Machine Learning for Subsurface Characterization

Front Cover
Machine Learning for Subsurface Characterization
Copyright
Dedication
Contents
Contributors
Preface
	Machine learning-driven success stories
	Challenges and precautions when using machine learning
	Recommendations for harnessing the power of machine learning
	Concluding remarks
	References
	Further reading
Acknowledgment
Chapter 1: Unsupervised outlier detection techniques for well logs and geophysical data
	1. Introduction
		1.1. Basic terminologies in machine learning and data-driven models
		1.2. Types of machine learning techniques
		1.3. Types of outliers
	2. Outlier detection techniques
	3. Unsupervised outlier detection techniques
		3.1. Isolation forest
		3.2. One-class SVM
		3.3. DBSCAN
		3.4. Local outlier factor
		3.5. Influence of hyperparameters on the unsupervised ODTs
	4. Comparative study of unsupervised outlier detection methods on well logs
		4.1. Description of the dataset used for the comparative study of unsupervised ODTs
		4.2. Data preprocessing
			4.2.1. Feature transformation: Convert R to log(R)
			4.2.2. Feature scaling: Use of robust scaler
		4.3. Validation dataset
			4.3.1. Dataset #1: Containing noisy measurements
			4.3.2. Dataset #2: Containing bad holes
			4.3.3. Dataset #3: Containing shaly layers and bad holes with noisy measurements
			4.3.4. Dataset #4: Containing manually labeled outliers
		4.4. Metrics/scores for the assessment of the performances of unsupervised ODTs on the conventional logs
			4.4.1. Recall
			4.4.2. Specificity
			4.4.3. Balanced accuracy score
			4.4.4. Precision
			4.4.5. F1 score
			4.4.6. Receiver operating characteristics (ROC) curve and ROC-AUC score
			4.4.7. Precision-recall (PR) curve and PR-AUC score
	5. Performance of unsupervised ODTs on the four validation datasets
		5.1. Performance on Dataset #1 containing noisy measurements
		5.2. Performance on Dataset #2 containing measurements affected by bad holes
		5.3. Performance on Dataset #3 containing shaly layers and bad holes with noisy measurements
		5.4. Performance on Dataset #4 containing manually labeled outliers
	6. Conclusions
	Appendix A. Popular methods for outlier detection
	Appendix B. Confusion matrix to quantify the inlier and outlier detections by the unsupervised ODTs
	Appendix C. Values of important hyperparameters of the unsupervised ODT models
	Appendix D. Receiver operating characteristics (ROC) and precision-recall (PR) curves for various unsupervised ODTs on th ...
	Acknowledgments
	References
Chapter 2: Unsupervised clustering methods for noninvasive characterization of fracture-induced geomechanical alterations
	1. Introduction
	2. Objective of this study
	3. Laboratory setup and measurements
	4. Clustering methods for the proposed noninvasive visualization of geomechanical alterations
		4.1. K-means clustering
		4.2. Agglomerative clustering
		4.3. DBSCAN
	5. Features/attributes for the proposed noninvasive visualization of geomechanical alteration
		5.1. Feature engineering
		5.2. Dimensionality reduction
	6. Results and discussions
		6.1. Effect of feature engineering
		6.2. Effect of clustering method
		6.3. Effect of dimensionality reduction
		6.4. Effect of using features derived from both prefracture and postfracture waveforms
	7. Physical basis of the fracture-induced geomechanical alteration index
	8. Conclusions
	Acknowledgments
	Declarations
	References
Chapter 3: Shallow neural networks and classification methods for approximating the subsurface in situ fluid-filled pore  ...
