Multi-objective evolutionary algorithms (MOEAs) have been the choice for generating a set of Pareto-optimal (PO) solutions in one run. However, these algorithms sometimes suffer slow and poor convergence toward the PO front. One of the remedies to improve their convergence is to couple global search of MOEAs with local search. However, such coupling brings other implementation challenges, such as what, when, and how many solutions can be chosen for local search with MOEAs? In this paper, these challenges are addressed by developing a local search module that can choose solutions for local search using a set of reference lines. The heuristic strategies are also developed with the module for determining the frequency of executing local search and for terminating MOEA adaptively using a statistical performance indicator. The proposed algorithm, which is referred to as \({\text {RM}}^2\)OEA, is tested on 2-objective ZDT and 3-objective DTLZ test problems. Results demonstrate faster and improved convergence of \({\text {RM}}^2\)OEA over a benchmark MOEA from the literature.

选择多目标进化算法（MOEA）一次即可生成一组帕累托最优（PO）解。但是，这些算法有时会向PO前端缓慢缓慢收敛。改善融合的一种方法是将MOEA的全球搜索与本地搜索结合起来。但是，这种耦合带来了其他实施方面的挑战，例如，使用MOEA进行本地搜索时可以选择什么，何时以及有多少解决方案？在本文中，通过开发一个本地搜索模块可以解决这些挑战，该模块可以使用一组参考线为本地搜索选择解决方案。启发式策略也与模块一起开发，该模块用于确定执行本地搜索的频率以及使用统计性能指标自适应地终止MOEA。提出的算法\（{\ text {RM}} ^ 2 \） OEA已针对2目标ZDT和3目标DTLZ测试问题进行了测试。结果表明，与文献中的基准MOEA相比，\（{\ text {RM}} ^ 2 \） OEA的收敛速度更快且得到改善。