一种基于分量热力学迁移策略的并行多种群GEP

A Parallel Multipopulation Gene Expression Programming Based on Component Thermodynamical Migration Strategy

针对传统并行多种群GEP存在着优良个体的传播和种群多样性之间的冲突问题,提出一种基于分量热力学迁移策略的并行多种群GEP算法(CTDPGEP)。该算法在当前子种群中选择出若干个优良个体和若干个随机个体组成精英子空间,并将精英子空间传送至其他各子种群的迁移区中;其他各子种群异步地将其迁移区中的个体采用分量热力学替换规则接收到自己的种群中。通过这种机制不仅有效地传播了各子种群中的优良个体,而且保持了各个子种群的多样性,定量地平衡优良个体的传播与种群多样性之间的冲突,在加快收敛速度的同时保持种群的多样性,减少陷入局部最优的概率。对比实验结果表明该算法表现出更高的求解精度和更快的收敛速度。

Aiming at the disadvantage of traditional parallel multipopulation gene expression programming,namely,the conflict between migration of individuals and diversity of subpopulation,a parallel multipopulation gene expression programming based on component thermodynamical migration strategy（CTDPGEP） was proposed.In this algorithm,an elite subspace of each subpopulation,consisting of some excellent individuals and some random individuals,was selected to migrate to the migration-buffer of any other subpopulation.The other subpopulations received the individuals in its own migration-buffer asynchronously using component thermodynamical replacement method.This mechanism not only ensured the excellent individuals being propagated quickly among the subpopulations,but also maintained the diversity of each subpopulation.Thus,it harmonized the conflict between migration of individuals and diversity of subpopulation quantitatively,accelerated the convergence speed as well as preserved the diversity of population to decrease the probability of trapping into local optimum.Experimental results indicated that the proposed algorithm outperformes some newly relevant algorithms both in solution precision and convergence speed.