不同路径的车辆会在某一交叉口聚集、拥堵与排队，导致全局减排增效困难的问题，研究提出了考虑减排的多交叉口分布式协同控制方法。在分布式控制框架下，研究将大规模路网问题拆分为小区域局部子问题，使各交叉口能够基于局部车辆路径信息独立完成多目标优化。进一步地，将小规模交叉口区域局部子问题拆解为相邻交叉口的供需匹配问题与单交叉口的多目标优化问题。二者在局部子问题中互为上下游，可以实现数据与控制方法的互通，二者协同可以避免局部瓶颈的转移与扩散。微观仿真实验结果表明，本研究提出的控制方法相较于传统的周期-绿信比-相位差协同配时方法，可降低路网平均排放7.64%至32.15%，降低平均延误5.84%至25.01%，提升路网通过量9.15%~16.30%。
本文以“调控车辆微观速度，实现系统减排增效”为研究主线，从面向减排增效的交通控制基础问题解析出发，系统梳理了交叉口信号控制的对象、场景与控制框架，构建了面向减排增效的车辆与交通流协同、车辆与信号协同、多交叉口协同的控制方法，初步形成了考虑减排的多交叉口多目标协同控制基本架构，具有重要的理论意义与实用价值。
关键词：城市交通，节能减排，网联环境，协同控制，多目标优化
英文摘要
ABSTRACT
Against the backdrop of global efforts to address climate change, China has explicitly set the strategic targets of “carbon peaking” and “carbon neutrality.” As a major emission source, the road transportation sector accounts for approximately 8%–10% of the country’s total carbon emissions, making it a priority area for emission reduction. Owing to its high cost-effectiveness, traffic control has long been regarded as a core approach to mitigating road traffic problems. Traditional control strategies primarily rely on macroscopic models based on aggregated traffic flow (e.g., isolated intersection signal control, green-wave coordination for multiple intersections). However, their homogeneity assumptions neglect the heterogeneity of vehicle-level emissions, resulting in insufficient accuracy in emission estimation and limiting the effectiveness of emission-oriented optimization. With the rapid advancement of connected communication and autonomous driving technologies, vehicle-level microscopic control strategies (such as eco-driving and speed guidance) have been increasingly adopted for emission reduction. Nevertheless, vehicle-side methods are executed through onboard perception for individual decision-making, lacking holistic awareness and modeling of the macroscopic traffic system. Consequently, a disjunction emerges between macroscopic traffic-flow control and microscopic vehicle control: macroscopic strategies are difficult to implement at the microscopic level, whereas microscopic strategies fail to respond to macroscopic objectives. As a result, relying on a single type of control measure cannot fully unlock the optimization potential of the system, hindering the dual goals of emission reduction and efficiency improvement.
To address this gap, this study focuses on urban road networks with multiple intersections and advances the concept of “macro–micro coordination,” wherein vehicle-level speed regulation is harnessed to achieve system-wide emission reduction and efficiency improvement. The research spans three levels—upstream road segments, single intersections, and multiple intersections—through three types of coordination: vehicle–traffic flow coordination, vehicle–signal coordination, and multi-intersection coordination. Together, these form a methodological framework for macro–micro coordinated traffic control to support emission reduction and efficiency gains.
First, to tackle the disjunction between macroscopic and microscopic control in existing methods—which impedes emission reduction and efficiency improvement—this study develops a macro–micro coordinated control framework. Based on real trajectory datasets, the emission characteristics of multiple intersections are identified with high precision, and the control scenarios and objects for emission reduction are systematically clarified. Pre-experiments are conducted on microscopic vehicle emission characteristics and macroscopic system-level optimization, confirming vehicle-level control strategies that reduce emissions through acceleration pattern optimization, and identifying three system-level control needs: vehicle–traffic flow coordination, vehicle–signal coordination, and multi-intersection coordination. On this basis, the fundamental requirements, preconditions, and optimization objectives for emission-oriented urban intersection signal control are summarized, leading to the definition of the overall theoretical framework for coordinated control.
