简介:
简介:ThisletteranalyzesthereasonswhytheknownNeuralBackPromulgation(NBP)networklearningalgorithmhasslowerspeedandgreatersampleerror.Basedontheanalysisandexperiment,thetraininggroupdescendingEnhancedCombinationAlgorithm(ECA)isproposed.TheanalysisofthegeneralizedpropertyandsampleerrorshowsthattheECAcanheightenthestudyspeedandreduceindividualerror.
简介:AdmissioncontrolplaysanimportantroleinprovidingQoStonetworkusers.Motivatedbythemeasurement-basedadmissioncontrolalgorithm,thisletterproposedanewadmissioncontrolapproachforintegratedservicepacketnetworkbasedontrafficprediction.Intheletter,FARIMA(p,d,q)modelsintheadmissioncontrolalgorithmisdeployed.AmethodtosimplifytheFARIMAmodelfittingprocedureandhencetoreducethetimeoftrafficmodelingandpredictionissuggested.Thefeasibility-studyexperimentsshowthatFARIMAmodelswhichhavelessnumberofparameterscanbeusedtomodelandpredictactualtrafficonquitealargetimescale.Simulationresultsvalidatethepromisingapproach.
简介:SurvivabilityisoneoftheimportantissuesinATM-basednetworkssineevenasinglenetworkelementfailuremaycauseaseriousdataloss.Thispaperintroducesanewrestorationmechanismbasedonmulti-layerATMsurvivablenetworkmanagementarchitecture.ThismechanismintegratesthegeneralcontrolandrestorationcontrolbyestablishingtheWorkingVPslogicalnetwork,BackupVPslogicalnetworkandsparelogicalnetworkinordertooptimallyutilizethenetworkresourceswhilemaintainingtherestorationrequirements.
简介:Aclassoflarge-sealesystems,wheretheoverallobjectivefunctionisanonlinearfunctionofperformanceindexofeachsubsystem,isinvestigatedinthispaper.Thistypeoflarge-scalecontrolproblemisnon-separableinthesenseofconventionalhierarchicalcontrol.Hierarchicalcontrolisextendedinthepapertolarge-scalenon-separablecontrolproblems,wheremultiobjectiveoptimizationisusedasseparationstrategy.Thelarge-scalenon-separablecontrolproblemisembedded,under;ertainconditions,intoafamilyoftheweightedLagrangianformulation.TheweightedLagrangianformulationisseparablewithrespecttosubsystemsandcanbeeffectivelysolvedusingtheinteractionbalanceapproachatthetwolowerlevelsintheproposedthree-levelsolutionstructure.Atthethirdlevel,theweightingvectorfortheweightedLagrangianformulationisadjustediterativelytosearchtheoptimalweightingvectorwithwhichtheoptimaloftheoriginallarge-scalenon-separablecontrolproblemisobtained.Theoreticalbaseofthealgorithmisestablished.Simulationshowsthatthealgorithmiseffective.
简介:Anewmethodforanalyzingdynamicsofcontinuousneuralnetworksisproposed,andthenecessaryconvergenceconditionsforaclassofassociativenetworksareobtained.Basedonthestabilitycriterionandtheequationsofequilibriumsetofthenetwork,synthesisofaclassofassociativeneuralnetworksisgiven.Thestabilitycontrolmodelofasymmetricunstablenetworksissuggested,whichisalsoavalidwayforoptimizationanddynamiccontrolofstableneuralnetworks.
简介:DespiteextensiveresearchonR-trees,mostoftheproposedschemeshavenotbeenintegratedintoexistingDBMSowingtothelackofprotocolsofconcurrencycontrol.R-linktreeisanacceptabledatastructuretodealwiththisissue,butproblemslikephantomstillexist.Inthispaper,wefocusonaconflictdetectionschemebasedonR-linktreeforcompleteconcurrencycontrol.Anin-memoryoperationcontrollistisdesignedtosuspendconflictingoperations.Themainfeaturesofthisapproachare(1)itcanbeimplementedeasilyanddoesnotneedanyextrainformation;(2)Nodeadlocksareinvolvedinlockingscheme;(3)Non-conflictingoperationsarenotrestricted;and(4)PhantomproblemsinR-linktreeareavoidedthroughbeforehandpredication.Theexperimentresultsshowthatthisschemeiscorrectandgainsbettersystemperformance.
