申请同济大学工学博士学位论文


城市快速路交通异常事件的检测与识别 方法研究

  国家高技术研究发展计划(863计划)专项课题 编号2007AA11Z245

 

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:张 淼

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

Study on Traffic Incident Detection and

Identification Method of Urban Expressway

  Supported by a grant from the National High Technology Research and Development Program of China (863 Program) No: 2007AA11Z245 Grant No: 50738004)

 

School: School of Transportation Engineering

Discipline: Transportation Engineering

Major: Transport Planning and Management

Candidate: Zhang Miao

Supervisor: Prof. Yang Dong-yuan

June, 2010

 

摘  要

城市快速路是城市路网的基本骨架,其正常运行对保证路网的运行效率有 着重要的意义。交通异常事件是指影响正常交通的偶然事件,本文通过调查发 现,异常事件对交通流产生影响的关键因素是持续占用道路资源,因此本研究 将交通异常事件定义为持续一定时间,占用车道资源,对交通流产生负面影响 的非常发性事件。由于快速路承担了大量中长距离出行,流量较大,异常事件 容易引发较大的交通影响,导致城市交通的整体运行效率降低;同时,由于城 市路网的复杂性,微小的扰动有可能对整个系统产生影响,追溯交通异常事件 也是对全路网进行研究的基础问题之一,因此研究异常事件的自动检测与识别 具有重要意义。

现有的大部分异常事件自动检测方法是基于国外高速公路的交通流特性开 发的,应用到城市快速路时存在着检测率低,误报率高,检测时间较长等缺 点,而且算法参数的标定流程通常较为复杂,影响其实用性。针对已有算法存 在的问题,本研究确定的快速路异常事件自动检测方法的总体设计要求如下: 1)检测算法的核心应当基于交通流理论,对中国城市快速路交通流进行建模, 使得自动检测算法适用于我国城市快速路的交通流状况;2)在保证检测率和误 报率的要求下,尽量缩短检测时间,为相关部门的快速响应创造条件,减小异 常事件对交通的影响;3)简化算法的标定流程,以利于检测算法在大规模路网 上的应用。

城市快速路的结构与高速公路类似但存在有许多不同,造成了城市快速路 的交通流特性与高速公路有一定差异。本文分析了城市快速路的结构,检测器 布设规律以及交通流的横向,纵向特性;并以上海市高架路为例,统计和分析 了城市快速路异常事件发生的规律。在此基础上,提出针对城市快速路进行设 计的交通异常事件检测算法需要注意的几个方面及具体设计指标:1)通过实际 调查发现,快速路异常事件的持续时间较短,大部分在 5 分钟以内,因此检测 算法需要对异常事件快速反应,平均检测时间应当明显小于 5 分钟;2)城市快 速路交通流随机波动比较大,而且背景交通流时常接近道路通行能力,交通状 况比较复杂,对算法的鲁棒性有较高的要求;3)城市快速路路段的结构差异导 致交通流特点的不同,检测算法建立的交通流模型应当考虑路段结构差异,分 别建模。

在确定了城市快速路异常事件自动检测算法的设计要求,分析了城市快速 路的特性的基础上,本研究根据城市快速路的结构特点,将其主线划分为 5 种 单元路段,即基本路段,车道数不守恒路段,上匝道路段,下匝道路段和交织 路段,分别建立了基于交通流密度的模型,提出了描述路段上下游交通流密度 差的变化规律的正常态公式作为异常事件检测的基础,此公式适用于正常情况 下的交通流,不满足此公式的情况即为异常事件检测的出发点。随后在此公式 基础上提出基于路段密度差模型的快速路异常事件检测算法,设计了异常事件 判断的逻辑流程。本研究提出的检测算法的参数标定不依赖于历史数据,参数 根据交通状况实时调整,克服了大多数现有检测算法在参数标定方面的弱点。

