NOVASINERGIA (Jun 2018)
Efecto de los coeficientes de aceleración de PSO en el desempeño de una Red Neuronal Artificial aplicada a la Estimación de Costos
Abstract
The particle metaheuristics Particle Swarm Optimization (PSO) since its appearance has proven to be efficient in solving optimization problems, the variation of its parameters has allowed to improve its efficiency. The present work is focused on performinga comparative study of the effect of the acceleration coefficients c1and c2, on the performance of PSO to solve a problem of cost estimation, through an Artificial Neural Network (ANN) sigmoidal feedforward. A range of values was evaluated in the acceleration coefficients, the other parameters, in this case inertial factor and the swarm size were worked with fixed values. The validation of the solution was carried out by means of a pipeline data set for fluid transfer used in the industry, coming from a real case, with information related to weight, welding type, diameter and the corresponding cost. The objective function used is the Mean Square Error (MSE), calculated between the observed values and the values estimated by the ANN. From the results it can be seen that very small values of c1and c2obtain low accuracy in the estimation of pipe manufacturing costs, while the best accuracy is achieved by means of acceleration coefficients with values greater than or equal to 0.5
