Abstract: An adaptive hybrid multi-objective estimation of distribution evolutionary algorithm based on the change rate of objective function is put forward for overcoming the defect in global convergence of multi-objective estimation of distribution algorithm. The basic idea of new method is that differential evolution algorithm is introduced into multi-objective estimation of distribution algorithm. When the change rate of function is large, new population is generated with estimation of distribution algorithm, and otherwise differential evolution algorithm is used to generate new population. Theoretical analysis and numerical results show that the hybrid algorithm has better global convergence, and the distribution and uniformity of solutions is improved to a certain extent compared with algorithm without considering the change rate of objective function.

Key words: multi-objective optimization, estimation of distribution algorithm, differential evolution algorithm, adaptive, change rate of function

摘要： 针对多目标分布估计算法全局收敛性较弱的缺陷，提出了一种自适应混合多目标分布估计进化算法。其基本思想是：在多目标分布估计算法中引入全局收敛性较强的差分进化算法，当函数变化率较大时，用分布估计算法产生新种群；当函数变化率较小即算法可能陷入局部收敛时，用差分进化算法产生新种群。理论分析和数值实验结果表明，这种混合算法不仅具有良好的全局收敛性，而且解的分布性和均匀性较没有考虑目标函数变化率的混合多目标分布估计算法也有了一定程度的提高。

关键词: 多目标优化, 布估计算法, 差分进化算法, 自适应, 函数变化率