Application of Machine Learning to Oil Production Forecast under Uncertainties-The Linear Model

Abstract

In this paper, we propose an alternative approach to the problem of oil-production forecast based on the most straightforward feature-based machine-learning algorithm: the linear model. The method can be successfully applied to forecast both oil-rate and liquid-rate in oil fields under (i) water injection, (ii) gas injection, and (iii) simultaneous water and steam injection. Our data-driven algorithm learns the underlying reservoir dynamics from 3 sets of time-series, namely, (i) injection-rate, (ii) liquid and oil-rate, and (iii) number of producers. That is all the data we need to make reliable forecasts, no geological model or numerical reservoir simulators were used.