Adaptive memory-augmented Artificial Protozoa Optimization with competitive operator control for complex numerical and constrained problems
Year: 2026
Type: article
Abstract: Reliable optimization of complex numerical and constrained problems requires algorithms that can exploit accumulated search experience without sacrificing population diversity. This paper proposes MAO-APO-RC, an enhanced Artificial Protozoa Optimizer that couples memory-augmented learning with adaptive competitive control. The method combines historical-solution guidance, quasi-oppositional sampling, Lévy-flight exploration, stagnation-triggered rescue, success-history parameter adaptation, and feedback-based operator scoring within a unified multi-candidate search framework. At every iteration, alternative candidates are produced by complementary operators, evaluated under a common selection rule, and reinforced according to their recent contribution, enabling dynamic regulation of exploration and exploitation. The performance was assessed in 30 independent runs on 15 benchmark functions including shifted, rotated, ill-conditioned, multimodal, hybrid and composition landscapes in 30 and 50 dimensions, and four constrained engineering design problems. The comparisons were made to APO, PSO, DE, GWO, WOA, SCA, CMAES-lite, and SHADE-lite. MAO-APO-RC achieved the best overall Friedman mean rank of 2.3667 on the numerical benchmark suite. Six of the eight competing algorithms were statistically significantly different with Wilcoxon–Holm analysis, whereas APO and SHADE-lite were statistically comparable. For the engineering problems, MAO-APO-RC obtained a joint-best mean rank of 1.875 with DE. Ablation analysis revealed the APO + QOD as the best reduced configuration with a mean rank of 1.8500 compared to 1.9167 for MAO-APO-RC whereas the complete framework significantly outperformed the Memory and Energy variants. Despite measurable constant-factor runtime overhead, MAO-APO-RC preserves APO’s asymptotic time and space complexity. These findings demonstrate that MAO-APO-RC provides competitive aggregate performance across the evaluated numerical and engineering problems, while its statistical advantage remains competitor- and problem-dependent.
Source: Scientific Reports
Author: M. A. Elsabagh
Institution: Kafrelsheikh University
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: gold
Estimated APC paid: $2,690
OpenAlex ID: W7236700411
