Análise paramétrica em algoritmo de inteligência de enxame utilizando funções Benchmark

Optimization problems are present in applications in the scientific, financial, industrial and management areas, and in recent years several methodologies have emerged that aim to obtain their solution. One of these techniques is known as Swarm Intelligence (SI), based on the behavior of relatively simple beings, but who manage to solve complex problems when they are placed in a collective. In the literature, several SI algorithms were found, but there was a lack of in-depth studies on the behavior of each of its parameters. Therefore, this work presents a parametric analysis of two widely used SI algorithms, namely Particle swarm optimization (PSO) and Firefly Algorithm (FA). For this, a benchmarking study was carried out using benchmark functions in three search intervals of different sizes, in order to evaluate metrics such as accuracy, precision and average processing time. For this purpose, respectively, 315 and 207 scenarios were developed for PSO and FA. Furthermore, to compare the SI in relation to the traditional heuristic, 171 scenarios were developed for the Random Walk (RW) algorithm. With results, sets of parameters were obtained with accuracy and precision around 100% in the best scenarios of the SI algorithms, demonstrating the importance of a good parameterization for an optimal performance of the method.