Enhanced Particle Swarm Optimization Using Fuzzy C-Means Clustering for Improved Convergence and Diversity

Traditional Particle Swarm Optimization (PSO) algorithms, despite their computational efficiency, frequently suffer from premature convergence and rapid diversity loss in complex search spaces, leading to suboptimal solutions especially in multimodal optimization problems. This paper presents a novel hybrid approach integrating Fuzzy C-Means (FCM) clustering with PSO to address these fundamental limitations. Our proposed FCM-PSO framework dynamically applies clustering analysis at the conclusion of each iteration to partition the swarm into distinct clusters based on particle similarities. This partitioning identifies promising regions and outlier particles. Dense clusters employ exploitation-focused mechanisms, while outliers and sparse regions use exploration-enhancing perturbations. Experimental validation across a suite of 20 benchmark functions demonstrates that FCM-PSO achieves significantly improved performance metrics compared to traditional PSO. Results show that FCM-PSO outperformed or matched standard PSO on 17 of 20 test problems, showing marked gains on complex functions like Ackley, Rosenbrock, and Levy. It also better maintained exploration and swarm diversity, reducing premature convergence and improving overall robustness. The dynamic clustering approach provides an effective balance between global exploration and local exploitation without requiring problem-specific parameter tuning. This research establishes a promising framework for addressing complex optimization challenges in diverse domains including neural network training, engineering design, and resource allocation problems.