简介:Amajordifficultyinmultivariablecontroldesignisthecross-couplingbetweeninputsandoutputswhichobscurestheeffectsofaspecificcontrollerontheoverallbehaviorofthesystem.Thispaperconsiderstheapplicationofkernelmethodindecouplingmultivariableoutputfeedbackcontrollers.Simulationresultsarepresentedtoshowthefeasibilityftheproposedtechnique.
简介:Theforcesensingresistor(FSR)anditsconstructionandcharacteristicaredescribed.Byusingtheoptimalelectronicinterface,theendresultwhichisadirectproportionalitybetweenforceandvoltageisobtained.Thecircuitsofapplicationforforceandpositionmeasurementsintheroboticcontrolaregiven.TheexperimentthatFSRsareplacedonthefingersofBH-1dexteroushandastactilesensorstomeasurethecontactingforcesshowsFSR'sforcesensitivityisoptimizedforuseinthecontrolofrobotcontactingwithenvironment.
简介:Astate-dependentroutingalgorithmbasedontheneuralnetworkmodel,whichtakesadvantageofotherdynamicroutingalgorithmforcircuit-switchednetwork,isgivenin[1].ButtheAlgorthmin[1]isacentralizedcontrolmodelwithcomplexO(N^7),therefore,isdiffculttorealizebyhardware.Asimplifiedalgorithmisputforwardinthispaper,inwhichroutingcanbecontrolleddecentralizekly,anditscomplexityisreducedtoO(10N^3).Computersimulationsaremadeinafullyconnectedtestnetworkwitheightnodes.TheresultsshowthatthecentralizedcontrolmodelhasveryeffectiveperformancethatcanmatchRTNR,andthecentralizedcontrolmodelisnotasgoodasthecentralizedonebutbetterthanDAR-1.
简介:Anoptimalpreviewmethodisappliedtothedesignofterrainfollowingcontrollerforcruisemissile.Inthismethod,trackingerrorsandcontrolincrementsarebothconsideredinthequadraticcostfunction.Integratingthegeneraloptimalservosystemwithapreviewfeedforwardcompensationthatfeedsforwardfuturecommandandfuturedisturbanceproducesanoptimalpreviewservosystem.Intheterrainfollowingsystem,theflightaltitudeofthecruisemissileisacommandsignal,anditsfutureinformationcanbeknownapriori.Hence,wehavedesignedaterrainfollowingcontrollerwithabasicstatefeedbackandafeedforwardcompensationforfuturealtitudeinformation.Simulationresultsshowthattheperformanceoftheterrainfollowingsystemwithsuchanoptimalpreviewcontrollerhasbeenimproveddramatically.
简介:ThisletterproposesaratecontrolalgorithmforH.264videoencoder,whichisbasedonblockactivityandbufferstate.ExperimentalresultsindicatethatithasanexcellentperformancebyprovidingmuchaccuratebitrateandbettercodingefficiencycomparedwithH.264.ThecomputationalcomplexityofthealgorithmisreducedbyadoptinganovelblockactivitydescriptionmethodusingtheSumofAbsoluteDifference(SAD)of16×16mode,anditsrobustnessisenhancedbyintroducingafeedbackcircuitatframelayer.
简介:Thispaperdescribesamodifiedspeed-sensorlesscontrolforinductionmotor(IM)basedonspacevectorpulsewidthmodulationandneuralnetwork.AnElmanANNmethodtoidentifytheIMspeedisproposed,withIMparametersemployedasassociatedelements.TheBPalgorithmisusedtoprovideanadaptiveestimationofthemotorspeed.Theeffectivenessoftheproposedmethodisverifiedbysimulationresults.TheimplementationonTMS320F240fixedDSPisprovided.