Optimalizace pomocí roje částic

Abstract

This master's thesis deals with application of Evolutionary algorithms on optimization problems and parallelization of those algorithms on GPU. The aim of this work was to implement Particle swarm optimization on a graphics card with CUDA API, then make performance benchmarks to test it and showcase it in the application, where particles of this algorithm are visualized. For better comparison, algorithms like Differential evolution and Ant colony optimization were implemented too, all used algorithms were implemented in CUDA, C++ with OpenMP and Python. In the best-case scenario, the GPU implementation was 15200 times faster, in the worst-case, the speedup was only 86 times faster than Python.