简介：Artificialintelligencetechniqueshavebeenusedtopredictbasicoxygenfurnace(BOF)end-points.However,themainchallengeistoeffectivelyreducetheinputnodesastoomanyinputnodesinneuralnetworkincreasecomplexity,decreaseaccuracyandslowdownthetrainingspeedofthenetwork.Simplypicking-upvariablesasinputusuallyinfluencevalidityofmodel.Itisquitenecessarytodevelopaneffectivemethodtoreducethenumberofinputnodeswherebytosimplifythenetworkandimprovemodelperformance.Inthisstudy,avariable-filtratingtechniquecombiningbothmetallurgicalmechanismmodelandpartialleast-squares(PLS)regressionmethodhasbeenproposedbytakingtheadvantagesofbothofthem,i.e.qualitiveandquantativerelationshipsbetweenvariablesrespectively.Accordingly,afuzzy-reasoningneuralnetwork(FNN)predictionmodelforbasicoxygenfurnace(BOF)end-pointcarboncontentbasedonthistechniquehasbeendeveloped.Thepredictionresultsshowedthatthismodelcaneffectivelyimprovethehitrateofend-pointcarboncontentandincreasenetworktrainingspeed.Thesuccessfulhitrateofthemodelcanreachupto94.12%withabout0.02%errorrange.