资源描述：
《Neural network and fuzzy logic techniques for》由会员上传分享，免费在线阅读，更多相关内容在学术论文-天天文库。
1、NEURALNETWORK&FUZZYLOGICTECHNIQUESFORTIMESERIESFORECASTINGGeorgiosLezos,andMonteTu11GEORGIOSLEZOSGraduateStudentSchoolofElectrical&ComputerEngineeringUniversityofOklahoma414CarsonEngineeringCenter,202Wesidoyd,Room219NO~I~XI,OK.73019-1023PHONE:(405)325-8620F
2、AX:(405)325-7066E-Mail:pelezos@?otmail.comMONTETULLAssociateProfessorSchoolofElectrical&ComputerEngineeringUniversityofOklahoma414CarsonEngineeringCenter,202WestBoyd,Room219NO~XI,OK.73019-1023PHONE:(405)325-4278FAX:(405)325-7066E-Mail:tull@,ou.edu191NEUMLNE
3、TWORK&FUZZYLOGICTECHNIQUESFORTIMESERIESFORECASTINGABSTRACTPredictionisatypicalexampleofageneralizationproblem.Thegoalofpredictionistoaccuratelyforecasttheshort-termevolutionofthesystembasedonpastinformation.Neuralnetworkandfuzzylogictechniquesareusedbecause
4、theybothhavegoodgeneralizationcapabilities.Theembeddingdimension(numberofinputs)andthetimelagselectionproblemistreatedinthispaper.Itisproposed,thattheselectionoftheappropriateembeddingdimensionandtimelagfortheinput/outputspaceconstructionplaysanimportantrol
5、eintheperformanceoftheabovenetworks.Itisshownthatthe“traditionallyaccepted”choicesfortheembeddingdimensionandtimelagarenotoptimal.Theproposedmethodoffersanimprovementoverthetraditionallyacceptedparameterchoices.Differentanalytxaltechniquesforthedeterminatio
6、noftheseparametersareused,andtheresultsareevaluated.192NEURALNETWORK&FUZZYLOGICTECHNIQUESFORTIMESERIESFORECASTINGI.INTRODUCTIONTimeseriesanalysisincludesthreeimportantspecificproblems:prediction,modeling,andcharacterizations.Onlythepredictionproblemistreate
7、dinthispaper.Atimeseriesisasequenceofobservablequantitiesxl,x2,x3,...,x,takenfromasystematregularintervalsoftime.Therearedifferenttypesoftimeseries:linearhonlinear,stationaryhonstationaryorchaotictimeseries.Wearemoreconcernedwithnonlinear,chaotictimeseriesb
8、ecausetheyaremoredifficulttopredictandareoftenencounteredinreallifeproblems.Thegoalofpredictionistoaccuratelyforecasttheshort-termevolutionofthesystembasedonpastinformation.Ifaneuralnetworkisusedforpre