Second, for upstream road segment control, given that existing vehicle-level microscopic control strategies inadequately account for macroscopic system benefits—and may induce traffic flow fluctuations, causing excessive acceleration/deceleration emissions—a “vehicle–traffic flow coordination” method is proposed. This method constructs a system dynamics model in which the “phantom density” quantifies the impact of connected and automated vehicles (CAVs) on traffic flow. Phantom density serves as a unified state variable simultaneously governing decisions on both the vehicle and infrastructure sides: vehicle coordination speed is constrained by the phantom density of its segment, while signal control uses phantom density as the state variable for dynamic phase optimization. Microscopic simulations evaluate performance under various conditions (different CAV penetration rates, traffic demand levels), compared with eco-driving. Results show that the proposed method reduces emissions by 0.51%–21.52% and increases intersection throughput by 11.83%.
Third, for single-intersection control, considering that conventional macroscopic signal control methods ignore microscopic vehicle progression and underutilize green times—causing frequent stop-and-go behavior and additional emissions—a bi-level “vehicle–signal coordination” framework is developed. By decoupling objectives for vehicles and signals, a bi-level optimization model is designed: the upper level optimizes signal timing to minimize delay and enhance efficiency, while the lower level optimizes vehicle speed guidance. Two formulations are proposed for the lower level: (i) a direct optimization model minimizing global emissions, and (ii) an indirect optimization model minimizing the number and duration of stops. Experiments demonstrate that the indirect model sacrifices only 2.3%–2.8% performance but achieves greater computational efficiency to support real-time implementation. Microscopic simulations across scenarios (different guided-vehicle penetration rates, demand levels, and demand imbalance) reveal substantial benefits: emissions reduced by 5.24%–17.60% and delays reduced by 22.82%–52.62%. Furthermore, recognizing the practical difficulty of implementing vehicle speed guidance, this study innovatively proposes a carbon incentive mechanism in connected environments. By providing economic incentives, the mechanism enhances vehicle compliance with guidance. The average carbon incentive reward per vehicle per intersection is 0.0015–0.016 CNY, corresponding to a carbon reduction cost of 7.5–31.5 CNY per ton.
Finally, for multi-intersection control, traditional macroscopic coordination methods struggle to reconcile vehicle progression needs along different paths, often resulting in vehicle clustering, congestion, and queues at certain intersections—hindering global emission reduction and efficiency improvement. To address this, a distributed multi-intersection coordination method is proposed. Under the distributed framework, the large-scale network problem is decomposed into smaller regional subproblems, allowing each intersection to independently perform multi-objective optimization using local vehicle path information. Further, each subproblem is decomposed into two interdependent components: (i) upstream–downstream supply–demand matching between adjacent intersections, and (ii) single-intersection multi-objective optimization. These components exchange data and coordinate decisions, thereby preventing the propagation of local bottlenecks. Simulation results show that, compared to conventional cycle–green ratio–offset coordination, the proposed method reduces average network emissions by 7.64%–32.15%, average delays by 5.84%–25.01%, and increases network throughput by 16.30%.
In summary, this study advances the research paradigm of “regulating vehicle-level speeds to achieve system-level emission reduction and efficiency gains.” Beginning with the fundamental issues of emission-oriented traffic control, it systematically analyzes the objects, scenarios, and frameworks of intersection signal control, and develops coordinated control methods for vehicle–traffic flow, vehicle–signal, and multi-intersection coordination. Collectively, these efforts establish a foundational architecture for multi-objective, emission-oriented coordinated control of urban intersections, offering both theoretical insights and practical value.
