A Method of Self-Adaptive Inertia Weight for PSO

Abstract

The particle swarm optimization algorithm (PSO) has successfully been applied to many engineering optimization problems. However, the most of existing improved PSO algorithms work well only for small-scale problems on low-dimensional space. In this new self-adaptive PSO, a special function, which is defined in terms of the particle fitness, swarm size and the dimension size of solution space, is introduced to adjust the inertia weight adaptively. In a given generation, the inertia weight for particles with good fitness is decreased to accelerate the convergence rate, whereas the inertia weight for particles with inferior fitness is increased to enhance the global exploration abilities. When the swarm size is large, a smaller inertia weight is utilized to enhance the local search capability for fast convergence rate. If the swarm size is small, a larger inertia weight is employed to improve the global search capability for finding the global optimum. For an optimization problem on multi-dimension complex solution space, a larger inertia weight is employed to strengthen the ability to escape from local optima. In case of small dimension size of solution space, a smaller inertia weight is used for reinforcing the local search capability. This novel self-adaptive PSO can greatly accelerate the convergence rate and improve the capability to reach the global minimum for large-scale problems. Moreover, this new self-adaptive PSO exhibits a consistent methodology: a larger swarm size leads to a better performance.