为了对所提出的检测算法性能进行综合评价,本研究利用上海市南北高架 交通异常事件的调查记录和感应线圈数据,选择目前实用性能较好的加州算法 和 DELOS 算法作为对比算法,进行详细的比较实验。实验对检测结果按照事 件的持续时间和事件发生路段类型进行分类统计分析,并给出了本文算法与对 比算法的检测率-误报率曲线。实验结果表明,本文提出的检测算法的性能显著 优于对比算法:在同样的误报率条件下,在所有路段类型上都具有更高的检测 率和更短的平均检测时间,达到了算法的设计要求;本文提出的算法省去了离 线参数标定的步骤,算法参数可以利用实时数据在线标定,有利于算法的在城 市范围内的应用。实验结果还从检测效果上验证了路段密度差的统计时间窗长 度的合理性。

提最后,总结了本文的研究结论并提出了今后的研究方向。

关键词: 城市快速路,交通异常事件,自动检测,密度差,单元路段

 

ABSTRACT

Urban expressway is the basis of urban transportation network, which has a significant effect on keeping the network operate efficiently. Traffic incident means random accidents that affect normal traffic state, like vehicle breakdown, traffic accident, load drop-off, lane occupation and so on. Since mid-long distance travel takes great part of urban expressway traffic and high travel speed plus large volume are key features of expressway traffic, traffic incidents are easily cause congestion that reduces the efficiency of the whole network. Also, traffic decision making analysis under information environment usually takes common traffic state into consideration, which means analyze common traffic rules with no interference of traffic incidents. Thus, research on automatic incident detection (AID) is very important and meaningful.

The existing AID methods are developed mostly based on foreign freeway, while their low detection rate, high false alarm rate, long detection time and complicated calibration are not quite fit for Chinese urban expressway. According to these analyses, this research concludes urban expressway AID should fulfill following requirements. ① Detection algorithm should be traffic flow theory based methods, and need to build models for better result on Chinese urban expressway. ② Minimizing detection time while under the required detection rate and false alarm rate, so that related team can have faster response to reduce incident interference on normal traffic. ③ Simplify the calibration procedure to help the proposed detection method application on massive network.

Urban expressway and freeway have different structures, resulted in different features of traffic flow. This research analyzed expressway structure, detector placement and cross-section feature and horizontal feature of traffic flow, and the expressway incidents occurrence pattern is statistically analyzed using Shanghai data. According to the analysis, the design of AID should take following aspects into consideration. ① After field investigation, the duration of expressway traffic incidents are shorter than freeway, most of which are within 5 minutes, so fast response is required, that is significantly less than 5 minutes. ② Compared to freeway, expressway has more random interference and back ground traffic flow is usually close to capacity, and proposed method should have high robustness. ③ Different segments have different feature, proposed method should build models separately.

After the above analysis, this research divided urban expressway into 5 kinds of segment according to their facility feature, that is, basic segment, unbalanced lane number segment, entrance ramp segment, exit ramp segment and weaving segment. 5 models are built for each segment unit based on traffic flow density, and a random variable equation of upstream and downstream density difference also established. This equation is the foundation of incident detection, reflecting the normal traffic state, while unsatisfied this equation means abnormal state; the unbalanced equation is the initial step of incident detection. Then the incident detection algorithm of urban expressway based on the proposed segment density difference model is proposed, and the logic determining incidents procedure is also provided. The calibration of variables is according to real-time data and traffic states, which overcomes the weakness of most existing algorithm.

And then experiments and tests are taken using Shanghai data for evaluating proposed method, test results are compared with two practical methods, California algorithm and DELOS algorithm. The comparing scenarios include incident duration, incident location and detection rate vs. false alarm rate. Results show that the proposed method has better performance than other two algorithms: at the same false alarm rate level, on all the types of expressway segment, proposed method has high detection rate, short detection time, and meet the design requirements; the proposed method does not need off-line calibration while can be calibrated online by real-time data, which is really helpful to application in cities. Meanwhile, the test also proved reasonable statistical time window for segment density model to get a better detection performance.

After all, the conclusion and future work are discussed in this research.

Key Words:  Urban Expressway, Automatic Incident Detection Method, Density Difference, Expressway Segment Unit