基于人工神经网络和统计模型的需水量预测(IJISA-V11-N9-5)

I.J.IntelligentSystemsandApplications,2019,9,40-55PublishedOnlineSeptember2019inMECS()DOI:10.5815/ijisa.2019.09.05Copyright©2019MECSI.J.IntelligentSystemsandApplications,2019,9,40-55PredictionofWaterDemandUsingArtificialNeuralNetworksModelsandStatisticalModelMohammedAwadDepartmentofComputerSystemsEngineering,ArabAmericanUniversity,PalestineE-mail:[e-mail]-AlkelaniDepartmentofComputerScience,ArabAmericanUniversity,PalestineE-mail:[e-mail]:19March2019;Accepted:09May2019;Published:08September2019Abstract—Thepredictionoffuturewaterdemandwillhelpwaterdistributioncompaniesandgovernmenttoplanthedistributionprocessofwater,whichimpactsonsustainabledevelopmentplanning.Inthispaper,weusealinearandnonlinearmodelstopredictwaterdemand,forthispurpose,wewillusedifferenttypesofArtificialNeuralNetworks(ANNs)withdifferentlearningapproachestopredictthewaterdemand,comparedwithaknowntypeofstatisticalmethods.Thedatasetdependsonsetsofcollecteddata(extractedfrommunicipalitiesdatabases)duringaspecificperiodoftimeandhenceweproposinganonlinearmodelforpredictingthemonthlywaterdemandandfinallyprovidethemoreaccuratepredictionmodelcomparedwithotherlinearandnonlinearmethods.TheappliedmodelscapableofmakinganaccuratepredictionforwaterdemandinthefuturefortheJenincityatthenorthofPalestine.Thispredictionismadewithatimehorizonmonth,dependingontheextracteddata,thisdatawillbeusedtofeedtheneuralnetworkmodeltoimplementmechanismsandsystemthatcanbeemployedtopredictsashort-termforwaterdemands.Twoappliedmodelsofartificialneuralnetworksareused;MultilayerPerceptronNNs(MLPNNs)andRadialBasisFunctionNNs(RBFNNs)withdifferentlearningandoptimizationalgorithmsLevenbergMarquardt(LM)andGeneticAlgorithms(GAs),andonetypeoflinearstatisticalmethodcalledAutoregressiveintegratedmovingaverageARIMAareappliedtothewaterdemanddatacollectedfromJenincitytopredictthewaterdemandinthefuture.TheexecutionresultsappearthattheMLPNNs-LMtypeisoutperformedtheRBFNN-GAsandARIMAmodelsinthepredictionthewaterdemandvalues.IndexTerms—Prediction,FutureWaterDemand,MultilayerPerceptronNNs,LevenbergMarquardtAlgorithm,RadialBasisFunctionNNs,GeneticAlgorithms,ARIMA.I.INTRODUCTIONThemajorityofthecountriesintheMiddleEastaresufferingproblemstheincreasingdemandforwaterinlightofthescarcityofresourcestoobtainsufficientquantitiesandsatisfytheneedsofcitizensofdifferentneedsindifferentfields[1].Ingeneral,thewaterdemandandsupplydependsontheinfrastructureofsupply,distributionsystems,andfuturestrategicplansthathavethecapacitytomeettheneedsandsustainthesuccessofthedevelopment[2].SowecandescribetheWaterDemandForecastingasatotalamountofusedwater,measuredorpredictedbasedonacertainapplicationtoknowthegeneraltrendofconsumptionsoastoevaluatetheabilityofexistingresourcestomeetfutureneedswithinageographicareaandtoprovidethebasisforplanningfuturesystemandimproveittolimittheuncertaintiesforfuturedemand.Thewatersectorisanimportantsectorofsustainabledevelopmentatthenationallevel.Thehighdemandforwaterandthesignificantgapbetweendemandandsupplyinthewatersectorisoneofthemajorchallengesfacingthesectoroverthenextfewyears.Thewaterdemandisincreasingbecauseofnaturalpopulationgrowthandnationaldevelopmentrequirements.Thisisagreatchallenge,anditisnecessarytofindcreativesolutionstosupplythenecessaryquantitiesofwatertodifferentsectorsandachievebalanceforsupplyoptimalwaterinPalestine.Theexistingmodelsandapplicationsthatcanpredictthewaterdemandeffectivelyisausefulelementinstrategicplanningandtheprocessesofscheduling,maintenance[3].Predictionstrategiesofwaterdemandareveryimportanttosupportandhelpthewaterauthoritiesandmunicipalitiesinidentifyingfutureneedsandtodevelopthenecessaryplanstofindrealsolutions.ThewatercircumstanceinnorthernPalestine,suchasthecityofJenin,issimilartotherestofPalestinecities.ButinJenincity,therearemorePredictionofWaterDemandUsingArtificialNeuralNetworksModelsandStatisticalModel41Copyright©2019MECSI.J.IntelligentSystemsandApplications,2019,9,40-55difficultiesbacktotheamountofwaterleakinginthegroundduetoweakofnetworksandinterruptionsinsupplyandconsumptionduringperiods.ThewaterdepartmentinthemunicipalityofJeninhasnoprogramsorapplicationsforestimatingthefuturedemandforwaterandthecalculatingandestimatingtheexpecteddemandisdependentonsimplestatisticalcalculationmethods,evenintheothergovernoratesalsotherearenomodernmethods,hence,simplestatisticalcalculationsarenotenoughandnoteffectiveandnotreliabletoprovidepredictsandestimatesthatareappropriateforthenatureandcharacteristicsofthedifferentregions.Timeseriesmodelsdependonthepremisethatanytimeserieshaveahistoricalparticularrecurringstatisticalthatcanbemanipulatedforpredictpurposes[4].TheunknowntimeseriesfunctionF(Xt)tobuildamodel,thatallowsobtainingaccuratepredicts[5].TimeseriesisacombinationthatrepresentsthedataXt,listedandregisteredoverthedurationoftimeforinstance,daily,weekly,monthly,yearly,…etc.[6].Overtheyears,awideresearchefforthasbeenundertakenbyresearcherstodevelopeffectiveandpowerfulmodelswiththeaimofimprov