一种基于拉丁超立方抽样和聚类分析的改进MOPSO算法

针对基本的MOPSO算法的可能导致局部最优和难以输出代表性非支配解的问题,在MOPSO算法的基础上,运用拉丁超立方抽样和聚类分析对算法进行改进。应用拉丁超立方抽样指导MOPSO算法种群初始化,使初始化种群可以均匀遍布整个空间,避免了基本的MOPSO算法可能会导致的局部最优等问题。为了加强非支配解集的分布性和多样性,同时考虑在众多非支配解中自动挑拣代表性非支配解,增加聚类分析环节,对输出解集进行聚类处理,以挑选代表性非支配解。与基本的MOPSO算法相比较,改进的MOPSO算法求解的Pareto解集在寻优效果及代表性解筛选方面具有一定优势。

Improved MOPSO algorithm based on latin hypercube sampling and cluster analysis

Aiming at the problem that the basic multi-objective particle swarm optimization(MOPSO)algorithm may lead to local optimum and the inability to output representative non-dominated solutions,this paper uses latin hypercube sampling and clustering analysis to improve the algorithm based on MOPSO algorithm.Firstly,the latin hypercube sampling is used to guide the MOPSO algorithm population initialization,so that the initial population can be evenly distributed throughout the space and avoids the local optimization problems that the basic MOPSO algorithm may cause.Secondly,the output solution sets are clustered to select representative non-dominant solutions,which purpose is to enhance the distribution and diversity of non-dominant solution sets.Meanwhile,this method can automatically select the representative non-dominant solutions among many non-dominant solutions.Compared with the basic MOPSO algorithm,the Pareto solution set for improved MOPSO algorithm has certain advantages in the effect of optimization and the selection of representative solutions.