Implementation and Interpretation of Machine and Deep Learning to Applied Subsurface Geological Problems

Prediction Models Exploiting Well-Log Information

1st Edition - February 18, 2025
Imprint: Elsevier
Author: David A. Wood
Language: English
Paperback ISBN:
9 7 8 - 0 - 4 4 3 - 2 6 5 1 0 - 5
eBook ISBN:
9 7 8 - 0 - 4 4 3 - 2 6 5 1 1 - 2

Implementation and Interpretation of Machine and Deep Learning to Applied Subsurface Geological Problems: Prediction Models Exploiting Well-Log Information explores machine and de… Read more

Purchase options

World Book Day Celebration!

Save up to 30% on books and eBooks!

Shop now

Institutional subscription on ScienceDirect

Request a sales quote

Implementation and Interpretation of Machine and Deep Learning to Applied Subsurface Geological Problems: Prediction Models Exploiting Well-Log Information explores machine and deep learning models for subsurface geological prediction problems commonly encountered in applied resource evaluation and reservoir characterization tasks. The book provides insights into how the performance of ML/DL models can be optimized—and sparse datasets of input variables enhanced and/or rescaled—to improve prediction performances. A variety of topics are covered, including regression models to estimate total organic carbon from well-log data, predicting brittleness indexes in tight formation sequences, trapping mechanisms in potential sub-surface carbon storage reservoirs, and more.

Each chapter includes its own introduction, summary, and nomenclature sections, along with one or more case studies focused on prediction model implementation related to its topic.

