A parallel frequently coverage random-drift particle swarm optimization algorithm for extremely high-dimensional problems

• 1、School of Internet of Things Engineering, Jiangnan University, Wuxi 214122

Abstract：In order to meet the requirements of solving extremely high-dimensional problems on the computational speed and solution accuracy, a parallel frequently coverage random-driftparticle swarm optimization (P-FC-RDPSO) algorithm based on GPU is proposed in this paper. In P-FC-RDPSO algorithm, the storage structure for population and mapping approach for individuals are designed by the fine-grain parallelism strategy, which can easily adjust the increasing of dimensions and the GPU structure. The fine-grain parallelism strategy is realized by mapping the thread with problem dimension and mapping multi-blocks with an individual. Through the frequently coverage strategy, the extremely high-dimensional problem can be optimized as a low-dimensional problem which helps to improve the search ability. By evaluating the proposed algorithm on nine benchmark functions with 10000, 100000, and 1 million dimensions, the experimental results show that the algorithm can solve the extremely high dimensional problems with higher solution accuracy in faster computational speed.

Keywords： extremely high-dimensional problem, particle swarm optimization, GPU