Key Words:Urban Traffic, Energy Conservation and Emission Reduction, Connected Environments, Cooperative Control, Multi-Objective Optimization
张梦雅
入学时间：2020级
答辩时间：2025年
论文题目：网联环境下高速道路交通运行状态感知和风险分析研究
中文摘要
摘要
高速道路作为现代交通系统的核心动脉，其高速度、大流量的运行特征使得交通安全问题尤为突出。交通事故不仅造成严重的人员伤亡和财产损失，更会对整个交通网络的运行效率产生深远影响。虽然交通事故的发生具有概率性特征，但其潜在的社会影响和经济损失不容忽视。随着车联网技术的快速发展，V2X（Vehicle-to-Everything）通信、智能感知等新技术为交通风险的主动防控提供了新的技术手段，使得从风险这一概率的视角研究交通安全问题成为可能。然而，在网联环境中，交通系统的运行状态更加复杂，交通风险的传播路径和影响因素亦发生重大变化。特别是，网联自动驾驶车辆（Connected and Autonomous Vehicles, CAV）与人类驾驶车辆（Human-Driven Vehicle，HDV）的混合运行环境，引入了新的风险形态与安全挑战。当前针对网联环境下交通运行状态感知与风险分析的研究已取得一定进展，但在运行状态由稳定向不稳定演化的动态过程中，不同类型车辆驾驶行为的差异及其对风险演变的具体影响仍缺乏系统刻画，状态感知的有效性、评估维度的合理性及管控策略的适应性等方面尚有进一步优化的空间。
本研究以网联环境下高速道路新型混合交通流为研究对象，围绕“特征解析—状态感知—风险量化—主动管控”的全链条研究框架，旨在揭示新型混合交通流运行状态与风险演化机理并构建主动风险防控体系。研究通过探讨车辆退化特性、交通状态不确定性、行为交互复杂性以及极端环境干扰等核心问题，力求为网联交通系统的安全优化提供新的研究视角与技术路径。具体研究框架如下：
（1）针对网联环境下不同车辆类型间交互复杂的问题，对新型混合交通流运行风险特征进行解析。基于传统与网联环境下的事故数据，分析了追尾事故的特征变化，提炼出影响运行安全的关键因素。从CAV退化现象出发，构建了由正常CAV、降级CAV与HDV共同构成的新型混合交通流模型。为准确刻画不同类型车辆的纵向行为特征，引入感知反应时间的动态修正机制，改进了现有跟驰模型，以分别表征三类车辆在不同行驶模式下的控制响应。结合多种安全替代指标，进一步分析了CAV退化现象、渗透率、排队强度等因素对交通流稳定性与运行风险的影响关系，揭示了新型混合交通流在关键参数变化下的风险演化机制。
（2）为提升复杂交通场景下运行状态的感知能力并量化其与风险之间的关系，构建了基于多维特征融合的交通运行状态感知模型。基于三相交通流理论和聚类算法，对自由流、同步流、拥堵流及其过渡状态进行阶段划分。结合速度、速度变异系数、上下游速度差与拥堵指数等多维交通参数，构建基于非负矩阵分解的运行状态感知模型，并提出运行状态感知指数，用于动态表征交通状态与风险等级之间的映射关系。通过与其他方法对比，展现出更强的可解释性和更优的风险指示效果。基于高精度车辆轨迹数据及不同CAV渗透率条件下的仿真轨迹，开展多工况下的模型应用与状态识别，结果表明该指数可有效识别瓶颈段、交织段等复杂区域的高风险状态分布特征，并揭示渗透率变化对交通状态结构的动态影响。
（3）针对新型混合交通流中驾驶行为多样、单一维度风险指标难以有效捕捉潜在冲突的问题，提出融合纵向跟驰与横向换道特征的综合风险指标。基于轨迹数据，采用小波变换与聚类算法提取减速点、车头时距等微观行为特征，识别高风险跟驰车辆与换道行为样本。通过参数敏感性分析确定指标中各要素的权重结构；结合经典安全替代指标，进行相关性分析，设定指标的风险识别阈值。针对跟驰与换道场景，设定不同感知反应时间、CAV渗透率、交通流量与排队强度等条件，应用该指标开展单车、车队及交通流三个层面的风险评估，并与单一维度安全替代措施进行对比分析以验证其有效性。
（4）针对复杂交通结构下风险管控策略在响应及时性与环境适应性方面的挑战，设计基于极端天气触发的双层响应主动风险管控策略。分析了CAV显著退化区与极端天气对新型混合交通流运行稳定性和风险水平的影响，提出了基于领航车协同控制的CAV专用道策略，并构建了融合运行态势感知与驾驶行为风险识别的双层响应机制。设计了响应机制中的动态阈值调节方法，以适应不同交通运行状态与环境扰动条件。设定多种渗透率与极端天气组合场景，开展仿真模拟，评估该机制在多种复杂交通情境下的风险识别能力、响应流程及环境适配性，并分析策略实施对新型混合交通流运行风险的调控效果。
关键词：网联环境，高速道路，交通运行状态，多维风险评估，主动风险管控
英文摘要
ABSTRACT
High-speed roads are the main arteries of modern transportation systems. Their high-speed and high-volume characteristics make traffic safety a particularly pressing concern. Traffic accidents cause serious casualties and economic losses, and also reduce the operational efficiency of the entire traffic network. Although accident occurrence is probabilistic, its potential impact on society and the economy cannot be overlooked. With the rapid development of connected vehicle technologies such as V2X (Vehicle-to-Everything) communication and intelligent sensing, new tools have emerged for proactive risk prevention. These tools make it possible to study traffic safety from a risk-based perspective. However, in connected environments, traffic operations have become more complex. The propagation paths and influencing factors of traffic risks have also changed significantly. In particular, the mixed presence of Connected and Autonomous Vehicles (CAVs) and Human-Driven Vehicles (HDVs) introduces new forms of risk and safety challenges. Although some progress has been made in the perception and analysis of traffic states and risks, the dynamic transition from stable to unstable states and the behavioral differences among vehicle types remain insufficiently addressed. Improvements are still needed in state perception accuracy, evaluation dimensions, and the adaptability of control strategies.
