【作者】 文孟飞；

【导师】 彭军；

【作者基本信息】 中南大学 ， 交通信息工程及控制， 2013， 博士

【摘要】 随着城市化进程的加快和汽车工业的发展,现有城市道路的通行能力与不断增长的交通需求之间的矛盾变得日益尖锐,交通拥挤现象日益突出。通过使用智能交通系统对交通流进行控制和诱导,以缓解交通拥堵,提供畅通和有序的交通环境,是目前各个国家大力发展和应用的重要手段。本文以大规模的城市复杂路网作为研究背景,采用协同控制与学习机制、模糊控制、动态重规划等理论和方法,研究智能交通系统中相序切换、区域负载均衡、交通流全局优化、车载传感网络的信息收集与处理、交通诱导任务的分解与最优路径选择等问题,建立一套全面高效的智能交通系统的信息收集、信号控制和自主诱导的管理和通行机制。主要研究工作如下：论文首先针对交通路网的连续车流和离散交通灯信号建立能够描述交通车流量的智能交通模型,然后分析了城市智能交通的控制与诱导的技术难点,针对城市智能交通系统的特点和应用背景给出了城市智能交通系统协同优化的控制与诱导设计方案。在该方案中,车载无线网络节点和道路上安装的感应节点采集车流和路况信息,若干车载无线网络节点经过转发和中继收集实时交通信息,送到交叉路口控制器,然后利用收集的数据提出基于置信水平的自适应粒子滤波算法,对道路交通流量进行预测,并以此为依据进行交通信号的协同控制和自主交通流的协同诱导。为了给交通控制和交通诱导提供实时、充分和可靠的动态路况信息和交通车流信息,利用车载无线节点和感应装置构成车载无线传感器网络采集数据,针对车载无线网络的移动特性并考虑节点能耗,提出了基于同心均衡分簇的数据收集算法,以均衡网络负载,减少网络延时,降低节点能耗,提高数据传输的可靠性；针对车载传感器网络的感知特性,在自适应粒子滤波算法的基础上,利用置信区间和估计状态的方差,提出一种基于置信水平的自适应粒子滤波算法,剔除冗余粒子数,实时获得下一时刻的最少粒子数目,从而降低算法实现的复杂度和运算量,对交通流量进行在线实时准确的估计,提高车载传感器网络交通流预测的准确性。为平衡大规模城市交通的负载,并达到交通通行性能的全局最优,提出一种具有智能学习能力的交通信号的协同控制策略。首先考虑车流量密度的非连续性,给出城市交通道路的分段仿射车流量优化指标,利用模糊规则进行交通信号的相序切换,然后利用系统辨识得到分段线形参数；再利用相邻交叉路口的流量信息,构造协同控制项,以平衡区域负载；为实现流量的优化,引入模糊Q学习机制,根据交通网络的状态,同时把粒子滤波得到的状态预测值作为学习及时回报,实时调整协同控制增益和本地反馈控制增益,由协同控制得到优化的信号灯控制命令,提高交通控制的实时性,得到全局最优的交通优化指标。具有学习能力的协同控制策略可以通过各个交通控制节点自身的反馈控制行为和邻接交通控制节点之间的协同控制行为达到路网的流量最优。为了提高出行效率,论文研究基于自主形式的智能交通诱导,提出一种动态不确定环境下交通诱导分解协调和路径选择的方法。首先引入时序约束的带权与或树对大规模交通诱导任务进行描述,然后将带权与或树修剪转变为AOE网,在AOE网的基础上将复杂交通诱导任务分解,并考虑诱导任务的时序约束,从而增强算法对大规模路网的可扩展性,以保证大规模交通诱导任务执行的一致性和动态实时要求；再提出一种增量动态重规划方法求解诱导问题的最优路径集合,先采用逆向多目标启发式搜索进行全局规划,然后以增量的方式对全局规划所保留的部分信息有效地重用,可快速调整变化位置与目标位置之间新的移动路径。所提出的方法具有较大的可扩展性,并能满足交通诱导的实时性要求。更多还原

