Skip to content

Multi-Hybrid Swarm Optimization | Master thesis in Computer Science - Data Science at AGH University of Cracow, Department of Computer Science

Notifications You must be signed in to change notification settings

purbancz/MEng_AGH

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

15 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Multi-Hybrid Swarm Optimization

Piotr Urbańczyk

MEng thesis in computer science -- data science, written under the supervision of Aleksander Byrski

@mastersthesis{urbanczyk2025multihybrid,
  title={Multi-Hybrid Swarm Optimization},
  author={Urbańczyk, Piotr},
  year={2025},
  school={AGH University of Krakow}
}

My thesis explores informed diversity-enhancing strategies for Particle Swarm Optimization. To gain a deeper understanding of my research, I invite you to look through the piece or read the abstract provided below.

github twitter linkedin website

Abstract

This thesis addresses the premature convergence limitation of Particle Swarm Optimization (PSO), especially in complex, high-dimensional problems. It introduces and evaluates novel informed diversity-enhancing strategies to improve PSO's exploration capabilities. The proposed methodology involves a family of mechanisms based on problem-specific landmarks (best/worst solutions) that apply attraction or repulsion forces within velocity updates. These strategies are categorized into opposing-best (repulsion), attraction-to-worst (negative learning), and opposing-worst (reverse learning) behaviors, forming both single-role algorithms and three multi-hybrid PSO paradigms: disjoint-role, component-specific, and fully flexible. Rigorous empirical testing on diverse benchmark functions up to 1000 dimensions demonstrates that these strategies, particularly multi-hybrid variants like component-specific HybridPartialDisjointPSO, significantly outperform standard PSO and simple perturbation methods in solution quality, convergence, and robustness. Variants employing attraction-to-worst strategies also showed strong, consistent performance. This research establishes that informed, role-based diversity mechanisms are, in most cases, more effective than random perturbations, offering scalable and robust PSO enhancements for complex optimization by better balancing exploration and exploitation.

Keywords

evolutionary computation, swarm intelligence, hybrid metaheuristics, Particle Swarm Optimization (PSO)

About

Multi-Hybrid Swarm Optimization | Master thesis in Computer Science - Data Science at AGH University of Cracow, Department of Computer Science

Topics

Resources

Stars

Watchers

Forks