86級畢業碩士　劉俊麟　發表論文

 在這篇論文中，針對一個梯形模糊集合輸入的類神經網 路知識庫系統(KBNN/TFS)提出模糊規則之驗證、不一致規則 化解以及加入新規則的方法。並使用梯形LR模式的模糊區 間，以方便處理模糊資訊。在此我們使用必要性支持度與可 能性支持度(此稱為支持對)來檢驗與去除不一致性。並且利 用支持對，以及初始學習點的觀念，尋找並加入新規則。我 們利用一個知識庫評估器 (Knowledge Base Evaluator) 的 例子來驗證我們所提出的方法。經過不一致規則的化解後， 不一致規則被去除，且學習結果也有所提升。除此之外，針 對因衝突而被刪除的規則，使用初始學習點產生新規則並加 入系統，其結果更是大幅度的改進僅做不一致規則化解的系 統。

 In this paper, a fuzzy rule inconsistency resolution and insertion method is proposed on a fuzzy neural network, named Knowledge-Based Neural Network with Trapezoidal Fuzzy Set inputs (KBNN/TFS). To facilitate the processing of fuzzy information, trapezoidal LR-type fuzzy interval is employed. Necessity support and possibility support (named support pair) are applied to detect and remove inconsistency. In addition to the support pair, the concept of initial learning point is used to handle rule insertion. We show the works of the proposed methods on an example named Knowledge Base Evaluator (KBE). After inconsistency resolution operations, inconsistencies are resolved, and the learning result is improved. Moreover, to set initial learning point on deleted conflict rule, a new rule is generated and inserted to the system. And the resu much better than only applying inconsistency resolution.