Hierarchical Segments Model for Time Series Streams
Shi Baile · Journal of Chinese Computer Systems · 2009
The work of this paper aimed at such a type of time series streams which have following characteristics:(1)Sequence data are continuously generated,potentially forever.(2)Sequence data are only scanned one pass and through the scan an approximation synopsis is built.The subsequent process is based on the synopsis.(3)The usefulness of data decay with age.In order to work out these problems,a generalized framework,named as Hierarchical Segments Model,is proposed.In the framework,a time series stream is dynamically divided into a set of subsequences,and each subsequence is represented as a segment and all segments are hierarchically organized.By segments,the traditional time series approximation representation techniques can be embedded into our framework for streaming time series.We experimentally evaluate the proposed framework.