李锐
入学时间：2020级
答辩时间：2023年
论文题目：非阻塞目标的道路交叉口交通主动控制方法研究
中文摘要
摘要
城市道路交通供需不平衡而产生的交通阻塞现象已成为中国城市交通中的突出问题。交通阻塞不仅表现为车辆排队和延误，还随着交叉口之间的关联关系，以点-线-面形式，不断扩大影响范围，由单点交叉口扩散至各路段和整个路网，对交通运行产生巨大影响。本论文旨在准确剖析车辆在道路中的通行需求与通行条件及其供需关系，估计交通运行状态，并据此主动地以非阻塞为目标，通过交叉口信号控制手段防止交通阻塞发生，形成了一套非阻塞主动控制方法。该方法不仅有利于提高交叉口交通通行效率，降低排队与延误，并使得交通控制更具主动性与实用性，可进一步赋能新一代交通控制与管理系统的发展。
在交通运行状态解析的基础上，本论文将相邻交叉口中间路段进行分段，分别为进口道路段与交织段，并分别对其交通通行特点进行了分析，形成了进口道通行能力概率计算模型与交织段通行能力函数模型，对不同通行条件下的道路通行能力进行了计算。进一步根据典型的实际交通感知数据检测条件，利用上述模型形成了一套交通运行状态感知与估计方法，亦即结合实测和估计的交通流量数据及通行能力模型计算结果，分析流量与通行能力之间的供需关系，界定道路通行条件，判断交通阻塞的发生及其程度，提高了应用场景中方法对不同数据基础的适用性。
关于交叉口非阻塞主动控制问题研究，本论文重点根据单交叉口场景特点，考虑信号控制输入输出高效性与控制策略决策的及时性，将阻塞感知估计与交通主动控制相结合，基于感知数据和基本控制逻辑建立了一个结合模糊规则与强化学习，对道路饱和度、流量、排队与延误多目标进行考量的交通控制方法，具有科学性。该方法一方面通过模糊规则，以实测和估计的道路交通状态参数，计算对应相位的绿灯时长并形成信号控制配时策略；另一方面通过强化学习，在模糊规则形成的方案基础上，进行配时调整，实现非阻塞主动控制。这种方法避免了模糊规则存在结果非最优和难以评价的缺点。同时，考虑到多进口感知结果、不同阻塞优先级与控制方案调整频率，对控制方法进行了优化和效果仿真验证。其次，本论文将控制场景拓展至与单交叉口存在关联的上下游交叉口，构建了一个多交叉口场景。根据多交叉口场景特点，根据已有参数，形成了多交叉口“量入为出”非阻塞协同控制策略，即针对下游的通行条件对上游的车辆放行进行限制，避免阻塞发生，同时对更下游交叉口的通行条件进行感知，有目的性地控制目标交叉口的放行，从系统角度避免阻塞的发生，并建立了通过强化学习进行模糊规则参数自学习的多交叉口交通主动控制方法，旨在通过实测数据对模糊规则方法中的隶属度函数参数和模糊规则进行自学习，以加强控制效果。最后对方法进行了仿真验证。
最后，选择实际的典型场景，本论文利用实地视频卡口采集的交通流量数据，对提出的非阻塞主动控制方法进行实地实验验证，证明了理论的正确性和方法的有效性。
综上所述，本论文建立了针对进口道阻塞和路段阻塞感知的交通运行状态感知方法，形成了对单交叉口和多交叉口场景的非阻塞主动控制方法，并进行验证，但同时，本论文在对随机事件的考虑、多源数据的使用、平台与硬件开发实践上存在提升空间。本论文研究成果在交通供需关系分析与非阻塞主动信号控制的应用方面具有一定的实际指导作用。
关键词：主动控制，通行能力，交通状态感知，模糊规则，强化学习
英文摘要
ABSTRACT
The traffic blockage caused by the imbalance between supply and demand of urban road traffic has become a prominent problem in urban transportation in China. Traffic blockage not only manifests as vehicle queues and delays, but also continuously expands the scope of impact as the correlation between intersections, spreading from single point intersections to the entire road network, exerting a huge impact on traffic operations. This paper aims to accurately analyze the traffic demand and conditions of vehicles on the road, as well as their supply-demand relationship, estimate the traffic operation status, and based on this, actively aim to prevent traffic blockage through intersection signal control measures, forming a set of non-blockage active control methods. This method is not only beneficial for improving the traffic efficiency of intersections, reducing queuing and delays, but also making traffic control more proactive and practical, which can further empower the development of a new generation of traffic control and management systems.
On the basis of analyzing the traffic operation status, this paper divides the middle section of adjacent intersections into two sections, namely the entrance road section and the weaving section, and analyzes their traffic characteristics. The probability capacity calculation model of the entrance road section, and the weaving section capacity function model are formed, to calculate the road capacity under different traffic conditions. Based on typical actual traffic perception data detection conditions, a set of traffic operation state perception and estimation methods have been developed using the above model. This involves combining measured and estimated traffic flow data with the calculation results of the capacity model, analyzing the supply-demand relationship between flow and capacity, defining road traffic conditions, and determining the occurrence and degree of traffic blockage.
Regarding the research on non-blockage active control at intersections, this paper first elaborates on the basic principles of active traffic control and the control data and information requirements. The focus is on the characteristics of single intersection scenarios, considering the efficiency of signal control input and output and the timeliness of control strategy decisions, and combining blockage perception estimation with active traffic control. Based on perceptual data and basic control logic, a scientific traffic control method has been established, which combines fuzzy rules and reinforcement learning to consider multi-objective factors such as road saturation, flow, queuing, and delay. On the one hand, this method uses fuzzy rules to calculate the green light duration of the corresponding phase based on the measured and estimated road traffic state parameters, and forms a signal control timing strategy. On the other hand, through reinforcement learning, based on the strategy formed by fuzzy rules, timing adjustments are made to achieve non-blockage active control. This method avoids the drawbacks of non-optimal results and difficulty in evaluating fuzzy rules. At the same time, considering the blockage of the entrance road section and the frequency of control strategy adjustment, this paper optimized the control method and verified its effectiveness through simulation. Secondly, this paper extends the control scenario to upstream and downstream intersections that are associated with a single intersection, and constructs a multi-intersection scenario centered around the target intersection. Based on the characteristics of multiple intersection scenarios and existing parameters, a non-blockage collaborative control strategy of "output according to input" for multiple intersections has been formed, which restricts the release of upstream vehicles based on downstream traffic conditions to avoid blockage. At the same time, it perceives the traffic conditions of downstream intersections and purposefully controls the release of target intersections to avoid blockage from a system perspective, and a multi-intersection traffic active control method is established, which using reinforcement learning for fuzzy rule parameter self-learning, aiming to self-learning the membership function parameters and fuzzy rules in the fuzzy rule method through measured data to enhance control effectiveness. Finally, simulation verification was conducted on the method.
Finally, this paper selecting a typical actual scenario and using traffic flow data collected from on-site video checkpoints, the proposed non-blockage active control method was experimentally validated, proving the correctness of the theory and the effectiveness of the method.
In summary, this paper establishes a traffic operation state perception method for perception of entrance road section blockage and road segment blockage, forming a non-blockage active control method for single intersection and multi-intersection scenarios, and verifies it. However, at the same time, there is room for improvement in the consideration of random events, the use of multiple data sources, and the development of platforms and hardware in this paper. The research results have practical guiding significance in the analysis of traffic supply and demand relationship and the application of non-blockage active signal control.
Key Words:active control, capacity, traffic state perception, fuzzy rules, reinforcement learning
