您好,欢迎访问三七文档
I.J.WirelessandMicrowaveTechnologies,2018,4,66-77PublishedOnlineJuly2018inMECS()DOI:10.5815/ijwmt.2018.04.05Availableonlineat(LSTM):RecurrentNeuralNetworksMeghnaSharmaa*,ManjeetSinghTomerbaTheNorthCapUniversity,Gurugram,Haryana,122017,IndiabYMCAUniversity,Faridabad,Hayana,121006Received:13October2017;Accepted:17January2018;Published:08July2018AbstractPredictionoflocationhasgainedlotofattentionindifferentapplicationsareaslikepredictingthepathoranydeviationliketaxi-route,busroute,humantrajectory,robotnavigation.PredictionofthenextlocationoranypathdeviationinRFIDenabledsupplychainpathfollowedintheprocessisquiteanovelareafortherelatedtechniques.ThepaperdefinesthearchitecturefortheoutlierdetectioninRFIDenabledSupplyChainPathbasedonhistoricaldatasets.Giventhetrainingdatasets,differentclassificationmodelsarecomparedfortheaccuratepredictionoftheoutliernessofthepathfollowedbythetaggedobjectsreadbyRFIDreadersduringthesupplychainprocess.ComparisonofHiddenMarkovModel(HMM),XGBoost(decisiontreebasedboosting),RecurrentNeuralNetwork(RNN)andstateofarttechniqueinRNNknownasLongShortTermMemory(LSTM)isdone.ToourknowledgeLSTMhasneverbeenusedforthisapplicationareaforoutlierprediction.Forthelongerpathsequences,LSTMhasoutperformedoverothertechniques.Thetrainingdatasetsusedhereareintheformoftherecordoftheoutlierpositionsinparticularpathandatparticulartimeandlocation.IndexTerms:Outlierprediction,RecurrentNeuralNetworks(RNN),LongShortTermMemory(LSTM),SupplyChain,RadioFrequencyIdentification(RFID).©2018PublishedbyMECSPublisher.Selectionand/orpeerreviewunderresponsibilityoftheResearchAssociationofModernEducationandComputerScience1.IntroductionThedomainareadiscussedinthispaperisrelatedtoRadioFrequencyIdentification(RFID).Radio-frequencyidentification(RFID)isatechnologytodetectthepresenceofanobjectwithradiowavesasmedium.Unlikebarcodereadingmechanism,lineofsightisnotrequiredinRFIDsystemandthuscanbe*Correspondingauthor.E-mailaddress:meghnasharma@ncuindia.eduPredictiveAnalysisofRFIDSupplyChainPathUsingLongShortTerm67Memory(LSTM):RecurrentNeuralNetworksreallyhelpfulforlongdistanceautomaticsensing/reading/identificationofanytaggedobject,personorpackagewithinreaders‘range.ThisisdoneusingRFIDtags.Thesearesmalltransponderswhichconsistofradioreceiverandtransmittertotransmitidentityinformationoveradistance,whenasked.RFIDtagreaderisusedtoreadthetagusingradiowaves.Thereadersareconnectedtocomputersystemstorecordthedatareadfromthetaggedobject.MostRFIDtagscontainatleasttwoparts.Oneisanintegratedcircuitforstoringandprocessinginformation,modulatingandde-modulatingaradio-frequency(RF)signal,andotherspecializedfunctions.Thesecondisanantennaforreceivingandtransmittingthesignal.ThisTechnologycanbeusedinvariousapplications[21][22]whichinvolvetrackingandmonitoringofobjectsprovidedwithuniqueidentificationnumbere.g.patientmonitoring,vehicletrackingintollpath,objectstrackinginsupplychainetc.InthispapertrackingandmonitoringofRFIDtaggedobjectsorgroupofobjectswithuniqueelectronicproductcode(EPC)insupplychainpathisdiscussed.ItstartswiththebriefintroductionRFIDenabledsupplychainprocess,thenexplanationofthepreprocessingandfeatureextractionfromtherawdatareadbythereaders.Variousclassificationtechniquesarecomparedforthepredictivemodelling.1.1.RFIDEnabledSupplyChainProcessInthewholeprocessoftheRFID-basedsupplychain,allreadingsfromRFIDreadersareautomaticallycaptured.Inthecompleteprocess,eachproductitemistaggedwithanelectronicproductcode(EPC)intheproductionlineandrelatedproductspecificationsarewrittenintotags.Thenthetaggedproductsarepackedintocasesinsupplierwarehouses,wheretheEPCtagsofboththecasesandthecontainingpalletsarescannedbyRFIDreaderswhentheobjectscomeintherangeofreadersandthenthepalletsareloadedintotrucks,tobedistributedtodealers.Unloadingofthepalletsfromthetrucksisdoneattheretailstorestobefinallypurchasedbyconsumers.Thepositionoftheobjectsisreadinformoflatitudeandlongitudebythereaders,GPSsensorsareattachedonEPCtags.TheprocessofreadingtagsbyRFIDreadersatvariouspositionsinthesupplychainprocesscanbevisualizedasinFigure1:Fig.1.RFIDSupplychainProcess[1]68PredictiveAnalysisofRFIDSupplyChainPathUsingLongShortTermMemory(LSTM):RecurrentNeuralNetworks1.2.DataModelDatareadbyreadersfromsupplierstoconsumersinrawformis:Tagid,Readerid,Timestamp.RFIDreaderskeeponcontinuouslyreadingthesametagidmultiplenumbersoftimestillthetimeanyobjectisintherangeofthereaderwhichleadstogenerationofredundantdata.Compressionofsuchdatacanbedonebydividingthetimestampintotimeinandtimeout.SoaftercompressionthedataformatisintheformofTagId,Readerid,Intime,OutTimeasshowninfigure2.Fig.2.CompressedrawDataFormatHereTagIdrepresentsuniqueEPCofthetagontheobject,Readeridrepresentsthereader’sLocationwhichismappedtolongitudeandlatitudeforfurtherprocessing,IntimeisthetimewhenataggedobjectcomesinrangeofreaderandOutTimeisthetimewhenataggedobjectgoesoutoftherangeofreader.Dataisintegratedfromthelocalserve
本文标题:基于长周期记忆的RFID供应链路径预测分析:递归神经网络(IJWMT-V8-N4-5)
链接地址:https://www.777doc.com/doc-7750687 .html