以模糊類神經網路實模型參考適應控制架構

The realization of MRAC structure via Fuzzy Neural Network

陳宗胤(Zong-Yin Chen)

指導教授 ： 涂世雄 ； 賴玲瑩

中原大學/工學院/電機工程研究所/碩士(2005年)

https://doi.org/10.6840/cycu200500100

摘要

本文使用模糊類神經網路(Fuzzy Neural Network，簡稱FNN)系統實現間接式模型參考適應控制(MRAC)架構[11]內迴路的線上穩定鑑定器(Fuzzy Neural Network Identifier，簡稱FNNI)與控制器(Fuzzy Neural Network Controller，簡稱FNNC)。而MRAC整體架構有控制器、受控體、參考模型與鑑別器。參考模型根據輸入命令提供受控體欲追隨的響應，使用FNNI線上鑑別受控體的系統模型，並提供受控體的靈敏度函數(Sensitivity)給FNNC，經由FNNC輸出控制量進行線上控制受控體的輸出，使其追隨參考模型的輸出。FNNC與FNNI的輸入尺規因子、規則庫及高斯型歸屬函數的頂點與寬度參數使用倒傳遞法則[11]進行線上調整。倒傳遞法則的學習速率因子則使用離散李亞普諾夫方程式(discrete-type Lyapunov function)加以推導成為動態自調參數。從穩定性分析可證明所有參數調整的方向確實讓FNNC與FNNI趨於漸進穩定。模擬部分採用參考文獻[1]中的範例2進行本文理論驗證。模擬過程系統參數的初值不予最佳化，以達到完全線上參數自調的目的。模擬結果顯示本文FNNC與FNNI的輸出響應最終都呈現穩定收斂，性能上也有不錯的表現。

關鍵字

模型參考適應控制架構 ； 模糊類神經網路

並列摘要

In the thesis, the FNN systems are used to realize a stable identifier (FNNI) and a stable controller (FNNC) in the inner loop of the indirect model reference adaptive control(MRAC) structure. There are a controller, a plant, a reference model and an identifier in the indirect MRAC structure. The reference model generates the output the plant wants to follow according to the plant input, the FNNI identifies on-line the model of the plant and supplies simultaneously the sensitivity of the plant for the FNNC, and the FNNC will generates the control needed to make the plant output track the reference model output. The scaling factors, the rule bases, and the centers and widths of the Gaussian membership functions of the FNNC and the FNNI are adjusted on-line by the back-propagation algorithm. On the other hand, the learning factors of the back-propagation algorithm are dynamic self-tuning parameters decided by discrete Lyapunov function. It is proved that all the changing directions of the adjustable parameters make the FNNC and the FNNI exponentially stable. In the thesis, the second example of the reference paper [1] is used to verify the theoretical design procedure. The initial values of the adjustable parameters of the FNNC and the FNNI are not necessary to be optimized in order to achieve the on-line purpose adjustment. The simulation results show that the output responses of the FNNC and the FNNI converge stably and the performances of both systems are also good.

並列關鍵字

MRAC ； Fuzzy Neural Network

參考文獻

[8] 陳愈仁。「直流無刷馬達模型參考模糊位置控制」。碩士論文，中原大學電機工程研究，民92。

[11] 黃臺民。「適應性類神經網路的強健性探討」。碩士論文，中原大學電機工程研究，民91。

[1] Y.C. Chen and C.C. Teng ,”A Model Reference Control Structure Using a Fuzzy Neural Network”, Fuzzy Sets and Systems 73(1995) 291-312.

[2] C.C. Ku and K.Y. Lee ,”Diagonal Recurrent Neural Networks for Dynamic Systems Control”, IEEE Trans. Neural Networks, VOL. 6, NO. 1, January 1995.

Systems using Recurrent Fuzzy Neural Networks”, IEEE Trans. Fuzzy Systems, VOL. 8, NO. 4, August 2000.