Non-dominated Bézier Curve-Based Optimization: A Multi-objective Extension for Engineering Design Problems

Buch, Dhara, Buch, Hitarth
Operations Research Forum · Springer Nature · 2026-11-05 · DOI: 10.1007/s43069-026-00719-x

Аннотация

Multi-objective optimization requires balancing conflicting objectives while maintaining convergence and diversity. This paper proposes NDBCO, a multi-objective extension of Bézier Curve-based Optimization (BCO) that retains the mathematical forms of the original linear, quadratic, and cubic Bézier search operators and their inherited parameter schedules. NDBCO introduces four structural changes: NSGA-II environmental selection, rank-1 guide selection through a crowding-distance tournament, a bounded accumulation archive used only for retention and reporting, and reuse of rank/crowding-distance state from environmental selection. All algorithms are evaluated on a common terminal-population basis. The study covers 18 ZDT, DTLZ, and UF benchmark problems against six established competitors and 12 constrained engineering design problems against seven competitors, including Two-Arch2. Six indicators (IGD, IGD $$^+$$ + , GD, HV, SP, and NR) are analysed using Friedman ranking with Holm-corrected post-hoc comparisons. On the benchmark suite, NDBCO ranks third overall, with no statistically significant difference from NSGA-II or NSGA-III on IGD, IGD $$^+$$ + , HV, or SP. On the engineering suite, NDBCO attains the best overall Friedman rank, with no significant difference from NSGA-II, NSGA-III, or Two-Arch2. Cumulative ablation identifies environmental selection as the dominant measurable incremental contributor to terminal quality; the accumulation archive is retention-only, while rank/crowding-distance reuse changes some trajectories without a significant aggregate quality effect. Archive-capacity sensitivity shows identical terminal populations for $$A_{\max }\in \{N,2N,4N\}$$ A max ∈ { N , 2 N , 4 N } under matched seeds. These results support NDBCO as a competitive multi-objective optimizer while clarifying the roles of its components.

Читать у издателя →

Похожие статьи