New search strategy for multi-objective evolutionary algorithm

Abstract

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.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.