《improved supply chain management based on hybrid demand forecasts》由會(huì)員上傳分享，免費(fèi)在線閱讀，更多相關(guān)內(nèi)容在工程資料-天天文庫(kù)。

1、Int.J.ProductionEconomics143(2013)198–206ContentslistsavailableatSciVerseScienceDirectInt.J.ProductionEconomicsjournalhomepage:www.elsevier.com/locate/ijpeIntermittentdemandforecastswithneuralnetworksnNikolaosKourentzesDepartmentofManagementScience,LancasterUniversity

2、ManagementSchool,Lancaster,LancashireLA14YX,UKarticleinfoabstractArticlehistory:Intermittentdemandappearswhendemandeventsoccuronlysporadically.TypicallysuchtimeseriesReceived24February2012havefewobservationsmakingintermittentdemandforecastingchallenging.Forecasterrors

3、canbeAccepted9January2013costlyintermsofunmetdemandorobsolescentstock.IntermittentdemandforecastinghasbeenAvailableonline20January2013addressedusingestablishedforecastingmethods,includingsimplemovingaverages,exponentialKeywords:smoothingandCroston’smethodwithitsvarian

4、ts.Thisstudyproposesaneuralnetwork(NN)Intermittentdemandmethodologytoforecastintermittenttimeseries.TheseNNsareusedtoprovidedynamicdemandNeuralnetworksrateforecasts,whichdonotassumeconstantdemandrateinthefutureandcancaptureinteractionsCroston’smethodbetweenthenon-zero

5、demandandtheinter-arrivalrateofdemandevents.ThisovercomestheForecastinglimitationsofCroston’smethod.Inordertomitigatetheissueoflimited?ttingsample,whichisSlowmovingitemscommoninintermittentdemand,theproposedmodelsuseregularisedtrainingandmedianensemblesovermultipletra

6、ininginitialisationstoproducerobustforecasts.TheNNsareevaluatedagainstestablishedbenchmarksusingbothforecastingaccuracyandinventorymetrics.The?ndingsofforecastingandinventorymetricsarecon?icting.WhileNNsachievedpoorforecastingaccuracyandbias,allNNvariantsachievedhighe

7、rservicelevelsthanthebestperformingCroston’smethodvariant,withoutrequiringanalogousincreasesinstockholdingvolume.Therefore,NNsarefoundtobeeffectiveforintermittentdemandapplications.Thisstudyprovidesfurtherargumentsandevidenceagainsttheuseofconventionalforecastingaccur

8、acymetricstoevaluateforecastingmethodsforintermittentdemand,concludingthatattentiontoinventorymetricsisdesirable.&2013Elsevi