Predicción de problemas en el proceso de perforación de pozos petroleros aplicando aprendizaje de máquina supervisado : generación de una base de datos a través del análisis de carpetas de pozos y codificación del modelo de aprendizaje de máquina.

Abstract

The implementation of machine learning in drilling has enabled time savings and procedure optimization, as algorithms can swiftly extract information and vividly visualize vast and intricate datasets. Optimization of operational parameters, problem prediction, trajectory optimization, and automated task control are some of the fields in which machine learning has been efficiently applied. This research employed the supervised machine learning algorithm XGBoost to predict operational problems during oil well drilling in the Ecuadorian East. The gathered information comprises daily drilling reports, final drilling fluid reports, and corresponding lithological records for 104 wells in Block 60 Sacha Field. Principal Component Analysis (PCA) facilitated the visualization of problem grouping based on the drilling section, representing the dataset in a two-dimensional graph. The proper implementation of XGBoost allowed the generation of an initial model with specific hyperparameters: n_estimators=38, learning_rate=0.1, max_depth=15, and random_state=42, which demonstrated 100% prediction accuracy. In order to mitigate potential overfitting of the model, the hyperparameters were adjusted to n_estimators=20, learning_rate=0.05, max_depth=3, and random_state=42, resulting in a prediction accuracy of 93% on the training set. The objective is to assess which of these models performs better on a test dataset. This approach enables secure operations and thereby facilitates decision-making processes.