Adaptive division of labor particle swarm optimization

• A new population division-based particle swarm optimization variant is proposed.
• Both swarm diversity and fitness are used to adaptively assign the search task of each particles.
• Two operators are applied on the best solution to further improve the algorithm’s convergence speed.
• A stagnation prevention module is also proposed to mitigate the premature convergence issue.
• The proposed algorithm outperforms its peers in term of searching accuracy and convergence speed.

Although evident progress and considerable achievements have been attained in developing a new particle swarm optimization (PSO) algorithm, successfully balancing the exploration and exploitation capabilities of PSO to determine high-quality solutions for complex optimization problems remains a fundamental challenge. In this study, we propose a new PSO variant, namely, adaptive division of labor (ADOL) PSO (ADOLPSO), to overcome the demerits of our previous work. Specifically, an ADOL module is developed in ADOLPSO to adaptively regulate the exploration and exploitation searches of swarm. To achieve this purpose, both criteria of swarm diversity and fitness are considered during the task allocation process of the ADOLPSO current swarm. Two new operators, namely, convex operator and reflectance operator, are adopted to generate new particles from the memory swarm of ADOLPSO to further enhance the searching accuracy and convergence speed of the proposed algorithm. These two operators are activated to evolve the memory swarm only if a fitness improvement is observed in the current swarm of ADOLPSO to prevent excessive computational complexity. The proposed ADOLPSO is applied to solve 18 benchmark functions with various characteristics. Simulation results of ADOLPSO are compared with those of other nine well-established PSO variants. Experimental findings reveal that ADOLPSO significantly outperforms the other PSO variants in terms of searching accuracy, reliability, and convergence speed.