Title of Book
Cover image
Title page
Table of Contents
Copyright
Preface
Chapter 1. Overview of artificial intelligence methods and data analysis techniques suitable for subsurface datasets
Abstract
1.1 Introduction
1.2 Types and purpose of artificial intelligence models
1.3 Statistical characterization of AI datasets
1.4 Dataset preprocessing and the identification of extreme data-record values
1.5 Feature influence and selection
1.6 Model configurations and control variable optimization
1.7 Prediction assessment methods and metrics
1.8 Potential future directions
1.9 Summary and conclusions
Nomenclature
Appendix 1.1 Python code snippet for summarizing the fundamental characterization data for a structured dataset
Appendix 1.2 Python code snippet for displaying useful graphics for visualizing key characteristics of a structured dataset
References
Chapter 2. Regression models to estimate total organic carbon (TOC) from well-log data
Abstract
2.1 Introduction
2.2 Empirical methods for calculating TOC from petrophysical well logs
2.3 Machine learning approaches to estimating TOC from well-log data
2.4 Types of well-log attributes worth calculating
2.5 Regression error prediction assessment
2.6 Multi-K-fold cross-validation analysis for SML model selection
2.7 Normalizing data variable distributions
2.8 SML control/ hyperparameter optimization
2.9 Feature selection using a suite of optimizers
2.10 Case study: SML applied to predict TOC from Barnett Shale Well A
2.11 Summary and conclusions
Declarations
Nomenclature
Appendix 2.1 Python code for leave-one-out loop applied with TOB Stage 1 model
References
Chapter 3. Predicting brittleness indexes in tight formation sequences
Abstract
3.1 Introduction
3.2 Mineralogical and geomechanical influences on brittleness of tight formations
3.3 Brittleness indexes and how they are calculated
3.4 Value of well-log data for predicting brittleness
3.5 Overcoming the challenges of sparse well-log data
3.6 Supervised machine learning methods applied to predict BI
3.7 Outlier detection and interpretation
3.8 Case study: assessing outlying data records in a BIm prediction dataset
3.9 Feature importance analysis to specific supervised machine learning model solutions
3.10 Case study: supervised machine learning applied to predict BIm from a sparse suite of well logs
3.11 Case study: supervised machine learning applied to predict BIg in Appalachian Basin shale gas wells
3.12 Summary and conclusions
Declarations
Nomenclature
Appendix 3.1 Python code for calculating Pearson’s (R) and Spearman’s (p) correlation coefficients simultaneously
Appendix 3.2 Python code for Permutation Feature Importance Calculation
References
Chapter 4. Classifying lithofacies from well logs using supervised machine learning, cluster, and principal component analysis plus stacking model combinations
Abstract
4.1 Introduction
4.2 Lithofacies variations observed in clastic sedimentary sequences
4.3 Lithofacies variations observed in carbonate sedimentary sequences
4.4 Machine learning methods applied to lithofacies classification using well logs
4.5 Prediction performance metrics for classification/ misclassification assessment
4.6 Annotated confusion matrices assist in the visualization of classification prediction performance
4.7 Case study: lithofacies prediction methods for synthetic clastic sequence
4.8 Case Study: Lithofacies prediction of mixed clastic and nonclastic sequences
4.9 Summary and conclusions
Declarations
Nomenclature
Appendix 4.1 Python code for confusion matrix and classification accuracy metrics
Appendix 4.2 Python code for conducting Bartlett’s test of equal variance
References
Chapter 5. Permeability, porosity, and water saturation relationships and distributions in complex reservoirs
Abstract
5.1 Introduction
5.2 Definition of hydraulic flow units of complex reservoirs
5.3 Relevance of pore size and specific surface area to permeability distributions
5.4 Convolutional neural networks prediction models configured to use image inputs
5.5 Case Study 1: Outlier analysis considering the Mahalanobis distance
5.6 Case study 2: Robustness testing of machine learning models by applying noise
5.7 Case Study 3: Partial dependence between model target and input variables
5.8 Case Study 4: Shapley additive explanations for feature importance and selection
5.9 Summary and conclusions
Declarations
Nomenclature
Appendix 5.1 Python code for Mahalanobis distance calculation
Appendix 5.2 Python code for applying Gaussian noise for SML robustness testing
Appendix 5.3 Python code for SML partial dependence analysis
References
Chapter 6. Trapping mechanisms in potential subsurface carbon storage reservoirs and their prediction by machine learning
Abstract
6.1 Introduction
6.2 Subsurface geologic reservoirs for storing captured CO2
6.3 The role of supervised machine and deep learning in evaluating potential CCS/CCUS reservoirs
6.4 Prediction of CO2 residual and solubility trapping indexes in saline reservoirs
6.5 Case Study 1. Multifold cross-validation for insight into RTI prediction models
6.6 Case Study 2: aFr-index prediction performance and data mining metric
6.7 Case Study 3: Regression error characteristic curves for improved visualization of model performance
6.8 Case Study 4: Techniques and metrics able to reliably assess SML model overfitting/underfitting performance
6.9 Summary and conclusions
Nomenclature
Appendix 6.1 Excel cell formula for calculating aFr-indexes
Appendix 6.2 : Python code snippet for calculating regression error characteristic (REC) curve and area under curve (AUC)
References
Chapter 7. Autocorrelation and accurate picking of formation boundaries using well-log data from multiple wells drilled in complex geological sequences
Abstract
7.1 Introduction
7.2 Autocorrelation of well-logged sections across multiple wellbores
7.3 Deep learning methods for accelerating well-log correlation
7.4 Exploiting well-log attributes for formation boundary/thickness determinations
7.5 Case Study: applying binary classification to predict a formation boundary
7.6 Summary and conclusions
Nomenclature
Appendix 7.1 1D convolutional autoencoder Python configuration
Appendix 7.2 Reconfiguring XGB models in Python to combine with RF capabilities
References
Chapter 8. Assessing formation loss of circulation risks with mud-log datasets: resampling and classifying imbalanced datasets
Abstract
8.1 Introduction
8.2 Machine and deep learning approaches to predicting lost circulation
8.3 Class imbalance challenges for lost circulation predictions
8.4 Synthetic minority class oversampling techniques
8.5 Local outlier factor (LOF) technique with data mining applications
8.6 Classification prediction performance metrics appropriate for imbalanced datasets
8.7 Case Study 1. Prediction of an imbalanced lost circulation dataset with regression and classification models
8.8 Case Study 2: Prediction of large imbalanced lost circulation classification dataset
8.9 Summary and conclusions
Nomenclature
Appendix 8.1 Python code snippet for dataset resampling algorithm configurations
Appendix 8.2 Python code snippet for applying local outlier factor (LOF) algorithm
References
Chapter 9. Predicting formation fracture characteristics derived from borehole image data with petrophysical well-log variables
Abstract
9.1 Introduction
9.2 Formation characteristics that can be detected and quantified with BHI data
9.3 BHI logging tool features and configurations
9.4 Disparities between feature distributions identified in cores and BHI
9.5 SML/SDL research studies addressing the prediction of reservoir fracture properties
9.6 Leverage and influence metrics to identify atypical data records
9.7 Case study: two-step prediction of reservoir fracture density from BHI and petrophysical logs
9.8 Summary and conclusions
Nomenclature
Appendix 9.1 Python code snippet for calculating leverage and influence metrics
References
Chapter 10. Quantifying reservoir microfacies characterization using thin-section, scanned, computed tomography, and electron microscope image data
Abstract
10.1 Introduction
10.2 Thin-section image analysis
10.3 Preparation and enhancement steps for thin-section image analysis
10.4 High-resolution thin-section scan images of core plugs and cuttings samples
10.5 Integrated reservoir characterization based on thin-section image analysis
10.6 Computed tomography and micro-CT (μCT) images analysis
10.7 SEM image analysis to provide detailed kerogen characterization
10.8 Case Study 10.1. 2D thin-section image characterization with cluster analysis models
10.9 Case Study 10.2: Pore network analysis of shales from FIB-SEM images segmented with a KiU-Net convolutional model
10.10 Summary and conclusions
Nomenclature
Appendix 10.1 Python code snippet for image edge detection
Appendix 10.2 Python code snippet for K-means cluster analysis to quantify 2D porosity from a thin-section image of a reservoir formation
References
Chapter 11. Diverse machine learning applications for coal property characterization of coalbed methane and mining resources
Abstract
11.1 Introduction
11.2 Coal’s fundamental properties
11.3 Coal gross calorific value
11.4 In situ gas content of coal
11.5 Coal mining safety considerations
11.6 Distinguishing coal versus noncoal lithologies
11.7 Coal bed methane production forecasting
11.8 Coal permeability
11.9 Coal seams as potential CO2 storage reservoirs
11.10 Natural gas emissions from subsurface coal deposits
11.11 Surface environmental contamination associated with coal mines
11.12 Coal geomechanical properties and stress conditions
11.13 Coal pore-scale networks, fractal dimensions and gas adsorption isotherms
11.14 Interpolation of spatial value distributions for coal seams
11.15 Case study: applying a geographic quantile regression forest interpolation model to predict the spatial distribution of coal-seam properties
11.16 Summary and conclusions
Nomenclature
Appendix 11.1 Python code snippet for applying the Scipy interpolate griddata function
Appendix 11.2 Python code snippet for generating a semivariogram and applying it for ordinary kriging
References
Index

Life Sciences

Physical Sciences & Engineering

Social Sciences & Humanities

Health

Implementation and Interpretation of Machine and Deep Learning to Applied Subsurface Geological Problems

Prediction Models Exploiting Well-Log Information

Purchase options

World Book Day Celebration!

Institutional subscription on ScienceDirect

David A. Wood

Related books