High performance implementation of APSO algorithm using GPU platform

Optimization can be defined as the act of getting the best result under given circumstances. Evolutionary algorithms are widely used for solving optimization problems. One of these evolutionary algorithms is Particle Swarm Optimization (PSO). Different kinds of PSO such as Adaptive Particle Swarm Optimization (APSO), have been presented to improve the original PSO and eliminate its disadvantages. Although APSO can overcome the problem of premature convergence and accelerate the convergence speed at the same time, it is computationally intensive because of its nested loops. The goal of this paper is high performance implementation of APSO algorithm based on GPU. In order to analyze this algorithm and evaluate its computational time, we have implemented APSO on both CPU and GPU. Different parallelisms such as loop-level parallelism have been exploited and we have achieved significant speedup up to 152x compared to CPU based implementation.