	1. Introduction
	2. Methodology
		2.1. Hydrocarbon-bearing shale system
		2.2. Petrophysical basis for the proposed data-driven log synthesis
		2.3. Data preparation and statistical information
		2.4. Categorization of depths using flags
		2.5. Fitting the T2 distribution with a bimodal Gaussian distribution
		2.6. Min-max scaling of the dataset (features and target)
		2.7. Training and testing methodology for the ANN models
	3. ANN model training, testing, and deployment
		3.1. ANN models
		3.2. Training the first ANN model
		3.3. Testing the first ANN model
		3.4. Training the second ANN model
		3.5. Testing the second ANN model
		3.6. Petrophysical validation of the first ANN model
		3.7. ANN-based predictions of NMR T2 distribution for various depth intervals
	4. Conclusions
	Appendix A. Statistical properties of conventional logs and inversion-derived logs for various depth intervals
	Appendix B. Categorization of depths using flags (categorical features)
	Appendix C. Importance of the 12 conventional logs and 10 inversion-derived logs
	Appendix D. Estimations of specific reservoir parameters from NMR T2 distributions
	References
Chapter 4: Stacked neural network architecture to model the multifrequency conductivity/permittivity responses of subsurf ...
	1. Introduction
	2. Method
		2.1. Data preparation
		2.2. Methodology for the dielectric dispersion (DD) log synthesis
		2.3. Evaluation metric/measure for log-synthesis model
		2.4. Data preprocessing
		2.5. ANN models for dielectric dispersion log generation
	3. Results
		3.1. Sensitivity analysis
		3.2. Generalization capability of the DD log synthesis using the SNN model
		3.3. Petrophysical and statistical controls on the DD log synthesis using the SNN model
	4. Conclusions
	References
Chapter 5: Robust geomechanical characterization by analyzing the performance of shallow-learning regression methods usin ...
	1. Introduction
	2. Methodology
		2.1. Data preparation
		2.2. Data preprocessing
		2.3. Metric to evaluate the log-synthesis performance of the shallow-learning regression models
		2.4. Shallow-learning regression models
			2.4.1. Ordinary least squares (OLS) model
			2.4.2. Partial least squares (PLS) model
			2.4.3. Least absolute shrinkage and selection operator (LASSO) model
			2.4.4. ElasticNet model
			2.4.5. Multivariate adaptive regression splines (MARS) model
			2.4.6. Artificial neural network (ANN) model
		2.5. Clustering techniques
			2.5.1. K-means clustering
			2.5.2. Gaussian mixture model clustering
			2.5.3. Hierarchical clustering
			2.5.4. DBSCAN clustering
			2.5.5. SOM followed by K-means clustering
	3. Results
		3.1. Prediction performances of shallow-learning regression models
		3.2. Comparison of prediction performances of shallow-learning regression models in Well 1
		3.3. Performance of clustering-based reliability of sonic-log synthesis
	4. Conclusions
	References
Chapter 6: Index construction, dimensionality reduction, and clustering techniques for the identification of flow units i ...
	1. Introduction
		1.1. Geology of the shale formation
		1.2. Literature survey
		1.3. Objectives
	2. Properties influencing EOR efficiency of light-hydrocarbon injection
	3. Methodology to generate the ranking (R) index
		3.1. Description of the R-index
		3.2. Calculation of the R-index
	4. Methodology to generate the microscopic displacement (MD) index
		4.1. Description of the MD-index
		4.2. Calculation of the MD-index
			4.2.1. Step 1: NMR decomposition using factor analysis
			4.2.2. Step 2: Determination of pore volumes of various fluid phases represented by the decomposed factors
			4.2.3. Step 3: Correction of miscible, free-oil volume for pore confinement effect
			4.2.4. Step 4: Compute the MD-index
	5. Methodology to generate the K-means clustering (KC) index
		5.1. Description of the KC-index
		5.2. Calculation of the KC-index
	6. Limitations
	7. Conclusions
	References
Chapter 7: Deep neural network architectures to approximate the fluid-filled pore size distributions of subsurface geolog ...
