基于知识引导的自适应动态多模态差分进化算法

为充分利用问题求解过程知识,提升动态多模态优化算法的计算资源利用效率,提出一种基于知识引导的自适应动态多模态差分进化算法.首先,利用自组织映射神经网络实现种群自聚类,形成稳定的小生境;然后,通过对种群全局知识和个体邻域知识的综合学习,设计一种基于知识引导的自适应差分进化算法,在对种群进化状态进行实时监测和分析的基础上,逐层递进地引导不同种群个体自适应地选择最符合当前进化需求的变异方式,提升种群搜索效率,平衡种群多样性与收敛性;最后,针对问题动态特性,设计一种基于历史动态过程知识引导的自适应动态响应机制,通过对历史寻优经验的自适应学习,预测生成新环境下的潜在精英个体,引导种群实现精准快速的多峰定位.实验结果表明,所提出算法能够有效解决动态多模态优化问题,且在不同动态环境设置下其求解性能均优于对比算法.

Adaptive dynamic multimodal differential evolution algorithm based on knowledge guidance

To fully use the knowledge of problem-solving process and improve the computational resource utilization efficiency of dynamic multimodal optimization algorithms, an adaptive dynamic multimodal differential evolution algorithm based on knowledge guidance is proposed. Firstly, a self-organizing mapping(SOM) neural network is used to realize the self-clustering of population and form some stable niches. Secondly, through the comprehensive learning of the population global knowledge and the individual neighborhood knowledge, a knowledge-guided adaptive differential evolution(KADE) algorithm is designed to layer by layer guide the individuals to adaptively choose the mutation strategies that best meets their evolutionary demands. The proposed algorithm can improve the search efficiency of population and balance the diversity and convergence. Finally, when a change happens, an adaptive dynamic response strategy based on the historical experience learning is proposed to predict the positions of the elite individuals in the new environment to achieve a fast convergence. Experimental results show that the proposed SOM-KADE shows superior performance compared with the state-of-the-art algorithms.