Mid-term forecasting of oil production on oilfield with SARIMAX model

Abstract

The article considers the possibility of applying time series analysis in the oil industry. A model of Seasonal Auto Regression Integrated Moving Average with external variables (SARIMAX) for medium-term forecasting (up to 1 year) of integrated oil production in oil fields is built. As a training sample, monthly data on oil production for 10 years of development of several fields of Rosneft PJSC have been studied; monthly data on the number of producing wells were used as an exogenous variable. As a result of the consistent application of the extended Dickey-Fuller test to the time series, the first order of integration of the random process was substantiated. The correctness of the inclusion of an exogenous variable in the model was confirmed by testing the hypothesis of the presence of co-integration between the variables of oil production and the number of producing wells. The analysis of the autocorrelation and private autocorrelation functions of the studied time series for oil production, as well as the selection of models based on the Akayke and Schwartz information criteria, made it possible to determine the best specification of the SARIMAX model. The obtained forecast values were checked against the actual values of oil production at the field. Based on the predicted and actual values, the model quality metrics have been calculated: the average approximation error (MAPE) was 0.78%. The application of the methodology for predicting oil production proposed in the article and the use of the accumulated volume of data in a structured form led to a qualitative forecast. This, in turn, will allow in the future making more informed business decisions, since a high-quality medium-term forecast allows you to save company resources.