	1. Introduction
		1.1. Log-based subsurface characterization
		1.2. Deep learning
		1.3. NMR logging
	2. Introduction to nuclear magnetic resonance (NMR) measurements
		2.1. NMR relaxation measurements
		2.2. Relationships between NMR T2 distribution and conventional logs
	3. Data acquisition and preprocessing
		3.1. Data used in this chapter
		3.2. Data preparation
	4. Neural network architectures for the NMR T2 synthesis
		4.1. Introduction to NMR T2 synthesis
		4.2. VAE-NN architecture, training, and testing
		4.3. GAN-NN architecture, training, and testing
		4.4. VAEc-NN architecture, training, and testing
		4.5. LSTM architecture, training, and testing
		4.6. Training and testing the four deep neural network models
	5. Application of the VAE-NN model
	6. Application of the GAN-NN model
	7. Application of the VAEc-NN model
	8. Application of the LSTM network model
	9. Conclusions
	References
Chapter 8: Comparative study of shallow and deep machine learning models for synthesizing in situ NMR T2 distributions
	1. Introduction
	2. Dataset
	3. Shallow-learning models
		3.1. Ordinary least squares
		3.2. Least absolute shrinkage and selection operator
		3.3. ElasticNet
		3.4. Support vector regressor
		3.5. k-Nearest neighbor regressor
		3.6. Artificial neural network
		3.7. Comparisons of the test accuracy and computational time of the shallow-learning models
	4. Deep learning models
		4.1. Variational autoencoder assisted neural network
		4.2. Generative adversarial network assisted neural network
		4.3. Variational autoencoder with convolutional layer assisted neural network
		4.4. Encoder-decoder long short-term memory network
		4.5. Comparisons of the accuracy and computational time of the deep learning models
		4.6. Cross validation
	5. Comparison of the performances of the deep and shallow models
	6. Discussions
	7. Conclusions
	References
Chapter 9: Noninvasive fracture characterization based on the classification of sonic wave travel times
	1. Introduction
		1.1. Mechanical discontinuities
		1.2. Characterization of discontinuities
		1.3. Machine learning for characterization of discontinuities
	2. Objective
		2.1. Assumptions and limitations of the proposed data-driven fracture characterization method
		2.2. Significance and relevance of the proposed data-driven fracture characterization method
	3. Fast-marching method (FMM)
		3.1. Introduction
		3.2. Validation
			3.2.1. Numerical model of homogeneous material at various spatial discretization
			3.2.2. Material with large contrasts in compressional wave velocity
			3.2.3. Material with parallel discontinuities
			3.2.4. Material with smoothly varying velocity distribution
		3.3. Fast-marching simulation of compressional wavefront travel time for materials containing discontinuities
	4. Methodology for developing data-driven model for the noninvasive characterization of static mechanical discontinuities ...
		4.1. Classification methods implemented for the proposed fracture characterization workflow
			4.1.1. K-nearest neighbors (KNN) classifier
			4.1.2. Support vector machine (SVM) classifier
			4.1.3. Decision tree (DT) classifier
			4.1.4. Random forest (RF) classifier
			4.1.5. AdaBoost classifier
			4.1.6. Naïve Bayes (NB) classifier
			4.1.7. Artificial neural network (ANN) classifier
			4.1.8. Voting classifier
	5. Results for the classification-based noninvasive characterization of static mechanical discontinuities in materials
		5.1. Characterization of material containing static discontinuities of various dispersions around the primary orientation
			5.1.1. Background
			5.1.2. Model accuracy
		5.2. Characterization of material containing static discontinuities of various primary orientations
			5.2.1. Background
			5.2.2. Model accuracy
			5.2.3. Sensor/receiver importance
		5.3. Characterization of material containing static discontinuities of various spatial distributions
			5.3.1. Background
			5.3.2. Model accuracy
			5.3.3. Sensor importance
	Acknowledgments
	References
	Further reading
Chapter 10: Machine learning assisted segmentation of scanning electron microscopy images of organic-rich shales with fea ...
