多精英采样与个体差分学习的分布估计算法

提出了基于多精英采样和差分搜索的分布估计算法EDA-M/D (Estimation distribution algorithm based on multiple elites sampling and individuals differential search)。EDA-M/D利用多精英个体独立采样生成子代来提升算法全局搜索能力,利用精英群体分布的σ²约束采样半径,实现种群从全局搜索逐步过度到局部搜索。当精英群体停滞时,劣势个体借助精英群体的μ和种群历史最优解进行差分搜索,帮助种群跳出局部最优解。通过多精英采样与差分搜索的自适应协同实现种群宏观信息与个体微观信息的有机融合。实验结果表明EDA-M/D在稳定性和搜索能力方面均表现出明显的优势。

An Estimation of Distribution Algorithm Based on Multiple Elites Sampling and Individuals Differential Search

An estimation distribution algorithm based on the multiple elites sampling and the individuals differential search (EDA-M/D) is proposed. In EDA-M/D, the elites carry out the sampling to generate the offspring independently and enhance the exploration. Meanwhile, the variance of the population distributionis selected to control the sampling radius. Thus, the target of the population can be gradually transited from exploration to exploitation. If the elite population stagnates, the nonentities will choose the mean value of the elites distribution μ and the population historical best solution as the two exemplars to execute a differential search operator, and then help the population jump out of a potential local optimum. Based on the adaptive strategy, two generation methods for the offspring, i.e., basing on the multiple elites sampling and the differential search, can be hybridized. Hence, the macro information of population and the micro information of individuals can be organically integrated. Experimental results show that EDA-M/D outperforms the other peer algorithms in the algorithm stability and the global optimal search capability.