Permeability is a critical parameter for understanding subsurface fluid flow behavior, managing reservoirs, enhancing hydrocarbon recovery, and sequestering carbon dioxide. In general, permeability is measured in the laboratory based on subsurface core samples, calculated from well logs or estimated from well tests. However, laboratory measurements and well tests are expensive, time-consuming, and usually limited to a few core samples or wells in a hydrocarbon field or carbon storage site. Machine-learning techniques are good options for generating a rapid, robust, and cost-effective permeability prediction model because of their strengths to recognize the potential interrelationships between input and output variables. Convolutional neural networks (CNN), as a good pattern recognition algorithm, are widely used in image processing, natural language processing, and speech recognition, but are rarely used with regression problems and even less often in reservoir characterization. We have developed a CNN regression model to estimate the permeability in the Jacksonburg-Stringtown oil field, West Virginia, which is a potential carbon storage site and enhanced oil recovery operations field. We also evaluate the concept of the geologic feature image, which is converted from geophysical well logs. Five variables, including two commonly available conventional well logs (the gamma rays [GRs] and bulk density) and three well-log-derived variables (the slopes of the GR and bulk density curves, and shale content), are used to generate a geologic feature image. The CNN treats the geologic feature image as the input and the permeability as the desired output. In addition, the permeability predicted using traditional backpropagation artificial neural networks, which are optimized by genetic algorithms and particle swarm optimization, is compared with the permeability estimated using our CNN. Our results indicate that the CNN regression model provides more accurate permeability predictions than the traditional neural network.

Porosity is a fundamental property that characterizes the storage capability of fluid and gas-bearing formations in a reservoir. An accurate porosity value can be measured from core samples in the laboratory, however core analysis is expensive and time-consuming. Well log data can be used to calculate porosity, but the availability of log suites is often limited in mature fields. Therefore, robust porosity prediction requires integration of core-measured porosity with available well log suites to control for changes in lithology and fluid content. A support vector machine (SVM) model with mixed kernel function (MKF) is used to construct the relationship between limited conventional well log suites and sparse core data. Porosity is the desired output and two conventional well log responses (gamma ray and bulk density) and three well log derived parameters (the slope of gamma ray, the slope of density, and V_sh) are input parameters. A global stochastic searching algorithm, particle swarm optimization (PSO), is applied to improve the efficiency of locating the appropriate values of five control parameters in MKF-SVM model. The results of SVM with different traditional kernel functions were compared, and the MKF-SVM model provided an improvement over the traditional SVM model. To confirm the advantage of the hybrid PSO-MKF-SVM model, the results from three models: 1) radial basis function based least square support vector machine, 2) multilayer perceptron artificial neural network, and 3) radial basis function artificial neural network, are compared with the result of the hybrid PSO-MKF-SVM model. The results indicate that the hybrid PSO-MKF-SVM model improves porosity prediction with the highest correlation coefficient (γ of 0.9560), the highest coefficient of determination (R² of 0.9140), lowest root mean square error (RMSE of 1.6505), average absolute error value (AAE of 1.4050) and maximum absolute error (MAE of 2.717).

Within mudrock reservoirs, brittle zones undergo failure during hydraulic stimulation, creating numerous artificial fractures which enable hydrocarbons to be liberated from the reservoir. Natural fractures in mudrock reduce the tensile strength of the host rock, creating planes of weaknesses that are hypothesized to be reactivated during hydraulic stimulation. Combined, brittleness and natural fractures contribute to creating more abundant and complex fracture networks during hydraulic stimulation. Research efforts toward quantifying rock brittleness have resulted in numerous mineral-/compositional-based indices, which are used during petrophysical analysis to predict zones most conducive to hydraulic stimulation. In contrast, investigations on the relationship between chemical composition and core-scale natural fractures are limited. For this study, we collected high-resolution energy-dispersive X-ray fluorescence (XRF) data, calibrated with a wave-dispersive XRF, from a Marcellus Shale core. Additionally, we characterized corescale natural fractures in terms of length, width, in-filling material or lack thereof, and orientation. Following the characterization, we transformed the natural fracture data into a continuous P10 (lineal fracture intensity) curve, expressed as the number of fractures per a one-half foot window. Using these data sets, we investigated the relationship between rock composition and natural fracture intensity. Regression analyses recorded positive relationships between natural fracture intensity and calcium, silicon/aluminum, and total organic carbon (TOC), and negative relationships with silicon and aluminum. Aluminum recorded the strongest (negative) relationship ([ ]) with natural fracture intensity. To access the degree to which natural fractures can be predicted based on chemical composition, we applied a partial least-squares analysis, a multivariate method, and recorded an [ ]. Our study illustrates that although numerous factors are responsible for natural fracture genesis, such fractures predictively concentrate in areas of similar chemical composition, largely in zones with low aluminum concentrations.

This study utilizes machine learning to quantify CO₂ plume extents by analyzing microseismic data from the Illinois Basin Decatur Project (IBDP). Leveraging a unique dataset of well logs, microseismic records, and CO₂ injection metrics, this work aims to predict the temporal evolution of subsurface CO₂ saturation plumes. The findings illustrate that machine learning can predict plume dynamics, revealing vertical clustering of microseismic events over distinct time periods within certain proximities to the injection well, consistent with an invasion percolation model. The buoyant CO₂ plume partially trapped within sandstone intervals periodically breaches localized barriers or baffles, which act as leaky seals and impede vertical migration until buoyancy overcomes gravity and capillary forces, leading to breakthroughs along vertical zones of weakness. Between different unsupervised clustering techniques, K-Means and DBSCAN were applied and analyzed in detail, where K-means outperformed DBSCAN in this specific study by indicating the combination of the highest Silhouette Score and the lowest Davies–Bouldin Index. The predictive capability of machine learning models in quantifying CO₂ saturation plume extension is significant for real-time monitoring and management of CO₂ sequestration sites. The models exhibit high accuracy, validated against physical models and injection data from the IBDP, reinforcing the viability of CO₂ geological sequestration as a climate change mitigation strategy and enhancing advanced tools for safe management of these operations.