一种应用动态区域划分的多种群异构进化算法

A Heterogeneous Multi-population Evolutionary Algorithm Using Dynamic Region Partitioning

研究通过多种群进化算法进行优化的过程,并利用云模型理论以及进化算法困难度二个指标进行评价,构建得到一种可以实现动态区域分类的多种群进化算法。从CEC2015函数库内选择5个函数作为测试对象再对各算法进行了性能猜测试,以DD-MEA算法对算法各项性能指标进行了测试。在同样的种群规模下时,DD-MEA可以获得最小的平均收敛代数并显著降低收敛时间,由此可以推断DD-MEA的寻优速率明显优于其它算法。F5函数表现出明显的非对称与旋转特征,存在很多的局部最优解。在所有维度下DD-MEA都具备更优异的性能指标,并且表现出更高的通用性以及稳定性。

This paper studies the optimization process of multi-population evolutionary algorithm,and evaluates the cloud model theory and the difficulty of the evolutionary algorithm,and constructs a multi-population evolutionary algorithm that can real⁃ize dynamic region classification.Five functions are selected from CEC2015 function library as test objects,and the performance of each algorithm is tested with DD-MEA algorithm.With the same population size,DD-MEA can obtain the minimum mean conver⁃gence algebra and significantly reduce the convergence time.Therefore,it can be inferred that the optimization rate of DD-MEA is significantly better than other algorithms.F5 shows obvious asymmetric and rotational characteristics,and there are many local opti⁃mal solutions.DD-MEA has better performance in all dimensions and shows higher versatility and stability.