Page 37 - 2025S

30 UEC Int’l Mini-Conference No.54

achieved an average HV improvement of 8. 26% [4] N. Beume et al., ”SMS-EMOA: Multiobjec-
in FDA test problems 1-5, with maximum gains tive Selection Based on Dominated Hyper-
of 18.7% in FDA 5 under rapid changes (τ t = volume,” EJOR, 2007.
10). Diversity-guided re-initialization proved
particularly effective for disconnected Pareto [5] A. Ghosh et al., ”Random Immigrants
fronts (FDA 2, FDA 5), improving the recovery for Dynamic Environments,” IEEE TEVC,
ratio by 25% compared to standard approaches. 2021.
Furthermore, adaptive mutation significantly re- [6] S. Jiang et al., ”A Memory-Based NSGA-II
duced HV volatility (σ HV ) by 32% in problems for Dynamic Multiobjective Optimization,”
featuring rotational landscape changes (FDA 3). IEEE CEC, 2018.
Statistical validation confirmed significant supe-
riority of the hybrid approach (p < 0.01) across [7] A. Zhou et al., ”Prediction-Based Popula-
all test scenarios. The framework demonstrated tion Re-initialization for Dynamic Multiob-
robust performance across change frequencies, jective Optimization,” SEAL, 2019.
with particularly strong results in rapid-change
environments (τ t = 10). [8] X. Chen et al., ”Directed Mutation for Dy-
namic Multiobjective Optimization,” IEEE
TEVC, 2020.
8 Future Work
[9] H. Wang et al., ”Change-Responsive
Crossover for Dynamic Optimization,”
Future research should explore: (i) Many- GECCO, 2022.
objective extension (m > 3) through special-
ized dominance handling, (ii) Real-world valida- [10] F. Li et al., ”Environmental Inheritance in
tion in dynamic scheduling and energy manage- Dynamic Optimization,” IEEE TEVC, 2021.
ment, (iii) Proactive adaptation via time-series
forecasting, (iv) GPU-accelerated hyper-volume [11] G. Karafotias et al., ”Parameter Control
computation, and (v) Meta-optimization of in Evolutionary Algorithms,” IEEE TEVC,
adaptation parameters for automated tuning. 2015.

[12] Y. Wang et al., ”Population Size Mod-
ulation for Dynamic Optimization,” IEEE
Acknowledgments
CEC, 2019.

This research was supported by the JASSO [13] M. Aziz et al., ”Multi-population for Dy-
Scholarship Program. The authors thank the namic Multiobjective Optimization,” Swarm
PlatEMO development team for their optimiza- and Evolutionary Computation, 2020.
tion framework and the UEC High-Performance
Computing Center for computational resources.

References

[1] K. Deb et al., ”A Fast Elitist NSGA-II,”
PPSN VI, 2000.

[2] E. Zitzler et al., ”Performance Assessment
of Multiobjective Optimizers,” IEEE TEVC,
7(2), 2003.

[3] K. Deb et al., ”Dynamic Multiobjective Op-
timization Problems,” IEEE CEC, 2006.