請用此 Handle URI 來引用此文件:
http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/88782
標題: | 探討多目標組合優化中目標間的交互關係:多目標基因池最佳混合進化式演算法的加強變體 Investigation on Inter-relationship among Objectives for Multi-objective Combinatorial Optimization: an Enhanced Variant of MO-GOMEA |
作者: | 廖勗辰 Hsu-Chen Liao |
指導教授: | 于天立 Tian-Li Yu |
關鍵字: | 多目標最佳化,基因遺傳演算法,建構模塊,供體選擇, Multi-objective optimization,Model building,Genetic algorithm,Donor selection, |
出版年 : | 2023 |
學位: | 碩士 |
摘要: | 多目標基因池最佳混合進化式演算法 (MO-GOMEA) 是一種強大且無需參數的基於模型的基因演算法,擅長解決多目標組合優化問題。這篇研究論文從雙目標優化問題獲取啟發,進而提出自適應捐贈者選擇混合(ADSM)機制,此機制的構思是基於目標間錯綜複雜的相互關係。此機制設計特別針對群集導向和菁英導向混合對於目標在雙目標優化中的一致性、孤立性和不一致性改善的影響進行深入探討。將ADSM整合到MO-GOMEA中後,產生了一個新的變體,即ADSM-MO-GOMEA。對多種基準問題的實證評估顯示,ADSM-MO-GOMEA在反轉世代距離(IGD)和前沿佔用率(FO)兩個指標上均優於原始的MO-GOMEA。在ADSM-MO-GOMEA超越MO-GOMEA的問題中,我們觀察到IGD指標平均提升約10%,且在多目標背包問題的FO指標上也有顯著的提升。此外,本研究將範疇進一步擴大至涵蓋三至五個目標的問題,並考慮群集導向與菁英導向混合在改善目標一致性上的差異。在這種情況下,一致性的程度由顯示改善的目標數量決定。此延伸研究的實證結果揭示了,在處理具有更多目標的優化問題時,將一致性概念與修訂的ADSM結合起來將能有效改進原始MO-GOMEA的演化效率。 The multi-objective gene-pool optimal mixing evolutionary algorithm (MO-GOMEA) is a powerful, parameterless genetic algorithm that effectively solves multi-objective combinatorial optimization problems. Drawing insights from bi-objective optimization, this study introduced the adaptive donor selection mixing (ADSM) mechanism, devised in view of the intricate inter-relationship among objectives. It particularly focuses on the effects of cluster-guided and elitist-guided mixing on three types of improvements in objectives: coherent, solitary, and incoherent. The integration of ADSM into MO-GOMEA leads to the development of ADSM-MO-GOMEA. Empirical assessments on varied benchmark problems underscore that ADSM-MO-GOMEA outperforms its predecessor MO-GOMEA, excelling in both inverted generational distance (IGD) and front occupancy (FO) metrics. An average enhancement of about 10\% in the IGD metric is observed where ADSM-MO-GOMEA outperforms the original, alongside a significant boost in the FO metric for the multi-objective knapsack problem. The research scope is further extended to tackle problems with three to five objectives, factoring in the impact of the coherency of improvement across objectives when employing cluster-guided and elitist-guided mixing. In this context, the degree of coherence is determined by the count of objectives that exhibit improvement. The empirical results from this broader study underscore the advantages of embedding the coherence concept within the refined ADSM in addressing optimization problems with a greater count of objectives. |
URI: | http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/88782 |
DOI: | 10.6342/NTU202303309 |
全文授權: | 同意授權(全球公開) |
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-111-2.pdf | 4.47 MB | Adobe PDF | 檢視/開啟 |
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。