Comparative study of Particle Swarm Optimization and Differential Evolution Algorithms on a Graphics Processing Unit

Abstract

Bio-inspired algorithm such Differential Evolution (DE) and Particle Swarm Optimization (PSO) algorithms are well known alternative options for hard to optimize problems settled with bio-inspired heuristics. Both algorithms have low computational complexity, good performance, and need only a few working parameters and have a good performance. This paper shows a comparative study for parallel implementations of these two well known heuristics, as long as these are population-based algorithms their coding an implementation on a Graphics Processing Unit device using CUDA as base of parallel programming are now common topics. Our main objective is to obtain the algorithm performance of both DE and PSO algorithms operating on a GPU and compare both algorithms with their sequential and parallel implementations. The result of our research shows that executing a parallel algorithm in a GPU changes the convergence behavior to the global optimum and it will present a decrease in computation time and its performance may be very different, with respect to the same algorithm but programmed in a sequential programming.