城市路网交通流分析预测及事故预警方法研究

发布时间：2018-03-19 22:36

本文选题：城市道路交通网络　切入点：空间相关性　出处：《北京交通大学》2017年博士论文　论文类型：学位论文

【摘要】：近年来,机动车、驾驶人数量及道路交通流量保持着迅猛增长的趋势,给人们生产生活带来便捷的同时,也带来不容忽视的安全隐患。分析人、车、路、环境等多方面因素对交通事故风险的影响和作用机理,构建高效的交通事故预警模型,制定行之有效的城市道路交通主动安全保障措施,为居民的出行保驾护航已成为亟待解决的问题。交通流因素是影响城市道路交通事故风险的关键因素之一,利用交通流数据,研究交通流动态特征对交通事故风险的影响,是对城市道路交通事故进行精准研判和提高整个城市道路交通系统安全水平的主要途径。此外,面向城市路网深入分析和掌握交通流变化规律,提高交通流预测的实时性、可靠性和自适应性是目前关注和研究的热点,为交通事故预警提供重要的数据保障。本文按交通流特性分析—交通流预测—交通事故预警的逻辑层次展开研究,重点解决如何分析城市路网交通流空间相关特性、如何利用交通流空间互相关特性实现城市路网多断面交通流短时预测以及如何利用交通流因素进行交通事故预警等问题。本文针对这些主要问题进行了深入研究,并通过场景实例或数值实例对相应的理论方法进行了验证,主要研究成果具体体现在以下几个方面:(1)基于复杂网络理论提出了一种体现路段交通流空间互相关性的城市道路交通网络建模和分析方法。利用复杂网络理论对城市道路交通系统进行建模,用以体现路网中路段交通流互相关分布的复杂性。将城市道路交通系统表示为一个由系统内路段上的交通流空间相互关联而生成的网络。其中,网络的节点代表路网中路段,节点之间的边是否存在取决于节点对上交通流序列空间相关的程度。针对构建的城市道路交通网络,提出一种考虑地理权重的PageRank算法(GWPA),用于确定路网中路段的重要度,为交通流相关性空间聚类的划分提供支持。(2)从复杂网络社区发现角度探索路网中路段交通流空间相关性分布规律,提出了一种基于GWPA-k-means的城市路网交通流空间相关性分析方法。该方法从两方面对传统的k-means算法进行改进:针对初始种群选择问题,提出了一种改进的基于密度峰值的聚类中心选取算法,利用节点的GWPA值和最短路径长度确定初始种群;针对节点相似度矩阵构造问题,基于节点GWPA值,提出了一种加权的信号传递方法,用于测量网络中节点的相似性,为k-means聚类提供依据。实验结果表明,GWPA-k-means方法能较好地揭示城市路网中路段交通流空间互相关模式。(3)基于交通流空间相关特性,提出了一种基于社区发现和长短期记忆神经网络的多断面交通流短时预测方法。方法首先根据GWPA-k-means算法划分的路段交通流空间互相关模式将路网划分为若干个区域,位于同一个区域内的路段之间交通流空间相关性较强;然后针对每个区域路网,以路段交通流序列构成的二维矩阵序列为输入,利用长短期记忆神经网络对选取交通流数据进行时空特征学习,进而实现多断面交通流短时预测。针对预测模型参数设置问题,提出了一种基于自适应正交遗传算法的模型参数优选算法。实验结果表明,考虑城市路网交通流空间相关性可以提升交通流预测精度,该方法在传感器故障造成数据缺失情况下适应性良好,具有较强的鲁棒性。(4)以交通流数据为基础,提出了基于交通流因素的城市道路交通事故预警方法。在实时获取交通流状态的基础上,分析交通流因素对交通事故风险的影响,并分别从交通事故检测和交通事故风险预测两方面对交通事故预警方法进行研究。以一种域划分的角度去阐明交通流变量与交通事故风险的关系,给出了交通安全域的概念;提出了一种基于序列向前选择和主成分分析的特征提取算法,用于提取影响交通事故风险的显著特征变量;提出了一种基于最小二乘支持向量机的安全域估计方法,用交通安全域实时检测交通事故。此外,基于实时获取的交通流数据,将交通安全域和可靠性分析理论相结合,提出了一种交通可靠性模型,该模型可同时对交通事故风险进行宏观的统计评价和面向单个交通事故的实时风险预测。
[Abstract]:In recent years, the number of motor vehicles, and road traffic driver to maintain rapid growth trend, bring convenience to people's production and life, but also brings security risks can not be ignored. Analysis of people, vehicles, roads, and many factors affecting the environment influence on traffic accident risk mechanism, construction of the road traffic accident, early warning model the development of effective active city road traffic security measures for the residents travel escort has become a serious problem. The traffic flow is a key risk factors of city road traffic accident factor, using traffic flow data, the dynamic characteristics of traffic flow on the impact on traffic accident risk, is the main way to accurate judgments the road traffic system and improve the security level of the whole city for city road traffic accidents. In addition, the city road network analysis and grasp the traffic flow The law, improve the real-time traffic flow prediction, reliability and adaptability is the hot topic of research, provide an important guarantee for the data traffic accident warning. Research on the logical level according to the traffic flow characteristics of traffic flow forecasting traffic accident warning, focus on how to analyze the spatial correlation characteristics of city traffic flow and how to use the traffic spatial cross-correlation realization of city road network multi section traffic flow forecasting and how to use the traffic flow of traffic accident early warning and other issues. This paper studies the main problems in the process, and the corresponding theory and method are verified by numerical examples or instances of the scene, the main research achievements embodied in the the following aspects: (1) the complex network theory put forward a reflection of road traffic flow spatial correlation of city road based on Through the network modeling and analysis method. By modeling the city road traffic system based on complex network theory, used to reflect the complexity of road section traffic flow correlation distribution. The