Aplicação de Modelos Nebulosos Univariados e Multivariados na Previsão de Preços de Minério De Ferro: Um Estudo Comparativo

  • Hamilton Tonidandel Jr. Programa de Pós-Graduação em Instrumentação, Controle e Automação de Processos de Mineração, Universidade Federal de Ouro Preto e Instituto Tecnológico Vale, Ouro Preto, MG
  • Frederico Gadelha Guimarães Departamento de Engenharia Elétrica, Universidade Federal de Minas Gerais, Universidade Federal de Minas Gerais, Belo Horizonte, MG
Keywords: Time series, Forecasting, Iron ore, Fuzzy logic, Multivariate series, Operational planning


Iron ore ranks second among the highest value items exported by Brazil in 2020, with Vale S.A. being the largest exporter of this commodity in the world. At this point, estimating the future behavior of a short-term iron ore price time series is an important tool at mining projects, especially in decision-making related to operational planning. The present work evaluates, in this context, the accuracy of the fuzzy models PWFTS (Probabilistic Weighted Fuzzy Time Series) and FDT (Fuzzy Decision Trees) in iron ore price prediction, by presenting a comparative study with the predictive models: ARIMA, Multilayer Perceptron (MLP) and Xgboost. In order to ensure the variability of input patterns, the data are distributed into subsets, using the sliding windows technique. In a multivariate context, predictor variables are selected through correlational analysis with the target time series, with emphasis on the inclusion of the iron ore surplus production from Vale S.A. Results for the univariable and multivariable modeling indicate superiority of the fuzzy models, according to RMSE and MAPE metrics.