【作者】 姚智胜；

【导师】 邵春福；

【作者基本信息】 北京交通大学 ， 交通运输规划与管理， 2007， 博士

【摘要】 2006年我国开始实施的“国家中长期科学与技术发展规划纲要(2006-2020)”中的交通科学问题研究专题中,明确提出了“发展一个系统,解决三个热点问题”的思路。其中,“一个系统”指的是综合运输系统,“三个热点”指的是交通能源和环境、交通安全和大城市交通拥堵。智能交通系统被认为是缓解道路交通拥堵、减少汽车尾气排放污染和交通事故等交通问题的有效方法之一。短时交通流预测是智能交通系统的核心内容和交通信息服务、交通控制与诱导的重要基础,能够给出行者提供实时有效的信息,帮助他们更好地进行路径选择,实现路径诱导,达到节约出行者旅行时间,缓解道路拥堵,减少污染、节省能源等目的。目前,道路交通数据采集设备的性价比越来越合理,道路上交通数据的采集设备不断完善,使得短时交通流状态的分析处理和预测成为可能。本论文以基于实时数据的道路网短时交通流预测理论与方法研究为题,在交通流数据分析的基础上,对道路网中多个断面交通流状态之间的相关性进行分析,从而选择预测的范围和对象,对道路网中多个断面的短时交通流预测理论和方法进行了深入探讨和研究,并且根据实际数据验证提出的预测模型。本论文首先阐述短时交通流预测的概念、基本流程、特点及要求,按照单一断面的短时交通流预测方法和多个断面的短时交通流预测方法两个方面综述了国内外短时交通流预测相关研究。第二,分析道路网中多个断面交通流的数据,介绍了包括删除错误数据和补充缺失数据的交通流数据预处理方法,解决数据来源质量问题,以此为基础,分析道路网中多个断面交通流数据两两之间的相关系数矩阵,应用多元统计分析中的多维标度法将道路网中多个断面交通流的相关程度映射到一张二维拟合构图,从而判别断面之间相关性的强弱,选择出道路网研究范围。第三,提出了基于状态空间模型的道路网多断面短时交通流的预测模型和方法,采用状态空间模型和时间序列分析相结合,应用EM算法估计参数。第四,建立了基于支持向量机回归的道路网多断面短时交通流的预测模型和方法,在选择研究范围的基础上,利用最小二乘支持向量机回归算法建立模型,将回溯系数的选择过程和利用遗传算法优化支持向量机回归模型参数选择的过程组合起来进行了预测模型参数的交替组合优化。第五,对道路网中多个断面的短时交通流数据进行了混沌时间序列分析,进行了多维交通流时间序列数据的相空间重构,建立了基于混沌时间序列分析的道路网多断面短时交通流预测模型,利用粒子群优化算法优化模型的参数选择。根据城市快速路多个断面交通流数据,对提出的各类道路网多断面短时交通流的预测模型,验证其有效性。本论文的主要研究成果包括:1、在交通流数据相关性分析的基础上,将道路网多个断面短时交通流作为一个整体来同时预测,基于交通流数据的相关性分析进行了道路网多断面短时交通流预测理论和方法的研究。2、引入多元统计分析中的多维标度法,将道路网中不同空间的多个断面短时交通流的相关性表示为一张二维拟合构图,通过二维拟合构图可以判别道路网中多个断面交通流的整体相关性,从而为下一步建立考虑时间和空间变化的多断面短时交通流预测的相关模型与方法界定了研究范围、研究对象和数据分析基础。3、根据目前短时交通流预测理论的性质,分别从线性系统理论、人工智能方法、非线性系统理论三个角度将短时交通流预测从单一断面预测方法向道路网多断面同时预测方法扩展:(1)在选择研究范围的基础上,结合状态空间模型和时间序列分析,应用EM算法估计参数,提出了基于状态空间模型的道路网多断面短时交通流的预测模型和方法;(2)在选择研究范围的基础上,利用最小二乘支持向量机回归算法建立模型,将回溯系数的选择过程和利用遗传算法优化支持向量机回归模型参数选择的过程组合起来进行了预测模型参数的交替组合优化,提出了基于支持向量机回归的道路网多断面短时交通流的预测模型和方法;(3)在选择研究范围的基础上,对道路网中多个断面的短时交通流数据进行了混沌时间序列分析,进行了多维交通流时间序列数据的相空间重构,建立了多维混沌时间序列预测模型,利用粒子群优化算法优化模型的参数,提出了基于混沌时间序列分析的道路网多断面短时交通流预测的模型和方法。更多还原

