资源描述:
《A New Deterministic Data Aggregation Method ForWireless Sensor Networks一种新的确定性数据聚集方法 无线传感器网络》由会员上传分享,免费在线阅读,更多相关内容在学术论文-天天文库。
1、ANewDeterministicDataAggregationMethodForWirelessSensorNetworksHuseyinAkcanandHerv¨eBr´onnimann¨Computer&InformationScienceDepartmentPolytechnicUniversity,Brooklyn,NY11201hakcan01@cis.poly.eduandhbr@poly.eduAbstractTheprocessingcapabilitiesofwirelesssensornodesenabletoaggregater
2、edundantdatatolimittotaldataflowoverthenetwork.Themainpropertyofagoodaggregationalgorithmistoextractthemostrepresentativedatabyusingminimumresources.Fromthispointofview,samplingisapromisingaggregationmethod,thatactsassurrogateforthewholedata,andonceextractedcanbeusedtoanswermulti
3、plekindsofqueries(suchasAVG,MEDIAN,SUM,COUNT,etc.),atnoextracosttothesensornetwork.Additionally,samplingalsopreservescorrelationsbetweenattributesofmulti-dimensionaldata,whichisquitevaluableforfurtherdatamining.Inthispaper,weproposeanovel,distributed,weightedsamplingalgorithmtos
4、amplesensornetworkdataandcomparetoanexistingrandomsamplingalgorithm,whichistheonlyalgorithmtoworkinthiskindofsetting.Weperformpopularqueriestoevaluateouralgorithmonarealworlddataset,whichcoversclimatedataintheU.S.forthepast100years.Duringtesting,wefocusonissuessuchassamplequalit
5、y,networklongevity,energyandcommunicationcosts.Keywords:Sampling;datareduction;in-networkdataaggregation;datastreams;sensornetworks0ResearchoftheauthorshasbeensupportedbyNSFCAREERGrantCCR-0133599PreprintsubmittedtoElsevier15May20071IntroductionInthispaper,wefocusonsamplingdatafr
6、omasetofsensornodeslinkedbyanetworkandconsidertheproblemofextractingthesamplefromthenetworktoacentralDBMSandlaterqueryingittofindanswerstoconditionsinsidethesensornetwork.Thesequeriesmaybebasedonsnapshots,ormaybecontinuous.Thedatacollectedbythesensorsisusuallyhighlyredundant,andt
7、husoneneednotcollectandprocessallofit,approximatequeryresultsareusuallysufficient.Hence,substantialsavingsmaybeobtainedbyeitheraggregatingorprocessingthedatain-network,orbyobtainingamuchsmallerbutrepresentativesamplewhichcanthenbeprocessedbyacentralizedmorepowerfulunit.Oneparticu
8、larkindofdatawefocusonisthemulti-dimensionalcou