基因演算法在知識庫整合之應用

Abstract

知識整合是發展一套專家系統時相當重要的技術，但是他有時花費不少時 間。知識整合的過程也往往相當令人厭煩，特別是整合從多個相同領域的 專家擷取而來的或各種不同學習演算法歸納而來的規則集知識庫。在本篇 文章中，我們提出一種自動整合規則集知識庫的整合方法。在我們的自動 整合方法裡，分成兩個發展的皆段，分別是規則集知識庫的編碼和整合。 知識庫的編碼階段中，我們將每個規則集知識庫編成以二進位表示的零壹 字串，此後再拿去當做整合階段中的初始成員。在整合階段裡,我們採用 基因演算法技巧，把多個知識庫整合，規納和尋找出最佳的知識庫。在此 同時，我們採用診斷腦瘤的專家領域來測試我們的方法，並且比較整合前 後的準確率高低及控制整合後的知識庫複雜度。實驗的結果顯示出，我們 的方法能獲得較高的準確率且整所花費的時間明顯的較低。 Knowledge-integration is a very important technique in developing the expert system, but it sometimes takes much time. Especially, when the multiple rule-sets are constructed by multiple experts or induced by various learning algorithms, the integrating process is tedious. In this paper, we will propose an automated knowledge-integration approach to integrate multiple rule-sets. Our approach consists of two phases: rule- sets encoding and rule-sets integrating. For the encoding phase, each rule-set is encoded to a bit-string as a member of an initial population. For the integrating phase, an adaptive searching technique (genetic algorithm) is used to induce the optimal concept description from the multiple rule-sets. In the mean time, experiments in diagnosing brain tumor (DBT) are schemed to compare the accuracy of knowledge integration with that of the original rule sets. Experimental results show that the accuracy concept description can be obtained from the above mentioned approach and the time consumption of the integrating process is obviously reduced.