規則式系統之雙向近似推理方法

Abstract

本論文針對規則式系統提出兩個新的雙向近似推理方法， 其中第一個方 法是以匹配函數為基礎所發展出來的區間值雙向近似推理的新方法； 第 二個方法是針對稀疏性規則式系統所提出之乏晰插補雙向近似推理新方法 。 我們所提出的第一種新方法比現有的推理方法有更合理的推論結果且 推理能力更強， 此乃因我們是根據匹配函數的匹配方向來作雙向近似推 理之緣故。 在本論文中所提的第二個方法中，我們根據目前乏晰插補推 理方法的缺點， 提出了更具完整性的乏晰插補雙向近似推理新方法，它 能保證在稀疏性規則式系統中，當規則庫內前題與結論都定義為三角形歸 屬函數之 乏晰集合，而且事實也定義為三角形歸屬函數之乏晰集合時， 乏晰插補推理的結論也是具三角形歸屬函數之乏晰集合， 反之亦然。 In this thesis, we propose two new bidirectional approximate reasoningmethods for rule-based systems. The first method is developed to handlebidirectional approximate reasoning for rule-based systems usinginterval-valued fuzzy sets based on the matching function and the second oneis developed to handle fuzzy interpolative bidirectional approximatereasoning for sparse rule-based systems. The first proposed method is morereasonable and powerful than the existing ones due to the fact that the proposed method can perform bidirectional approximate reasoning based on thedirection of matching of the matching function. The second proposed methodcan guarantee that the interpolated conclusion will also be triangular- typefor a triangular-type observation and vice versa.