【Abstract】 With the accelerated urbanization and the development of the automobile industry, there is the increasing requirement for the traffic throughout of the urban in contrast to the current urban road capacity. The traffic congestion problem has become serious gradually for many metropolises. Thus it is an important and extensive method in many countries to use the intelligent transport system to control traffic and to induce vehicle flows so that the road congestion may be mitigated and the traffic may be more efficiency. Taking the large-scale urban complex traffic net as research background, the paper adapts the theories and methods such as cooperative control and optimization, fuzzy control and dynamic re-planning to research the critical problems in intelligent transport system such as the signal phase switch, region traffic balance, global traffic optimization, the data acquisition and the data process for vehicle sensor network, the optimal path planning and task decomposition of the traffic guidance and so on. By this research, the paper tries to construct a completed and efficient traffic management and passing mechanism. The main work is as follows:The paper firstly constructs the intelligent transport model to analyze the traffic flow of the transport network including continuous vehicle flows and discrete traffic signal. Then the difficulty and problem of the control and guidance in urban intelligent transport system is analyzed based on the characteristics and the application background of urban transport network. To solve these problems, a design framework of cooperative control and guidance is presented. In the design, the vehicle wireless nodes collect the transport information of sensor nodes along the urban roads. Then a new confidence level based adaptive particle filter algorithm is proposed to predict the short-term flow of traffic using the collected data. After it is transferred or relayed by vehicle wireless nodes, which can be used for cooperative traffic signal control and traffic guidance.In order to provide the real-time, fully reliable information about dynamic traffic report and vehicle flows for the traffic control, a data collection algorithm based on the concentric trees by using vehicle wireless nodes considering the mobility and the power efficiency. This algorithm can balance the networks load, reduce the power consuming of the nodes and increase the reliability for data transmitting. Then a confidence-level-based new adaptive particle filter (CAPF) algorithm is proposed in this paper to increase the prediction accuracy of traffic flow. In this algorithm the idea of confidence interval is utilized. The least number of particles for the next time instant is estimated according to the confidence level and the variance of the estimated state. CAPF can effectively reduce the computation while ensuring accuracy of online real-time and accurate estimates of traffic flow and improving the accuracy of the traffic flow forecasting.The cooperative control strategy with the intelligent learning ability was proposed to induce the urban traffic load and optimize global traffic property. Considering the discontinuity of the vehicle flows density, the paper proposes the affine traffic optimization index for the subsection of the urban traffic road and made the signal phase switch by using fuzzy rules, then obtain the subsection linear parameter. Secondly the traffic information of the adjacent Crossroads was used to construct the cooperative control items and balance the area load. Finally the global optimal traffic index is obtained by introducing the fuzzy Q-learning mechanism to adjust the gains of the cooperative control and the local feedback control in order to achieve optimism of traffic. In this paper, the learning reward is derived from the estimated value of traffic flow. The optimal cooperative control strategy with learning can obtain optimal traffic flow through the feedbacks of the traffic control nodes and the cooperative control acting of the neighbor nodes.Apart from research about controlling the urban vehicle flows by using traffic signal, this paper also analyzed an intelligent Traffic stream guidance, discussed the key function of traffic distribution path planning and finally proposed a way to address the traffic guidance decomposition and path selection issue. First, the large-scale traffic stream guidance is described by introducing WAndOrTree. Then the WAndOrTree is turned into AOE network. On the basis of AOE network, the traffic stream guidance is decomposed, considering the timing constraints of the guidance task, which can ensure the consistency of the task execution and meet dynamic real-time requirement for task planning. Furthermore, a group of optimal path set was obtained by using reverse multi-objective search. Then the Traffic stream guidance reused the remaining unchanged planning information with incremental way. By this way, the move path between the current position and destination location can be adjusted faster. The proposed traffic guidance is more scalable for large-scale city with real-time requirement.更多还原