標題: 使用靜態和動態類神經網路做系統鑑別和控制設計
System Identification and Control Using Static and Dynamic
作者: 林炳榮
Ping-Zong Lin
王啟旭
李祖添
Chi-Hsu Wang
Tsu-Tian Lee
電控工程研究所
關鍵字: 模糊類神經網路;非對稱性高斯歸屬函數;架構適應演算法;適應性控制;系統鑑別;動態類神經網路;霍普菲爾類神經網路;里阿普諾夫準則;Fuzzy neural network;asymmetric Gaussian membership function;structure adaptation algorithm;adaptive control;system identification;dynamic neural network;Hopfield neural network;Lyapunov criterion
公開日期: 2007
摘要: 針對非線性動態系統,本論文發展一個新的模糊類神經控制器和一個新的霍普菲爾動態類神經網路鑑別器。第一個設計是提出一個適應性自我建構的非對稱性模糊類神經網路控制器,此控制器是由一個自我建構的模糊類神經網路控制器和一個強健控制器組成。自我建構模糊類神經網路控制器具有架構和參數學習功能的自我建構模糊類神經網路,因此可用以模仿一個理想控制器。強健控制器是用來補償自我建構模糊類神經網路控制器和理想控制器之間的模仿誤差。提出的適應性自我建構非對稱性模糊類神經網路控制器應用到二階的混沌系統,模擬的結果顯示提出的控制器可以達到不錯的追跡效果。對於第二個設計,提出一個新的基於霍普菲爾的動態類神經網路,用以執行非線性動態系統的鑑別。應用Lyapunov方法調整神經網路的權重值。藉著似Lyapunov的穩定準則,執行穩定性的分析,且可以保證系統鑑別的誤差收斂性。最後,為了說明此方法的有效性,所提出的設計機構用以鑑別兩個非線性動態系統。模擬的結果顯示,使用Lyapunov方法訓練的動態類神經網路可以得到好的鑑別效果,且符合文中所推導的收歛作用。
In this dissertation, a novel fuzzy neural network control law and a new Hopfield-based dynamic neural network identifier is developed for nonlinear dynamic systems. For the first control design, an adaptive self-structuring asymmetric fuzzy neural-network control (ASAFNC) system which consists of a self-structuring fuzzy neural-network (SFNN) controller and a robust controller is proposed. The SFNN controller uses a SFNN with structure and parameter learning phases to mimic an ideal controller in a real-time environment. The robust controller is designed to compensate for the modeling error between the SFNN controller and the ideal controller. The proposed ASAFNC system is applied to a second-order chaotic dynamics system. The simulation results show that the proposed ASAFNC can achieve favorable tracking performance. For the second scheme, a new dynamic neural network based on the Hopfield neural network is proposed to perform the nonlinear system identification. The weighting factors of the proposed neural network are adjusted by the Lyapunov approach. Stability analysis is performed by the Lyapunov-like criterion to guarantee the error convergence during identification. Finally, in order to illustrate the effectiveness of this method, the proposed scheme is applied to identify two nonlinear systems. The simulation results demonstrate that the proposed dynamic neural network trained by the Lyapunov approach can obtain good identified performance which is consistent with the convergent analysis proposed in this dissertation.
URI: http://140.113.39.130/cdrfb3/record/nctu/#GT009212820
http://hdl.handle.net/11536/69401
Appears in Collections:Thesis


Files in This Item:

  1. 282001.pdf