Please use this identifier to cite or link to this item: http://hdl.handle.net/11455/2732
標題: 模擬退火之應用與改進
The Applications and Improvements of Simulated Annealing
作者: 蘇志傑
Su, Jyh-Jye
關鍵字: Simulated Annealing;模擬退火法;Optimization;Structure;最佳化;結構
出版社: 機械工程學系
摘要: 
本文是對模擬退火法( Simulated Annealing )的應用和效率的提昇進行研究,模擬退火法具有較佳的全域搜尋能力,並可以處理各種變數型態(連續、整數、離散)的最佳化問題,而且搜尋法則極為簡單,因此近年來廣受重視。
本文對模擬退火法的基本搜尋方式如多軸向同時搜尋和單軸向依序搜尋,以及在可接受區(feasible region)內搜尋和使用懲罰函數(penalty function)的搜尋方式進行廣泛的測試。為了提昇搜尋的效率,本文提出兩種方法,一是應用設計靈敏度分析來訂定每一搜尋溫度的可接受區並以此為搜尋空間,以減少不必要的搜尋。另一是從變數的上限區開始搜尋的策略,以達到搜尋效率的提昇。
七個實例被用來測試各種模擬退火法的搜尋方式和本文所提出的作法,所得到的結果被詳細分析,比較和討論。模擬退火法與其他零次法相同,必須進行大量的函數計算,因此對大型結構最佳化而言,必須慎選使用的時機。

The purpose of this thesis is to improve the efficiency of simulated annealing method. The simulated annealing method has a better capability of finding global optimum. Because the method can treat optimization problems with different types of design variables, it is used frequently in recent years.
Two traditional searching approaches include searching in the feasible region and searching with penalty function. During the searches, two strategies of changing design variables are used. One is to change the design variable one at a time and the other is to change them simultaneously. To increase the searching efficiency, two methods are proposed. One is to estimate the feasible region using sensitivity analysis and then proceed to search in this region. The other one is to begin the search from the subregion where the design variables have higher values.
Seven structural optimization problems are employed to test the methods proposed. Their results have been compared, analyzed and discussed. Owing to huge number of function evaluations needed, the application of SA to solve complex structural optimization problems has to be carefully evaluated.
URI: http://hdl.handle.net/11455/2732
Appears in Collections:機械工程學系所

Show full item record
 
TAIR Related Article

Google ScholarTM

Check


Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.