city road traffic system is a system composed of sections of the traffic flow space correlation generated network. The network node represents road section, the existence of edges between nodes depends on the node of traffic flow on the spatial correlation degree. According to the sequence of city road traffic network construction, this paper presented a geographical weighted PageRank algorithm (GWPA), is used to determine the importance of road section, to provide support for the division of traffic flow correlation of spatial clustering. (2) from the complex network community discovery of road section traffic flow distribution spatial correlation angle, this paper proposes a GWPA-k-means based on city traffic space The correlation analysis method. The method was improved from the two aspects of the traditional K-means algorithm for the initial population selection problem, proposed an improved algorithm based on clustering center density, the initial population was established using the node GWPA and the length of the shortest path problem; for node similarity matrix is constructed, based on node GWPA value and put forward a weighted signal transmission method for measuring similarity of nodes in the network, to provide the basis for k-means clustering. The experimental results show that the GWPA-k-means method can better reveal the city road section traffic spatial correlation model. (3) based on the spatial correlation characteristics of traffic flow, proposes a community discovery and long term memory neural network multi section traffic flow forecasting method based on GWPA-k-means algorithm. The method according to the classification of road traffic flow spatial cross-correlation mode will The road network is divided into several regions, located between the same area of road traffic flow strong spatial correlation; then for each regional network, a two-dimensional matrix sequence formed by road traffic flow sequence as input, selection of traffic flow data using the spatial and temporal characteristics of learning and long-term memory neural network, so as to realize the multi section traffic flow short term forecasting. According to the prediction model parameters, this paper puts forward a model of adaptive parameter optimization algorithm based on orthogonal genetic algorithm. The experimental results show that the city traffic flow spatial correlation to enhance traffic flow prediction accuracy, this method resulted in good adaptability in case of missing data in sensor fault, has strong robustness. (4) with traffic flow data as the foundation, proposed the city road traffic accident early warning method based on the traffic flow factors. In real-time traffic flow. State on the basis of analysis of influence factors on the traffic flow of traffic accident risk, and separately from the traffic accident detection and traffic accident risk prediction on the two aspects of early warning methods of traffic accidents. The relationship in a domain division perspective to clarify the traffic flow and traffic accident risk variables, gives the concept of traffic safety domain; put forward a feature analysis based on sequence forward selection and principal component extraction algorithm for feature variables extraction significant traffic accident risk; proposes a security domain least squares support vector machine estimation method based on real-time detection of traffic accident, traffic safety domain. In addition, the real-time traffic flow data based on the traffic safety domain and reliability analysis theory, puts forward a traffic reliability model, the model can also to the traffic accident risk assessment and macro statistics Real-time risk prediction to a single traffic accident.

【学位授予单位】：北京交通大学
【学位级别】：博士
【学位授予年份】：2017
【分类号】：U491

【参考文献】