基于模式特征保持的数据流降载方法-北京大学机器感知与智能教育部.PDFVIP

ISSN 计算机研究与发展 1000—1239[CN1卜1777／TP Journalof Researchand Computer Development 基于模式特征保持的数据流降载方法 苗高杉 李红燕 (北京大学信息科学技术学院北京 100871) (机器感知与智能教育部重点实验室(北京大学)北京100871) (miaogs@gmail．corn) PatternFeature forLoad overDataStreams PreservingStrategy Shedding MiaoGaoshanandLiHongyan (SchoolElectronics and 100871) of Engineering ComputerScience，PekingUniversity。Beijing Machine 100871) (KeyLaboratoryof Perception(PekingUniversity)，MinistryofEducation，Beijing AbstractDatastream often realistic and domain patternspossessgood meaningshigh significance． Pattern thefeatureofdata load methodsare diversityrepresents streams．Existingshedding mostly focusedoneach or ofdata oftenleadsto singlepointtuple streams．Deletingpoints patternintegrity loss，thereforemakes ofdatastreamfeaturesdiscardedload part by sheddingprocess．Thispaper conceivedanefficientand structure，andanovelmethodfordata compressivepatternstorage proposed streamload onfeature newmethodconsistsoffour shedding，basedpreservingstrategy．The steps： data into extraction thenew— streams Firstly，split subsequencesbypattern process．Secondly，match arrival with and thedata makes subsequenceexistingpatternsupdate structure．Thirdly，thesystem load decisionwhenmostofthe is coefficientfor shedding