摘要
随着城市化的快速发展,城市空间结构愈发复杂,城市功能区的快速有效识别对资源的有效配置和城市规划具有重要意义。传统的功能区识别缺乏对居民这一城市空间活动主体的动态表征,而长时间序列的出租车数据能动态表征居民出行行为,进而反映城市空间结构。动态时间扭曲(DTW)距离比传统的欧氏距离更能有效挖掘高维数据,泛化后的LB_Keogh距离和LB_Hust距离相继克服了DTW距离时间复杂度高和不对称的缺点。为了探究基于时间相似性度量的聚类算法在识别城市功能区方面的可行性,首先基于OpenStreetMap(OSM)路网数据获取研究单元,再通过滴滴订单数据提取上下车点、构建研究单元内的居民出行时间序列,然后利用PAM算法结合4种相似度度量方法进行聚类,最后结合兴趣点(POI)数据识别城市功能区,并对结果进行精度验证。结果表明,基于LB_Hust距离的PAM算法能有效挖掘高维时间序列数据,应用于城市功能区识别的精度高达86%,为应用时间序列数据进行城市研究提供了一种新的方法。
With the rapid development of urbanization,the urban spatial structure is becoming much more complex,the rapid and effective identification of urban functional areas is of great significance to the effective allocation of resources and urban planning.Traditional functional area identification doesn’t take residents’ activities into consideration.Long-term serial taxi data can dynamically characterize residents’ travel behavior,and then reflect urban spatial structure.In the case of high-dimensional data mining,the dynamic time warp(DTW)distance has a great advantage over the traditional Euclidean distance.However,it has the disadvantages of high time complexity and asymmetry.These shortcomings are respectively overcome by LB_Keogh distance and LB_Hust distance.To explore the feasibility of clustering algorithm based on time series similarity measure in urban functional area identification,we used the road network data of OpenStreetMap(OSM) to divide the study area,mapping the origins and destination of taxi order data into the units and counting the number by hour separately to constructs the travel time series of residents.Then,we combined four distances with PAM clustering algorithm.Finally,we used Points of Interest(POI) data to identify urban functional areas,and verified the accuracy of the functional zoning results.The result shows that PAM algorithm based LB_Hust distance can be applied to urban functional area identification,and the recognition accuracy of functional area is up to 86%,which can provide a new method for functional area identification based on other time series data.
出处
《地理空间信息》
2021年第1期22-29,47,I0005,共10页
Geospatial Information
基金
国家自然科学基金资助项目(41771432)。