结合轨迹数据的居民出行模式及功能识别研究

Study on travel pattern and function identification based on trajectory data

城市人群的出行特征通过车辆轨迹数据隐含的行为信息可以体现,但传统的单维度模型将不再适用于轨迹数据隐含的多维信息的挖掘。本文将海口市中心城区根据路网划分区域,使用能够挖掘多维信息的非负稀疏约束下张量分解基于“滴滴出行”轨迹数据,从时空维度挖掘居民出行规律并进行区域功能特征识别。结果表明:居民出行时间符合工作日,休息日的早高峰、日间、晚高峰、夜间的出行时间模式;居民出行空间包含6种上/下车出行空间模式,同时发现时间模式与空间模式之间存在交互联系,即居民在不同时间模式下的不同出行空间模式都有不同的上下车冷热点活动,基于该交互联系可以识别出空间功能特征,同时在此基础上基于兴趣点(point of interest,POI)功能识别结合分析,不仅表明居民出行行为识别空间功能特征的可行性,也提高了功能识别结果的准确性,可以为城市管理者解决城市问题如交通拥堵、功能布局失衡等做出决策提供帮助。

The travel characteristics of urban population can be reflected by the behavior information implied by the vehicle trajectory data,but the traditional single dimensional model is no longer suitable for the mining of the multidimensional information implied by the trajectory data.In this paper,the downtown area of Haikou city was divided into regions according to the road network.Based on the trajectory data of"Didi Chuxing",the non-negative sparse constraint tensor decomposition was applied to mine the multidimensional information,and the residents'travel rules were introduced from the space-time dimension and the regional functional features are identified.The results showed that the travel time of residents conformed to the modes of morning peak,day peak,evening peak and night travel on working days,rest days,and residents travel space contained six kinds of on/off the travel space pattern.It was also found that there was interaction relationship between model of time and space,namely the residents travel mode in different time different spatial patterns had different getting cold hot spots.The interaction among identify space feature kept consistent with the POI recognition results and improved the accuracy.It also demonstrated that there were certain behavior patterns of residents'travel,and it was feasible to identify spatial functional characteristics according to residents'travel behavior,so as to provide help for city managers to solve urban problems and make decisions.