Yuejun Liu
Software School of Anyang Normal University, Anyang, 455002, Henan, China.
Key Laboratory of Oracle Bone Inscriptions Information Processing, Ministry of Education, Anyang, 455002, Henan, China.
Heliyon. 2024 Dec 5;10(24):e40917. doi: 10.1016/j.heliyon.2024.e40917. eCollection 2024 Dec 30.
To address the problem of low search efficiency of multi-objective evolutionary algorithm during iterations, we proposed a new idea which considering a single individual to generate better solutions in a single iteration as a starting point to improve the search performance of multi-objective evolutionary algorithm and designedthe neighbor strategy and guidance strategy based on this improved approach in this paper. We used our proposed new search strategy to improve NSGA-III algorithm(named as NSGA-III/NG) and MOEA/D algorithm(named as MOEA/D-NG). On ZDT, DTLZ and WFG public test sets, the NSGA-III/NG algorithm using the new search strategy was compared with NSGA-II algorithm, NSGA-III algorithm, ANSGA-III algorithm and NSGA-II/ARSBX algorithm. The MOEA/D-NG algorithm using the new search strategy was compared with MOEA/D algorithm, MOEA/D-CMA algorithm, MOEA/D-DE algorithm and CMOEA/D algorithm. Experimental results indicate that the performance of NSGA-III/NG algorithm using our search strategy is superior to NSGA-II, NSGA-III,ANSGA-III and NSGA-II/ARSBX algorithm and the performance of MOEA/D-NG algorithm using our search strategy is superior toMOEA/D, MOEA/D-CMA,MOEA/D-DE and CMOEA/D algorithm. Our proposed search strategy can improve the convergence speed of NSGA-III algorithm and MOEA/D algorithm by 12.54 %,the accuracy of the non dominated solution set by 3.67 %. This situation indicates that our search strategy could significantly improve the search capability of the multi-objective evolutionary algorithm. In addition, this strategy has excellent applicability and could be combined with mainstream multi-objective evolutionary algorithms.
为了解决多目标进化算法在迭代过程中搜索效率低的问题，本文提出了一种新的思路，即以考虑单个个体在单次迭代中生成更好的解为出发点来提高多目标进化算法的搜索性能，并基于此改进方法设计了邻域策略和引导策略。我们使用提出的新搜索策略对NSGA-III算法（命名为NSGA-III/NG）和MOEA/D算法（命名为MOEA/D-NG）进行了改进。在ZDT、DTLZ和WFG公共测试集上，将采用新搜索策略的NSGA-III/NG算法与NSGA-II算法、NSGA-III算法、ANSGA-III算法和NSGA-II/ARSBX算法进行了比较。将采用新搜索策略的MOEA/D-NG算法与MOEA/D算法、MOEA/D-CMA算法、MOEA/D-DE算法和CMOEA/D算法进行了比较。实验结果表明，采用我们搜索策略的NSGA-III/NG算法的性能优于NSGA-II、NSGA-III、ANSGA-III和NSGA-II/ARSBX算法，采用我们搜索策略的MOEA/D-NG算法的性能优于MOEA/D、MOEA/D-CMA、MOEA/D-DE和CMOEA/D算法。我们提出的搜索策略可以将NSGA-III算法和MOEA/D算法的收敛速度提高12.54%，将非支配解集的精度提高3.67%。这种情况表明我们的搜索策略可以显著提高多目标进化算法的搜索能力。此外，该策略具有出色的适用性，可以与主流的多目标进化算法相结合。
