【作者】 王亚琴；

【导师】 朱扬勇；

【作者基本信息】 复旦大学 ， 计算机软件与理论， 2007， 博士

【摘要】 研究交通流的各种形态及其运行规律,建立快速、稳定、高效的交通流模型是智能交通系统的重要研究内容。随着智能交通系统的发展,智能交通系统中积累了海量交通流数据,于是研究者开始研究利用先进的数据挖掘技术分析智能交通系统中的交通流信息,发现交通流信息中隐含的交通模式及规则。本文针对交通流信息的特点以及智能交通系统的新的数据挖掘应用需求,对交通流数据预处理、交通流量预测、交通状态识别、交通流空间聚类以及实时交通流的查询等若干问题进行了研究,设计了适合的数据挖掘模型和算法。这些问题的研究对于智能交通系统的交通信号管理与控制、交通流诱导、动态交通分配等方面有着重要的意义。本文的主要研究内容和成果包括以下几个方面:(1)智能交通系统是一个非常庞大的系统,其复杂性和稳定性使交通流数据的采集质量难以保证,对交通流数据进行异常检测及预处理对于后续的数据分析、挖掘结果的质量和预测的准确性具有重要意义。本文根据交通领域的流量—时间占有率的倒“V”字型曲线模型,提出了一种基于曲线拟合的交通流异常检测方法,利用三次多项式的最小二乘法拟合流量/时间占有率曲线,并且利用分箱的思想对拟合好曲线上下部分分别采用基于统计的方法划分上下界,有效的识别异常交通流数据。(2)道路网络上运行的交通流具有不同的空间分布模式,如城市主干道的交通流具有“线”性模式、繁华路段的交通流具有“面”状模式等,根据交通流运行的空间分布特性,对城市道路交通网络进行实时、动态的交通区域划分是当前智能交通系统的研究热点之一。利用聚类分析方法对分布在道路网络空间中的环形感应线圈检测器检测的交通流数据进行空间聚类(Spatial Clustering)分析,使具有相似性质且具有空间关联性的交通流数据对象聚成一类,可以发现道路交通流的空间分布模式。本文基于凝聚的层次聚类算法思想,设计了一个高效的交通流空间聚类算法ESCA-TF(Efficient Spatial Clustering Algorithm ofTraffic Flow),自底向上的生成道路交通流的空间聚集类。ESCA-TF无需执行复杂的空间连接和空间合并操作,实验证明具有良好的时间效率。(3)对道路交通流状态进行分析研究,及时、准确地识别和预测道路交通流的状态是智能交通系统实现动态交通管理的重要前提。交通流状态的识别和预测包括交通流量短时预测和交通状态的实时识别。对于路口短时交通流量预测,本文提出了基于二次聚类的交通流量序列分割和BP神经网络的组合模型的路口短时交通流量预测方法,实验证明基于二次聚类和BP神经网络组成的组合模型提高了神经网络模型的预测精度;对于道路交通状态的实时识别,本文提出了基于聚类分析的交通状态动态识别模型,基于该模型我们不需任何先验知识就可以识别道路交通状态,且具有较高的拥挤判别率和较快的判别反应时间。(4)随着微电子技术、无线通信、移动定位技术的发展,在智能交通系统中,许多具有普适计算功能的移动装置(如PDAs、cellphone及各种GPS装置)可以跟踪人或车的实际位置,获取和传输与用户位置相关的各种有用信息,因此对于道路网络上的移动交通流提供基于位置(Location-Based Services,LBS)的服务也是当前交通信息化和智能交通系统的一个研究方向。Skyline查询提供了一种重要的基于位置服务的功能,本文设计了道路网络上移动对象的skyline连续查询算法。算法分为两个部分:独立查询点的Skyline查询算法RNASQ(Absolute Skyline Query)和Skyline连续查询算法RNCSQ(Continuous SkylineQuery)。RNASQ算法无需计算所有对象到查询点的网络距离,具有较好的时间效率。在RNASQ算法的基础上,本文提出了道路网络的上Skyline连续查询算法RNCSQ。在RNCSQ算法中,Skyline连续查询转化为对查询路径的顶点和查询对象与查询点距离的交叉点的有限个独立查询,可以快速地判断连续分段的分割点,有效地计算Skyline连续查询的连续分段。(5)建立统一、开放、可扩展的智能交通系统数据挖掘平台是交通流数据挖掘研究的重要内容。本文提出了一个四层的ITS数据挖掘平台体系结构,主要划分为:数据层、数据挖掘算法工具层、分析逻辑层和应用系统层。这种层次的系统应用平台模式便于数据挖掘算法、分析功能的设立,方便数据挖掘系统的开发与配置,可以使用户轻松地根据实际应用的需要使用数据挖掘技术,基于此结构的数据挖掘系统具有良好的可扩展性及与实体的独立性,便于二次开发。在可扩展的智能交通系统数据挖掘应用平台体系架构的基础上,本文设计实现了一个基于SOA技术的智能数据挖掘平台UTDD(Urban TrafficData-Mining Development),实现了本文提出的交通流数据挖掘方法。更多还原

