海洋数据集的质量检查验收抽样方法翻译.docx

海洋数据集的质量检查验收抽样方法翻译.docx

《海洋数据集的质量检查验收抽样方法翻译.docx》由会员分享，可在线阅读，更多相关《海洋数据集的质量检查验收抽样方法翻译.docx（19页珍藏版）》请在冰豆网上搜索。

海洋数据集的质量检查验收抽样方法翻译

Acceptancesamplingplanofqualityinspectionforoceandataset

Comparedwiththedatasetofindustrialproducts,oceandatasetshaveseveraldistinctcharacteristics,suchaslargequantitiesandbeingmulti-source,multi-dimensionandmulti-type.Basedontheacceptancequalitylevel（AQL）andlimitqualitylevel（LQL）,wedesignedanacceptancesamplingplanofqualityinspectionforoceandatasets（ASP-OD）,usedthisplantoinspectoceandatasetquality,andevaluateditsadvantage.ASP-ODhasaconsistentandstablediscriminatorypowerindependentoflotsize’whichsolvestheproblemof‘strictnessforlargelotsize,tolerationforsmalllotsize’inthepercentsamplingplan.ASP-ODestablishesarelationshipbetweenlotsizeandsamplingsize,andprovidesaplanforagivenlotsize.ThisplanovercomesthedeficiencyofISO2859-basedsamplingplans,differentlotsizecorrespondingtothesamesamplingplan,inthequalityinspectionofoceandatasets.Collectively,thisstudysuggeststhatASP-ODisasuitablesamplingplanfortheinspectionofoceandatasetquality.

Keywords:

oceandataset;qualityinspection;AQL;LQL;acceptancesamplingplan

1.Introduction

Withtherapiddevelopmentofoceanmonitoringtechnology,hugeamountsofoceandatahavebeencollectedfromvarioussources,suchasremotesensingimages,buoys,cruisedataandunderwaterobservationdata.Thus,oceandatasetshavegraduallybecomeaclassicexampleofmulti-modalbigdata.However,thebiggestobstacleforpreparinganoceanatlasishowtocontrolthequalityofdata.Thequalitycontrolofoceandatasetsisanimportantpartofanyoceananalysis/forecastingsystem.Usingoracceptingerroneousdatacouldleadtoaninvalidconclusionoranincorrectanalysis.Bycontrast,rejectingextremebutvaliddatasometimescouldcausethemissingofkeyeventsandanomalousfeatures.Todate,agrowingnumberofscientistshavebeguntofocusonthequalityinspectionofoceandata.Anautomatedqualitycontrolsystemwasproposedtoinspectoceanictemperatureandtemperature-salinityprofiles.TheSurfaceOceanCO2Atlas（SOCAT）projectwasperformedtoinvestigatetheglobaldatasetofmarinesurfaceCO2.Duringthisproject,alldataweredesignedtobeputinauniformformatfollowingastrictprotocol.Qualitycontrolwasconductedaccordingtoclearlydefinedcriteria.Inaddition,thequalityandconsistencyofNASAoceancolourdata,includingspectralwater-leavingreflectance,chlorophyll-αconcentration,anddiffuseattenuation,wereexaminedusingcommonalgorithmsandimprovedinstrumentcalibrationknowledge.Thesestudieshaveputforwardseveralqualityinspectionplansforoceandata,especiallyforoneorafewelements.Oceandatasetsareusuallycomposedofmulti-element,multi-scaleandmulti-temporalgeo-informationelements.Moreover,thereisapotentialinterplaybetweendifferentelementsinanoceandataset.Thus,itisrequiredtoproposeanovelacceptancesamplingplantoinspectthequalityofoceandataasacompleteandindivisibledataset.

Thegoalofqualityinspectionistojudgewhetherthedatareachtherequiredqualitythroughasamplingplan.Currently,theoptimisationofacceptancesamplingplanshasbeenconductedtosatisfythebalancebetweeninspectionriskandinspectioncostforthequalityinspectionofindustrialproducts.Someacceptancesamplingplanshavebeendesignedbasedoninspectionrisk,whichaimedtominimiseeithertheproducer’sriskortheconsumer’srisk.Someotheracceptancesamplingplansweredesignedbasedontheinspectioncost,whichaimedtoreducethesamplingnumber.Theseexistingplansaremainlyusedtoinspectthequalityofindustrialproducts.Generally,industrialproductsareproducedinacontrolledandconsistentmanner,andusuallyhavecertainitemsanduniformunits.Comparedwithindustrialproducts,oceandatahavesomedistinctcharacteristics,suchasbeingmulti-source,multi-dimensionalmulti-type,multi-time-state,withdifferentaccuracyandnonlinearity.Thus,theseexistingacceptancesamplingplansarenotsuitableforthequalityinspectionofoceandatasets.

Inthispaper,wedesignedanacceptancesamplingplanofqualityinspectionforanoceandataset（ASP-OD）.Insection2,theconceptualframework,derivationprocessandtheformulasofASP-ODareshown.Insection3,weapplytheASP-ODtoinspectthequalityofoceandata,andcompareitsadvantagesoverexistingacceptancesamplingplans.Insection4,wesummarisethisstudy,andproposethatASP-ODisasuitableacceptancesamplingplanforthequalityinspectionofoceandatasets.

