FFOA: An adaptive swarm optimizer with centered-optimum-bias assessment

Abstract

The Fennec Fox Optimization Algorithm (FFOA) is introduced as a fennec-inspired adaptive swarm optimizer for continuous optimization tasks. Its primary proposal is algebraically a current-to-promising differential move; successful-hunt memory and remembered food-cache reuse provide success-history adaptation and archive-assisted variation, while population reduction, local refinement, diversity recovery, and stagnation recovery are generic numerical support operators. The contribution is therefore an explicitly positioned integration and empirical audit, not a wholly new variation primitive. Evaluation uses exact objective-call accounting. Under one deterministic SHA-256 shift per function–dimension cell, FFOA outperforms Chaotic Evolution Optimization (CEO) in 130 of 150 rounded average-FE cells, ties in one, and is slower in 19; this provides no evidence of centered-optimum bias under the tested shifts, but does not establish general translation invariance. In a 14-algorithm analysis, FFOA obtains the best CEO-style cell-level average rank but is not significantly separated from LSHADE. A CEC-2017/2022 campaign covering 82 function–dimension cells, 17 algorithms, and 41,820 seed records ranks LSHADE first overall (3.000) and FFOA second (3.037); a paired cell-rank test finds them practically indistinguishable ( p Holm = 0.527 , r rb = − 0.081 ). A separate five-variant minimal-core ablation over the same 82 cells and 30 seeds per cell ranks the complete model first and shows that six of seven prespecified Holm-adjusted Wilcoxon contrasts are significant. A sensor-network localization case study likewise shows competitive, not dominant, performance.