資源描述:
《時滯系統(tǒng)的rbf神經(jīng)網(wǎng)絡(luò)預(yù)測控制》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、浙江理工大學(xué)碩+學(xué)位論文摘要時滯現(xiàn)象普遍存在于工業(yè)過程中,且時滯系統(tǒng)難以控制。許多補償方法在理論上能克服純滯后帶來的動態(tài)品質(zhì)影響,但由于工業(yè)現(xiàn)場的不確定性和干擾的隨機存在,使得建立精確的數(shù)學(xué)模型很難實現(xiàn)。預(yù)測技術(shù)具有適應(yīng)性強、響應(yīng)速度快、超調(diào)量小等特點,可以很好的消除時滯和不確定性的影響,而且對模型的精確度要求不高,因此可以很好的控制時滯系統(tǒng)。隨著工業(yè)過程的復(fù)雜化,系統(tǒng)的非線性程度越來越高,基于線性模型的預(yù)測控制己很難應(yīng)用。由于神經(jīng)網(wǎng)絡(luò)能對任意復(fù)雜的非線性函數(shù)充分逼近,因此非線性時滯系統(tǒng)的神經(jīng)網(wǎng)絡(luò)預(yù)測控制得到了迅速的發(fā)展。本文研究了動態(tài)矩陣控制(DMC)和廣義預(yù)測控制(GPC)的基本原理,
2、將它們應(yīng)用于時滯系統(tǒng),取得良好的控制效果,并通過仿真深入探討了優(yōu)化時域長度、控制時域長度、柔化因子以及控制加權(quán)系數(shù)等預(yù)測控制參數(shù)對系統(tǒng)性能的影響。文中對BP算法深入研究,基于“重新息,輕老息”的思想,經(jīng)過二二次指數(shù)平滑處理,對BP算法提出一種改進,仿真表明改進算法具有收斂速度快、預(yù)測精度高的特點。針對BP網(wǎng)絡(luò)易陷入局部收斂,訓(xùn)練速度慢,通過對同一輸入函數(shù)的靜態(tài)辨識仿真,驗證RBF網(wǎng)絡(luò)能克服以上缺點?;诖耍疚膶BF神經(jīng)網(wǎng)絡(luò)和動態(tài)矩陣控制算法相結(jié)合,提出了基于RBF的動態(tài)矩陣控制。仿真研究表明,該方法運用于時滯系統(tǒng)中不僅顯示良好的控制效果,還具有自適應(yīng)的調(diào)整功能。關(guān)鍵詞:H、J‘滯系統(tǒng);
3、預(yù)測控制:神經(jīng)};b《絡(luò);二次平滑指數(shù):RTRL算法浙江理。r大學(xué)碩十學(xué)位論文AbstractThetime—delayphenomenonwithintheindustryprocessiswidespread,butthesystemsaredifficulttobecontrolled.Manymethodscanovercomethebadeffectofthetime.delayqualityintheory,buttheyneedtheprecisemodelsofthesystemswhicharedifficulttobeproducedbecauseoftheindete
4、rminationintheindustryprocessandthestochasticdisturber.Atthesametime,thepredictivecontrolcancontrolthetime—delaysystemsbecauseithasthecharactersasfollowed:goodself-tuningability,fastresponse,littleovershot,andSOon.Besides,thepredictivetechniquedoesnothaveastrictrequesttothesystemsmodels.Withthecom
5、plicationandmoreandmorenon.1inearizationoftheindustryprocess,itisdifficultforthepredictivecontrolbasedonlinearsystemstobeapplied.Duetotheuniversalapproximationofneuralnetworkforarbitrarynonlinearmapping,neuralnetworkpredictivecontrolfornonlineartime—delaysystemsrapidlydevelopedinrecentyears.Inthis
6、paper,theDMCandGPCstudiedandusedintime-delaysystemshasgoodcontroleffect.Furthermore,theimpacttothesystemsresponseoftheparametersofpredictivecontrolisstudiedthroughsimulation,suchastheoptimizationtimedomain,controltimedomain,softengeneandcontrolweightfactor.TheBParithmeticisstudiedandimprovedbasedo
7、nthethoughtofthinkingalotofnewinformationandtakingtheoldinformationlightlythroughdealingwiththeoriginaldataaccordingtodoublesmoothing.Andthesimulationresultssuggestthattheimprovedarithmetichasagoodrapidityofconve