基于凝聚层次聚类的域内交通流周期模式发现  

Method of discovery of periodic pattern in traffic flow based on agglomerative hierarchical clustering

作　　者：田嫦丽 赵霞[1] 于重重[1] 吴子珺[1] 

机构地区：[1]北京工商大学计算机与信息工程学院,北京100048

出　　处：《计算机应用研究》2015年第8期2261-2264,2271,共5页Application Research of Computers

基　　金：北京市自然基金重点项目B类(KZ201410011014)

摘　　要：域内交通流量数据是典型时序数据,对其进行周期模式挖掘能够促进区域交通资源调配,减少区域交通拥堵。针对原始交通数据存在潜在周期性的特点,提出一种基于凝聚层次聚类的域内交通流量周期模式发现方法——AHC→DOPP(agglomerative hierarchical clustering→discovery of periodic pattern)。该方法改进了动态时间弯曲技术以完成原有凝聚层次聚类算法初始邻近性距离矩阵的计算工作,并形成以Ward方法为核心的簇合并邻近度更新方法的凝聚层次聚类框架。采用AHC→DOPP方法对一段时间内的日交通流量时变图进行聚类,能快速得到层次聚类结果。运用该方法并结合所采集的交通数据集进行了相应的实验,得到了明显的域内交通流量周期模式。Domain traffic data is typical timing data, the mining of traffic flow periodic pattern can contribute to its regional transportation resource allocation and reduce regional traffic congestion. For the potential exists periodic pattern for the original traffic data, this paper proposed a discovery method periodic pattern of traffic flow based on agglomerative hierarchical clustering： AHC→DOPP（ agglomerative hierarchical clustering→discovery of periodic pattern）. This method improved the dynamic time warping （DTW） technology to complete the calculations of initial proximity distance matrix in original agglomerative hierarchical clustering algorithm and formed an agglomerative hierarchical clustering framework in which Ward method was the core of the cluster merger proximity update methods. Using AHC→DOPP method to cluster the daily traffic variation data in a period of time can quickly get results. A series of experiments on the collected traffic data show that it can get obvious periodic pattern of domain traffic by using this method.

关 键 词：时间序列 周期模式发现 凝聚层次聚类 交通流量 

分 类 号：TP391.7[自动化与计算机技术—计算机应用技术]