Data-driven decision-making for lost circulation treatments: A machine learning approach

Lost circulation is an expensive and critical problem in the drilling operations. Millions of dollars are spent every year to mitigate or stop this problem. In this work, data from over 3000 wells were collected from multiple sources. The data went through a processing step where all outliers were removed and decision rules were set up. Multiple machine learning methods (support vector machine, decision trees, logistic regression, artificial neural networks, and ensemble trees) were used to create a model that can predict the best lost circulation treatment based on the type of loss and the reason of loss. 5-fold cross-validation was conducted to ensure no overfitting in the created model. After using all the aforementioned machine learning methods to train models to choose the best lost circulation treatment, overall, the results showed that support vector machine had the highest accuracy among the other algorithms. Thus, it was selected to train the modeL The created model went through quality control/quality assurance (QC/QA) to limit the results of incorrect classification. Two treatments were suggested to treat partial loss, four to treat severe loss, and seven for complete loss, based on the reason of loss. In addition, a formalized methodology to respond to lost circulation was provided to help the drilling personnel handling lost circulation in the field.