交通流时间序列的研究主要采用数据挖掘和机器学习的方法,这些"黑箱"挖掘方法很难直观反映序列特性.为增强交通流时间序列及其特征分析的可视化性,结合可视图理论来构建交通流时间序列的关联网络,从复杂网络角度实现交通流时间序列的特性分析.在网络构建的过程中,考虑到不同交通状态下交通流表征具有的差异性,首先利用交通流参量的相关性对交通流状态进行分类,然后构建不同交通状态下的时间序列复杂网络,并对这些网络的特征属性给出统计分析,如度分布、聚类系数、网络直径、模块化等.研究表明,可视图法可为交通流时间序列映射到网络提供有效途径,并且不同状态下交通流时间序列构建的复杂网络的模块化、聚类系数和度分布等统计特征呈现一定的变化规律,为交通流运行态势的研究提供了可视化的分析角度.%A traffic flow time series is a sequence of traffic detection parameters in chronological order. This differs from a general quantitative data sequence in that the time series includes a time attribute that contains not only the data with time characteristics, but also the distribution of the data itself. To date, studies of traffic time series have pri-marily adopted data mining methods consisting of data mining and machine learning methods—similar sequence search, dimension reduction, clustering, classification, pattern analysis, prediction, etc. In order to improve the visualization of traffic flow time series and feature analyses, a proposed method builds the association networks of traffic flow time series by using visibility graph theory. This approach differs from traditional traffic flow theory as it performs feature analysis of traffic flow time series from the perspective of complex networks, and then analyzes the relationship between the characteristics of the structure in the visual network and the state characteristics of the traffic flow. The proposed method also takes into account the different traffic flow time sequences that correspond to different traffic states. In the network building process using the proposed method, the traffic flow is classified by correlating the traffic flow parameters to the structure of the complex time series networks under different traffic conditions through considering the changes in traffic flow characteristics under various traffic conditions. Next, statistical analyses of the signs and attributes of the networks (e.g. degree distribution, clustering coefficient, network diameter, and modularization) are conducted. The analysis results show that the proposed visibility graph method can provide an effective approach to mapping traffic flow time series to the network. Moreover, the modularity, clustering coefficient, and degree distribution of the traffic flow time series networks in different traffic states show specifically varying patterns, providing a way to visually analyze the trends in traffic flow operation. When the traffic condition is at level 1, the distribution of the scattered points of the network conforms to a power law distribution. When the traffic condition is at any other level, the distribution of the scattered points of the network is consistent with a Gaussian distribution. The modularity of the time series network also shows some statistical characteristics, that is, the number of modules grows rapidly when the traffic state switches from smooth to moderate congestion, but decreases slowly when the traffic state switches from moderate congestion to serious congestion. These characteristics can be used to distinguish different traffic states, providing more perspective to understand different traffic scenarios. In this work we preliminarily study the attributes of traffic time series based on the proposed visibility graph method. Future efforts will continue to compare various methods of time series network construction to determine the pros and cons of each method for further analysis.
展开▼
