基于正则性辅助的多目标优化进化算法研究

Research on Multi-objective Optimization Evolutionary Algorithm Based on Regularity Assistance

针对目前基于正则性辅助的多目标优化算法缺少局部信息以及模型参数设置对多目标优化算法的影响问题,本研究提出一种基于正则性辅助的多目标优化进化算法(Regularity Assisted Multi-objective Optimization Evolutionary Algorithm,RAMEA)。该方法将高斯采样和基于邻域的交配重组结合并用于子代重组,同时使用k-均值聚类方法获取流形结构信息,将种群划分为K个聚类,用K个聚类的均值向量建立高斯概率模型,从中抽取K个后代,然后将取样解作为父代添加到每个集群中去交配生成其他子代解。实验对比结果表明,研究提出的基于正则性辅助的多目标优化进化算法明显优于其他算法,其参数灵敏度和有效性表现更加突出。

At present,the multi-objective optimization algorithm based on regularity assistance lacks local information and model parameter settings have an impact on multi-objective optimization algorithm.To solve this problem,a multi-objective optimization evolutionary algorithm based on regularity assistance is proposed in this study.In this method,Gaussian sampling and neighborhood-based mating are combined and used for offspring recombination.At the same time,the k-means clustering method is used to obtain the manifold structure information.The population is divided into K clusters.The Gaussian probability model is established by using the mean vector of K clusters,from which K descendants are extracted,and then the sampling solution is added to each cluster as the parent to generate other offspring solutions.The Gaussian probability model is established by the mean vector of K clusters,and K offspring are extracted.Then the sampling solution is added to each cluster as a mating parent to generate other sub-generation solutions.Experimental comparison results show that the regularity assisted multi-objective optimization evolutionary algorithm proposed in the article is significantly better than other algorithms,and its parameter sensitivity and effectiveness are more prominent.