出租车OD序列聚类的城市功能区识别算法研究

Urban Functional Regions Recognition Algorithm for Taxi OD Series Clustering

提出一种由添加窗口约束的规整路径距离改进的K中心聚类算法(LDTW-K-medoids),通过构造时间序列、动态时间规整、时间序列聚类、聚类结果解读的流程来识别城市功能区。将算法应用在纽约市城市功能区识别中,对50万条出租车OD数据进行处理,将265个交通小区划分为五大功能区,并结合出租车全局流动模式、建筑物面积指数和富集指数来定量识别功能性质。与谷歌地图和卫星影像对比的结果表明,算法能够有效识别城市功能区,总体精度为83.8%。

In this paper,we provided a method that using warp path distance with window limits to improve K-medoids clustering algorithm(LDTW-K-medoids),which could identify urban functional regions by constructing orientation-destination(OD)series,using limited dynamic time warping,time series clustering and interpretation of clustering results.We applied the algorithm to identify urban functional regions in New York City,used 506625 taxi OD records to divide 265 traffic districts into five functional regions,and combined the global flow pattern of taxis,building area index and enrichment index to quantitatively identify the functional properties.Compared with Google maps and satellite images,the result shows that the algorithm can effectively identify urban functional regions,with overall accuracy of 83.8%.