【Abstract】 Special research on traffic science of "National Guideline on Medium-term and Long-term Program for Science and Technology Development(2006-2020)" implemented from 2006 clearly put forward Concepts of "develop a system, solve three hot spots". A system indicated comprehensive transportation system while three hot spots indicated traffic energy and environment, traffic safety as well as traffic congestion of the metropolis. Intelligent transportation systems are considered as one of effective methods to relieve the urban traffic problems such as traffic congestion, air pollution caused by motor vehicle exhausts and traffic accidents. Short-term traffic flow forecasting is not only a core element of intelligent transportation systems but also an important base of traffic information service, traffic control and guidance which can supple travelers with efficient information and help them to choose a optimal path so as to perform path guidance, to save travel time of travelers, to relieve traffic congestion, to reduce air pollution and to save energy. As the development of traffic science and technology, automatic vehicle identification devices set on the roads has improved constantly, which make it possible for us to perform short-term traffic flow analysis and forecasting. Theories and methods of short-term traffic flow forecasting of road network based on real-time data were studied in the dissertation. On the base of traffic flow data analysis, relevance of traffic flows of several road cross-sections in different spots in a road network was considered so as to choose range and object for forecasting. Then theories and methods of short-term traffic flow forecasting of several road cross-sections in a road network were discussed and investigated. Moreover, real-time data were used to test the efficiency of the proposed forecasting model.Firstly, concept, process, characteristic and request of short-term traffic flow forecasting were introduced and researches on short-term traffic flow forecasting up to now were summarized into two kinds of methods which were single road cross-section short-term traffic flow forecasting and multiple road cross-sections short-term traffic flow forecasting. Secondly, traffic flow data were analyzed. In order to solve the problem of data source, preprocessing techniques of traffic flow data were adopted include deletion of erroneous data and supplement of absent data. Matrix for the correlation coefficients of road cross-sections in a road network was examined. Then multidimensional scaling from theory of multivariate statistical analysis was used to associate the relevance of traffic flows data from road cross-sections with a two dimensional chart of derived stimulus configuration. According to the chart, the strong or weak of the relativity of each road cross-section can be distinguished so as to choose the range of road network under research. Thirdly, a short-term traffic flow forecasting model of multiple road cross-sections in a road network was proposed based on state space model in which state space model was combined with time series analysis. EM algorithm was applied to estimate the parameters of the proposed model. Fourthly, a short-term traffic flow forecasting model of multiple road cross-sections in a road network based on support vector regression was proposed. After choosing research range, a forecasting model was set up on the base of algorithm of least squares support vector regression. The process of selecting go-back dimension was combined with the process of using genetic algorithm to select parameters of support vector regression to perform optimization of parameters of the proposed forecasting model. Fifthly, after chaotic time series analysis and phase space reconstruction of multidimensional short-term traffic flow data, a short-term traffic flow forecasting model of multiple road cross-sections in a road network based on chaotic time series analysis was put forward. The parameters of forecasting model was optimized with particle swarm optimization algorithm. Meanwhile, the forecasting models and methods of multiple road cross-sections in a road network proposed in the dissertation were tested by real-time traffic flow data from urban expressway and the results were satisfying.The followings are the main conclusions of this dissertation.1. Based on traffic flow data analysis, traffic flows of multiple road cross-sections in a road network are considered as a whole to perform short-term traffic flow forecasting and theories and methods of short-term traffic flow forecasting of multiple road cross-sections in a road network based on traffic flow data analysis were studied.2. Multidimensional scaling from theory of multivariate statistical analysis was used to associate the relevance of traffic flows data from road cross-sections with a two dimensional chart of derived stimulus configuration. According to the chart, the strong or weak of the relativity of each road cross-section can be distinguished, which supplied research range, object and base of data analysis to short-term traffic flow forecasting models and methods considering the variety of time and space of traffic flow.3. According to characteristic of theories of short-term traffic flow forecasting up to now, traffic flow forecasting was extended from single-cross-section forecasting approach to multi-cross-sections forecasting approach from the aspects of linear system theory, artificial intelligence methods and nonlinear system theory: (1) On the base of choosing research range, state space model was combined with time series analysis to set up a short-term traffic flow forecasting model of multiple road cross-sections in a road network based on state space model. Moreover, EM algorithm was applied to estimate the parameters of the proposed model. (2) On the base of choosing research range, a short-term traffic flow forecasting model of multiple road cross-sections in a road network based on support vector regression was proposed. Algorithm of least squares support vector regression was used to set up the model. The process of selecting go-back dimension was combined with the process of using genetic algorithm to select parameters of support vector regression to perform optimization of parameters of the proposed forecasting model. (3) On the base of choosing research range, a short-term traffic flow forecasting model of multiple road cross-sections in a road network based on chaotic time series analysis was put forward on the base of chaotic time series analysis and phase space reconstruction of multidimensional short-term traffic flow data as well as the optimization of the parameters of forecasting model with particle swarm optimization algorithm.更多还原