A GPU accelerated PSO with application to Economic Dispatch problem

This paper investigates the use of Graphics Processing Units (GPUs) as general purpose parallel architectures, for the acceleration of the solution of the Economic Dispatch problem (ED) via stochastic search algorithms. The Comprehensive Learning Particle Swarm Optimizer (CLPSO) is used as host process to carry out the optimization task. At every time of the evolution a parallel graphics card speeds up the optimization process by calculating, in parallel, the fitness value of all particles. Two different approaches are investigated: a fine-grained parallelism and a coarse-grained one. The results demonstrate that GPUs can be applied with success to speed up computationally intensive problems in electric energy systems.