Journal of Transportation Systems Engineering and Information Technology

Journal of Transportation Systems Engineering and Information Technology ›› 2016, Vol. 16 ›› Issue (4): 178-184.

Previous Articles     Next Articles

Gaussian Mixture Model for Urban Link Travel Time Analysis

LI Rui-min1, QIAN Xiao-dong2, WU Hong-bin3   

  1. 1. Institute of Transportation Engineering, Tsinghua University, Beijing 100084, China; 2. Department of Civil Engineering, University of California, Davis, CA 95616, USA; 3. Traffic Management Research Institute of the Ministry of Public Security, Wuxi 214151, Jiangsu, China
  • Received:2016-01-04 Revised:2016-04-26 Online:2016-08-25 Published:2016-08-26

城市道路旅行时间高斯混合模型研究

李瑞敏*1，钱小冬2，武红斌3   

  1. 1. 清华大学交通研究所，北京100084；2. 美国加州大学戴维斯分校土木工程系，美国95616；3. 公安部交通管理科学研究所，江苏无锡214151
  • 作者简介:李瑞敏（1979-），男，山东莱州人，副教授，工学博士.
  • 基金资助:

    国家自然科学基金/ National Natural Science Foundation of China（71361130015）；北京市自然科学基金/Beijing Natural Science Foundation（8162024）

PDF

Abstract:

Travel time of urban roads is an important parameter for urban road transportation system. Based on the travel time data for signalized arterials derived from the matching of the vehicle license plate number identified from the license plate recognition system, this paper investigates the distribution estimation method for urban road travel time using Gaussian Mixture Model (GMM), and compares the difference of GMM, Normal, Log Normal and Weibull distributions and analyses the influence of the number of normal distribution in GMM on fitting performance of GMM. The result shows that for the data used in this study, the GMM can fit the signalized arterial link travel time distribution properly. Furthermore, the proper number of normal distribution in GMM is suggested to be 2 or 3 and more normal distributions in GMM don’t have substantial influence on travel time distribution fitting. This paper can provide sufficient support for the travel time reliability study for urban signalized arterials and others.

Key words: transportation engineering, travel time, Gaussian Mixture Model, expectation maximization algorithm

摘要：

城市道路旅行时间是城市交通系统的重要表征参数.基于利用车牌识别系统所检测到的高样本率的车牌数据进行匹配获得信号控制路段的旅行时间数据，本文应用高斯混合模型研究了城市道路旅行时间分布的估计方法，对比了高斯混合模型与正态分布、对数正态分布、Weibull 分布模型的差异，并在此基础上分析了高斯混合模型中密集峰的数量对拟合结果的影响.结果表明，针对本文应用的数据，城市信号控制路段的旅行时间可以用高斯混合模型进行良好的表征，恰当的密集峰值往往是2 个或3 个，更多的密集峰值数量并无实质性影响.研究结果可以为城市信号控制道路旅行时间可靠性等的进一步分析提供良好的支撑.

关键词: 交通工程, 旅行时间, 高斯混合模型, 最大期望算法

CLC Number: