05864993

a

05864993 Tan, Ying Zhou, You Parallel particle swarm optimization algorithm based on graphic processing units. Panigrahi, Bijaya Ketan (ed.) et al., Handbook of swarm intelligence. Concepts, principles and applications. Berlin: Springer (ISBN 978-3-642-17389-9/hbk; 978-3-642-17390-5/ebook). Adaptation, Learning, and Optimization 8, 133-154 (2011). 2011 Berlin: Springer EN particle swarm optimization graphic processing units parallelization CUDA

doi:10.1007/978-3-642-17390-5_6

Summary: A novel parallel approach to implement a particle swarm optimization(PSO) algorithm on graphic processing units (GPU) in a personal computer is proposed in this chapter. By using the general-purpose computing ability of GPU and under the software platform of compute unified device architecture (CUDA), which is developed by NVIDIA, the PSO algorithm can be executed in parallel on the GPU. The process of fitness evaluation as well as the updating of the velocity and the position of all the particles in the swarm are parallelized and described in details in the context of this chapter. Experiments are conducted by running the PSO both on the GPU and the CPU, respectively, to optimize several benchmark test functions. The running time of the PSO based on GPU (GPU-PSO, for short) is greatly shortened compared to that of the PSO based on CPU (CPU-PSO, for short). A $40\times$ speedup can be obtained by our implemented GPU-PSO on a display card of NVIDIA Geforce 9800GT, with the same optimizing performance. Compared to the CPU-PSO, the GPU-PSO has special speed advantages on large-scale populations and high-dimensional problems which often happen and which are widely used in a lot of real-world optimization applications.