應用基因演算法於微波濾波器之最佳化設計

Abstract

本研究結合基因演算法及糢糊理論的歸屬函數,提出一多目標基因演算 法,來搜尋濾波器在設計過程中最佳參數之組合.在濾波器的設計過程中, 通常使用Gradient,Quasi-Newton及Random等方法來搜尋最佳電子元件參 數的組合.然而,利用這幾種搜尋方法所找出來的參數組合常無法滿足微波 濾波器之多項規格,且容易陷入局部最佳解.此外,上述方法的搜尋結果易 受搜尋起始解的影響且需對各項規格的績效指標分別給予不同的權重,利 用加權平均法將之整合成單一指標才可應用. 為了改善上述缺失,本研究 利用糢糊理論的歸屬函數,藉由最大最小法(maxmin),將多目標的問題轉為 單一目標的問題,來解決不同的品質特性和單位不同之績效,給予權重的難 題.經由本論文所提之多目標基因演算法可找到完全符合規格的參數組合, 且此法不受起始解及搜尋範圍設定之影響,較以往之方法更具穩健性. This thesis presents a multiple objective genetic algorithm for finding a set of near-optimum component parameters in the design of a microwave filter.In giving the component parameters of a microwave filter,most previous works model this prolem as a space-search problem and apply three major approaches,known asthe Gradient,the Quasi-Newton,and the Random,in order to find an optimum setting from the solution space. These three approaches are deficient in three aspects. First, the proposed solution may be trapped to a local optimum point before the global optimum setting is found. Second, for a design to meet multiple objectives, it is difficult to give appropriate weight to each objective, and therefore the proposed solution may not meet all the criteria. Third,the proposed solution is dependent on the initial settng for space search. To overcome the above three issues,this research proposes a genetic algorithm including a novel membership function and applies the max-min operator to convert the multiple objectives space search problem into a single objective one. A large number of experiments has been performed and shown satisfactoryresults.