between Time Series
Fusheng Yu, Keqiang Dong · 2005
Inthispaper, a granulation-based methodfor finding similarity between twotimeseries isproposed. Firstly, for eachtimeseries X={x1,x2, ...X, theapproach develops a granular timeseries induced bytheoriginal timeseries, anda trendgranular timeseries inducedby thetrendtime series aX = {x2 - xl, X3-X2, IXn- XnI}.Secondly, itcompares thetwooriginal timeseries bycomparing thecorresponding two (trend) granular timeseries. Inordertocompare two(trend) granular timeseries, anindex, nameddegree ofsimilarity, is defined toreflect thesimilarity ofthem. Bythegranulation-based method, wecandeal withthetemporal datamining tasks suchas similar subsequence searching, clustering andindexing etc. on thegranular level. Experiments showthat ourmethod iseffective andapplicable.