APPLICATION OF MACHINE LEARNING FOR FORECASTING FORMATION PRESSURE IN OIL FIELD DEVELOPMENT

Abstract

The relevance of the study is caused by the fact that the advent of artificial intelligence in the oil industry has led to an increase in its use in exploration, development, production, field design and management planning to speed up decision-making, reduce costs and time. Machine learning has gained immense popularity in correlating complex nonlinear datasets and has demonstrated its superiority over regression methods in petroleum engineering in terms of large data prediction errors, processing power and memory. This article discusses the use of machine learning to assess its effectiveness and potential for determining and predicting reservoir pressure values in oil field development, compared with conventional statistical models of oil and gas engineering. The main aim of the study is to assess the possibilities of calculating and predicting reservoir pressure using the «random forest» machine learning method. Object: dynamics of reservoir pressure during the development of terrigenous deposits of oil fields in the Perm Krai. Methods: methods of probabilistic-statistical analysis and machine learning «random forest regression». The results. The presented study proposes a new method for predicting reservoir pressure using machine learning, based on a nonparametric multidimensional model that links well performance over time. The proposed method takes into account the dynamics of indicators characterizing the operation of wells, and the predicted reservoir pressure is well correlated with the values measured using hydrodynamic studies. It was found that the «random forest» machine learning method provides better performance in terms of reservoir pressure prediction accuracy than the linear regression method. The prospects for further development are additional «training» of the «random forest» model and assessment of the possibility of using other machine learning methods to solve the problem, including expanding the set of factors for more accurate modeling of reservoir pressure.

Keywords

• artificial intelligence
• machine learning
• multidimensional model
• random forest
• well operation indicators
• current reservoir pressure
• correlation