This study focuses on mixed traffic flows on high-speed roads under connected environments. A full-chain research framework is proposed, covering four aspects: feature analysis, state perception, risk quantification, and proactive control. The aim is to reveal the mechanisms behind traffic state evolution and risk propagation, and to establish a proactive risk management system. The main research components are as follows:
(1) The characteristics of traffic risks in mixed traffic flows are analyzed. Based on accident data from both traditional and connected environments, the patterns of rear-end collisions are studied, and key influencing factors are identified. A new traffic model is built, which includes normal CAVs, degraded CAVs, and HDVs. A correction mechanism for perception–reaction time is introduced to improve traditional car-following models. This allows different vehicle types to be accurately represented under various driving conditions. Several surrogate safety measures are applied to explore the effects of CAV degradation, penetration rate, and queuing intensity on traffic stability and risk. The mechanisms of risk evolution under key parameter changes are revealed.
(2) A multi-dimensional traffic state perception model is developed to enhance the recognition of complex traffic conditions and their relation to risk. Based on the three-phase traffic theory and clustering methods, traffic states are divided into free flow, synchronized flow, congested flow, and transitional phases. Parameters such as speed, speed variation coefficient, upstream–downstream speed difference, and congestion index are integrated into a non-negative matrix factorization (NMF) model. A Traffic State Perception Index is proposed to map traffic conditions to risk levels. The model is validated using high-resolution trajectory data and simulations under various CAV penetration rates. Results show that the model effectively identifies high-risk areas such as bottlenecks and weaving zones, and captures structural changes in traffic states caused by penetration rate variations.
(3) A composite risk metric is proposed to address the limitations of single-dimensional measures. This metric integrates longitudinal car-following and lateral lane-changing behaviors. Using vehicle trajectory data, wavelet transformation and clustering techniques are applied to extract deceleration points and headway patterns. High-risk behaviors are identified. Weight structures are determined through sensitivity analysis, and thresholds are set based on correlations with traditional surrogate safety measures. The metric is used for hierarchical risk assessment at vehicle, platoon, and flow levels under different traffic and behavior scenarios. Its effectiveness is verified through comparison with existing methods.
(4) A dual-layer proactive risk control strategy is designed for extreme weather and complex traffic conditions. The effects of degraded CAV zones and adverse weather on traffic stability and risk levels are analyzed. A dedicated CAV lane strategy based on cooperative control is proposed. This strategy combines traffic state perception and driving behavior risk identification. A dynamic threshold adjustment method is used to adapt to different traffic states and environmental disturbances. Simulation scenarios with various CAV penetration levels and weather conditions are tested. The results demonstrate improved risk detection, responsive control, and environmental adaptability.
Key Words: connected environment, high-speed roads, traffic operational state, risk assessment, proactive risk management
