A Parallel Multi-swarm Particle Swarm Optimization Algorithm Based on CUDA Streams

Since the Compute Unified Device Architecture (CUDA) has been proposed, some swarm intelligence algorithms were migrated to the GPU. The release of the Fermi architecture allows NVIDIA's GPUs to use CUDA streams to launch multiple kernel functions simultaneously. Many forms of particle swarm optimization algorithms for CUDA without streams were proposed. In this paper, in order to further utilize GPU performance, we propose a method of applying the multi-swarm particle swarm optimization algorithm (PSO) on CUDA streams. Aiming to further enhance the search ability of PSO, the multi-start local search algorithm was applied to multi-swarm PSO. The experimental results show that the algorithm has better performance.