摘要
随机优化的PSO只利用了进化过程中的上一时刻t的速度v(t)和位置x(t)信息,以及个体最优值Pi和群体最优值Pg,缺乏对待优化目标函数特征的充分认识,导致了后期进化过程的长期停滞现象。PSO在长期进化过程中,尤其是在经历了大量函数评估次数的进化后期,待优化的目标函数的性态特征可以从进化迭代过程信息中得到了解。通过采集学习PSO进化过程中的目标函数的解分布特征信息,使PSO可以利用这些特征信息来控制部分粒子的重新初始化过程和交叉选择过程,以及在参数选择中平衡探索模式和开采模式。实验结果表明,利用了进化过程信息的PSO可以增加种群的多样性,从而获得更高的优化精度和更少的期望迭代次数,虽然其轻微地增加了进化过程特征采集的时间和空间复杂性。
Particle swarm optimization(PSO) easily falls into the stagnation at the late evolutionary period because it does not know about the characteristics of the objective function completely.In the classic PSO,the finite information,such as the velocity v(t),the location x(t),the individual extremum Pi of the particle and the global extremum Pg of the swarm at the prior time t,is employed to drive the evolutionary process.But in the evolutionary of PSO,the distribution characteristics of solutions of the objective function are hidden in the many and many function evaluations while the evolutionary is iterating.The novel PSO based on evolutionary learning(L-PSO) balances the exploration and the exploitation process and controls the re-initialization and crossover selection of particles through the distribution characteristics of solutions extracted statically from the historical evaluations.The experimental results show that the L-PSO can improve the precise of solution and reduce the expected iterations although the time and space complexity is increased lightly.
出处
《计算机科学》
CSCD
北大核心
2012年第4期193-195,213,共4页
Computer Science
关键词
粒子群优化
进化过程学习
分布特征
智能粒子
Particle swarm optimization
Evolutionary process learning
Distribution characteristics
Intelligent particle