简介:Thesimulationofaone-dimensionalrivernetworkneedstosolvetheSaint-Venantequations,inwhichthevariableparametersnormallyhaveasignificantinfluenceonthemodelaccuracy.ATrial-and-Errorapproachisamostcommonlyadoptedmethodofparametercalibration,however,thismethodistime-consumingandrequiresexperiencetoselecttheappropriatevaluesofparameter.Consequently,simulatedresultsobtainedviathismethodusuallydifferbetweenpractitioners.ThisarticlecombinesahydrodynamicmodelwithanintelligentmodeloriginatedfromtheGeneticAlgorithm(GA)technique,inordertoprovideanintelligentsimulationmethodthatcanoptimizetheparametersautomatically.Comparedwithcurrentapproaches,themethodpresentedinthisarticleissimpler,itsdependenceonfielddataislower,andthemodelaccuracyishigher.Whentheoptimizedparametersaretakenintothehydrodynamicnumericalmodel,agoodagreementisattainedbetweenthesimulatedresultsandthefielddata.
简介:Theknowledgeofflowregimesisveryimportantinthestudyofatwo-phaseflowsystem.AnewflowregimeidentificationmethodbasedonaProbabilityDensityFunction(PDF)andaneuralnetworkisproposedinthispaper.Theinstantaneousdifferentialpressuresignalsofahorizontalflowwereacquiredwithadifferentialpressuresensor.ThecharactersofdifferentialpressuresignalsfordifferentflowregimesareanalyzedwiththePDF.Then,fourcharacteristicparametersofthePDFcurvesaredefined,thepeaknumber(K1),themaximumpeakvalue(K2),thepeakposition(K3)andthePDFvariance(K4).Thecharacteristicvectorswhichconsistofthefourcharacteristicparametersastheinputvectorstraintheneuralnetworktoclassifytheflowregimes.Experimentalresultsshowthatthisnovelmethodforidentifyingair-watertwo-phaseflowregimeshastheadvantageswithahighaccuracyandafastresponse.Theresultsclearlydemonstratethatthisnewmethodcouldprovideanaccurateidentificationofflowregimes.