Answers are grounded in this paper and cite their sources.
Proper prediction of the length of perforation penetration plays an essential role in maximization of well productivity and improved reservoir connectivity, but the standard empirical models are still limited in their scope, often necessitating recalibration and inadequate coverage of gun and well parameters. This study mitigates these shortcomings by building an advanced machine learning model that is trained on 1,648 API-19B real-world standard perforation tests, which constitute 16,480 data points under a variety of operating conditions. Routinely measured input features were used to evaluate ten machine learning algorithms. These features include shot density, casing grade, casing nominal weight, casing outer diameter, explosive type, temperature rating, explosive weight, cement compressive strength, and gun diameter. Ensemble and neural network models recorded the best accuracy with AdaBoost, Random Forest, Gradient Boosting, and Neural Network recording root mean square errors of 0.059, 0.072, 0.078, and 0.084, respectively. These findings confirm that machine learning has a profoundly better performance compared to traditional empirical models; not only does it provide effective, low-cost, and real-time estimations of the length of perforation penetration, but also it is more adaptable to a wide range of perforation guns and well completion parameters.
No takes yet. Share an insight, caveat, or question.
Nashed et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: