杨航
随着我国各大城市快速路建设的发展，由快速路和普通道路组成的混合路网逐渐成为大城市道路网的主要形式。混合路网在缓解城市交通拥堵压力，增强城市交通咽喉的流量转换能力方面发挥了重要的作用。但由于快速路和普通道路交通流的异质性，且现有的管控措施往往采取快速路和普通道路各自运行的优化模式，并未充分考虑其异质交通流特性而采取针对性的同步协调管控措施，因而在整体路网层面上对交通流的集中疏导和调控效果非常有限，导致两个路网的衔接区域及其毗邻区在高峰时段内极易发生交通拥挤，该区域的拥堵以及由此引发的大规模拥堵扩散会极大地降低整个混合路网的运行效率。同时，现有的管控逻辑多遵循的是以路网管理者为主导的优化模式，管理者与用户之间存在着较为单一的领导-跟随关系，用户往往只能被动接受管理者提供的最优分配方案，这在一定程度上忽视了用户对于管理者方案的选择和反馈对整体管 控系统的影响。本文以降低总体出行成本为全局优化目标，以协同SoS（Collaborative System of System）的集中化框架为协同理念，兼顾均衡不同路段的交通压力，设计了管理者与用户角色可变的协同模型，对混合路网环境下的交通状态预测技术、集中化协同优化技术以及基于多智能体（Multi Agent System, MAS）的协同群集技术展开研究，实现了连续流与间断流、交通控制与交通分配、管理者与用户这三个视角下的同步协调优化，克服了以往通过非集中化优化逻辑制定城市混合路网管控优化策略的局限性。
本文首先以SoS系统架构和混合路网拓扑结构的关联性为切入，建立了协同SoS视角下的混合路网系统架构。通过管控对象、协同范围、优化目标、约束条件论述了混合路网协同管控在集中化程度、协同层次、建模求解上的技术难点，进而对协同问题本质进行剖析，将复杂的大规模路网交通流协同优化问题抽象为协同SoS视角下的信号控制-交通分配的组合问题（CAC问题）。
在问题抽象的基础上，本文构建了基于协同SoS的混合路网协同管控优化体系，体系包括功能与定位、实现的管控框架、构建的理论支撑与运行的技术保障。其中，体系功能与定位将研究聚焦于指导混合路网交通状态的识别与预测以及交通拥堵的协同疏解和管控；集中化管控框架建立了CAC问题下混合路网交通控制、交通分配与路网监测终端三个模块间的交互机制，并明确了本文需要研究的三大关键技术与出行成本、流量模式与管控策略间的更新间的关系；构建理论支撑与运行技术保障则为三大关键技术的建模实现和优化奠定了基础。
交通状态的识别与预测以行程时间作为体现整体路网流量模式的效率指标，用以评估路网的总体出行成本和通行效能，提出了基于确定性状态成分、非确定性状态成分和噪声修正的短时行程时间组合预测模型，并通过实际路网数据验证了模型的预测精度和鲁棒性，同时该模型也具有捕捉交通状态突变拐点的能力。
集中化协同优化技术是为解决快速路与普通道路间异步协同的非集中化问题。本文设计了同一时空维度下的混合路网底层交通流模型，并结合管理者和用户在对策模型中同时存在非对称性和对称性的特征，论述了二者在协同SoS视角下的角色关系，基于Stackelberg博弈建立了管理者与用户角色可变的混合路网集中化协同管控优化模型，并通过简易路网（toy network）和实际路网两个案例论证了在混合路网协同管控中引入角色可变机制的合理性和有效性。
基于MAS的协同群集技术是为了进一步提升管控优化策略变化过程中的路网运行稳定性。基于协同群集控制理论将混合路网系统分解为多个要素Agent，将交通流底层模型及集中化协同管控优化模型分别置入对应功能的Agent中，最后通过各Agent间的博弈关系建立起基于MAS的混合路网协同群集模型。通过李雅普诺夫第二方法，论述了协同群集模型下的路网运行稳定性表现，同时讨论了不同网联车渗透率对于模型优化结果的影响。
鉴于大规模路网实际运行数据在设备布设、采集和处理上的技术难度以及在路网现场进行实际场景验证的限制，研究尚存在一定的不足和缺陷，如未能在混合路网实际运行场景下对相关管控策略和模型效果进行现场有效性分析、尚需探索更多元化和体现全局性的均衡指标来表征路网交通压力分布、要针对更大规模的混合路网设计对应的高效算法等，这些都是后续研究的重要内容。同时，混合路网道路交通协同管控平台的开发也是未来的重要发展方向。
关键词：混合路网，CAC问题，交通运行，角色可变，协同管控，最优化方法
With the development of expressway in metropolis of China, the mixed network, which consists of expressway systems and urban surface streets，tends to be the major component of urban road network. The mixed network plays a very important role in ease the urban congestion strain and raising the flow transfer ability of urban traffic throat. However, the fact of the heterogeneous traffic flow between expressways and surface streets, as well as the individual operation or non-synchronous coordination between the two road sub systems on the present control and management, leading to the limited performance of centralized dispersion and control for the traffic flow in the entire network. It is more likely to arise traffic congestion in the connection part of expressways and urban roads and its vicinity. The coming congestion and the serious congestion propagation will lower the operation efficiency of the entire network dramatically. Meanwhile, the present optimization logistic of control and management mainly relies on the authority-oriented format, in which travelers should receive the optimal strategy provided by the authority passively. The effect of users’ choice and response to the authority strategy in the holistic management and control system is neglected in such optimization framework. This study aims to lower the travel cost of the entire network, the related coordination idea is mainly based on the centralized framework of collaborative SoS (System of System). Combined with the goal of balancing the utilize of the road traffic resources, the integrated model which enable the level change procedure between the traffic authority and road users is established. Three key technologies including traffic state prediction, centralized integrated optimization and MAS (Multi Agent System) based collaborative clustering is proposed under the condition of mixed network. The designed model fulfills the synchronous coordination and optimization among the multi views including continuous flow and interrupted flow, traffic control and traffic assignment, authority and users, overcomes the limitation of un-cooperative control and management strategies for the mixed network.
