|Theme||Visible||Selectable||Appearance||Zoom Range (now: 0)|
Abstract With the recent tremendous development in algorithms, computations power and availability of the enormous amount of data, the implementation of machine learning approach has spurred the interest in oil and gas industry and brings the data science and analytics into the forefront of our future energy. The idea of using automated algorithms to determine the rock facies is not new. However, the recent advancement in machine learning methods encourages to further research and revisit the supervised classification tasks, discuss the methodological limits and further improve machine learning approach and classification algorithms in rock facies classification from well-logging measurements. This paper demonstrates training different machine learning algorithms to classify and predict the geological facies using well logs data. Previous and recent research was done using supervised learning to predict the geological facies. This paper compares the results from the supervised learning algorithms, unsupervised learning algorithms as well as a neural network machine learning algorithm. We further propose an integrated approach to dataset processing and feature selection. The well logs data used in this paper are for wells in the Anadarko Basin, Kansas. The dataset is divided into training, testing and evaluating wells used for testing the model. The objective is to evaluate the algorithms and limitations of each algorithm. We speculate that a simple supervised learning algorithm can yield score higher than neural network algorithm depending on the model parameter selected. Analysis for the parameter selection was done for all the models, and the optimum parameter was used for the corresponding classifier. Our proposed neural network algorithm results score slightly higher than the supervised learning classifiers when evaluated with the cross-validation test data. It is concluded that it is important to calculate the accuracy within the adjacent layers as there are no definite boundaries between the layers. Our results indicate that calculating the accuracy of prediction with taking account the adjacent layers, yield higher accuracy than calculating accuracy within each point. The proposed feed-forward neural network classifier trains using backpropagation (gradient descent) provides accuracy within adjacent layers of 88%. Our integrated approach of data processing along with the neural network classifier provides more satisfactory results for the classification and prediction problem. Our finding indicates that utilizing simple supervised learning with an optimum model parameter yield comparable scores as a complex neural network classifier.
Chawshin, Kurdistan (Norwegian University of ScienceTechnology) | Berg, Carl Fredrik (Norwegian University of ScienceTechnology) | Varagnolo, Damiano (Norwegian University of ScienceTechnology) | Lopez, Olivier (Equinor ASA)
Abstract CT scan images provide valuable three-dimensional information on the mineralogical composition and overall internal structure of cores. X-ray computerized tomography (CT) imaging of whole cores has therefore become a routine step in core analysis workflows. This new data type gives new possibilities in reservoir characterization. Lithological classification of reservoir rocks, in its turn, is an essential step to better understand the depositional environment and for subsequent effective reservoir characterization: the chemical composition of the minerals, combined with their grain size, sorting and pore size distribution is known to highly affect the transport properties of reservoir rocks. Lithological classification on the extracted whole core material is thus consequential; however, it also requires significant investments, being traditionally conducted through visual inspection performed by expert geologists. This manual process is time consuming, and prone to subjective interpretations and human errors. Therefore, a current research and development trend is to find automated methods for computer-assisting the assessment of this type of data, eventually reducing time and costs of core analysis, and improving the overall business decisions. In this study we explore the application of Convolutional Neural Networks (CNN) to automatically classify lithofacies. We propose a workflow for high resolution lithofacies classification using whole core three-dimensional CT images, and we assess the validity of our approach on a field-example from the Norwegian continental shelf. The novelty of our approach is thus learning, through a CNN, the relationship between convolution-derived three-dimensional features and expert-derived lithofacies classes. We thus extend approaches working on two-dimensional images into a workflow that uses high-resolution three-dimensional CT images as direct input. In our work the training data set includes information obtained from manual core description. Prior to training, the three-dimensional CT images are pre-processed so that undesired artefacts are automatically flagged and removed before being fed into the network. The approach is validated using the trained CNN classifier to predict lithofacies in a set of unseen three-dimensional CT data. The trained model can predict lithofacies classes with high accuracy, with a misclassification rate of about 3%. We found that these misclassifications are mainly associated with the presence of high-density material such as pyrite nodules and drilling mud invasions. Dipping fractures and missing values, not completely removed by image pre-processing, are additional reasons for model deficiency in some of the incorrectly classified images. Overall, the trained classifier exhibits higher pixel-wise precision and captures the high-resolution heterogeneities more accurately compared to the manual core descriptions.
Formation evaluation and production design is often challenging in organic-rich mudrocks due to complexities in petrophysical and compositional properties, and post-depositional hydrocarbon generating mechanisms such as thermal maturation over time. Petrophysical parameters such as porosity, permeability and fluid saturations are important, but not sufficient to fully characterize organic-rich mudrocks. Integration of petrophysical, geochemical and geomechanical data is therefore required for a reliable rock classification in source rocks. This paper focuses on integrated rock classification in the Eagle Ford Shale in South Texas, consisting of organic-rich fossiliferous marine shale deposited in Late Cretaceous.
