Optimization of well trajectory with machine learning algorithms for geosteering directional drilling
Оптимизация траектории скважины с использованием алгоритмов машинного обучения для геонавигации при направленном бурении
2025-04-30
SCID: 54.1/mavmu8uc
Discuss with AI
SVM-GWOgeosteering directional drillingmachine learningrotary steerable systemwell trajectory optimization
Figures from the paper
Abstract (AI)
Abstract Drilling operations in the oil and gas sector demand precise planning and strategic decision-making. This study introduces a comprehensive real-time drilling solution that combines well trajectory planning, drilling simulation with rotary steerable system (RSS) technology, and machine learning (ML) models for enhanced steering decisions. The grey wolf optimization (GWO) algorithm and 10-fold cross-validation were employed for hyperparameter tuning to further improve model performance. The research focuses on evaluating the performance of five ML models applied to Well 14 in the Grane oil field. The primary objective is to enhance the accuracy, reliability, and efficiency of directional drilling operations, using the Grane field as a representative case study. In evaluating the ML models for steering decisions, the support vector machine with GWO (SVM-GWO) emerged as the most effective, achieving an accuracy of 96.68% and a standard deviation of 0.0046 across 10-fold cross-validation. The confusion matrix revealed true positive counts of 109, 113, and 113 for the hold (0), steer left (1), and steer right (2) classes, respectively, highlighting the model's ability to accurately identify positive instances across all categories. Receiver operating characteristic (ROC) curve and area under the curve (AUC) analyses further confirmed the SVM-GWO model's robustness, with an impressive AUC-ROC of 0.99 for all classes. Following the simulation using the SVM-GWO model, the results showed that the critical buckling load, weight on bit, torque, drag force, dogleg angle, and rate of penetration are 50 520.52 lb, 59 390.29 lb, 85 513.68 ft-lb, 14 754.43 lbf, 2.06°/100 ft, and 193.9 ft/h, respectively. All these values were within industry standards. The SVM-GWO model was further validated with real-world logging-while-drilling data from the Dora Martin-1H Well in the Eagle Ford Shale, where it achieved an accuracy of 84%, confirming its reliability in making geosteering decisions under real subsurface conditions. This study successfully integrates well trajectory planning, dynamic drilling simulation, and ML evaluations, establishing SVM-GWO as a powerful model for steering decisions in diverse geological formations.
Key Findings
1
A real-time drilling framework integrates well trajectory planning, RSS-based drilling simulation, and machine-learning steering decisions for directional drilling.
2
Among five evaluated models on Well 14 in the Grane field, SVM-GWO performed best, achieving 96.68% accuracy and a 0.0046 standard deviation in 10-fold cross-validation.
3
SVM-GWO accurately classified hold, steer-left, and steer-right decisions, with true-positive counts of 109, 113, and 113, respectively, and an AUC-ROC of 0.99 for every class.
4
Simulation using SVM-GWO produced drilling parameters within industry standards, including a 2.06°/100 ft dogleg angle and a 193.9 ft/h rate of penetration.
5
Validation on real-world logging-while-drilling data from the Dora Martin-1H Well achieved 84% accuracy, supporting the model’s reliability for geosteering decisions.
Research Object
real-time directional drilling and geosteering of oil wells, represented by Well 14 in the Grane oil field
Research Subject
machine-learning-based steering-decision performance and drilling-trajectory optimization, including accuracy, reliability, efficiency, and simulated drilling parameters
Publication Details
Publication Date
2025-04-30
Journal
Publisher
ISSN
Cited by
3
Open access PDF
Access Type
Author Information
Download PDF
Subscribe to digest