使用模糊類神經系統在動態系統的鑑別與控制

Abstract

近年來,由於模糊邏輯與類神經網路的結合,使得產生的模糊類神經系 統,不僅具有模擬人類做決定的能力, 而且具有學習的能力.所以,本篇論 文先討論順向模糊類神經網路(FNN).然而, 現今的記憶元件與狀態的反 饋,是廣泛運用於各種系統當中,於是我們以FNN為基礎,去延伸發展一具有 遞回性質的遞回模糊類神經網路(RFNN).這裡研究的順序分成三部份,第一 個部份為修改FNN的架構;第二個部份為參數調整演算法的推導;第三個部 份是證明RFNN為廣泛的近似器.由於所得到的FNN與RFNN同樣具有近似非線 性函數的能力,為了暸解彼此的性能,於是分別將它們運用於非線性動態系 統的鑑別與控制.最後,經過模擬之後,FNN與RFNN在精確度要求上,都有良 好的模擬效果, 但是那一種的性能較好是不容易判斷的. Recently,due to the the combination of fuzzy logic and neural network,the resultingfuzzy neural system not only has the capability of simulation for human making decision, but also has the capability of learning.So we study first the feedforward fuzzy neural network(FNN) in the thesis. However,memory device and state feedback are widely applied on many systems.Therefore, we extend the FNN to develop the recurrent fuzzy neural network( RFNN).This research is divided into three parts.First,the structure of FNN is modifiedand in the second part,the algorithm of adjusting parameters is derived.Then,we prove the RFNN as a universal approximator.Because the resulting RFNN and FNN have the ability of approximating nonlinear functions,they are individually applied on identification and control of nonlinear dynamical systems.Finally,simulation showed that the FNN and RFNNhave good performance,but it is difficult to say which one is better.