We first performed joint inversion of triple-combo, spectral gamma ray and elemental capture spectroscopy (ECS) logs to estimate depth-by-depth volumetric concentration of minerals, porosity, and fluid saturations. In the absence of acoustic measurements, concentrations and shape (i.e., aspect ratio) of minerals were used as inputs to the Self-consistent Approximation (SCA) model, to estimate depth-by-depth effective elastic properties such as Young's Modulus (YM) and Poisson's Ratio (PR). We then classified the rocks based on geologic texture and geochemical properties, as well as well-log based estimates of petrophysical and geomechanical parameters.
We successfully applied a well-log based rock classification to two wells located in the oil window of Eagle Ford formation. Well no. 1 produced an additional 20% of hydrocarbons in the first 90-day of its production. Through the analysis of the results, we observed similar petrophysical properties and organic content of the reservoir quality classes in both wells. However, we noticed differences in estimates of elastic parameters such as Young's Modulus and Poisson's Ratio between the two wells. For the interbedded wackestone-limestone facies, YM average estimate in well no. 1 was approximately 10% higher than well no. 2, which can be the reason for the difference in their production.
Chawshin, Kurdistan (Norwegian University of Science and Technology) | Gonzalez, Andres (The University of Texas at Austin) | Berg, Carl F. (Norwegian University of Science and Technology) | Varagnolo, Damiano (Norwegian University of Science and Technology) | Heidari, Zoya (The University of Texas at Austin) | Lopez, Olivier (Equinor ASA)
Summary X-ray computerized tomography (CT) is a nondestructive method of providing information about the internal composition and structure of whole core reservoir samples. In this study we propose a method to classify lithology. The novelty of this method is that it uses statistical and textural information extracted from whole core CT images in a supervised learning environment. In the proposed approaches, first-order statistical features and textural grey-levelco-occurrence matrix (GLCM) features are extracted from whole core CT images. Here, two workflows are considered. In the first workflow, the extracted features are used to train a support vector machine (SVM) to classify lithofacies. In the second workflow, a principal component analysis (PCA) step is added before training with two purposes: first, to eliminate collinearity among the features and second, to investigate the amount of information needed to differentiate the analyzed images. Before extracting the statistical features, the images are preprocessed and decomposed using Haar mother wavelet decomposition schemes to enhance the texture and to acquire a set of detail images that are then used to compute the statistical features. The training data set includes lithological information obtained from core description. The approach is validated using the trained SVM and hybrid (PCA + SVM) classifiers to predict lithofacies in a set of unseen data. The obtained results show that the SVM classifier can predict some of the lithofacies with high accuracy (up to 91% recall), but it misclassifies, to some extent, similar lithofacies with similar grain size, texture, and transport properties. The SVM classifier captures the heterogeneity in the whole core CT images more accurately compared with the core description, indicating that the CT images provide additional high-resolution information not observed by manual core description. Further, the obtained prediction results add information on the similarity of the lithofacies classes. The prediction results using the hybrid classifier are worse than the SVM classifier, indicating that low-power components may contain information that is required to differentiate among various lithofacies.
Abstract Wrong manual interpretation from the log data about the formation type and other important information can be catastrophic for the company-operator. With Machine-Learning (ML) (a branch of Artificial Intelligence) algorithms, the interpretation of formation type from the log data has been addressed. As a result, we have successfully developed a program able to accurately predict the type of formation. Using the conventional Machine Learning technique of splitting the data into training, validation and test sets, we tried six different ML algorithms to fit with the training part of the data and then verify their prediction accuracy with cross-validation scores and cross-validation predictions which tests the performance of the classifiers (ML algorithms) on the validation set. The three best performing classifiers were selected and further improved by a search of classifier's best hyperparameters. These improved classifiers are further tested on unseen data to produce a comparative analysis. Our prediction accuracy with Receiver Operating Characteristic (ROC) scores and ROC-Area Under-the-Curve (ROC-AUC) for each type of formation from the log data lies in the range of 95-99%, except for formations such as shaly sandstone and shale (50% and 84% respectively). The reason for this seemed to be under-fitting i.e., during the training, the classifiers did not see enough instances of these types of formation to know exactly what characteristics of the data make the type of formation to be shaly sandstone or shale. The issue of under-fitting was verified by skimming through the data. To resolve this problem, we suggest training classifiers with a larger data with more targets (types of formation). Furthermore, during the data cleaning (prior to classifier training) and data analysis phases we have discovered important relationships between well logs and defined relative importance of each well log for different formations. This observation can be investigated further to help eliminate the use of multiple well logs while dealing with some formations (based on prior geological knowledge) and reduce the cost of the well logging operations. Using our program with a larger well log data consisting of more formation type instances, we can train the classifiers to accurately predict the formation type irrespectively of differences in formation type. Our program is dynamic in the sense that with different targets, i.e., type of formation fluid instead of type of formation or both together, it can successfully predict either or both targets. Increasing the numbers of data instances resulted in a better training and thus, more accurate predictions. Utilization of the program will make the formation-evaluation process easier, faster, automated and more-precise.