多峰值函数优化的改进粒子群算法

Multimodal Function Optimization Using an Improved Swarm Optimizer

多峰值函数优化中,基本粒子群算法进化后期收敛速度较慢,且可能出现最优解粒子在全局最优解附近'振荡'的现象,导致优化精度降低。为此,提出一种具有可控速度因子的改进粒子群算法。在完全随机、部分可控与完全可控三种速度调控策略下,对比研究几种具有不同变速特性的寻优轨迹下算法的精度及运算效率。试验结果表明,通过采用可控的寻优速度因子,优化性能得到改进,特别是采用完全控制策略,不仅可获得较高的优化精度,而且收敛速度更快,表现出更好的综合性能。

In multimodal optimization, convergence of the basic particle swarm optimizer （BPSO） is relatively slow at the late evolution. And, particle with the best fitness may fluctuate around the globally-optimal solution, which decreases optimization precision. Therefore, an improved swarm optimizer with controllable velocity factor is proposed. On the basis of the definition of three strategies for velocity control of evolved particles, i.e. the completely random one, the partial controllable one and the completely controllable one, optimization precision and computation expense of the modified optimizers are researched comparatively by using several tracks for optimization with different velocity-changing features. Experiments show that performance of the BPSO algorithm is improved to some extent by these controllable modes for velocity-updating. Especially, those improved swarm optimizers using the completely controllable strategy are not only of high precision, but also of faster convergence, both of which imply their better overall performance in multimodal optimization.