基于大货车GPS数据的轨迹相似性度量有效性研究

目前,中国货车上全球定位系统(GPS)的强制安装,使得利用包含时间、空间和速度等信息的货车轨迹数据来研究货车运行模式成为可能。基于距离的轨迹相似性度量算法,采用全国道路货运车辆公共监管与服务平台获取的货车GPS轨迹数据,对比分析其在货车轨迹模式识别中的应用。选用文献中最常用的4种基于距离的轨迹相似性度量算法,分别为离散弗雷歇距离(DFD)、动态时间规整(DTW)、最长公共序列(LCS)和实序列编辑距离(EDR)。试验结果表明:当使用二维地理空间轨迹数据(即经度和纬度)时,4种基于距离的轨迹相似性度量算法都能很好地对相似轨迹进行分类(正确率均高于85%),这与现有文献的结论一致。虽然一般认为二维轨迹相似性算法可以直接应用到多维轨迹数据,但是解决具体问题时可能出现的误差以及各种轨迹相似性算法的适用性仍然不确定。目前几乎没有文献对三维及其以上的多维轨迹数据进行实例分析研究,因而,通过相同路线上的三维GPS货车轨迹数据(包括经度,纬度和速度)对4种基于距离的轨迹相似性度量算法进行验证。将第3维速度加入到二维空间轨迹上后发现LCS算法对基于地理空间轨迹的速度模式分类效果优于其他3种基于距离的轨迹相似性度量算法。这说明运用LCS轨迹相似性度量算法来识别基于三维GPS轨迹的货车运行模式是可行的,LCS算法在货车运营管理等方面将有很大的应用潜力。

Effectiveness of Trajectory Similarity Measures Based on Truck GPS Data

The widely enforced installation of global positioning system (GPS) devices on trucks enables the study of truck travel patterns through a large number of trajectory data, including data on time, location, and speed. In this study, truck travel patterns were clustered using trajectory similarity measures. Four distance-based trajectory similarity measures, i.e. the discrete Frechet distance (DFD), dynamic time warping (DTW), longest common subsequence (LCS), and edit distance on real sequence (EDR), were used. These measures were applied to the three-dimensional GPS trajectory data-latitude, longitude, and speed-and were produced by trucks traveling along the same route repeatedly. The results show that all the four similarity measures are capable, with correction rates greater than 85%, to find similar trajectories on the same route when two-dimensional geospatial (i.e. latitude and longitude) trajectories are used. This is consistent with the conclusions in the literature. Although it has been claimed that the similarity measures are directly applicable to multidimensional trajectory data, potential issues, such as possible errors, and the applicability of different methods, are still uncertain. There have been few studies on similarity measure methods for high-dimensional trajectories. Therefore, the trajectory similarity methods were classified and verified based on three-dimensional GPS trajectory data (i.e. longitude, latitude, and speed). When the third dimension of speed was added to the spatial dimensions to obtain three-dimensional trajectory data, it was found that the LCS method is the most suitable trajectory similarity measure among the four measures to distinguish speed patterns on the basis of the geospatial trajectory. The results indicate that it is feasible to identify the driving behavior patterns through three-dimensional GPS trajectory similarity measures. This method has great potential for application in tasks such as the operation and management of truck fleets.