Data-driven surrogate model approach for recommending lost circulation treatments

Lost circulation is a significant challenge in drilling operations, often resulting in increased non-productive time (NPT) and higher operational costs. Traditional methods for selecting lost circulation treatments rely heavily on trial-and-error, typically involving multiple field attempts before achieving success. This process is time-consuming, costly, inefficient, and has a low success rate. To overcome these limitations, we developed an innovative data-driven surrogate model that predicts the expected effects of lost circulation treatments. By integrating the surrogate model into a treatment decision-making framework, the model systematically evaluates various treatment options and predicts the probabilities of different effects, such as success, partial success, and failure. Based on comprehensive insights into the treatment outcomes, enabling more informed decisions and selecting the most optimal treatment option. The surrogate model was trained using a dataset of 813 lost circulation treatment cases, incorporating 15 key parameters such as well conditions, geological features, and operational drilling parameters. Built on the CatBoost algorithm, the model achieves an AUC consistently above 0.90, demonstrating high accuracy in predicting treatment effects. The model was integrated into the treatment decision-making framework and tested on three field cases. The results showed that the model's predictions aligned closely with actual field outcomes. Additionally, it provided comprehensive analyses of various treatment options, enabling engineers to enhance treatment success and improve decision reliability. In summary, the proposed intelligent decision-making framework offers a systematic, scientific approach to lost circulation management, reducing reliance on field experience, improving treatment success rates, enhancing drilling safety and efficiency, and lowering operational costs.