Fitness peak clustering based dynamic multi-swarm particle swarm optimization with enhanced learning strategy

Particle Swarm Optimization (PSO) is a well-known swarm intelligence algorithm and its performance primarily depends on the tradeoff between exploration and exploitation. In order to well balance the exploration and exploitation, this paper presents a fitness peak clustering based dynamic multi-swarm Particle Swarm Optimization (FPCMSPSO) with enhanced learning strategy. In the presented FPCMSPSO, first, FPC-based partitioning method is utilized to divide the initialized population into several sub-swarms so as to avoid crossover evolution caused by random partitioning. These sub-swarms evolve independently based on comprehensive learning strategy and along with further evolution they would merge into a global swarm according to their own stagnancy information. Second, an enhanced learning strategy is exploited to some particles, and their velocities are updated based on learning exemplars alternately generated by comprehensive learning or dimensional learning strategies according to their stagnancy information. Extensive experimental results demonstrate that the solution accuracy, convergence speed and stability of FPCMSPSO are remarkably improved due to the usage of above strategies. The comparative results of FPCMSPSO with other existing PSO variants on various optimization problems show that FPCMSPSO statistically outperforms other PSO variants with significant difference.

Automated summary

This paper proposes a dynamic multi-swarm Particle Swarm Optimization algorithm, FPCMSPSO, based on fitness peak clustering to balance the tradeoff between exploration and exploitation. The algorithm utilizes enhanced learning strategy and partitioning method to improve solution accuracy, convergence speed and stability, outperforming other PSO variants statistically on various optimization problems.