基于多神經(jīng)網(wǎng)絡(luò)融合的鋁帶坯晶粒度軟測(cè)量技術(shù)研究.pdf

1、中南大學(xué)碩士學(xué)位論文基于多神經(jīng)網(wǎng)絡(luò)融合的鋁帶坯晶粒度軟測(cè)量技術(shù)研究姓名：郭燕萍申請(qǐng)學(xué)位級(jí)別：碩士專業(yè)：控制科學(xué)與工程指導(dǎo)教師：凌玉華20100603ABSTRCTDuringtheprocessofaluminumelectromagneticcasting，thegrainsizeisaimportantindexformeasuringthealuminumstrip’SqualityInordertocontrolthealumi

2、numelectromagneticcastingsystemoptimally，thekeyistomeasurethegrainsizeofaluminiumstriponlineHoweveruntilnowthereisnoanyinstrumentCanbeusedtomeasurethevalueofgrainsizedirectly，onlyrelyonsomeoffiinemethodsuchasmicrostructu

3、reanalysis，whosebiglagwilllargelyaffectthesystem’Soptimalcontr01Withmaturityofsoftsensortechnologyinsolvingparametermeasurementissues，thismeasurementtechniqueisintroducedtoachievingonlinemeasurementofgrainsizeBasedonthep

4、rocessmechanismofaluminumelectromagneticcasting，theworkingprincipleofelectromagneticcastingsystemisanalysedindetail，andtheinfluencingfactorsofgrainsizeisdeeplystudiedwhichiscontributedtoselectingsecondaryvariablesforsoft

5、sensormodelAccordingtothecomplexmechanismandhi曲nonlinearcharacteristicofelectromagneticcastingprocess，multipleneuralnetworks(MNN)isadoptedtoestablishsoftmeasurementmodelforaluminumstrip’SgrainsizeAfternormalizationandpri

6、ncipalcomponentanalysis(PCA)，sampledataisdividedintofoursamplesubspacesthroughK—meansclusteringalgorithm，whichareusedtocreatedifferentsub—neuralnetworks，andfinallyobtainedthepredictivevaluebyweightedfusingthesub—networko

7、utputSub—networkusesBPorRBFneuralnetworkDifferentcombinationsmodelsofMNNgrainsizesoft—sensoraresimulatedbyMATLAB，andtheresultsshowthatcomparingwithsingleneuralnetwork，MNNmodelCanreducethepredictiveriskandimproveforecasti