2.DesignofASP-OD

ThetheoryofASP-OD

TheacceptancesamplingplanofqualityinspectionforoceandatasetswasdesignedasS（N,n,c）.Here,Nisthelotsizeandcomprisesallinspectedoceandatafromwhichthesampleistobetaken;nisthesamplesizeandconsistsofanumberofsamplingunitsselectedfromthelotsize,whichisacompromisebetweentheaccuracyofproductinspectionandthecostoftheinspection;c,theacceptancenumber,isusedtojudgewhethertheinspectedoceandatameettherequirementoftheoceandataconsumer.Theprocessofqualityinspectionisshownasbelow:

（1）n-sampleddataareextractedfromthelotsizeN;

（2）thequalityofextracteddataisinspectedonebyone;（3）ifthenumberofnon-conformingdata（d）islargerthantheacceptancenumber（c）,thequalityofinspectedoceandataisconsideredtobenon-conforming.Otherwise,thequalityofinspecteddataisconsideredtobeconforming.

BasedontheacceptancesamplingplanS（N,n,c）,thepercentnon-conforming（P）iscalculatedby

（1）

whereDisthenumberofnon-conformingoceandatainthetotaloceandataset.

Generally,itisdifficulttoobtainthevaluesofDandPunlessthetotaldataare100percentinspected.Sampledoceandataareusedtoestimatetheparametersforlotsize.Thus,Pisusuallyestimatedusingthepercentnon-conformingestimator（p）;piscalculatedby

（2）

wheredisthenumberofnon-conformingoceandatainthesampleddataset.

Basedontheabove-mentionedparameters,theacceptancequalityprobability（L（p））oftheacceptancesamplingplanS（N,n,c）canbecalculatedby

（3）

（0≦d≦n,d≦D,n-d≦N-Np）

Operatingcharacteristiccurves（OC-curve）arepowerfultoolsinthefieldofqualitycontrol,astheydisplaythediscriminatorypowerofanacceptancesamplingplan.Here,weconsideredthequalitylevelasthehorizontalaxisandthecorrespondingacceptanceprobabilityastheverticalaxis.TherelationshipbetweenL（p）andtheproportionpofnon-conformingitemswasrepresentedastheOC-curveofsamplinginspectioninarectangularcoordinatesystem.

Generally,consideringtheinterestsofboththeproducersandconsumers,acceptancequalitylevel（AQL）andlimitingqualitylevel（LQL）wereadoptedtodesigntheacceptancesamplingplan.LQLisamaximumqualitylevelofdefectivestoleratedintheinspectiondata.WhenthequalitylevelisworsethanLQL,theconsumerstendtorejecttheinspecteddata.AQLrepresentsameanqualitylevelofdefectivesamplestoleratedintheinspection.IfthequalityleveloftheinspecteddataisbetterthanAQL,theproducerstendtoaccepttheinspecteddata.

Tomeettherequirementofbothproducersandconsumers,AQLandLQLweretakenintoaccountintheASP-ODdesign,whichwasshownastwopointsintheOC-curve（Figure1）.Thefirstpointisdenotedas

Figure1.OC-curveoftheacceptancesamplingplan

（p0,1-α）p0,i.e.AQL,istheproportionofnon-conformingitemsthatcanbetoleratedtojudgethattheentirelotcanbeaccepted.α,theproducer’srisk,istheprobabilityofrejectionoftheinspectedloteventhoughthequalitylevelofthelotisequaltoorbetterthanAQL.Thesecondpointisdenotedas（p1,β）.p1,i.e.LQL,istheproportionofnonconformingitemsthatcanbetoleratedtojudgethattheentirelotcanberejected.β,theconsumer’srisk,istheprobabilityofacceptanceoftheinspectedloteventhoughthequalitylevelofthelotisequaltoorworsethanLQL.

UndertheconditionofthetwopointsontheOC-curve,therelationshipbetweenthelotsize,thesamplesizeandtheacceptancenumberiscalculated.Theproblemcouldbeformulatedasanonlinearprogrammingproblem.

TheASP-ODmodel

Fromtheperspectiveoftheproducer,theacceptancesamplingplanshouldsatisfythefollowingcondition:

（4）

D1,apositiveinteger,isthenumberofnon-conformingdataelementsintheinspectedoceandataset.Whentheproportionofnon-conformingdataisequaltoAQL,thevalueofD1iscalculatedby

D1=round（N·p1）（5）

Fromtheperspectiveoftheconsumer,theacceptancesamplingplanshouldsatisfythefollowingcondition

（6）

D2,apositiveinteger,isthenumberofnonconformingdataelementsintheinspectedoceandataset.Whentheproportionofnonconformingdataisworsethanthelimitingqualitylevel（LQL）,thevalueofD2iscalculatedby

D2=round（N·p2）（7）

Thetotalresidualerror,ε,meansthesumofresidualerrorsoftheacceptanceprobabilityatbothAQLandLQL.Theroleofεisusedforthecalculationofthevalueofnandcintheacceptancesamplingplan.Here,wechosetheminimalεtodeterminetheoptimalnandcfortheacceptancesamplingplanatAQLandLQL.Theoptimalacceptancesamplingplanisformulatedasthefollowingnonlinearoptimisationproblem

s.t.

（8）

（9）

ε1istheresidualerroroftheacceptanceprobabilitybasedontheproducer’srisk.ε2istheresidualerroroftheacceptanceprobabilitybasedontheconsumer’srisk.Thenonlinearoptimisationproblemissolvedbasedontheiterativealgorithm.TheiterativealgorithmisimplementedinMatlabsoftware.

3.Casestudy

Inthissection,weemployedASP-OD,thepercentsamplingplan（PSP）andtheISO2859-basedsamplingplan（I