基于概率评估差分进化的多峰值优化

Multimodal function optimization based on DE algorithm of probabilistic evaluation mechanism

多峰值优化问题要求算法同时找到一个问题的多个全局最优解。近年来,演化算法已被广泛用于求解多峰值优化问题。然而,如何在极其有限的适应值评估次数内找到问题的多个全局最优解依然为演化算法带来了巨大的挑战。通过分析个体的历史更新经验,为每个个体赋予双层适应值评估概率,对个体进行选择性评估,从而减少算法运行过程中无效或低效的适应值评估,提出了一种基于概率评估差分进化的多峰值优化算法。实验结果显示,概率评估机制可以为算法节省更多的适应值评估次数,增加迭代过程,效果远好于其他主流的多峰值优化算法。

Multimodal optimization problems(MMOPs)require algorithms to simultaneously determine multiple global optima.Recently,evolutionary algorithms(EAs)have been widely used to solve MMOPs.However,there is still a great challenge for EAs to determine multiple global optima within very limited fitness evaluation(FE)times.To solve the inefficient FE,this paper proposes a multimodal function optimization algorithm based on the differential evolution algorithm of the probabilistic evaluation mechanism for solving MMOPs.In this algorithm,each individual will be assigned with the two-level FE probability according to its historical update experience to determine whether it needs to be evaluated.The experimental results show that the probabilistic evaluation mechanism can reduce FE times for the proposed algorithm and increase its iterative process,and its effect is much better than that of other mainstream mechanisms.