非线性动态加速系数对粒子群算法的影响

The Influence of Dynamic Nonlinear Acceleration Coefficients on PSO

粒子群算法(PSO)中的加速系数影响着粒子的个体认知和群体认知,而传统算法中的加速系数一般取常量。据研究发现,粒子的个体认识和群体认识分别主导着粒子的全局搜索能力和局部搜索能力,提高粒子个体认识可以有效增强算法的全局搜索能力,而提高粒子的群体认识可以有效增强算法的局部搜索能力。为进一步研究加速系数对粒子群算法的影响,本文在时变线性加速系数的基础上,提出了三种动态自适应非线性加速系数,并利用四个基准函数进行对比仿真实验。实验结果表明:非线性时变加速系数PSO的寻优效果较线性策略有一定的提高,且加速系数以反正切函数动态改变的PSO寻优效果最佳。

The acceleration coefficients in particle swarm optimization have impact on the cognitive and social learning of particles, which are always the constants in traditional algorithms. According to some research, cognitive and social learning factors are two important parameters associated with the global and local search performance of particle swarm optimization respectively. A larger cognitive coefficient provides better global search ability, and the better local search ability can be obtained with a larger social coefficient. In order to study the impact of the acceleration coefficients on particle swarm optimization further, three types of dynamic nonlinear acceleration coefficients based on a linear time variation strategy are proposed in the paper, and four benchmark functions are used to test the performance. Simulations show that the performance of nonlinear strategies is better than linear ones, and the best performance can be obtained with the arctangent variation acceleration coefficients.