Mega busca harmônica: algoritmo de busca harmônica baseado em população e implementado em unidades de processamento gráfico

Abstract

This work propose a new approach for the metaheuristic Harmonic Search (HS), by using a population of solutiona and other strategies inspired in another metaheuristics. This new model was implemented using a parallel architecture of a graphical processing unity (GPU). The use of GPU for general-purpose processing is growing, specially for scientific processing. Its use is particularly interesting for populational metaheuristics, where multiple operations are executed simultaneously. The HS is a metaheuristic inspired by the way jazz musicians search for a perfect harmony. In the proposed model a population of temporary harmonies was included. Such population was generated at each iteration, enabling simultaneous evaluation of the objective function being optimized, and thus, increasing the level of parallelism of HS. The new approach implemented in GPU was named Mega Harmony Search (MHS), and each step of the algorithm is handled in the form of kernels with particular configurations for each one. To show the efficiency of MHS some benchmark problems were selected for testing, including mathematical optimization problems, protein structure prediction, and truss structure optimization. Factorial experiments were done so as to find the best set of parameters for the MHS. The analyzes carried out on the experimental results show that the solutions provided by MHS have comparable quality to those of the simple Harmony Search. However, by using GPU, MHS achieved a speedup of 60x, compared with the implementation in regular CPU. Future work will focus other improvements in the algorithm, such as the use of niches and species, as well a study of the interactions between them.