本篇論文提出以基因演算法為基礎之模糊建模方法。首先，對於一個未知的系統或是函數，給予此系統一組輸入資料，即可以得到所對應之輸出資料。再針對所收集到之輸入輸出對，將其當作訓練樣本，本篇所提出之演算法可以完善地訓練模糊系統，使其擁有近似未知系統的輸入輸出對應關係。本篇論文的主要考量除了在建立一個系統複雜度較低(使用較少參數/規則數目)、近似效果較好的模糊系統外，並且考慮到使用者的知識背景與建立模糊系統的困難度，而建立一自我組織、全自動化之系統建模方法。此外，對於各輸入變數的重要性與所應分配到的歸屬函數數目，以及不重要甚至是不正確的系統輸入變數的選取與刪除方法也提出了解決方法。各章節中皆與許多其他文獻做比較，並得到令人滿意的實驗結果。 In this dissertation, the GA-based fuzzy modeling algorithm is proposed. For an unknown system or function, giving a set of input data could generate a corresponding set of output data. The gathered input-output data pairs will be the training data set, and the fuzzy system could be effectively trained by the proposed algorithm to have an approximate input-output relation as the unknown system. This dissertation concerns about not only generating a less-complex (with few parameters/rules) fuzzy system with precise approximation accuracy, but also constructing a self-organized and full-automatic fuzzy modeling method. Moreover, this work has proposed the solutions about determining the significance of each input variable and its membership function number, and even the extraction and rejection methods of less-important or inaccurate input variables. In each chapter, abundant experimental comparisons are presented to prove the effectiveness of this work.
