Positive Linear Correlation Particle Swarm Optimization

摘要

Social component and cognitive component are important for updating particles' velocity. In classical particle swarm optimization, the social component and the cognitive component in the updating velocity equation are supposed to be independent. It is reasonable to consider that the dependence between objects reflects the underlying mechanisms. This paper presents a novel dependence model of particle swarm optimization, in which correlation coefficient is used to measure the dependence between the social component and the cognitive component. Further, a positively linear correlation particle swarm optimization is derived for the dependence model. The new algorithm uses a novel strategy that the beliefs of particles to the social component and the cognitive component are positive linear. This strategy could maintain diversity of the swarm and overcome premature convergence. Finally, the effect of three special dependence relations on the performance of particle swarm optimization is illustrated by simulation experiments. Results show that the completely positive linear correlation has better performance than completely negative linear correlation and independence.