Individual different state-based multi-swarm particle swarm optimization

Particle swarm optimization (PSO) is a renowned stochastic optimization algorithm that has attracted extensive attention from researchers worldwide. However, most PSO variants suffer from slow convergence and are susceptible to local optima. In this article, an individual different state-based multi-swarm PSO (IDSMPSO) is developed to address these problems. First, the swarm is dynamically divided into four sub-swarms based on fitness value of particles, and the number of particles in each sub-swarm varies randomly to enhance the swarm diversity. Second, different inertia weight strategies are applied to the corresponding sub-swarms, which can keep the balance between the exploration and exploitation. Subsequently, different dynamic learning mechanisms are used to each sub-swarm. To be specific, the best sub-swarm employs Levy flight and greedy strategy to boost particle’s search ability, the better sub-swarm utilizes cooperative learning scheme to strengthen information exchange between sub-swarms, the worse sub-swarm adopts a positive cosine learning factor to improve the convergence speed, and the worst sub-swarm leverages elite learning mechanism, random combination strategy and Gaussian mutation to enable particles escaping from local optima. Finally, the BFGS quasi-Newton method is utilized to refine the obtained optimal solution so as to further enhance the local convergence ability of the swarm. Extensive experiments on CEC2017 and CEC2022 benchmark test suites validate that the IDSMPSO proposed in this work surpasses several other peer algorithms in terms of global search ability, solution accuracy and convergence rate. Notably, the application of IDSMPSO to the three-dimensional uncrewed aerial vehicle (UAV) path planning has achieved highly satisfactory results, which further validate the effectiveness of our proposal.