采用種群進化的粒子群多模態函數優化

王棟浩 靳其兵 牛亞旭

摘? 要： 針對多模態問題中收斂速度慢，粒子種群容易早熟的問題，提出一種利用種群進化的改進粒子群算法（SRPSO）。該算法在經典多模態粒子群優化算法SPSO的基礎上，通過對初始種群進行均勻化空間拉伸更新，同時，對每個新粒子進行梯度進化，加快了粒子種群收斂速度。為了避免種群早熟，漏掉部分極值點，引入環形拓撲模型提高種群交流能力，同時對速度更新公式做出改進。最后利用6個經典的測試函數對三種經典算法做對比實驗，結果表明SRPSO具有加快收斂速度，提高尋優成功率的性能。

關鍵詞： 多模態函數; 粒子群算法; 小生境技術; 群智能; 環形拓撲; 粒子梯度進化

中圖分類號： TN911.1?34? ? ? ? ? ? ? ? ? ? ? 文獻標識碼： A? ? ? ? ? ? ? ? ? ? ? ? ? ?文章編號： 1004?373X（2020）13?0106?04

Particle swarm multi?modal function optimization adopting population evolution

WANG Donghao， JIN Qibing， NIU Yaxu

（College of Information Science & Technology， Beijing University of Chemical Technology， Beijing 100029， China）

Abstract： In view of the multi?modal related problems like slow convergence rate and particle population being prone to premature， a species ring?topology particle swarm optimization （SRPSO） is proposed. On the basis of the classical multi?modal species?based PSO （SPSO） algorithm， the proposed algorithm accelerates the convergence rate of the particle population by uniformization space stretching and updating of the initial population and gradient evolution of each new particle. In order to avoid population premature and missing some extreme points， a ring topology model is introduced to improve the communication ability of the population. Meanwhile， the speed updating formula is improved. The contrastive experiments were performed on the three classical algorithms by six classical test functions. The results show that SRPSO has the performance of accelerating the convergence rate and improving the success rate of optimization.

Keywords： multi?modal function; PSO algorithm; niche technology; swarm intelligence; ring topology; particle gradient evolution