New well water cut prediction using machine learning

Background: The drilling of new wells is one of the most effective geological and technical activities. In mature fields characterized by high production of reserves and high water availability, the selection of design points for drilling is a difficult task. Forecasting the parameters of new wells is possible by using geological and hydrodynamic models or analytical methods. In this paper, the authors propose the use of machine learning algorithms to predict the initial parameters of new wells based on an extensive set of geological and field data.

Aim: The article describes the process of developing machine learning algorithms and demonstrates the performance indicators of a complex model. As part of this work, testing of machine learning algorithms was performed to predict the start-up water cut of potential candidates.

Materials and methods: Within the framework of this work, various machine learning methods were applied on geological and technical field data.

Results: The developed complex model showed acceptable convergence results based on classification and regression metrics, which indicates its applicability for predicting the start-up water cut of project wells.

Conclusion: This method of predicting indicators is an alternative tool for predicting the start-up water cut of new wells, which makes it possible to clarify and supplement the forecast parameters calculated using a geological and hydrodynamic model or empirical dependencies of the initial water cut of new wells on geological parameters.