請用此 Handle URI 來引用此文件:
http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/72822
標題: | 基於分群與資源分配之全域和區域的自適應調整:一個嶄新的實數最佳化方法 Global and Local Adaptation with Clustering Embedded Resource Allocation (GLACERA): A Novel Approach for Continuous Optimization |
作者: | Ming-Chun Lu 呂明峻 |
指導教授: | 于天立(Tian-Li Yu) |
關鍵字: | 實數最佳化,多模態最佳化,分群,多臂吃角子老虎技術,IEEE CEC 2005,IEEE CEC 2013, Continuous Optimization,Multimodal Optimization,Clustering,Multi-armed Bandit Algorithm,IEEE CEC 2005,IEEE CEC 2013, |
出版年 : | 2019 |
學位: | 碩士 |
摘要: | 實數最佳化與現實生活中的問題息息相關。而這些問題因為多模態(multimodal)和崎嶇不平(rugged)等因素變得複雜,促使研究者發展更好和更有效率的解決方法。本論文提出一個名叫”基於分群與資源分配之全域和區域的自適應調整”的新方法。我們感興趣的是趨勢不被崎嶇不平所干擾的問題,也就是區域最佳解能引導我們找到全域最佳解。這些問題可以任意地旋轉、平移和放大縮小,並且可以擁有任意維度。為了能夠更有效率的解決這些問題,本論文提出一個新的分群方法來幫助分辨區域最佳解。接著,我們提出一個新的多臂吃角子老虎技術(multi-armed bandit)技術,來分配資源給各個區域最佳解,使得獲得全域最佳解的機率最大化。然而,在分配資源的過程中,會遇到探索(exploration)和開發(exploitation)的平衡問題。因此,我們提出了一個新的算式來平衡開發和探索的比例。最後,我們探討族群大小來增進我們演算法的表現,並且將測試在兩套標準IEEE CEC 2005和IEEE CEC 2013的結果與四個目前最佳最常用的演算法進行比較。 Continuous optimization problems, which are highly related to real-world problems are complex as they are multimodal and rugged that motivate re-searchers to develop better and more efficient problem-solving methods. In this thesis, we introduce a new algorithm called Global and Local Adaptation with Clustering Embedded Resource Allocation (GLACERA) for continuous optimization. We are interested in problems that their ruggedness do not affect tendency, which means that local optima can give us clues to find global optimum. The problem’s dimension, rotation, shift and scale can be arbitrary.To solve problems more efficiently, a new clustering technique is proposed in this thesis to help identify potential local optima. Then, a new multi-armed bandit technique, aiming to reach global optimum with greater probability,is presented to allocate resources for each local optimum. However, allocating resources to exploit different local optima leads to the common dilemma between exploration and exploitation. As a result, we proposed a new formula to calibrate the ratio between exploration and exploitation according to remaining function evaluations. Finally, the population size is discussed to improve the performance of our algorithm. We compare our algorithm with four milestone algorithms that are commonly used for continuous optimization. The results are evaluated with the continuous optimization benchmark problems of IEEE CEC 2005 and IEEE CEC 2013. |
URI: | http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/72822 |
DOI: | 10.6342/NTU201901855 |
全文授權: | 有償授權 |
顯示於系所單位: | 電機工程學系 |
文件中的檔案:
檔案 | 大小 | 格式 | |
---|---|---|---|
ntu-108-1.pdf 目前未授權公開取用 | 1.81 MB | Adobe PDF |
系統中的文件,除了特別指名其著作權條款之外,均受到著作權保護,並且保留所有的權利。