简介:Anewfuzzysupportvectormachinealgorithmwithdualmembershipvaluesbasedonspectralclusteringmethodisproposedtoovercometheshortcomingofthenormalsupportvectormachinealgorithm,whichdividesthetrainingdatasetsintotwoabsolutelyexclusiveclassesinthebinaryclassification,ignoringthepossibilityof'overlapping'regionbetweenthetwotrainingclasses.Theproposedmethodhandlessample'overlap'efficientlywithspectralclustering,overcomingthedisadvantagesofover-fittingwell,andimprovingthedataminingefficiencygreatly.Simulationprovidesclearevidencestothenewmethod.
简介:AfterreviewingtheanalyticaltheoriesofT-Scurve,somemethodsofT-Srelationship,andfuzzysetsforstudyingwatermasses,newmethodsoffittingthemembershipfunctionofoceanicwatermassesarepresentedbasedonthecharacteristicsofT-Scurvefamilyofoceanicwatermasses.ThemembershipfunctionsofoceanicSubsurfaceWaterMasswithhighsalinityandIntermediateWaterMasswithlowsalinityarefittedanddiscussedusingthenewmethods.Theproposedmethodsareusefulinanalyzingthemixingandmodifyingprocessesofthesewatermasses,especiallyintracingtheirsources.Theprinciplesandformulaeofthenewmethodsandexamplesaregiven.