【Abstract】 The research and application of Intelligent Transportation System has developed rapidly due to the demand on safe, convenient, comfortable and information-based modern transportation. It is important part of the research of Intelligent Transportation System to study different forms and operation rules on traffic flow and establish rapid, stable and effective traffic flow model. With the development of Intelligent Transportation System, mass traffic flow data have been accumulated in Intelligent Transportation System. More and more researchers have started to analyze the information of traffic flow by use of advanced data-mining technique, and discover hidden transportation mode and regulation amongst the information of traffic flow.This paper has made the research on several questions such as traffic flow forecasting, traffic state identification, traffic spatial clustering and real-time inquiry on traffic flow etc. in light of the characteristics of the information of traffic flow and the application demand on new data-mining of Intelligent Transportation System. The research on these questions is of great significance to traffic signal management & control, traffic flow induction, dynamic traffic allocation of Intelligent Transportation System. In general, the main contents and achievements of this paper consist of the following aspects:(1) Based on the characteristics of the transportation and the classic flow-occupancy inverse "V" model, implement polynomial fitting using least-squares algorithm and statistics method on flow curves to detect outliers which are proved to be not accord with practice through the actual implement, then use the moving average model to recorrect the outliers and absent.(2) The traffic flows operated in road network have different distribution models in space, for example, the traffic flow in urban main roads has "line" model and that in busy downtown area has "plane" model etc. it is one of current research issues on Intelligent Transportation System to divide the urban road traffic network into dynamic real-time traffic areas according to the distribution features of operated traffic flow in space. This paper introduces the Spatial Clustering analysis method on traffic flow data from the loop induction coils arranged in road network space to collect the traffic flow data with similar characteristics and spatial relevancy into one category and discover the distribution models of road traffic flow.(3) It is an important content of traffic flow data-mining to forecast up-to-date and accurate short-term traffic flow. The crossroad is intersected by several roads, which is critical component of road network and plays important role in the whole urban road transportation network. The research of short-time traffic flow forecasting in the crossroad may assist to optimize the real-time control and traffic flow induction on road transportation. The neural network model is an important classification forecasting model and different kinds of neural network models have been used for forecasting the short-term traffic flow in road transportation. This paper points out the forecasting method of short-term traffic flow in the crossroads based on relevance analysis and sequence partition by means of BP neural network to increase the accuracy of traffic flow forecasting.(4) For mobile traffic objects in road network, we design an algorithm for the skyline continuous query. The algorithm is divided into two parts: RNASQ (Absolute Skyline Query) and RNCSQ (Continuous Skyline Query). For RNASQ, it is unnecessary to calculate the network distance of all objects from query point, which has good time efficiency. On the basis of RNASQ, we point out RNCSQ (Continuous Skyline Query) in road network. For RNCSQ, continuous Skyline query is transformed into limited independent inquiry among inquiry routine apex and the point of intersection between inquiry object and inquiry point to help judge the division points of continuous segments rapidly and calculate consecutive segments of continuous Skyline query.(5) It is an important content of traffic flow data-mining research to establish unifying, open and extendable data-mining platform of Intelligent Transportation System. This paper introduces a four-layer system structure of ITS data-mining platform, which consists of data layer, data-mining algorithmic tool layer, logical analysis layer and application system layer. This system application platform model is served to establish the data-mining algorithm & analysis function and facilitate the development & configuration of data-mining system so that the clients may utilize the data-mining technique easily based on the needs of practical application. The data-mining system based on this structure has good extensibility and entity independence to facilitate the secondary development. On the basis of extensible system structure of Intelligent Transportation System data-mining application platform, we design an intelligent data-mining platform UTDD (Urban Traffic Data-Mining Development) based on SOA technique to realize the traffic flow data-mining method in this paper.更多还原