Firstly, based on the meeting point between the structure of SoS and the topology of the mixed network, the system element of the entire mixed network is determined by the collaborative SoS. Based on the settled cooperation structure, the object, range, targe, restrictions of control and management are demonstrated, the technical obstacle of the mixed network on the degree of centralization,the hierarchy of the coordination and the corresponding modelling and problem solving, as well as bringing the insight into the nature of the cooperation problem in the mixed network, the integration optimization problem for the traffic flow in the complexed large scale road network is abstracted to the CAC (The Combined Signal Control – Traffic Assignment) problem in view of collaborative SoS.
Secondly, this study structures a complete optimization system of integrated control and management of the mixed network under the collaborative SoS based on the above abstraction. The proposed system consists of function and orientation, related control and management framework, supporting theory and technologies. The corresponding focus of the system function and orientation is put on guiding the identification and prediction of the traffic state and the congestion cooperative mitigating and management. The concerned framework set the communication mechanism among the three modules of traffic control,traffic assignment and network monitoring terminal in mixed network under CAC problem condition, and the relationship between the three key technologies to be studied and the travel cost, the flow pattern, the updating strategy is then determined in the set framework. The supporting theory and technology lay the foundation of modelling and optimization of the mentioned three key technologies.
Travel time is used as the efficiency index for the entire network to evaluate the total travel cost in traffic state identification and prediction. A hybrid model consisting of deterministic component, un-deterministic component and noise correction is put forward for the short-term travel time prediction. The prediction precision and robustness of the designed hybrid model are verified by the actual data in a real network. The result also proves the ability of the break-point capture of the traffic state.
The centralized collaborative clustering technology is proposed to solve the uncentralized problem of non-synchronous coordination between expressway and surface streets. The basic traffic flow model in a unified temporal and spatial dimension is reconstructed for the holistic mixed network in this study. Combined with the existence of both asymmetry and symmetry features between authority and user aroused in the game model, the relationship between these two stakeholders in view of collaborative SoS is clarified through the equilibrium index. Further, a level-change Stackelberg Games Model for the centralized cooperation model is established, the rationality and effectiveness of the proposed model is verified by two cases.
To further improve the operation stability of the entire network during the optimization process. The mixed network is firstly decomposed into several element agents through collaborative clustering control theory. Then the basic traffic flow function and the centralized integrated management and control optimization model are put into the corresponding functional agents respectively. Finally, the MAS based collaborative clustering model for the mixed network is formed through the gaming among all the agents. The model performance is discussed by the stability with Lyapunov Second Method. Meanwhile, the impact of the connected vehicle (CV) under different penetration rate to the model performance is also demonstrated.
Limited by the technical complexity in facility locating、data gathering and handling, as well as the field experiment conditions, there are some shortcomings in this study, especially for the failure of testing and analyzing the effects of proposed model and strategies in field,exploring more diversified and globalized equilibrium index and designing more efficient algorithm for the larger scale network, which might be potential topics in further research. Meanwhile, the platform development for cooperative control and management in mixed network is also a vital direction.
Key Words: mixed network, CAC problem, traffic operation, level change, cooperative control and management, optimization theory
