里!垦!!!竺!121里!里2型尘1211兰!12里竺!!!坚坚竖!!里竺!兰塑!!
seriesdatapre-processing,tothelaststep,theinterpretationandevaluationoftemporalassociationrules.Ineverystep,theauthorcombstheexistingresearch,teststherelativetheoreticalmodels,offersimprovementandprovesit.Becausetheminingofmultivariatetimeseriesisahotissue,theauthordiscussesitinthelast
canpart.Theinnovationsofthedissertationbeincludedasfollows.
(1)Intimeseriesdatapre—processing,theauthorputsforwardsrecognition
onmethodofoutliernoisedatabaseddatavarianceratio.Timeseriesusually
containsnoisedata,whichwillaffecttheminingtemporalassociationrules,SOitshouldbecleanedoutbeforemining.Becausetimeseriescompressionisintoleranttooutliernoisedata,meanwhiletheexistenceofoutlierwillaffectthedivisionoftimeseriesandrepresentationoftimeseriespatterns,SOidentifyinganddeletingtheoutlierintimeserieswillbeoneoftheimportantworksintimeseriesdata
onpre processing.Whether
surroundingdata.The
estimatetheadatumistheoutlier,dependsUSeSitsvibrancywithauthordatavarianceratiooftimeseriesdatatovibrancy,andthenoffersrecognitionofoutliernoisedata.
(2)In
distance
bringstimeseriesandsimilaritymeasure,theauthorcomesupwithEuclidmethodtomeasurethesimilarityoftwodynamictimewarpingmeta—patternsand,andalsoforwarddistancemeanstomeasurethesimilarityoftWOtimeseriespatterns.Inminingtemporalassociationrules,themeta patternmonotonydistancemethod
notsuitableandthemeta-patternvectordistancemethodbothareforgettingfrequentpatternwhenmeasuringthesimilaritybetweentwometapatterns.Aimingthespecialtyoftimeseriespattern,thedissertationoffersweighted
timewarpingdistancemethodofmeta—pattern,andthencomesupwithdynamicmeans,whichCalldistancemeasurethesimilaritybetweentwosequentialpatterns.
(3)In
forwardsthetheacquirementoftemporalassociationrules,theauthorputslayeredmeans.The
rulestimerestrictionoftemporalassociationrulestheand...ofassociationdeterminesdifficulty
andofcanacquiringtemporalassociationrules.Inordertodecreasethedifficulty,wetemporalassociationrulesintodifferentlengthdividethebeforerofSOthenmine,thatiscalledthe
layeredminingoftemporalassociationrules.Becauseofthedifferenceindefiningthefrequentpatterns,themethodisdifferentfromotherminingways.Meanwhilebecausethemethodconsidersthebeforerofdifferentlength,ithastheunique4