基于生存分析的GPS轨迹缺失规律挖掘

近年来,智能交通系统(Intelligent Transportation Systems,ITS)已成为提高交通系统性能和增强出行安全性的有效方式。但随着系统数据量的增加,数据缺失问题日益严重,其中由于车载GPS信号丢失导致的轨迹数据缺失是主要的研究问题之一。引起GPS轨迹缺失的原因的多样性造成数据补全工作困难,且至今很少有关于轨迹缺失规律的研究。针对GPS信号丢失原因多样化的问题,基于大量真实数据,首次将生存分析应用于数据缺失领域,提出了基于生存分析的GPS轨迹缺失规律挖掘模型(Survival Analysis-Missing Trajectory Pattern Mining,SA-MTPM)。首先通过生存函数描述信号丢失时长与丢失原因的关系,然后利用Cox回归模型分析信号丢失的关键因素。使用上海市强生出租车公司一个月内13666辆车的数据进行实验,结果表明GPS轨迹缺失存在一定规律,据此可以方便地对信号丢失事件进行识别分类,为进一步对大数据进行研究提供了参考。

Pattern Mining of Missing GPS Trajectory Based on Survival Analysis

In recent years,intelligent transportation systems(ITS) has been an effective way to improve the traffic performance of transportation system and enhance the safety of travels.However,with the increase of data size in intelligent transportation system,the problem of data loss becomes increasingly serious.The trajectory data missing caused by vehicle-mounted GPS signal loss is one of the main research subjects.The reasons of GPS data missing are various,and they make the data completion difficult.However,there are few studies on the pattern of missing GPS trajectories.In this paper,based on large amounts of real data on diversification of GPS signal loss,the survival analysis was first applied into data missing field,and a survival analysis-missing trajectory pattern mining(SA-MTPM) model was proposed.The relationship between the length of signal loss and the regression causes of loss was described in the survival function,and the Cox model was used to analyze the key factors of signal loss.This paper performed experiments based on the GPS data of 13666 vehicles in Shanghai Qiangsheng Taxi Company for a month.The experimental results show that these signal loss events can be classified,which provides a further study for big data.