本計畫為去年度提出之兩年期計畫的第二年期計畫，主要目的在發展以粒子群法 (Particle Swarm Optimization method, PSO)結合差分演化法(Differential Evolution,DE)之高階啟發式混合演化法，期可有效地搜尋複雜、多極值之非線性最佳化問題的全域解，並應用於求解結構最佳化設計問題。PSO 是Kennedy 和Eberhat 於1995 年所提出的一種高階啟發式演算法(metaheuristic algorithms)，可以進行全域搜尋，然而過去許多研究結果顯示，PSO 的移動速度會隨迭代次數的增加而趨緩的現象，因而逐漸失去群體中粒子的多樣性(diversity)形成早熟現象(premature)。為了解決這個缺失，故本研究計畫提出兩種新型的混合搜尋法HNPSO 和PSODE，其中HNPSO 是第一年度所提出，目前正在執行中。今年度則將探討PSODE 的混合搜尋策略，其搜尋策略是將群體分成兩個子族群，分別以PSO 和DE 進行演化，並透過兩個子族群間訊息的傳遞與交換，來達到維持粒子群多樣性的目的。研究中，考慮到單一個體可能發生搜尋停滯的現象，故將導入突變機制來改善。此高階混合啟發式演化法將應用於連續、離散和混合變數之結構最佳化設計問題。 ; In this research, a novel PSO-based (particle swarm optimization-based) hybrid metaheuristic algorithms are proposed for finding the global optimum solution of complex and highly multimodal systems, with particular emphasis on structural optimization. The PSO, which is a global search method, was invented by Kennedy and Eberhat in 1995. However, many researchers have indicated that the main drawback of PSO is the problems of premature convergence and slow search speed. To resolve the drawback of the method, a PSO-based hybrid search algorithms, named as PSODE, is proposed. In this method, the particles in the swarm will be divided into two groups first. The particles in the two groups will evolve new positions for the particles by PSO and DE, respectively. During the evolution, the particles in the two groups will share information together such that the diversity of the swarm can be maintained. For preventing stagnation of each particle, a mutation operator will be considered in the algorithms, too. The proposed hybrid algorithm will be applied to solve for realistic problems of structural optimization that are characterized by non-linearity, non-convexity and by continuous, discrete design variables. ; 研究期間 9708 ~ 9807
