改良式粒子群演算法最佳化设计-德霖.PDFVIP

德霖學報「第27期」   2014年 1月 改良式粒子群演算法最佳化設計 改良式粒子群演算法最佳化設計 邱進東 德霖技術學院 創意產品設計系 副教授 摘要 粒子群演算法(Particle Swarm Optimization，PSO)為啟發式演算法，是仿自鳥類群體覓食行為， 已被廣泛運用於各個領域之最佳化問題，本文以粒子群演算法為基礎，藉由格點法空間分割的概念對搜 尋空間進行分割與擇優，利用格點法一維搜尋的特性檢視各個維度的關聯度，經由關聯度的大小來選擇 最佳化的機制，並基於粒子群演算法的架構，提出改良式粒子群演算法，以求解全域最佳解。 最後以五種標竿函數為例，測試粒子群演算法與改良式粒子群演算法收斂比較，結果顯示改良式粒 子群演算法有很好的收斂效果。 關鍵字： 粒子群演算法、標竿函數、最適化 Optimization Design of Modified Particle Swarm Method Jinn-Tong Chiu De Lin Institute of Technology, Department of Creative Product Design Associate Professor Abstract A new variant of PSO is developed to improve the performance of PSO, which has been widely used in various fields for optimization. The proposed PSO incorporates a space partitioning technique in grid method with PSO. In the searching process of the new algorithm, three position vectors are introduced to enhance the exploration of the particles in the population of PSO and hence helpful for the search of the global solution of a problem of interest. First, the proposed variant of PSO is verified by applying it on the seven benchmark men-made functions and thereafter proved from the results as a robust one. Next, we applied the algorithm on the optimization design of 5 benchmark functions. The convergence test is compared between Particle Swarm Method and Modified Particle Swarm Method. The results shows the Modified Particle Swarm Method have good convergence. Keywords: particle swarm optimization, be nchmark function, optimization 壹、緒論 粒子群演算法啟發於鳥群遷移的生態行為。文獻[1]提出 PSO 對於低維度問題與單極值問 題，大多能搜尋到全域最佳解，而處理高維度問題與多極值問題，往往搜尋不到全域最佳解。 1999 年，Suganthan[2]就線性遞減慣性權重(L.PSO)提出動態性增加鄰近粒子數的方法；將線性 遞減慣性權重函數的觀念，用