本論文提出一個完全以生物免疫系統為基礎的演算法則-免疫演算法(Immune Algorithm),並應用於多目標(multi-objective)最佳化、單目標多值域(multi-modal)最佳化與實際工程最佳化設計問題(如: 桁架, 結構拓樸及scheduling等)全域最佳解之搜尋。不同於其他演化式演算法,例如遺傳演算法(Genetic Algorithms)、演化策略法(Evolution Strategy),本免疫演算法具有較佳的多樣性與局部搜尋能力。藉由結合生物免疫系統中適應性免疫反應之特徵,例如抗原與抗體之專一性(specificity)與適應性(adaptiveness) 、抗原識別(discrimination)、抗體之株落增殖(clonal proliferation)、抗體之記憶性(memory)與抗體激素(cytokine)等,以及抗體片段重組和抗體多樣性機制,包含自體突變(somatic mutation)、自體重組(somatic recombination)、基因轉換(gene conversion)、基因倒置(gene inversion)、基因飄移(gene shift)與核甘酸插入(nucleotide addition)等,使得本免疫演算法於最佳化搜尋時,同時兼具全域與局部搜尋之能力, 並且能在全域與局部搜尋之間達到平衡。 為了驗證本免疫演算法之搜尋效能,本論文以無限制條件測試函數、具限制條件測試函數、實際工程結構設計等問題進行多目標與單目標多值域最佳解之搜尋。 在經由與其他演化式演算法比較後其結果顯示,以本免疫演算法搜尋之結果確實優於其他演算法,同時亦證實本論文所提之免疫演算法適用於最佳化搜尋問題。
THIS DISSERATTION focuses on developing a novel immune algorithm called for finding Pareto-optimal solutions simultaneously maintaining diversity to single- and multi-objective optimization problems (SOOPs and MOOPs) based fully on the features of a biological immune system. The applications in this dissertation include unconstrained/constrained test functions and truss-structure sizing multi-objective optimization, structural topology single-objective with multi-modally optimization, and single-objective job-shop scheduling optimization problems. The use of proposed immune algorithm as opposed to the evolutionary algorithm (e.g., genetic algorithm, GA, evolution strategy, ES) provides this methodology with superior diversification and local search abilities. Inter-relationships within the proposed algorithm resemble antibody- antigen relationships in terms of specificity and adaptiveness, antibody clonal proliferation, antigen discrimination, and the antibody memory characteristics of adaptive immune responses. Besides, the features for producing antibodies in biological immune system such as gene fragment rearrangement and several antibody diversification schemes (including somatic recombination, somatic mutation, gene conversion, gene reversion, gene drift, and nucleotide addition) are incorporated into the proposed immune algorithm in order to improve the balance between exploitation and exploration. Moreover the concept of cytokines is also combined to algorithm for constraint handling. By using several performance metrics and comparison with the other approaches, the effectiveness of proposed immune algorithm are evaluated by unconstrained/constrained test functions and several engineering applications (truss sizing, structural topology, and scheduling). The simulated results demonstrated that the proposed immune algorithm provides better effect than other methods and suitable for searching in optimizations.
為了持續優化網站功能與使用者體驗,本網站將Cookies分析技術用於網站營運、分析和個人化服務之目的。
若您繼續瀏覽本網站,即表示您同意本網站使用Cookies。