【作者】 刘爽；

【导师】 田玲；

【作者基本信息】 电子科技大学 ， 计算机应用技术， 2017， 硕士

【摘要】 随着定位技术精度的提高和各种车载传感器的广泛使用,越来越多的车辆轨迹数据被记录下来。这些海量的轨迹数据中蕴含着大量的有价值信息,因此,如何从这些海量数据中挖掘出符合一定场景的有价值信息是很有必要的。而车辆轨迹数据不同于一般的轨迹数据,它被限定在受约束的路网中,因此,常用的轨迹数据挖掘方法往往难以直接应用在车辆轨迹挖掘中。基于此,本文提出了轨迹时空相似性度量算法GTTSD和基于K-均值优化DBSCAN聚类算法,并设计实现了车辆轨迹数据综合挖掘系统,具体工作如下:(1)轨迹时空距离度量算法GTTSD。该算法在进行轨迹时空相似性度量时,首先通过动态网格划分法将传统基于欧式空间的路径表示转换为空间网格表示,随后根据轨迹序列上的断点进行轨迹分割,最后将分割后的子轨迹进行时空相似度计算。此算法既可以降低轨迹数据的存储开销又避免了相同轨迹数据在欧式空间中因采样时间不一致而造成的轨迹相似性度量出现偏差。通过实验证明了该度量方法相比DTW、LCSS、G_LCSS具有更高的执行效率和更为准确的度量结果。(2)基于K-均值优化的DBSCAN算法。针对传统DBSCAN聚类算法结果受人为设定参数值影响较大的弊端,本文提出了基于K-均值优化的DBSCAN算法。该算法首先通过K-均值聚类算法对数据对象进行快速聚类,并将聚类结果进行统计分析得出DBSCAN算法中初始邻域半径和邻域密度阈值的取值,基于此,对数据对象进行DBSCAN聚类并在聚类过程中动态调整邻域半径的取值。通过实验证明了该算法对轨迹数据具有较好的聚类效果。(3)车辆轨迹数据综合挖掘系统。该系统集轨迹数据提取、轨迹时空距离度量、轨迹聚类、路网热点区域挖掘功能于一体,可根据用户需求不同对轨迹数据进行不同操作,具有较强的实际应用价值。更多还原

【Abstract】 With the improvement of the precision of GPS and the widely use of various vehicle mounted sensors,more and more vehicle trajectories data are recorded.There is a large amount of valuable information in the massive trajectories data,so it is necessary to mine valuable information from the large amount of data.The vehicle trajectories data are different from the general trajectories data,which is limited in the constrained road network.Therefore,the commonly used trajectory data mining methods are often difficult to apply directly to the vehicle trajectory mining.Based on this,this thesis proposes a trajectory space-time similarity measure algorithm GTTSD and K-means optimization DBSCAN clustering algorithm,and implements a traffic hotspot area discovery system.The specific work is as follows1.Grid-Based trajectory time-space distance.When the trajectory spatiotemporal similarity measure is carried out,the traditional method based on the continental space is transformed into the spatial grid representation by the dynamic meshing method.after that,the trajectory segmentation is performed according to the breakpoint on the trajectory sequence.Finally,the sub-trajectories of the segmented are calculated by spatiotemporal similarity.This algorithm can reduce the storage cost of the trajectory data and avoid the deviation of the similarity between the same trajectory data in the European space due to the inconsistency of the sampling time.the experimental results show that the algorithm can obtain accurate trajectory space-time distance and high execution efficiency.2.DBSCAN algorithm based on K-means optimization.DBSCAN algorithm based on K-means optimization is proposed for the traditional DBSCAN clustering algorithm,which is influenced by the artificially set parameter value.Firstly,the K-means clustering algorithm is used to quickly cluster the data objects,and the clustering results are statistically analyzed to obtain the values of the initial neighborhood radius and neighborhood density threshold in the DBSCAN algorithm.Firstly,the K-means clustering algorithm is used to quickly cluster the data objects,and the clustering results are statistically analyzed to obtain the values of the initial neighborhood radius and neighborhood density threshold in the DBSCAN algorithm.Based on this,The algorithm performs DBSCAN clustering on the data object and dynamically adjusts the value of the neighborhood radius during the clustering process.Finally,The experimental results show that the algorithm has good clustering effect on the trajectory data.3.Traffic hotspot area discovery system.The system includes trajectory data extraction,trajectory space-time distance measurement,trajectory clustering,road network hotspot mining four modules.The system can be based on user needs different trajectory data for different operations and have a strong practical value.更多还原