high performance computing on graphics processing units: hgpu.org

Gravitation Field Algorithm (GFA) is a simple but very effective heuristic search algorithm. This algorithm has obvious advantages in multimodal function optimization problems compared with SA and GA. However, when we want to get a more precise global optimal value, it needs a lot of initial dusts involved in computing, which causes a low efficiency of the algorithm. In this paper, we propose Parallel Gravitation Field Algorithm (PGFA) based on the island model, taking advantage of the powerful computing ability of GPU and CUDA parallel computing architecture. In PGFA, dust points uniformly distributed in a plurality of islands, and we achieved a parallel movement and evolution of dusts on each island. We compared PGFA with GFA in the unimodal functions and multimodal functions. The experimental results show that PGFA can not only get an effective speedup, but also higher accuracy than the GFA.