基于时间窗口聚类的时序数据索引压缩  

A CLUSTERING-BASED INDEXING APPROACH FOR VARIABLE WINDOW QUERY OF TIME SERIES

作　　者：刘璐 王鹏[2] 汪卫[2] Liu Lu;Wang Peng;Wang Wei(School of Software,Fudan University,Shanghai 200120,China)

机构地区：[1]复旦大学软件学院,上海200120 [2]复旦大学计算机科学与技术学院,上海200120

出　　处：《计算机应用与软件》2022年第1期39-44,99,共7页Computer Applications and Software

摘　　要：子序列匹配是时间序列挖掘的经典课题,旨在发现大型数据集中的相似数据序列。很多文献关注固定时间段的序列的查询。但对于多种不同时间段的查询的问题仍然未解决好。基于时间段的查询含义是有时间窗口限制的查询。为了满足多时间段上的查询,简单地为每个时间段的子序列构建索引既耗时又耗存储空间。从目前的文献来看,已有的索引无法满足具有不同窗口限制的大量查询。提出一种基于聚类的轻量级的可变窗口索引方法(CBI),通过压缩不同窗口的索引来减少索引时间和空间开销。实验结果表明,该索引不仅节省了时间和空间开销,同时在支持多种类型的查询时都具有较高的效率。Subsequence matching is a classic topic of time series mining.Many works aim to discover similar data series among large datasets.Lots of literatures pay attention to queries over fixed time period.However,variable time windows query remains an unsolved problem.Window-based query means querying with window limitation.However,simply building indexes for each time period will be both time-consuming and space-consuming.In the variable window query situation,different indexes should be built for different windows.So,state-of-the-art indexes could not satisfy a large number of queries with different window limitations.In this paper,we propose CBI,a clustering-based variable-window indexing approach that reduces both indexing time consumption and space cost for variable-window queries.The experiment results show that our index saves time and space cost,at the same time,performs high efficiency on exact query,approximate query,and range query.

关 键 词：时间序列索引压缩 相似性查询 多时间窗口查询 

分 类 号：TP3[自动化与计算机技术—计算机科学与技术]