ADVANCING METAHEURISTIC ALGORITHMS: INNOVATIVE OPERATORS AND ADATIVE COOPERATIVE ISLAND MODEL FOR EFFECTIVE OPTIMIZATIO رسالة دكتوراة

Abstract

Swarm intelligence algorithms are renowned for their ability to tackle complex global optimization problems by mimicking natural processes. However, these algorithms, including the Crow Search Algorithm (CSA) and Capuchin Search Algorithm (CapSA), often suffer from inherent limitations such as low search accuracy and a tendency to converge to local optima. This thesis aims to develop advanced variants of these algorithms that could effectively handle a diverse range of theoretical and practical optimization problems. One widely explored approach is the structured population mechanism, which maintains diversity during the search process to mitigate premature convergence. The island model, a common structured population method, divides the population into smaller independent sub-populations called islands, each running in parallel. Migration, the primary technique for promoting population diversity, facilitates the exchange of relevant and useful information between islands during iterations. Enhancements to CSA and CapSA include introducing adaptive strategies and novel operators, enhancing them into variants named ECSA and ECapSA, respectively. Furthermore, it integrates these enhancements within an island-based model -iECSA and iECapSA- equipped with an adaptive migration policy designed to dynamically adjust migration rates based on real-time evaluations of population diversity and fitness. This innovative approach aims to avoid the limitations of traditional island models and enhance global search capabilities. The performance of the proposed models is evaluated using 53 real-valued mathematical problems and three practical applications: neural network training, multilevel thresholding for image segmentation, and software reliability growth modeling. It is also validated by conducting an extensive evaluation against a comprehensive set of well-established and recently introduced meta heuristic algorithms. Experimental results demonstrate that the enhanced variants of CSA and CapSA outperform their fundamental counterparts in the majority of test cases, providing more accurate and reliable outcomes. Furthermore, extensive experimentation consistently showcases that the iCapSA outperforms its comparable algorithms across a diverse set of practical applications.