ESTIMASIKECEPATANANGINPADAJARINGAN ANEMOMETER DISTASIUNMETEOROLOGI JUANDA SURABAYAMENGGUNAKAN COOPERATIVESENSING, TEMPORALCONVOLUTIONAL NETWORKDANKOREKSI SPASIAL TESIS Karyatulissebagaisalahsatusyarat untukmemperolehgelarMagisterdari InstitutTeknologiBandung Oleh HARYASSUBYANTARA WICAKSANA NIM.23822008 (ProgramStudiMagisterInstrumentasidanKontrol) INSTITUTTEKNOLOGIBANDUNG Juni2024 3 ABSTRACT WINDSPEEDESTIMATIONONMULSTISITES ANEMOMETER INJUANDASURABAYAMETEOROLOGICAL STATIONUSINGCOOPERATIVESENSING,TEMPORAL CONVOLUTIONAL NETWORKANDSPATIALCORRECTION By: HARYASSUBYANTARA WICAKSANA NIM.23822008 (INSTRUMENTATION ANDCONTROLSTUDYPROGRAM) Windisatmosphericairmassesmovementduetoanimbalanceinairpressure bothhorizontallyandvertically.Anemometerasatoolformeasuringwind parametersconsistsofsensorsforwindspeedanddirection.Windsensordamage isoftencausedbydamagetothetransducer,internalcomponentsandwiring. Replacingthewindsensorrequirestheavailabilityofsparepartsanda significantamountoftime,especiallysincethesensorinstallationisataheightof 10meters.Theunavailabilityofwindspeeddataisverycrucial,becausethe anemometernetworkisinstalledattheairport.Meanwhile,aviationactivities requirefastandaccuratemeteorologicalinformation.Thisresearchaimsto developamethodforestimatingwindspeedbasedoncooperativesensingand TemporalConvolutionalNetwork(TCN),aswellascorrectingitspatially.Itis hopedthatthisestimationmethodcanbeanalternativetowindspeeddatafor anemometernetworksatairports.Pre-processingofthedatasetiscarriedoutby detectingoutliersandcarryingouttemporarymissingdatatreatment.Estimation usingcooperativesensing.Thistechniquecombinessensordataina supplementarymannertomeasurecertainparameters.Thecooperativesensing sub-componentconsistsofsensors,models,estimationalgorithmsandoutput quantities.Sensorsinclude4anemometersatJuandaInternationalAirport, Surabaya,namelySB3,SB4,SB5andSB6.Thedatasetiswindspeeddataper minutefortheperiod1January2022-24December2023. Baselineestimationinputusesthedominantwinddirectionmethodandquadrant weighting.Thedominantwinddirectionmethoddividestheestimationmodel datasetbasedonthedominantwinddirection,namelywesterly,easterlyand transitionalmodels.Meanwhile,thequadrantweightingmethodconvertsthewind directionvalueforeachanemometerintoacertainweightaccordingtothe quadrantdivisionwithoutdividingthedataset.Quadrantsaredividedinto directionsapproachingthetarget,awayfromthetargetandtransitional.The inputbaselineforthequadrantweightingmethodconsistsofthe1-3closest anemometerlocations,historicaldata,andacombinationofhistoricaldataand 4 nearbyanemometers.TheSB4andSB5anemometersinthemiddleofthe configurationaresimulatedasestimationtargetsinturn. Totaldatasetperanemometeris1,041,120data.Windspeeddatasetissegmented into75%trainingdataand25%testdata.TheestimationalgorithmusesTCN. TheTCNhyperparametersconsistofafilterlengthof32,3filterkernelsanda dilationfactord=[1,2,4].TheTCNestimationresultsarethenverifiedandthen spatiallycorrected.Iftheestimatedresultsoftheanemometerwindspeedsensor exceedtwicethespatialstandarddeviation,thenacorrectionisappliedtothe estimationresults.Baselineinputcooperativesensingusingthequadrant weightingmethodismoreeffectivethandividingthedatasetusingthedominant winddirectionmethod.TheresultsshowthattargetanemometersSB4andSB5 canbeestimatedalternatelybasedoninputdatafromthenearestanemometer andhistoricalanemometerdata(datafromtheprevious1-5minutes). Spatialcorrectionwasabletoimprovetheaccuracyperformanceofwindspeed estimationwiththecorrelationcoefficientincreasingfrom0.87to0.88forboth SB4andSB5targets.RMSEdecreasesfrom0.86m/sto0.83m/sforSB4and0.90 m/sto0.88m/sforSB5.MAEdecreasesfrom0.60m/sto0.58m/sforSB4and 0.64m/sto0.62m/sforSB5.Overall,thecorrectedTCNperformancestillmeets WMOrequirementswithRMSE