摘要： 针对传统聚类算法多数无法对时序数据流进行聚类的问题,提出一种基于密度和网格的聚类算法。引入动态划分网格的方法,通过当前数据块内数据的特征动态地设置网格划分、网格密度阈值等参数并自适应地生成网格,将其转化为不同类型的图并分别进行聚类。分析某一个体的轨迹,采取按时间段的个体轨迹划分方法检测个体异常轨迹。实验结果表明,该算法可根据用户的需求得到不同时间段内数据的聚类结果,适用于处理轨迹信息等时序数据流的异常检测问题。

关键词: 数据流, 数据块, 聚类, 动态划分, 异常检测

Abstract: Aiming at the problem that most traditional clustering algorithms cannot cluster time-series data flow,a clustering algorithm based on density and grid is proposed.It introduces the method of dynamic grid partitioning,dynamically set parameters such as grid partition and grid density thresholds and adaptively generates grids through the characteristics of data within the current data block,convert it into different types of maps and performs clustering separately.The trajectory of an individual is analyzed,and the individual trajectory is detected by the individual trajectory division method according to the time period.Experimental results show that the algorithm can obtain the clustering results of data in different time periods according to the user’s requirements,and is suitable for the problem of abnormal detection of time-series data flow such as trajectory information.

Key words: data flow, data block, clustering, dynamic partitioning, anomaly detection

中图分类号: 

• TP391