Please use this identifier to cite or link to this item: `http://hdl.handle.net/11455/8002`
 標題: 一個可學習數值/語意式資訊的糢糊化類神經模糊推論網路A Fuzzified Neural Fuzzy Inference Network that learns from numerical/linguistic information 作者: 李俊毅 關鍵字: numerical;數值;linguistic;Fuzzified;Neural Fuzzy Inference Network;語意式;糢糊化;類神經模糊 出版社: 電機工程學系 摘要: 本文提出一個可處理糢糊資訊的模糊化TSK型類神經模糊推論網路 (FTNFIN)。FTNFIN的輸入和輸出是任意形狀的模糊數。在結構上，FTNFIN是由一系列後件部為TSK型的“如..則..” 法則所發展而成。在TSK型的後件部上，我們採用的是糢糊輸入變數的線性組合再加上一個任意的糢糊集合。這裡，所有的組合參數全都是任意的單值。網路的模糊化及後件部的計算主要是利用 -切割技術，這使得網路可同時處理數值化及口語化的資料。一開始FTNFIN本身並沒有任一條法則存在，他們是由線上且同步的架構與參數學習所建構而成。FTNFIN 具有網路小和高的學習精確度之特性,且能夠應用於語意式資料處理有關的各領域上。在這裡我們將其應用在多條模糊規則和具有模糊輸入輸出的數學方程式之學習及倒車入庫的控制問題。在上述應用上我們均可得到不錯的模擬結果。A Fuzzified TSK-type Neural Fuzzy Inference Network (FTNFIN) for handling fuzzy information is proposed in this thesis. The inputs and outputs of FTNFIN are fuzzy numbers with any shapes. In structure, FTNFIN is a fuzzy network constructed from a series of fuzzy if-then rules with TSK-type consequent parts. In the TSK-type consequence, a liner combination of fuzzy input variables plus a fuzzy set of any shape is adopted, where the combination coefficients are all crisp values. The technique of -cut is used in fuzzification and consequent part computation, which enables the network to deal with numerical and linguistic information simultaneously. There are no rules in FTNFIN initially; they are constructed on-line by concurrent structure and parameter learning. FTNFIN is characterized with small network size and high learning accuracy, and can be applied to linguistic information processing. The network has been applied on learning fuzzy if-then rules, a mathematic function with fuzzy inputs and outputs, and truck backer-upper control problem. Good simulation results are achieved on these applications. URI: http://hdl.handle.net/11455/8002 Appears in Collections: 電機工程學系所