自适应动态控制种群分组的自然计算方法

多种群优化方法可以解决数据量增大导致的优化难度增加的问题,现有的种群分组都采用随机分组或人为设定的方法,没有充分考虑粒子运动轨迹。针对此问题,提出一种种群分组自适应动态控制策略,使用高斯拟合函数作为种群分组的参考曲线,根据函数单调区间划分子种群;对于有越过子种群上界趋势的粒子采用逆向策略,保持种群多样性同时提高收敛速度。该策略不依赖于算法的具体进化过程,适用于所有基于种群优化的自然计算方法。验证实验结果表明了所提新算法的有效性和普适性。

Nature Computation of Self-Adaptive Dynamic Control Strategy of Population Grouping

Multi-population optimization method can solve the optimization difficulty caused by the increase of data volume, but the existing population grouping is carried out by means of random grouping or artificial setting, which doesn't take particle trajectories into full consideration. In view of the problem a self-adaptive dynamic control strategy of population grouping is proposed, which uses Gaussian fitting function as the reference curve of population grouping and divides sub populations according to the function's monotone interval. For particles with the trend of crossing the upper boundary of sub populations, the contrarian strategy is adopted to maintain the population diversity and improve the convergence speed. The tactics does not rely on concrete evolution procedure of the algorithm, and the strategy is applicable to all nature computation means, based on population optimization. The experimental results show the effectiveness and universality of the algorithm.