资源描述:
《论文翻译第3部分》由会员上传分享,免费在线阅读,更多相关内容在工程资料-天天文库。
1、(17)3.Designoftheneuro-fuzzycontrollerTheblockdiagramoftheNFCforthespeedcontrolofthePMSMisshowninFig.3.TheNFCiscomposedofapatternset,anoff-linelearningalgorithmwithbackpropagationandaneuro-fuzzynetwork.Theneuro-fuzzynetworkistrainedusingoff-linelearningalgorithm.Fortheoff-linelearn
2、ing,apatternsetisrealizedusingdynamicsignalanalysisofthemotor.Thepatternsetisalookuptablethatconsistof%(/),RHOand△/*(/)・Iisnumberofpatterninthepatternset(/=1,300).Thelookuptableisusedinoff-linelearning.Fig.3.BlockdiagramoftheNFCforthespeedcontrolofthePMSM.TheNFChastwoinputs,thespee
3、derrore(0andthederivativeofspeederror离叭TheoutputischangeofthecontrolcurrentA/.FortheNFC,afourlayerneuralnetworkasshowninFig.4isused.LayersI,II,IIIandIVrepresenttheinputsofthenetwork,themembershipfunctions,thefuzzyrulebaseandtheoutputsofthenetwork,respectively.Fig.4.Schematicdiagram
4、oftheneuro-fuzzynetwork.3.1.ThefeedforwardalgorithmLayerI:Inputsandoutputsofnodesinthislayerarerepresentedasnet;=%,#=.A1(net;)=net;=%net2=©g、兄=(net?)=net?=&切where^)areinputsand3?areoutputsofinputlayer.Inthislayer,theweightsareunityandfixed.LayerII:Inthislayer,eachnodeperformsafuzzy
5、setandtheGaussianfunctionisadoptedasmembershipfunction此「GgU)-exp(n&jJ問=)=exp(nei;:)(18)where加尙,〃必and吩,arethemeanandthestandarddeviationoftheGaussianfunction,respectively.Therearej+nodesinthislayer.LayerIII:ThislayerrepresentsrulebaseusedintheFLC.Inthislayer,aproductoperatorisusedto
6、representaruleineachnodenct#=(£xx;1:),球=(net;]1)=net;]1(〔刃咗(/)=啞卩-1)+从A喷⑴(25)wherethevaluesoflinkweightsbetweenmembershiplayerandrulebaselayerareunity.LayerIV:ThislayerincludesinferenceanddefuzzificationusedintheFLC.Fordefuzzification,thecenterofareamethodisused“力力啜堺心力力閲Jtjk碘'上老卅=£
7、v(nc世三(20)vvivwhere心iscentersoftheoutputmembershipfunctionsusedintheFLC・"hisoutputoftherulelayer・aandharethenumeratorandthedenominatorofthefunctionusedinthecenterofareamethod,respectively.IntheNFC,theaimoflearningalgorithmadjusttheweightsof,砲占anday,.Thesupervisedgradientdescentmeth
8、odisusedinthelearningalgor