高维多目标进化算法中的密度评估策略研究

Density estimation strategies in high-dimensional MOEAs

多目标进化算法中常引入密度评估策略来使算法获得更好的分布性和收敛性.但对于高维多目标问题,现有的密度评估策略却难于达到这一目的.为此更全面地考虑目标空间上各子目标的影响,提出了四种新的密度评估策略,并将其应用到经典多目标进化算法SPEA2中.在4～9个目标的多目标背包问题上的实验结果表明,采用新的密度评估策略的SPEA2算法能更有效地收敛到Pareto前沿.

A density estimation strategy is often adopted in order to guarantee better distribution and convergence in MOEA. But the current density estimation strategies cannot achieve this goal when the number of objectives become large. Each objective was more generally considered and four novel strategies of density estimation were proposed. Then, they were applied in SPEA2, which was one of the classical MOEAs. The experimental results of the test cases of MOKP with 4 to 9 objectives show that SPEA2 with the novel strategies have better convergence to the Pareto front on all test cases.