粒子群最佳化演算法改良之研究

Research on a Modified Particle Swarm Optimization Algorithm

李維平(Wei-Ping Lee) ； 王雅賢(Ya-Xian Wang) ； 江正文(Cheng-Wen Chiang)

《科學與工程技術期刊》 4卷2期 (2008/06) Pp. 51-62

https://doi.org/10.7117/JSET.200806.0051

摘要

標準PSO (standard particle swarm optimization)具有較少參數設定、快速收斂等優點，但粒子移動時僅跟隨pbest與gbest，使得標準PSO有容易落入區域最佳解的弱點。本研究提出一個分群式粒子群演算法的架構，將初始產生的粒子用K-means演算法分群劃分搜尋領域後，以實驗法得到較小的V(下标 max)以加強粒子的區域搜尋能力，再經由比較分群各自找到的分群最佳解gkbest，產生全域最佳解，此爲分群式粒子群演算法(K-means particle swarm optimization, KPSO)；另外，爲確保演算法的收斂性，本研究將文化演算法「知識空間」的概念帶入了KPSO中，由知識空間中的粒子來引導主群體粒子前往具良好解答區搜尋，此爲文化分群式粒子群演算法(culture K-means particle swarm optimization, CKPSO)。藉由此兩個PSO的改良演算法，以期提高粒子搜尋到之全域最佳解的準確度。由研究結果可得知KPSO與CKPSO在測試函數中的表現，整體來說均能優於過去學者提出之標準PSO、HPSO (hybrid particle swarm
Optimization)、FPSO(fuzzy adaptive particle swarm optimization)。

關鍵字

群體智慧 ； 粒子群最佳化演算法 ； 文化演算法 ； K-means演算法

並列摘要

According to Particle Swarm Optimization (PSO), all particle were guided to search by pbest and gbest. For preventing the particle trap into local minima, this article purpose K-main PSO. After swarming off the search region of initial particles with K-main Algorithm, the smaller V(subscript max) was got by experimentation to enhance searching ability. All gkbest were got by comparing particles in it's own group and the gbest was got comparing all gkbest. For keeping convergence, this article purpose CKPSO (culture K-means particle swarm optimization). All particle groups were guide to search by particles which were in knowledge space. The experimental results show that KPSO (K-means particle swarm optimization) and CKPSO are effective and gain better performance then SPSO (standard particle swarm optimization), HPSO (hybrid particle swarm optimization) and FPSO (fuzzy adaptive particle swarm optimization).

並列關鍵字

collective intelligence ； particle swarm optimization ； cultural algorithms ； K-means algorithms