Machine learning methods for selecting candidate wells for bottomhole formation zone treatment

Relevance. The fact that currently various technologies are widely used in oil fields to increase oil recovery and intensify the inflow, such as treatment of a bottomhole zone with hydrochloric acid. In relation to the widespread use of this technology, problematic issues are coming to the fore, including those related to the selection of the right candidate wells at a given time for carrying out well treatment. Aim. To optimize the search for candidate wells for carrying out treatment of the bottomhole zone. The work explores the possibility of using machine learning models to predict whether a well will be the right candidate for a well treatment. Object. Machine learning models of the sklearn library. Methods. To solve the problem of predicting whether a well is a candidate for bottomhole treatment, three machine learning models of the sklearn library were used: RandomForestClassifier, DecisionTreeClassifier, LinearRegression. To assess the quality of the constructed models, the following metrics from the same library were used: F1 -score, AUC-ROC-score. Results. The learning forest model showed the best results during training. Using the F1 -score metric, this model showed 99.5% convergence on the testing dataset, and using the AUCROC-score metric, the accuracy was 99.9%. The resulting accuracy indicates the correctness of using RandomForestClassifier model to solve the problem of identifying the correct candidate wells. Conclusion. The machine learning model was obtained that predicts with 99.5% accuracy whether a well will be the right candidate for a well treatment.