Previsão de Falta de Equipamento de Carga para Frota de Transporte com base em Aprendizado de Máquina e Cenários de Despacho

Abstract

A current challenge in fleet management in the mining industry is the prediction of process failure events, due to the high number of internal influence parameters (such as equipment breakage), external to the process (such as bad weather) and interface (such as shutdown of crushing). This paper deals with the development of a machine learning approach to predict the Lack of Load Equipment event for the transport fleet using dispatch data. The model applied was the Random Forest varying the number of trees between 20, 50, and 100. Backward Elimination was performed to reduce the number of attributes from 20 to 5. The metric used to evaluate the models was the correlation factor (R). Real operation data was used to build the model containing 860,000 records gathered for four months. The results of the Shapiro-Wilk and T-Welch tests showed that the models of 20, 50, and 100 trees were equivalent, the average correlation coefficients were respectively 0.768, 0.773, and 0.773; better than the value of 0.35 obtained with the linear regression method currently used by the area. Thus, the use of machine learning through Random Forest showed good results and applicability for predictions and classifications of events in fleet management in the mining industry.