模糊分類元系統及其應用

Abstract

模糊專家系統已經在很多方面有著不錯的表現。但模糊專家系統的建構有時有知識擷取方面的困難，因而有不少論文嘗試自動建構模糊專家系統，例如利用遺傳演算法。在這篇論文中，我們嘗試整合模糊專家系統和分類元系統，提出模糊分類元系統的模型，來自動建構模糊專家系統。 在模糊分類元系統的模型裡，我們簡化了原來分類元模型的報酬分配機制，使系統更加簡單穩定。另外選提出取代線的機制，使系統在只有稀少的資源下，也能有穩定優異的表現。 在這篇論文中，我們實作二個問題來驗證此模型，第一個是鳶尾花分類問題，第二個是動態系統的預測問題。由實作得知，模糊分類元系統不但穩定而且效能優異，所需要的資源也很少。另外藉著系統參數的改變，我們也觀察到模糊分類元系統的一些特性。

Fuzzy expert systems have been performing well in many fields. However, due to the problem of knowledge acquisition, the time and energy spent in constructing fuzzy expert systems are still much . Many papers have attempted to achieve automatic construction of fuzzy expert systems. For example, using genetic algorithm to solve this problem has been obtaining more and more attention. In this thesis, we try to integrate fuzzy expert systems and Holland's classifier system. We propose a fuzzy classifier system model for the automatic construction of fuzzy expert systems. In this model, we simplify the mechanism of credit assignment in classifier system. Consequently, the system is simpler and more robust. We also propose the concept of replacing line. By using this mechanism, the system performance is good and stable even when the supply of resource is low. In this thesis, we solve two problems to verify the proposed model. The first problem is the iris classification problem. The second one is the prediction of dynamical systems. When solving these two problems, we observed that fuzzy classifier system is not only effective but also stable. Furthermore, with the change of system parameters, we observed some important features of fuzzy classifier system.