CONCEPTION ET MISE EN ŒUVRE D'UNE MÉTAHEURISTIQUE ÉVOLUTIVE DE TYPE VEUVE NOIRE POUR L'OPTIMISATION DE FONCTIONS NUMÉRIQUES EN INGÉNIERIE

This study presents the design and implementation of an enhanced evolutionary metaheuristicinspired by the Latrodectus spider’s reproductive behaviour, tailored for high-precision optimization of complex numerical functions in engineering. While the canonical Black Widow Optimization Algorithm (BWOA) leverages aggressive cannibalistic selection to accelerate convergence, its propensity for premature solution elimination impedes robustness in engineering applications. To address this, we propose ACR-BWOA, a novel variant featuring three adaptive innovations : Dynamic Cannibalism Rate Control via exponentially decaying elimination intensityCR(t) = CR0 · e−λt , Elitist Solution Shielding preserving top-performing juveniles and probabilistic male survival and Hybrid Local Intensification using embedded Nelder-Mead refinement.Implemented in Python and tested across 15 benchmark functions (CEC 2017) and real-worldengineering cases – including photovoltaic array configuration and pressure vessel design ACRBWOA, demonstrates 99.1% retention of elite solutions (vs. 68.3 % in standard BWOA) whilemaintaining population diversity, superior precision with 0.05–2.3 % lower error rates on multimodal engineering functions. Accelerated convergence from 17 to 32 % through targeted exploitation, validated via Friedman statistical tests. The algorithm’s design eliminates the traditionaltrade-off between selective pressure and solution preservation, achieving 98.5 % faster constrainthandling in structural optimization problems. ACR-BWOA, establishes a new paradigm for bioinspired metaheuristics in engineering, where strategic cannibalism reduction enhances reliabilitywithout compromising computational efficiency.