| Summary: | 博士 === 國立成功大學 === 資訊管理研究所 === 104 === In spite of the varying position and fitness of each distinct particle, most of the PSO algorithms treat the given swarm of particles simply. This study aims to find good controls for facilitating exploration and exploitation movements to enhance the traditional particle swarm optimization (PSO) algorithm. In this sense, this study seeks improvements to PSO by introducing adaptive controls on inertia weight and acceleration coefficients according to their corresponding evolutionary states.
Two novel PSO strategies are proposed to facilitate the transitions between searches of exploration and exploitation on the corresponding evolutionary status instead of merely on time (number of iteration). The enhanced particle swarm optimization algorithms, referred as PSO-LGR (location gain regulator) and PSO-FWAC (fitness weighted acceleration coefficients), detect the evolutionary state based on the location and fitness of particles respectively.
Experimental results on widely used benchmark functions show that PSO-LGR and PSO-FWAC outperform the static and time-varying approaches in terms of the precision, success rate, and convergence speed of particle swarm optimization. It is considered as valuable contributions of this study that the proposed regulators are able to enhance traditional PSO by introducing appropriate turbulence depending on corresponding evolutionary states.
|
|---|
