標題: | 基因演算法於以電磁場論為基礎之微波濾波器合成之應用 EM-Based Microwave Filter Synthesis by Genetic Algorithms |
作者: | 梁逢烈 Liang, Fong-Lieh 莊晴光 C. K. C. Tzuang 電信工程研究所 |
關鍵字: | 基因演算法;濾波器;電磁場論;最佳化;微波;genetic algorithm;filter;EM theory;optimization;microwave |
公開日期: | 1996 |
摘要: | 基因演算法為解決問題的有效工具,並已廣範的應用於許多領域諸如人工 智慧,電路合成及最佳化.這是由於它在複雜的多模空間強而有效全域搜尋 能力.因此本篇論文採用了基因演算法的概念設計了結合了三維空間電磁 全波模擬軟體的全新濾波器設計工具.為了驗證基因演算法亦可解決傳統 最佳化問題,以及它本身即有著獨特的不同於傳統最佳化方法的特色,我們 設計了兩個基因演算法的不同用途,分別為:(一)用於連續變數最佳化上. 以一個帶拒濾波器的結構的幾何參數為變數,中心頻率在我們要求下,由15 GHz移到13GHz,說明基因演算法的概念可以用來找出和應用傳統最佳化搜 尋法所得到一樣的解;(二)用於離散變數的最佳化上.以被切割後的單位金 屬塊存在與否作為變數,使一個開路截線(open stub)的|S21|參數在頻率6 GHz處小於0.3;從其最佳化過程中,我們發現每一代最佳結構之間的變化遵 循著一個簡單的規則並發現和物理上的解釋吻合.這個結果說明了基因演 算法的最佳化應用也能像生物演化般的展現. Genetic algorithms (GA) have been widely applied to a lot of fields such as artificial intelligence, circuit synthesis and optimization as design tools and problem solvers because of theisr global versatility and ablity to optimize in complex multimodal search spaces. We adopt the concept of GA and program new filter design tools with the application of the 3-D full wave EM simulator as the simulation engine. To validate that GA could be used to solve the traditional optimization problems and the most characteristics it owns is superior than traditional methods, we design two different usgae of GA respectively: (1) For the optimization of continuous variables. We take the geometry parameters of the bandstop filter structure as the variables and the centerfrequency is shifted from 15 GHz to 13 GHz successfully under our specifications.This illustrates that the concept of GA could be implemented as the traditional optimizer to find the best design. (2) For the optimization of discrete variables. We take the existence of unit patches, which are gridded from the region needed to be optimized, as the optimization variables. We specify that the |S21| of the open stub is less than 0.3 at 6 GHz. From the intermediate results, we find a simple rule between the best design of the generation. We discover the phenomena corresponding to the results of physical explnation. Theconclusion is made to explain that state-of-the-art optimization application of GA could also be demonstrated as the evolution of real world creatures. |
URI: | http://140.113.39.130/cdrfb3/record/nctu/#NT850436022 http://hdl.handle.net/11536/62096 |
顯示於類別: | 畢業論文 |