基于改進(jìn)的回聲狀態(tài)神經(jīng)網(wǎng)絡(luò)的非線性預(yù)測

《基于改進(jìn)的回聲狀態(tài)神經(jīng)網(wǎng)絡(luò)的非線性預(yù)測》由會員上傳分享，免費在線閱讀，更多相關(guān)內(nèi)容在學(xué)術(shù)論文-天天文庫。

1、南京工業(yè)大學(xué)碩士學(xué)位論文基于改進(jìn)的回聲狀態(tài)神經(jīng)網(wǎng)絡(luò)的非線性預(yù)測姓名：王瑟申請學(xué)位級別：碩士專業(yè)：計算機(jī)應(yīng)用技術(shù)指導(dǎo)教師：蔚承建20060515摘要關(guān)鍵詞：回聲狀態(tài)神經(jīng)網(wǎng)絡(luò)；小波神經(jīng)網(wǎng)絡(luò)；小波分解；混沌時間序列預(yù)測；先驗性；PSO；集群智能II碩士學(xué)位論文ABSTRACTNonlinearsystempredictionusingneuralnetworksappearsgreatefficiencyandhasabundanceofapplications.Recurrentneuralnetworksshowsmoreadvancedadvantagesamongthem

2、againstthesepredictiontasks,althoughitslearningmethodshavenotimprovedmuchmoreforlongtime.Echostatenetworkisonenovelstructureofrecurrentneuralnetwork(RNN)alsoonenovellearningmethodforRNNaswell,it’ssimilarwiththosebio-neural-networksstructurally,andithastheperfectSTMcapabilityasoneRNN.Itempl

3、oysonelargescaleRNNasinformationreservoircalleddynamicalreservoir,thenminimizesthemeaningsquarederror(MSE)duringtrainingtogetthelearningusingcomputingsimpleregressionweightmatrixfrominternalstatestowardsoutputunit.However,thereisonecontradictionexistinginESN,itis:toemploynonlinearneuroncan

4、raisethenonlinearcapabilityofESNbutreducetheSTMofitsimultaneously.IthastoemployoneverylargescaleDRwhenfacethosetoughtaskwhichrequirenotonlyhighnonlinearitybutalsoniceMClikechaotictimeseriesprediction.ThiscausestherunningprocessofESNslowingdownandbecomingmoreinstableduringexploitationperiod

5、.AccordingtothetranscendentalknowledgetheoryofANN,theESNcanemployotherneuralnodetoimprovetheperformance,thewavelonusinginWNNchoseninthisthesis.Theinternalstatespaceisenlargedwheninputsometunedwavelon.TheSWHESNcanpredict46%furtherthantheoriginalESNwithouttypicaldeviationbutonlyconsumingonly

6、30%timeofwhatESNdowhenlearningsamedatasample.Wecan’tforgetthatESNhasimprovedthebestprevious[1]technologybyfactor700.Thisthesisshowstri-highlightviews:1.WeintroducewavelonintoRNNwhichappearsinforwardANNtraditionally.2.Wereducedthediversitybetweenwavelonforthereasontosmoothworkingconditionin

7、ESNratherthanaugmentingthemwhichforwardANNwhoneedlargerbasicvectorfunctionembedded.3.Theparametersinechostatenetworksinvolvedinapplicationaresetbyexpertofechostatenetworkscommonly,whichusuallywasteofcomputationresource,inthispaperwepresentonemethodthattooptimi