	1. Introduction
	2. Method
		2.1. SEM images
		2.2. Workflow
		2.3. Feature extraction
			2.3.1. Gaussian blur (one feature)
			2.3.2. Difference of Gaussians (one feature)
			2.3.3. Sobel operator (one feature)
			2.3.4. Hessian matrix (three features)
			2.3.5. Wavelet transforms (six features)
			2.3.6. Local information (three features)
			2.3.7. Other features investigated for this study
		2.4. Training dataset
		2.5. Classification of the feature vectors
	3. Results and discussions
		3.1. Four-component segmentation
		3.2. Multilabel probability-based segmentation
		3.3. Performance on testing dataset
		3.4. Feature ranking using permutation importance
		3.5. Deployment of the segmentation method on new, unseen SEM images
	4. Conclusions
	Acknowledgments
	Conflict of Interest
	References
Chapter 11: Generalization of machine learning assisted segmentation of scanning electron microscopy images of organic-ri ...
	1. Introduction
	2. Method
		2.1. Image preprocessing
		2.2. Segmentation performance evaluation
		2.3. Training dataset
		2.4. Hyperparameter optimization
		2.5. Testing dataset
	3. Results and discussion
		3.1. Task 1: Training and testing on Map-1
		3.2. Task 2: Training on Map-1 and testing on Map-2
		3.3. Task 3: Training and testing on Map-2
		3.4. Task 4: Training on Map-2 and testing on Map-1
		3.5. Task 5: Training on both Map-1 and Map-2 and testing on Map-1
		3.6. Task 6: Training on both Map-1 and Map-2 and testing on Map-2
		3.7. Feature ranking
	4. Conclusions
	Acknowledgments
	Declarations
	References
Chapter 12: Characterization of subsurface hydrocarbon/water saturation by processing subsurface electromagnetic logs usi ...
	1. Introduction
		1.1. Error-minimization methods
		1.2. Subsurface characterization problem
	2. Method
		2.1. Relevant EM logging tools
		2.2. Relevant log interpretation models
	2.3. Proposed deterministic inversion of EM logs
		2.3.1. Introduction
		2.3.2. Modified Levenberg-Marquardt nonlinear inversion
	2.4. Limitations and assumptions in the proposed multifrequency-EM log processing using modified Levenberg-Marquardt algo ...
	3. Results and discussion
		3.1. Sensitivity of the integrated WS, SMD, and CRI model
		3.2. Application of the proposed log processing to synthetic data
		3.3. Application of the proposed log processing to the 9 EM logs acquired in the upper Wolfcamp formation
		3.4. Petrophysical interpretation of the inversion-derived estimates in the upper Wolfcamp formation
		3.5. Error analysis for the inversion-derived estimates in the upper Wolfcamp formation
	4. Conclusions
	References
Chapter 13: Characterization of subsurface hydrocarbon/water saturation using Markov-chain Monte Carlo stochastic inversi ...
	1. Introduction
		1.1. Saturation estimation in shale gas reservoir
		1.2. Electromagnetic (EM) logging tools
		1.3. Conventional EM-log-interpretation models
		1.4. Estimation of water/hydrocarbon saturation based on the inversion of electromagnetic (EM) logs
	2. Method
		2.1. Mechanistic clay-pyrite interfacial-polarization model
		2.2. Proposed inversion-based interpretation method
			2.2.1. Bayesian formulation of the parameter estimation problem
				Bayesian framework:
				Prior models:
				Likelihood function:
			2.2.2. Metropolis-Hastings sampling algorithms
				Proposal distribution:
				Acceptance probability:
			2.2.3. Convergence monitoring
		2.3. Limitations and assumptions in the proposed inversion-based interpretation method
	3. Results and discussion
		3.1. Application of the MCMC-based stochastic inversion to synthetic layers
		3.2. Application of the MCMC-based stochastic inversion to process broadband EM dispersion logs acquired in a shale gas f ...
		3.3. Petrophysical interpretation and log analysis of the inversion-derived estimates in the shale gas formation
	4. Conclusions
	References
Index
Back Cover