一种双层可变子群的动态粒子群优化算法

摘要

Information sharing method between particles influences the convergence rate and global search capability in particle swarm optimization (PSO) algorithm. We have proposed an information sharing method based on two-layer sub-population by studying a-bout all-linked and ring topology model. It can coordinate approach the search capability and local capacity for optimum. And then, it changes the size of sub-population dynamically by calculating the convergence rate. Sub-population exclusive mechanism can increase the diversity of the particles which makes the sub-swarm away from the local best position. We compare the experiment performance of the proposed algorithm and other extended PSO using five benchmark functions for high-dimension solutions. The experiment shows that the proposed algorithm can avoid local best and improve global searching ability and accuracy while having fast convergence speed.%粒子间信息的共享方式对粒子群优化算法的收敛速度和全局搜索能力有重要的影响.针对全互联、环形拓扑结构,提出基于双层子群的信息共享方式,以收敛率作为子群规模变化的标识,实现子群规模动态变化,协调了算法的全局搜索能力和局部寻优能力.子群排斥机制使子群跳出局部最优解的束缚,提高解的多样性.选取目前比较流行的几种粒子群优化算法,通过五种经典的Benchmark高维函数优化问题进行实验仿真.结果表明基于双层可变子群的动态粒子群优化算法可以有效的避免算法陷入局部最优,在保证收敛速度的同时算法的全局搜索能力和精度有明显的提高.