模糊專家系統發展環境之研製

Abstract

傳統的專家系統缺少良好的人機界面與機制去擷取專家的知識，而對模糊 知識的擷取更是顯得無能為力。目前的專家系統主要有兩個主流: 法則式 的專家系統(rule-based system)與案例式的專家系統(case-based system)。而不論是法則式或案例式的專家系統都有它的優點和缺點。在 此篇論文中我們提出了一種建構模糊專家系統的方法，稱為"以表格為基 礎的模糊專家系統"(grid-based fuzzy expert system environment)， 簡稱GBFES。在知識擷取 (knowledge acquisition)與知識展現( knowledge representation)方面GBFES不僅含括了許多新的專家系統的方 法而且也引用了模糊 (Fuzzy)的觀念。在資料型態(data type)方面GBFES 也提供了許多傳統專家系統缺少的且難以處理的型式，如"顏色"、"圖型" 、純量模糊資料項及非純量模糊資料項等，此外在GBFES中我們也採用了 一種圖學的方法來定義隸屬函數(membership function)。而最重要的 GBFES的輸出不再是單一的答案而是輸出所有可能的答案選項，每個答案 選項均有三個評定量(measures)來讓使用者決定符合其需求的最佳答案。 最後在GBFES推理方面(reasoning)，我們也作了廣泛且深入的討論。讀者 將可由"動物選擇"的實驗中了解整篇論文的精神與要旨。 A traditional expert system shell has a poor interface to elicit experts' knowledge and it is difficult to represent fuzzy expertise. In the rule-based system, to capture em- bedding meanings may generate too many rules that expert system hard to cope with. Traditional expert system shells lack too many ingredients to form a good recipe of building a good expert system. In this dissertation, we proposed a grid-based fuzzy expert system environment -- GBFES, whose idea consists of repertory grid, EMCUD, fuzzy concepts and case-based reasoning. For knowledge representation and ac- quisition, GBFES supports many data types which contain sca- lar-fuzzy term, non-scalar fuzzy term, and color, etc. GBFES also employs a new way -- Bezier curve to define membership function of scalar-fuzzy term. On the other hand, two con- cepts were proposedin GBFES -- feasibility and reliability. These two measures and certainty factor construct the struc- ture of GBFES's reasoning mechanism. We will discuss a lot in reasoning and the most concerning is "Whst is the best way of reasoning." Finally, for end user, GBFES supports three measures of objects in goal set, which can assist the users in making better decisions in various circumstances.