動態(tài)神經(jīng)網(wǎng)絡(luò)與分?jǐn)?shù)階Fourier變換的研究及其應(yīng)用.pdf

1、大連理工大學(xué)碩士學(xué)位論文動態(tài)神經(jīng)網(wǎng)絡(luò)與分?jǐn)?shù)階Fourier變換的研究及其應(yīng)用姓名：田玉松申請學(xué)位級別：碩士專業(yè)：信號與信息處理指導(dǎo)教師：邱天爽20051201田玉松：動態(tài)神經(jīng)網(wǎng)絡(luò)與分?jǐn)?shù)階Fourier變換的研究及其應(yīng)用ResearchandApplicationofDynamicalNepalNetworksandFractionalFourierTransforrllAbstr。actn愴FractionalFourierTransf

2、ormiSageneralizationoftheFourierTransforioItiSatimefrequencyanalysisalgorithminessenceTheFractionalFourierTransformhasmanyexcellentpropertiesItiswidelyusedinnonstafionarysignalanalysisandprocessingInthisthesis，theFractio

3、nalFoYerTransformisusedtOextractappropriatespikefbatllresfromscalpEEGsignalsbasedonspikeshavingthepropertyofGaussfunctionBecausespikesandbackgroundsignalshavedifferentpropertiesinFractionalFoYerregion，firsttllewavelettra

4、nsformisusedtothescalpEEGsignalsteduceaffectofbackgroundnoise；theninproperFractionalFourierregiontheGaussfunctionisusedtosimulatethespikes；thegoodextractingfeaturesCanbeacquiredTheFractionalFoYerTransformissensitivetosec

5、ondaryphasicoflinearchirpletTheFractionalFoYerTransformisappliedtoparameterestimateofGausslinearchirplet’SadaptivedecomposealgorithmBecauseGausslinearchirplet’Smodelhasdifferentwidthandhi曲indifferentFractionalFourierregi

6、on，aGausslinearchirpletcomponentthatstandsoutbackgroundnoiseCallbeacqnimdtoaccuratelyestimateparameterinproperFractionalFourierregionBasicElmannetworkisoneofthedynamicallyrecurrentneuralnetworksksstructureisrelativelysim

7、ple；itsoperationisrelativelysmall；butitsdynamicalpropertyissufficientSoitiswidelyusedTraditionarily，BPalgorithmbasedOngradsdescendingandextendedKalmanfilteralgorithmareusedtoBasicElmannetwork’strainingButthesealgorithmsh

8、averespectivemeritsandlacksExtendedKalmanfilteralgorithmcarlquicklyconvergetoconvergenceButresultisnotadequatelyaccurateBPalgorithmbasedongradsdescendingcarladequatelyaccuratelyrealizenonlinearmappingofinoutsystemsButits

9、lacksrest、ⅣitIlconvergenceslowlyinpolarandthe“bad”propertyoftrackConsideredrespectivemeritsandlacksofBPalgorithmandextendedKalmanfilteralgorithm，anewextendedKalmanfilteralgorithmisadvancedThisnewalgorithmconsideroutputof

10、hiddenlayerasstatesofnonlinearsystem，realizestatesquicklytrackthroughExtendedKalmanfilteralgorithmandmodifiedweightsofnetworkthroughgradsdescendingDuringnetworktraining，precisionofconvergenceisimprovedthroughupdatingtrai