申请同济大学工学博士学位论文


城市快速路拥堵瓶颈及态势的识别与分析

 

 

培养单位:交通运输工程学院

一级学科:交通运输工程

二级学科:交通运输规划与管理

研 究 生:弓晋丽

指导教师:杨东援 教授

二○一○年六月

 

A dissertation submitted to

Tongji University in conformity with the requirements for

the degree of Doctor of Philosophy

The Identification and Analysis of Traffic Bottleneck and

Traffic State Evolution on Urban Expressway

 

 

School: School of Transportation Engineering

Discipline: Transportation Engineering

Major: Transport Planning and Management

Candidate: Gong Jinli

Supervisor: Prof. Yang Dong-yuan

June, 2010

 

摘  要

目前,城市交通拥堵问题已成为全球大城市面临的共同问题。发生在城市快速路上的交通拥堵如果不及时消除将可能导致路网大面积拥堵,因此,快速路拥堵的研究和治理变得尤为重要。近年来,随着信息采集与处理技术的不断进步,城市快速路交通监控系统采集获取了大量道路交通数据,为识别、评价和监测快速路交通拥挤提供了可能。如何利用这些大量的信息,增加人们对交通拥挤规律的认识,成为一个具有重要研究意义的问题。论文以城市快速路交通拥挤时空分布特性为目的,以数据挖掘技术和宏观交通流模型为方法,建立了交通瓶颈系统识别及交通拥挤态势模式辨识与监测的理论与方法,并实际应用于城市快速路上以检测线圈数据为信息来源的拥挤分析中,有效地得到了快速路拥挤空间集中分布位置和异常模式集中分布时段,具有支持快速路交通拥堵疏导措施决策的价值,可为提高道路网运行效率,改善人们出行条件提供参考依据。

对已有的交通状态分类与判别方法,交通状态常用描述指标进行了总结,分析了各个方法和指标的适用性。结合快速路基本交通参数特性分析,证实了快速路交通运行系统的随机性和不确定性,进一步指出传统使用单一交通参数阈值进行状态判别的不足。为此,构造了包含6个指标的状态特征集,采用模式识别中状态特征选择方法,确定了分线圈数据样本各自的最佳状态描述特征组合。从理论和实际相结合的角度提出了快速路交通状态划分理论,并在此基础上构造了基于FCM的交通状态分类器,实现聚类结果与交通状态的有效对应。以上海南北高架东侧43个线圈为例,发现存在3种不同类型的流量-占有率关系图,指出仅需对其中第一类型和第二类型进行交通状态判别。最后,采用Kappa一致性非参数检验方法,验证了判别结果与交通状态划分理论的一致性。

介绍了宏观CTM模型近年来出现的两种扩展模型:MCTM和ACTM。从理论角度剖析了不同路段交通状态模式下由于交通信息传播方式差异导致了路段 “可观测性”的不同,并结合路段交通状态模式划分方法,提出了分模式CTM模型。该模型更新了CTM模型中流量传输模型,将其表示为不同路段交通状态模式下的分段函数形式,并针对城市快速路特点构建了下匝道流量传输模型。以上海南北高架东侧DX02-DX05间共1302m长的路段为例,测试了3种CTM模型,结论为分模式CTM模型计算结果误差低于平均水平,性能最好。

根据实地观察,归纳了快速路上3种常见的固定瓶颈类型。对应用较广泛的两种瓶颈拥挤识别方法进行了总结,并借鉴其中“阈值法”中基于交通参数等高线图识别瓶颈拥挤的思想,开发了基于交通状态时空变化图的“瓶颈拥挤自动判别算法”。该算法集成了交通定性状态转化、瓶颈拥挤时空识别和拥挤严重性评价功能,实现从车公里长度、车小时长度、延误、通行能力损失4个方面进行瓶颈拥挤的评价。为实现瓶颈拥挤排序,以多元统计分析为方法,进行瓶颈拥挤评价指标的“降维”,并据此确定了“瓶颈拥挤度”值,将其作为瓶颈拥挤排序的依据。根据排序结果最终获得了快速路瓶颈拥挤的空间集中分布位置。

对交通拥挤态势问题进行了描述,将问题归结为3个子问题的研究,即:拥挤态势指数的确定、拥挤态势模式的辨识、异常模式的检测。从定性和定量两个角度分别定义了6个拥挤态势描述指标,并按时间顺序将它们排列形成了交通拥挤态势描述指标时间序列。以R/S分析为方法,验证了多个指标间的时间序列趋势一致性,找到了既能从定性角度又能从定量角度进行综合评价的指标作为拥挤态势指数。系统梳理了时间序列分割方法,开发了应用于交通拥挤态势模式辨识的分割算法。以算法运行结果为基础,将时间序列表示为5元组的分段线性形式。在此基础上,构造了5种常见的时间序列形状相似性距离,比较了它们在交通拥挤态势模式分离中的适用性。选取其中效果最好的模式距离与欧式距离组合,以凝聚分层聚类为方法,确定了模式辨识最终结果,并以其中拥挤最严重类为比较基准,提出了基于距离的“异常因子”概念,实现从形状和幅度值两方面检测异常模式。设计了基于滑动窗技术的异常模式实时检测算法,其核心思想为当滑动窗内时间序列异常因子超过阈值即认为有异常模式存在,此时需报警指示交通运行系统机制发生了变化。以上海南北高架东侧2009年9月30日检测线圈数据为例,监测得到了当天异常模式集中分布在中午12:10-13:20、13:40-14:30和下午17:10-17:15时间段内。

