臺灣博碩士論文加值系統

近年來﹐由於模糊邏輯(Fuzzy Logic) 在控制界上的廣泛應用，使得模糊
邏輯成為控制的一股新潮流。然而它所不能解決的一些問題﹐我們可以藉
由具有學習能力的類神經網路(Neural Networks) 和模糊邏輯做結合以解
決此問題。這個結合的網路稱為模糊類神經網路(Fuzzy Neural
Networks) 。所以本論文首先要提出一個具有模糊推理，模糊法則，及學
習能力的模糊類神經網路。此外我們對於模式參考控制架構一直很有興趣
研究。因為它是一個非直接適應控制系統，可以任由設計者來完成控制目
的。然後我們經由 Lyapunov 函數推導出最佳的學習速率﹐用以保證模糊
類神經網路會收斂。本文旨在利用模糊類神經網路(Fuzzy Neural
Networks) 設計模式參考控制系統。我們先介紹一個簡單的模糊類神經網
路﹐此網路具有模糊邏輯及神經網路的特性。其次利用此模糊類神經網路
來當做控制器及系統判別器﹐藉由系統判別器來提供控制器訊息以完成模
式參考控制系統。然後我們經由 Lyapunov 函數推導出最佳的學習速率﹐
用以保證模糊類神經控制器 (Fuzzy Neural Networks Controller) 和模
糊類神經系統判別器 (Fuzzy Neural Networks Identifier)會收斂。模
擬結果顯示﹐此模式參考控制系統具有即時控制﹐穩健﹐和學習的能力。

In this thesis, we present a design method for a model
reference control structure using fuzzy neural networks (FNN).
A simple fuzzy logic based neural networks system is first
studied. Knowledge of rules is explicitly encoded in the
weights of the proposed fuzzy neural networks and inferences
are executed efficiently high rate. Two proposed fuzzy neural
networks are utilized in the proposed model reference control
structure. One is a controller, called the Fuzzy Neural
Networks Controller (FNNC); the other is an identifier, called
the Fuzzy Neural Networks Identifier (FNNI). The control action
issued by the FNNC is updated by observing the controlled
results through the FNNI. Adaptive learning rates for both the
FNNC and FNNI are guaranteed to converge by a Lyapunov
function. We compare the proposed fuzzy neural networks with
the Horikawa's type I FNN in the simulation. The on-line
control ability, robustness, learning ability, and
interpolation ability of the proposed model reference control
structure using fuzzy neural networks are confirmed by
simulation results.