科学数据网格中数据挖掘技术研究

《科学数据网格中数据挖掘技术研究》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、中国科学院计算技术研究所博士学位论文科学数据网格中数据挖掘技术研究姓名：佟强申请学位级别：博士专业：计算机系统结构指导教师：阎保平20060601ResearchonDataMiningintheScientificDataGridQiangTong(ComputerArchitecture)DirectedByBaopingYanWiththeemergenceanddevelopmentof面dcomputing，itbecomespossibletosharedataandcollaborateinalargescalemodelofcross-organiz

2、ationandcross-legion．Intheareaofscientificresearch，theproblemofmodernscientificresearchbecomesmoleandmorecomplex，whichresultsinabrand·newscientificcollaborationmodelandthelargescienceproject，i．e．，theinfomationizationofscientificresearch(e-Science)．Inordertoshareresourcesandproducts，and

3、alsocollaboratetoaccomplishlargescalemodemscientificresearches，itisnecessarytoestablishallalliedvirtualresearchgroupviatheIntemetbasedon卯dcomputing．Byusingdataminingtechnologies，thispaperaimstoimprovetheserviceleveloftheScientificDataGridandtheScientificDatabase，basedontheirexistinglar

4、ge—scaledatastorageandpowerfulcomputingcapabilities．ThemainresearchcontentsandcontributionsarelistedasfoIlows．’(1)BasedondetailedanalysesofthedataminingpropertiesoftheScientificDataGrid,ascientificdataminingsystemisproposed．Thesystemconsistsofthleemaincomponents：theScientificDataMining

5、Architecture(SDMA)，theScientificDataMimngToolkit(SDMK),andtheScientificDataMiningService(SDMS)．SDMAdescribesthemulti-dimensionmodelarchitectureofdatamimngapplications；SDMKprovidesalargeamountofdatapreprocessinganddataminingalgorithms；SDMSpresentsadataminingschemetoaddresstheproblemsund

6、er画denvironmentthroughaformof鲥dservice．Comparedwithtraditionaldataminingsystems，theproposedsystemhasmanyexcellentproperties，andismolesuitabletotheenvironmentoftheScientificDataGridandtheScientificDatabase．Nowadays，ithasbeenappliedinsomerealdatabaseapplications．Besidesthesimplequeryands

7、earchfunctions，theproposedsystemCanalsoperformmoreadvancedfunctionssuchasdatastatistic，dataanalysis，andknowledgediscovery．Asaresult,theservicelevelofthedatabaseisimproved．(2)Clusteringindataminingisadiscoveryprocesswhichgroupsasetofdatasuchthattheintra-clustersimilarityismaximizedand