基于贝叶斯网络的稀疏出租车GPS轨迹路径还原方法

Bayesian network-based GPS path restoration for sparse taxi trajectories

为提高出租车GPS大数据的可用性,提出一种基于贝叶斯网络研究稀疏出租车GPS轨迹路径还原的方法.与传统仅基于时空变量的研究方法不同,新算法同时考虑天气条件、驾驶员特性、车辆行驶特性与出租车的载客状态等因素来进行路径还原预测.以宁波市体育中心周围的路网为例,将出租车服务信息管理平台的GPS轨迹数据作为测试对象,验证本文方法的适用性.结果显示,基于多因素的贝叶斯网络方法在还原精度方面(达到91.4%)优于Logit选择模型.此外,新算法尤其适用于出租车轨迹数据缺失率较高的场景,比如缺失轨迹点跨度在5 min左右.

In order to improve the availability of taxi GPS big data,a method based on Bayesian network for sparse taxi GPS path restoration is proposed.Different from the traditional research that is only based on spatiotemporal variables,the algorithm takes into account the weather conditions,driver characteristics,vehicle driving characteristics and taxi load status to calculate path restoration prediction.The applicability of the presented method is verified by taking the road network around Ningbo sports center as an example combined with GPS trajectory data collected from the taxi service information management platform for the testing purposes.The case study results show that the Bayesian network method based on multi factors is superior to the Logit selection model in restoration accuracy(up to 91.4%).In addition,the algorithm is especially suitable for the situation with high missing rate of taxi track data,such as the track points of about 5-minute missing span.