Machine Learning using synthetic data in the selection of artificial lift systems for oil wells

Authors

Carrion-Maldonado, Freddy

Lligüizaca-Dávila, Jorge

Abstract

The problem addressed the selection of hydrocarbon artificial lift methods, proposing techniques that can be more effective, whose results present the lowest possible degree of error, which should be automated as much as possible. The study proposed as a solution the implementation of computational techniques to solve the challenge of selecting artificial lift methods, and in this way presented a new alternative that provides support to the decision-making process, orienting the final criteria to be based on data.The methodology was implemented in 4 wells of interest, using python to code 4 supervised machine learning algorithms for classification. These models were trained with synthetic data. The training, evaluation, and execution of the algorithms were carried out through the use of the scikit-learn library, the whole project was developed and can be visualized in a Python IDE that can be accessed online.The selected algorithm was the decision tree algorithm, and its results were consistent with the quality of the training data. In the analysis, the influence of the distribution of training and test data was verified, which for the algorithm with the best performance exhibits an improvement when using the proportion 80 and 20. Finally, the comparisons made with the in-situ engineering evaluations show that the proposal provides a good data-driven guide, which could even be improved by adding new criteria and an improved database.