Model Development for Brittleness Index Estimation and Depth Determination in Hydraulic Fracturing Operations in Shale Gas Reservoirs Using Machine Learning Algorithms

SPE Journal, Год журнала: 2025, Номер unknown, С. 1 - 22

Опубликована: Май 1, 2025

Summary Accurate estimation of the brittleness index (BI) is critical for optimizing hydraulic fracturing operations in shale gas reservoirs, as it directly influences fracture propagation and recovery efficiency. The BI quantifies resistance rock to fracturing, a key factor determining optimal depth stimulation. Prior methods estimating BI, such empirical correlations other utilized machine learning (ML) techniques, often suffer from limited accuracy generalizability, particularly complex geological formations like Fuling field. To address these limitations, ML techniques have gained prominence due their ability capture complex, nonlinear relationships within large data sets, improving predictive accuracy. In this study, we propose novel approach that utilizes hybrid group method handling based on discrete differential evolution (GMDH-DDE) predict BI. GMDH-DDE model was compared with (GMDH), random forest (RF), multilayer perceptron (MLP). results demonstrate significantly outperforms models, achieving coefficient determination (R2) 0.9984, root mean square error (RMSE) 0.2895, absolute (MAE) 0.02543 unseen data. GMDH ranked second estimation, an R2 0.9805, RMSE 0.4635, MAE 0.04224. It followed by RF model, 0.9599, 0.6034, 0.0997. MLP however, had lowest performance, 0.9263, 0.9566, 0.1256. Additionally, demonstrates superior computational efficiency, requiring only 1.12 seconds. This significant advantage over methods, taking 4.82 seconds, 11.23 27.45 These findings highlight potential providing accurate computationally efficient estimations. improved efficiency are expected contribute more effective cost-efficient operations, ultimately enhancing economic viability reservoirs.

Язык: Английский

Advancing Reservoir Characterization with Machine Learning: A Multi-Well Predictive Analysis

Опубликована: Июнь 2, 2025

Abstract Reservoir characterization remains one of the most significant requirements in optimizing hydrocarbon recovery, yet traditional methods still struggle with data variability, complexity, and cross-well generalizability. This study presents a supervised machine learning (ML) approach to predict production profiles across three wells, focusing on dataset structure variabilities, specifically use whole compared perforated-zone subsets, impact model performance, predictive accuracy. In this study, 926 points from two training wells (A B) was used for validations, while 917 an additional well C out-of-distribution testing. Advanced processing techniques, including variance inflation factor (VIF) mitigating multicollinearity singular value decomposition (SVD) identifying hidden correlations were used. Ten different ML models trained via randomized search optimization, Light Gradient Boosting (LightGBM) achieving highest accuracy (MAE: 0.0679, R2: 0.88). Testing Well revealed deteriorated performance 1.1907, 0.66) poor generalizability, especially variables out range set. can be attributed inherent geological variability differences. The predictions deteriorate even further when perforated zones 0.866, 0.459), indicating that factors may require investigation enhance prediction these specific reservoir intervals.

Язык: Английский