基于动态粒子数的微粒群优化算法

Particle Swarm Optimization Based on Dynamic Population Size

提出了基于动态粒子数的微粒群算法,并建立了粒子数变化函数.该函数包含粒子数衰减趋势项和周期振荡项.衰减趋势项能够在种群向最优解不断收敛的过程中逐渐减少粒子数,以提高粒子效率.周期振荡项中的递增阶段代表了新粒子的随机出现,以增加粒子群的多样性,而周期振荡项中的递减阶段代表了探索性能差的粒子逐渐消亡,以提高优化效率.对4个标准函数进行测试,仿真结果表明该算法能有效地减少计算量,并显著提高全局搜索性能.

The dynamic particle population based particle swarm optimization algorithm （ DPPPSO ） is introduced, in which the time-variant population size function is constructed, which contains an attenuation term and an undulation term. The attenuation term makes the population decrease gradually when the particles are converging to the optimum in order to reduce the computational cost ; the undulation term consists of periodical phases of ascending and descending. In the ascending phase, new particles are randomly produced to avoid the particle swarm being trapped in the local optimal point; while in the descending phase, particles with lower ability gradually die so that the optimization efficiency is improved. The test on four benchmark functions shows that the proposed algorithm effectively reduces the computational cost and greatly improves the global search ability.