城市交通流数据优化感知关键技术研究

发布时间：2018-07-28 15:18

【摘要】：交通信息采集是智能交通系统的核心子系统,是交通应用的基础。通过先进的信息技术采集更高时空精度的交通流数据,并结合微观信号控制系统进行控制和诱导,均衡交通流在路网上的时空分布,是解决城市交通拥堵问题的关键。传统的感应线圈等交通监督技术只能检测固定点数据,实际应用中一般仅部署在干道的主要交叉口,路网上存在大量的信息“真空”,无法全面地感知交通流的动态变化,限制了信号控制系统的优化能力。近年来,移动互联网、传感网、车联网等新一代信息技术不断涌现,如果这些网络产生的数据与智能交通系统连接起来,将会为交通信息采集开辟新的技术途径。研究一种精度高、实时性好、维护成本低、适应大数据时代的交通信息采集技术,具有十分重要的意义。本文以城市交通大数据为背景,研究了城市交通信息采集中的一些优化问题。论文的创新性工作包括以下几个方面。第一,在单点数据采集方面,研究了基于无线传感器网络的交通流参数采集优化模型和算法。无线传感器网络可以进行大规模部署,在智能交通系统中具有很好的应用前景。本文在伯克利大学P. Varaiya等人提出的自适应阈值检测算法的基础上,针对阈值更新缓慢及分类算法未考虑车辆长度等问题,提出了一种基于信号相关性的车辆速度测量算法和一种基于邻接传感器网络的车辆分类算法。提高了车辆速度估计和车辆分类的精度,并且在阂值漂移、叠加干扰等条件下也具有较好的精度和鲁棒性。第二,研究了群体参与式感知在交通信息采集中的应用,提出了可以采集路段交通流数据的拉格朗日感知算法。该方法利用传感器数据来求解交通方程对交通流的内在的运行规律进行预测,同时使用参与式感知数据作为观测值,基于卡尔曼滤波算法综合交通方程和实际观测数据对交通流参数进行最优估计,获取连续的、具有更高时空精度的交通流数据。在此基础上,提出了道路的堵塞因子,对交通拥堵状况进行实时度量,并应用到路口交通信号配时优化场景中,结合粒子群优化对信号相序进行优化,达到避免交通拥堵形成的目的。第三,研究了参与式感知中的数据集选择优化问题。已有研究成果表明,相对于数据的数量,提供的数据所在的位置对于交通流估计的结果有更大的影响。在大规模的城市路网中,参与式感知的数据体量非常巨大,如何在大量数据中区分出数据价值、选择最优数据集是一个重要的问题。本文研究了给定传感器可选位置条件下的数据集选择优化问题,采用互信息熵作为目标函数、以均方根误差为约束条件建立了传感器数据集选择的多目标优化模型,提出了一种基于贝叶斯优化解决传感器数据集序贯选择的算法。第四,针对基于车联网和车载终端的参与式感知中传感器节点随着交通流移动的特征,研究了交通流变化及网络的拓扑时变带来的动态不确定性。本文采用时变网络模型对移动传感器网络的动态拓扑及数据价值的时变不确定性进行建模,基于传感器节点的数据效用定义了时变价值网络,并采用蚁群优化进行传感器数据集的并行优化选择。此外,针对传感器节点的移动性和交通流数据的时变特性,提出一种基于Internet的传输控制协议,使控制节点可以感知交通流模式变化并选择最优价值的数据,同时对传感器节点的数据传输进行反馈和控制优化。
[Abstract]:Traffic information collection is the core subsystem of the intelligent transportation system, and it is the foundation of traffic application. It is the key to solve the traffic congestion problem by collecting the traffic flow data with higher temporal and spatial accuracy through advanced information technology and combining the micro signal control system to control and induce the traffic flow on the road network. Traffic supervision technology such as induction coil, such as induction coil, can only detect fixed point data. In practical application, it is generally only deployed at main intersection of the main road. There is a lot of information "vacuum" on the road network. It can not fully perceive the dynamic changes of traffic flow, and limit the optimization ability of the signal control system. In recent years, mobile Internet, sensor network, car Federation The new generation of information technology, such as network, is constantly emerging. If the data generated by these networks are connected with the intelligent transportation system, it will open up a new technical way for the traffic information collection. It is of great significance to study a kind of traffic information collection technology with high precision, good real-time, low maintenance cost and adapt to the age of large data. This paper is based on the city. In the background of city traffic data, some optimization problems in urban traffic information collection are studied. The innovative work of this paper includes the following aspects. First, in the single point of data acquisition, the optimization model and calculation method of traffic flow parameter acquisition based on wireless sensor network is studied. In this paper, based on the adaptive threshold detection algorithm proposed by P. Varaiya and others in Berkeley University, this paper proposes a vehicle speed measurement algorithm based on signal correlation and a neighborhood based on the slow updating of threshold and the length of the vehicle. The vehicle classification algorithm of sensor network improves the accuracy of vehicle speed estimation and vehicle classification, and has good accuracy and robustness under the conditions of threshold drift and superposition interference. Second, the application of group participatory perception in traffic information collection is studied, and rag Lang, which can collect traffic flow data of sections, is proposed. The method uses the sensor data to predict the internal running rules of traffic flow by using the sensor data, and uses the participatory perception data as the observation value. Based on the Calman filtering algorithm, the traffic flow equation and the actual observation data are optimized to obtain the continuous and higher time. The traffic flow data of empty precision. On this basis, the congestion factor of the road is proposed, and the traffic congestion is measured in real time. In the optimization scene of traffic signal timing in the intersection, combined with particle swarm optimization to optimize the phase sequence of the signal, the purpose of avoiding traffic congestion is achieved. Third, the data in the participatory perception are studied. The existing research results show that the location of the data provided has a greater impact on the results of traffic flow estimation than the number of data. In the large-scale urban road network, the participatory data volume is very large. How to distinguish the value of data in a large number of data and select the best data set is one. An important problem. In this paper, the selection and optimization of data sets in the optional position of a given sensor is studied. Using the mutual information entropy as the objective function and the mean square root error as the constraint condition, a multi-objective optimization model for the selection of sensor data sets is established. A new method is proposed based on Bayesian optimization to solve the sequential selection of sensor data sets. Fourth, in view of the characteristics of the sensor nodes in the participatory perception of vehicle networking and vehicle terminal, the dynamic uncertainty caused by the change of traffic flow and the time-varying topology of the network is studied. In this paper, the time-varying network model is used to change the dynamic topology and the data value of the mobile sensor network. The time-varying value network is defined based on the data utility of sensor nodes, and the parallel optimization selection of sensor data sets is carried out by ant colony optimization. In addition, a transmission control protocol based on Internet is proposed to enable the control nodes to be aware of the mobility of sensor nodes and the time-varying characteristics of traffic flow data. The traffic flow pattern changes and selects the best value data, and feedback and control optimization for data transmission of sensor nodes.
【学位授予单位】：大连理工大学
【学位级别】：博士
【学位授予年份】：2015
【分类号】：U495;TP212.9;TN929.5

【参考文献】