An Improved Multi-swarm Particle Swarm Optimization Based on Knowledge Billboard and Periodic Search Mechanism

Multi-swarm particle swarm optimization has faster convergence rate, wider range of search, and higher convergence accuracy. However, the information among sub-swarms is not updated in time, which may decrease the search ability of the multiple swarms. An improved multi-swarm particle swarm optimization based on the periodic search mechanisms and the knowledge billboard (KBMPSO) is proposed. The swarm is divided into several sub-swarms using the improved K-means method. In a search cycle, one sub-swarm searches collaboratively and the remaining sub-swarms search independently. When the particles evolve independently to a certain generation, the global best value is periodically updated. The information stored in the knowledge billboard can help the sub-swarm jump out the local optimum. The KBMPSO algorithm will exchange the information between the adjacent sub-swarms every fixed number of generations. Once the sub-swarm is trapped into the local optimum during the search process, it will affect the convergence effect of its adjacent sub-swarm. Introducing the knowledge billboard to the sub-swarm during its searching avoids the sub-swarm trapping into the local optimum. To effectively keep the balance between the global exploration and the exploitation, the particle takes advantage of the shared information which stored on the knowledge billboard. In the simulation studies, several benchmark functions are conducted to verify the superiority of the KBMPSO algorithm.