Combination of forecasts applied to modeling operations: a case study in a port terminal

Abstract

The aim of this study was to adjust a combined forecasting model to verify the behavior of an ore discharge system at a port terminal. It was used observations of one month of operation, collected every hour, totaling 720 observations. The methodological procedures were performed in three stages: (i) adjustment of the competitors models by the Box-Jenkins methodology; (ii) combination of the models selected by arithmetic average methods, Less heavy squared and inverse mean squared; (iii) compare of the accuracy measures of each forecast to select the best model. Twenty concurrent models were obtained, of which three were selected for a combination stage. The best model was obtained by the Inverse Mean Squared method, with an average absolute error of 2.017%. The combination methods provide an increase in the guarantee of the forecasts, generating subsidies to the decision making for improvements in the planning of the productive process.