資源描述：
《基于神經(jīng)網(wǎng)絡(luò)直流無刷電機(jī)控制策略的研究》由會(huì)員上傳分享，免費(fèi)在線閱讀，更多相關(guān)內(nèi)容在學(xué)術(shù)論文-天天文庫。
1、摘要'’721163近年來，隨著稀土永磁材料、電力電子器件、微處理器、變頻器和控制理論的飛速發(fā)展，直流無刷電機(jī)得到了廣泛應(yīng)用。然而直流無刷電機(jī)驅(qū)動(dòng)系統(tǒng)受電機(jī)參數(shù)變化、外部負(fù)載擾動(dòng)和非線性等不確定性的影響，要獲得高性能的電機(jī)驅(qū)動(dòng)系統(tǒng)，必須研究先進(jìn)的控制策略以解決這些不確定性的影響，使系統(tǒng)具有較強(qiáng)的自適應(yīng)能力和抗干擾能力。本文在對(duì)神經(jīng)網(wǎng)絡(luò)結(jié)構(gòu)及算法研究的基礎(chǔ)上，利用神經(jīng)網(wǎng)絡(luò)在非線性、不確定性系統(tǒng)控制和辨識(shí)方面優(yōu)越的性能，提出了一種直流無刷電機(jī)高性能在線辨識(shí)和控制的神經(jīng)網(wǎng)絡(luò)控制器。在系統(tǒng)運(yùn)行時(shí)，兩個(gè)不同的神經(jīng)網(wǎng)絡(luò)控制電機(jī)的動(dòng)態(tài)性能。在第一個(gè)控制系統(tǒng)中采用三層神經(jīng)網(wǎng)絡(luò)來控制轉(zhuǎn)子的轉(zhuǎn)
2、速，第二個(gè)控制電機(jī)定子的電流，并采用Levenburg。Marquardt算法對(duì)三層前饋神經(jīng)網(wǎng)絡(luò)進(jìn)行在線訓(xùn)練，經(jīng)過較短訓(xùn)練時(shí)間后，直軸和交軸電流就可以成功的跟蹤預(yù)定軌跡。這種控制策略適用于電機(jī)的動(dòng)態(tài)不可確定性以及自身的非線性。仿真結(jié)果表明在外界有干擾時(shí)這種結(jié)合自適應(yīng)控制的神經(jīng)網(wǎng)絡(luò)控制器經(jīng)過訓(xùn)練后可以得到理想的結(jié)果。關(guān)鍵詞：直流無刷電機(jī)(BLDCM)；神經(jīng)網(wǎng)絡(luò)控制器；在線訓(xùn)練ABSTRACTInrecentyears，advancesinrare—earthpermanentmagneticmaterial，powerdevices，micro—processor,conve
3、rterdesigntechniqueandcontroltheoryhavemadeBLDCMplayavitalroleinmotion．contr01applications．ttowever,thecontrolperformanceoftheBLDCMdriveisstillinfluencebyuncertainties．whichusuallyfeaturesparametervariations。extemalloaddisturbancesandnonlineardynamics．Toachievehigll．performanceBLDCMdrive，w
4、hichhasgreatabilityofadaptationandbetterperformanceagainstdisturbances，advancedcontrolschemeshavetobedevelopedtodealwiththeseuncertainties．Thisthesis，onthebasisoftheresearch01"1structureandlearningalgorithmofneuraInetwork，ahigh．performanceneuro-controllerwithsimultaneousonlineidentificatio
5、nandcontroliSproposedforcontrollingBLDCM．Thedynamicsofthemotorarecontrolledusingtwodifierentneuralbasedidentificationandcontrolschemes．a(chǎn)ssystemiSinoperation．Int11efirstscheme．a(chǎn)naRemptiSmadetocontroltherotorangularspeed．utilizingasinglethree—hidden．1ayernetwork．thesecondschemeattemptstocont
6、rolthestatorcurrents．Thisschemesincorporatesthreemultilayeredfeedforwardneuralnetworksthatareonlinetrained，usingtheLevenburg-Marquardttrainingalgorithm．ThecontrolofthedirectandquadraturecomponentsofthestatorcurrentCansuccessfullytracedtrajectoriesafterrelativelyshortonlinetrainingperiods．T
7、hecontrolstrategyadaptstotheuncertaintiesofthemotordynamicsandtheirinherentnonlinearities．PromisingsimulationresultshavebeenobservedwhentheneuralcontrolleriStrainedinanenvironmemcontaminatedwithnoise．Keywords：BrushlessDCMotor；Neuro—controller；Onlinetraining．II
