基于聚类的目标约简高维多目标差分进化算法

Large-dimensional Multi-objective Differential Evolution Algorithm Using Clustering Based Objective Reduction

高维多目标优化是多目标优化中的难点,而目标约简是高维多目标优化中较为可行、有效的方法。本文提出了一种基于聚类的目标约简高维多目标差分进化算法,其基本思想是:以采用精英选择和排序策略的多目标差分进化算法为基础,根据近似的Pareto最优前沿,对所有目标进行基于相关距离的聚类,然后删除相似的聚类,得到约简的目标集。在优化过程中,周期性地在全目标集和约简目标集中搜索,以兼顾算法的收敛性和运行效率。采用高维DTLZ测试函数对新算法进行了性能测试,并与其它算法进行了比较,结果验证了新算法的收敛性和有效性。

Large-dimensional multi-objective optimization is a difficult problem in multi-objective optimization,and objective reduction is a feasible and effective method for large-dimensional multi-objective optimization.A clustering-based large-dimensional multi-objective differential evolution algorithm is proposed in this paper,and the basic idea is:based on the multi-objective differential evolution algorithm with elite selection and ranking strategy,according to the approximate Pareto optimal frontier,all objectives are clustered based on correlation distance,and then similar clustering is deleted.Finally the reduced objective set is obtained.In the process of optimization,the algorithm is searched periodically in the set of all objectives and the set of reduced objectives in order to ensure the convergence and operation efficiency of the algorithm.The new algorithm is tested by using large-dimensional DTLZ test function and compared with other algorithms.The results show that the new algorithm is convergent and effective.