一种基于手机信令的通勤OD 训练方法

针对基站定位精度低、信令采样间隔长、轨迹不连续的手机信令，提出一种职 住及通勤OD(origin-destination)计算框架.对用户单日手机轨迹按时间排序，标识每个轨 迹点的进出时间及停留时间，剔除长距离漂移轨迹点，对邻近轨迹点进行空间聚合.将全 天划分为多个时窗，叠加用户多日轨迹，计算稳定指数并识别用户在各时窗内的多日稳 定点.综合工作日与节假日稳定点判断用户居住地、工作地.采用基于常住人口的扩样方 法，对街道通勤OD矩阵进行扩样.模型结果与重庆主城常住人口分布、2014 年居民出行 调查结果吻合.

A Commuting OD Matrix Training Method Based on Mobile Phone Data

A methodology framework for training people’s residence and work place is presented to overcome defects in extracting people's activity pattern from mobile phone data like low precision positioning accuracy, irregular trajectory interval. First, each user’s daily trajectory is restructured to add the label of access time, departure time and delay time for each point. Spurious points are distinguished by given time and spatial threshold and then be clustered to filter out noise resulting from mobile service provider. Then, stable point for each time window is recognized by overlaying multi-days trajectory to get the most frequent location. User’s residence and workplace are determined by comparing stable points in workdays and holidays. Finally, commuting trips are constructed for each user between home and workplace. These trips are multiplied by expansion factors based on the street population. Analysis results of this methodology are in good agreement with the pattern revealed by the 2014 Census population and the 2014 Person Trip Survey in Chongqing, China.