Particle Swarm Optimization: A Comprehensive Survey

Analysis

Why This Paper Matters

Particle swarm optimization (PSO) remains one of the most widely used swarm intelligence algorithms due to its simplicity and effectiveness. However, its well-known premature convergence problem has spurred an enormous body of modifications. This comprehensive survey, published in IEEE Access with over 1300 citations, provides a much-needed structured overview of the PSO landscape. For AI practitioners, understanding which PSO variant to apply to a given optimization or feature selection task is critical, and this paper offers a clear taxonomy to navigate the options.

The paper's timing is significant: as machine learning models grow in complexity, efficient hyperparameter tuning and feature selection become paramount. PSO variants are increasingly used in deep learning contexts, and this survey consolidates the state of the art, making it an essential resource for both newcomers and experienced researchers.

Technical Contributions

The paper categorizes PSO modifications into four main strategies:

• Parameter control modifications: Adaptive inertia weight, acceleration coefficients, and constriction factors to balance exploration and exploitation.
• Hybridization with other metaheuristics: Combining PSO with genetic algorithms (GA), differential evolution (DE), simulated annealing, and others to leverage complementary strengths.
• Cooperative and multi-swarm techniques: Dividing the population into sub-swarms that communicate or co-evolve, improving diversity and avoiding local optima.
• Binary PSO and topology variants: Extending PSO to binary search spaces and analyzing neighborhood topologies (e.g., ring, star, von Neumann) that affect information flow.

The survey also dedicates a section to feature selection using PSO, a critical application in high-dimensional data analysis. It reviews how binary PSO and hybrid approaches have been applied to select relevant features while maintaining classification accuracy.

Results

While the paper is a survey and does not present new experimental results, it synthesizes findings from hundreds of studies. Key takeaways include:

• Hybrid PSO-GA and PSO-DE variants consistently outperform standard PSO on benchmark functions and real-world problems.
• Adaptive parameter control (e.g., time-varying inertia weight) improves convergence speed by 10-30% on many test functions.
• Multi-swarm PSO variants show superior performance on multimodal and high-dimensional problems, often achieving better final solutions than single-swarm PSO.
• Binary PSO with transfer functions (e.g., S-shaped, V-shaped) achieves competitive feature selection results, reducing feature sets by 50-80% while maintaining or improving accuracy.

Significance

This survey has become a standard reference in the optimization community, evidenced by its high citation count. It provides a roadmap for researchers to identify gaps and promising directions, such as PSO for dynamic and multi-objective optimization, integration with deep learning, and theoretical analysis of convergence. For practitioners, it offers a practical guide to selecting and tuning PSO variants for engineering and machine learning tasks. The eight future research directions outlined—including self-adaptive PSO, multi-objective PSO, and PSO for big data—will likely shape the next wave of PSO research.