关键词: 城市快速路,交通状态,交通瓶颈,分模式CTM模型,交通拥挤态势,时间序列,异常检测

 

ABSTRACT

Nowadays, the traffic congestion has become a common problem in the metropolises of the world. If it is unable to relieve traffic congestion on the city expressway efficiently, the larger traffic jam will ultimately emerge in city road network as a result. Thus it is significant important to study and to manage the congestion on urban expressway. In these years, with the progress of information collection and processing technology, there are large amounts of traffic data acquired by city monitoring systems, which made identifying, accessing and monitoring traffic congestion possibly. How to make use of the information to increase the knowledge on congestion has become a meaningful research. The main purpose of this paper is to study the spatial-temporal distribution characteristics of traffic congestion on urban expressway. Analyzed by data mining and macroscopic traffic flow model, the theories and methods for bottleneck systematic diagnosis and traffic state evolution are established and are applied to congestion analysis based on real Loop Detector data from urban expressway. The spatial distributed locations of congestion and concentrated periods of outlier patterns are obtained, which provide a support for decision of traffic congestion relief measure and for improving city road network’s efficiency and people’s travel conditions.

The traffic state classification and identification methods,and various indicators described the congestion were summarized. With the characteristic analysis of traffic parameters of urban expressway, the randomness and uncertainty of traffic operation system were proved. It is uncovered that using a single threshold for traffic state identification is not accurate. Therefore, six indicators were proposed to present a set of state features. The best combination of state features is selected using feature selection algorithm in Pattern Recognition. Through combination of theory and practical point of view, the traffic state classification theory is put forward. The classifier based on FCM is designed to identify the traffic state according to the theory. Taking forty-three Loop Detectors on the eastern of Shanghai North-South expressway as examples, it is show that there are three different types in flow-occupy relationship diagram. The traffic state identification should be only carried out on the first and second type,and their identification results are examined by Kappa nonparametric test, which represent the result is consistent with traffic classification theory.

Two new models of CTM: MCTM and ACTM are introduced, and then Pattern CTM is proposed based on the analysis about road traffic states’ observability under different transmission modes of traffic information which are decided by road traffic pattern. In Pattern CTM, the flow transmission model on main-road is updated to be expressed as the form of piecewise function and the flow transmission model on off-ramp is constructed against the characteristics of urban expressway. Both three CTM models are tested on road sections from DX02 to DX05 of Shanghai North-South expressway’s eastern side, which show that the performance of Pattern CTM is the best of all as its calculation errors are below average.

Three general types of static bottlenecks on urban expressway are summarized based on the actual observation, and two popular methods for bottleneck identification are compared. Drawn on the idea of one of the bottleneck identification method: ‘Threshold method’ which identified the bottleneck based on traffic parameters contours, the automatic identification algorithm for bottleneck on urban expressway is designed based on traffic state spatial-temporal changes. Three functions--transforming traffic qualitative state, identifying spatial-temporal scope of congestion, evaluating the congestion severity--are integrated in the algorithm, and those realize evaluate the congestion severity by four indicators: Vehicle Distance Traveled, Vehicle Hours Traveled, Delay and Productivity Loss. After reduced four indicators dimension reduced through multivariate statistical analysis, the Level of Congestion Severity for Active Bottleneck is acquired, and the spatial distributed locations of congestion are determined according to it at last.

The traffic state evolution problem is divided into three sub-problems: determining the index, pattern discrimination and outlier detection. Six indictors are defined from the two aspects of qualitative and quantitative, and they are arranged to form traffic state evolution time-series according to chronological order. Using Rescaled Range Analysis, the consistence of these indicators time-series is tested and the synthetic indicator which can evaluate the status from two aspects of qualitative and quantitative, is found as Traffic State Evolution Index. The segmented algorithm for traffic pattern discrimination is designed based on the summarization about existed algorithms of time-series data mining. Based on the segmentation result, the time-series is represented in piecewise linear form, and then time-series’ five shape-similarity distances are defined, whose performance on discriminate of different traffic state evolution patterns are tested through Cohesion Hierarchical Clustering. It is concluded that the Pattern distance is the best of all. Take the combination of the Pattern distance and Euclidean distance as similarity measurement methods, the pattern discrimination result is determined. Take the most serious class of the result as the benchmark, the Outlier Factor based on distance is defined to detect outlier pattern from its shape and values. Using it as the key technology, the outlier pattern real-detecting algorithm based on sliding windows is designed. The central idea of this algorithm is that: if the Outlier Factor of time-series during sliding windows exceeds the threshold, the alarm should be activated to indicate that traffic operation mechanism has changed. Take the Loop Detector data from the eastern of Shanghai North-South expressway at September 30, 2009 as an example, it is concluded that the outlier pattern are distributed at 12:10-13:20, 13:40-14:30 and 17:10-17:15.

Key Words:  Urban Expressway, Traffic State, Traffic Bottleneck, Pattern Cell Transition Model, Traffic State Evolution, Time Series, Outlier Detection