Parallelization of a Self-adaptive Harmony Search Algorithm on Graphics Processing Units

In recent years, in order to reduce the execution time, some evolutionary algorithms that run on GPUs using Compute Unified Device Architecture (i.e., CUDA) have been proposed. In these evolutionary algorithms, they compared the execution time and precision ofGPU versions with those of CPU versions. In this study, we parallelize aself-adaptive harmony search algorithm and compare with the existing evolutionary algorithms on the same GPU platform. The proposed algorithm is divided into four steps: initialization, improvising, sorting, and updating.In the experiments, we use eight well-known optimization problems to evaluate the proposed algorithm and the other existing algorithms. As a result, our algorithm achieves the best performances among all the algorithms on the single-objective optimization problems with more dimensions or populations.