无定向拓扑连接的并行遗传算法及其变异算子改进

Parallel genetic algorithm based on unoriented-connected topologyand the improvement of mutation operator

为解决巨量优化问题,在Internet平台下为并行遗传算法提出一个新的拓扑结构———无定向拓扑连接。该拓扑连接既允许驻留子种群的计算机节点中途退出,又允许新的计算机节点随时参与进化,增强了算法的鲁棒性和容错性能。针对传统浮点变异算子的不足,提出一种新的二元浮点补码变异算子,讨论了它在克服早熟收敛方面的作用。实验表明,提出的算法能显著提高寻优质量,节约寻优时间;新的变异算子能有效阻止遗传算法陷入局部极值,进一步提高了遗传算法的寻优能力。

In order to solve the massive optimization problems, a novel topology-unoriented-connected topology is presented based on Internet for parallel genetic algorithm. Some computers in which the subpopulations reside are allowed to quit in the midway without any limitation, and some other computers are allowed to participate in the evolution, such that the robustness and fault-tolerance of the algorithm are enhanced. At the same time, according to the disadvantages of the traditional floating-point mutation operator, a dyadic floating-point supplementary mutation operator is brought forward. The function of the novel mutation operator to prevent premature convergence is also discussed. Experimental results show that parallel genetic algorithm based on unoriented-connected topology can not only gain better optimization results but also largely save the optimization time of the problems. The mutation operator can effectively prevent genetic algorithm from being trapping in local extremum, and the searching performance of genetic algorithm is largely improved.