Enhancing reservoir characterization: A novel machine learning approach for automated detection and reconstruction of outliers-affected well log curves

Physics of Fluids, Journal Year: 2025, Volume and Issue: 37(3)

Published: March 1, 2025

The drilling process can result in irregular measurements due to unconsolidated geological formations, affecting the accuracy of wireline logging devices. This impacts precision elastic log measurements, such as velocity and density profiles, which are essential for reservoir characterization. reliability wireline-logging tool is crucial preventing inaccuracies when assessing rock properties. Previous studies have focused on applying machine learning (ML) techniques logging, but these methods limited applicability, particularly outlier detection reconstruction. In response, this study integrates both supervised unsupervised ML enhance responses Initially, density-based spatial clustering applications with noise was applied detection, followed by feature selection identify correlated logs reconstructing log. A random forest regression model, optimized particle swarm optimization (PSO), then trained using selected features. comparative analysis showed a significant improvement porosity estimation from reconstructed compared core data. Specifically, comparison between original bulk yielded an R2 0.95 root mean squared error (RMSE) 0.012. contrast, rebuilt resulted 0.98 RMSE 0.007. integration advanced PSO-optimized models represents considerable advancement field approach enhances also saves time reduces manual effort, highlighting potential petroleum exploration production.

Language: Английский

---

A Multi-Model Fusion Network for Enhanced Blind Well Lithology Prediction

Processes, Journal Year: 2025, Volume and Issue: 13(1), P. 278 - 278

Published: Jan. 20, 2025

Lithology identification is essential for formation evaluation and reservoir characterization, serving as a fundamental basis assessing the potential value of oil gas resources. However, traditional models often struggle with accuracy due to complexities nonlinear relationships class imbalances in well-logging data. This paper presents an effective multi-model ensemble approach lithology identification, integrating one-dimensional multi-scale convolutional neural networks (MCNN1D), Graph Attention Networks (GAT), Transformer networks. MCNN1D extracts local features lithological changes varying kernels, enhancing robustness complex geological The GAT assigns adaptive weights adjacent nodes, capturing spatial among samples interactions. Meanwhile, uses self-attention capture contextual sequences, improving global feature processing identification. fusion effectively combines strengths individual models, enabling comprehensive efficient modeling features. Experimental results show that proposed Multi-Model Fusion Network outperforms other accuracy, precision, recall, F1-score on Hugoton–Panoma oilfield dataset, achieving 95.06% lithologies. mitigates effects data imbalance enhances making it powerful tool reservoirs.

Language: Английский

---

Prediction of Lithofacies in Heterogeneous Shale Reservoirs Based on a Robust Stacking Machine Learning Model

Minerals, Journal Year: 2025, Volume and Issue: 15(3), P. 240 - 240

Published: Feb. 26, 2025

The lithofacies of a reservoir contain key information such as rock lithology, sedimentary structures, and mineral composition. Accurate prediction shale is crucial for identifying sweet spots oil gas development. However, obtaining through core sampling during drilling challenging, the accuracy traditional logging curve intersection methods insufficient. To efficiently accurately predict lithofacies, this study proposes hybrid model called Stacking, which combines four classifiers: Random Forest, HistGradient Boosting, Extreme Gradient Categorical Boosting. employs Grid Search Method to automatically search optimal hyperparameters, using classifiers base learners. predictions from these learners are then used new features, Logistic Regression serves final meta-classifier prediction. A total 3323 data points were collected six wells train test model, with performance evaluated on two blind that not involved in training process. results indicate stacking predicts achieving an Accuracy, Recall, Precision, F1 Score 0.9587, 0.959, respectively, set. This achievement provides technical support evaluation spot exploration.

Language: Английский

---

Gas well productivity prediction based on fractional Fourier transform

Journal of Petroleum Exploration and Production Technology, Journal Year: 2025, Volume and Issue: 15(5)

Published: April 19, 2025

Language: Английский

---

Optimizing seismic-based reservoir property prediction: a synthetic data-driven approach using convolutional neural networks and transfer learning with real data integration

Artificial Intelligence Review, Journal Year: 2024, Volume and Issue: 58(1)

Published: Nov. 30, 2024

Language: Английский

---

Geostatistics and artificial intelligence coupling: advanced machine learning neural network regressor for experimental variogram modelling using Bayesian optimization

Frontiers in Earth Science, Journal Year: 2024, Volume and Issue: 12

Published: Dec. 12, 2024

Experimental variogram modelling is an essential process in geostatistics. The use of artificial intelligence (AI) a new and advanced way automating experimental modelling. One part this AI approach the population search algorithms to fine-tune hyperparameters for better prediction performing. We Bayesian optimization first time find optimal learning parameters more precise neural network regressor goal leverage capability consider previous regression results improve output using three variograms as inputs one training, calculated from ore grades four orebodies, characterised by same genetic aspect. In comparison architectures, Bayesian-optimized demonstrably achieved superior Coefficient determination validation 78.36%. This significantly outperformed non-optimized wide, bilayer, tri-layer configurations, which yielded 32.94%, 14.00%, −46.03% determination, respectively. improved reliability demonstrates its superiority over traditional, regressors, indicating that incorporating can advance modelling, thus offering accurate intelligent solution, combining geostatistics specifically machine

Language: Английский

---