Acurácia dos métodos de previsão base e combinação via reconciliação temporal de demanda para tubos de revestimento de poços terrestres de petróleo: Rio Grande do Norte e Ceará

Abstract

The decline in production of mature oil and gas fields has led companies to further optimize their inventory and acquisition costs. This work aims to compare the accuracy of different methods of base and temporal hierarchy forecasting of demand for coating tubes for onshore oil wells in Brazilian states of Rio Grande do Norte and Ceará (RN & CE). The research method was exploratory used in the forecasting methods: naive, seasonal naive, ETS (ExponenTial Smooth), ARIMA (AutoRegressive Integrated Moving Average) and THETA, with accuracy metrics: MAPE (absolute mean error), sMAPE (synmetric mean absolute percentage error), RMSE (root mean squared error) and MASE (percentage error absolute mean of scale), using a training approach in the period from 2010 to 2016 and testing in the years 2017 and 2018. For the base forecasts, the ETS and ARIMA methods match the best of the results, the benchmark itself (naive). Considering the temporal reconciliation, there were several improvements in different methods and metrics applied, however, not being unanimous in all cases. It is highlighted the THETA method when using this approach, which occurred because the base accuracy is low, therefore being more efficient in this scenario of low accuracy. The different results between the frequencies can be explained by the difficulties of the reconciled models of capturing, with relative accuracy, as components of trend and seasonality of the models.