基于目标空间分区的稳态高维多目标进化算法

摘要 针对高维多目标优化中Pareto非劣候选解所占比例很大，常用的先考虑收敛性再考虑分布性的多目标进化算法面临选择压力衰减的问题，提出一种先考虑分布性再考虑收敛性的高维多目标进化算法——基于目标空间分区的稳态高维多目标进化算法(SS-OSP).该算法先采用目标空间分区策略将种群按照权重向量分为多个子空间，在每个子空间中按照分解方法中的聚合函数选择个体;然后，考虑到常规的PBI聚合函数的罚参数在进化过程中一直保持不变的情况，提出一种自适应PBI聚合函数;最后，仿真实验结果表明所提出的算法与其他三种算法相比，具有更好的收敛性和分布性.
Abstract：Due to the sharp increasing of the proportion of Pareto non-dominated candidate solutions for many-objective optimization problems， the commonly used many-objective evolutionary algorithms encounter the selection pressure deterioration problem considering the convergence-first-and-diversity-second selection approach. This paper proposed a many-objective evolutionary algorithm with the diversity-first-and-convergence-second selection strategy-steady-state many-objective evolutionary algorithm based on objective space partition (SS-OSP). Firstly， it divided objective space into a large number of subspaces using a set of weight vectors. Then， one individual in each subspace was selected via adopting aggregation function. In addition， since the penalty parameter of PBI aggregation function remained constant in evolutionary process， an adaptive PBI aggregation function was proposed. Finally， the experimental results show that better convergence and diversity can be obtained using the proposed algorithm.