Computational Geosciences最新文献

Over the past few decades, pore-network models (PNMs) have emerged as a pivotal tool in the investigation of fluid flow within porous media. The crux of PNM lies in the extraction of the topological structure of porous media, as abstracted from geological scans, commonly referred to as the pore network. Conventional methods for pore-network extraction rely on pixel-based techniques and necessitate high-quality images to accurately capture pore information. In recent times, the flashlight search medial axis (FSMA) algorithm has been introduced, offering a novel approach to extract pore networks within continuous spatial domains. This innovation enables the algorithm to operate independently of specific pixels, thereby significantly reducing computational complexity. Building upon the foundational principles of the FSMA algorithm, this paper presents an efficient search algorithm in conjunction with string methods. This algorithm facilitates the precise determination of pore and throat center locations within porous media using a minimal number of computational points and can accurately compute the positions of pore medians. Furthermore, this algorithm can effectively circumvent the issue of dead-end pores encountered in the FSMA algorithm, a feature of paramount importance in the study of multiphase flow within porous media.

The permeability characteristics of fractured rock significantly influence the efficient development of subsurface energy resources. Under subsurface environmental conditions, the evolution of fractured rock permeability is affected by both fluid flow and chemical reactions. Traditional models like the Kozeny-Carman model are no longer suitable for accurately predicting the permeability under hydraulic-chemical coupling. In this study, a two-dimensional fractured rock model is constructed and combined with the lattice Boltzmann method, a pore-scale hydraulic-chemical coupling model is developed to explore the evolution characteristics of the pore structure and permeability of fractured rocks under different conditions. The results show that due to the comprehensive influence of seepage flow and chemical reactions, the distribution characteristics of pore structure in fractured rocks are different during the evolution process, including diffusion-controlled, convection-controlled, and reaction-controlled. Additionally, pore structure variations result in permeability differences among fractured rock with identical porosity. Finally, an efficient model is proposed to predict the porosity–permeability relationships of fractured rocks under hydraulic-chemical coupling. The relative error between the predicted value and the value calculated using the formula falls within the range of ± 15%.

We propose data-driven models based on artificial neural networks (ANN) to predict changes in water-oil relative permeability curves given a salinity reduction in the injection water. The ANN consisted of a multilayer feedforward structure with backpropagation. For validation, a database from a semi-empirical correlation was created, and models with added noise were used to analyze the influence of the data dispersion. Then, a survey of experimental relative permeability curves was performed to produce a real database for sandstone and carbonate rocks, utilized in the training of the final models, with hyperparameter optimization and cross-validation. The initial model was able to consistently reproduce the original correlation, with a mean squared error (MSE) on the order of (10^{-6}). In the noise-trained model, the error measured was lower than the analytical error expected from random dispersion. In models trained with real data, the adopted strategy led to a final training MSE on the order of (10^{-3}), with better performance in networks with two hidden layers. The obtained models are useful in modeling relative permeabilities for low-salinity and engineered water injection projects. Training can be continuously updated with new data, and the methodology can be applied to other properties or even other multivariate regression problems.

We propose a smooth stochastic process for modeling the vertical well path uncertainty. This process describes the accumulation of measurement errors along the well path. We combine the stochastic process with a stochastic model for surfaces into a consistent framework for simultaneous prediction of well paths and surfaces. We show properties of the proposed stochastic process and provide examples of interaction between wells and surfaces.

We propose a nonisothermal reactive multicomponent multiphase flow model for simulating offshore geological carbon dioxide (CO(_2)) storage. The model considers CO(_2) hydration as well as CO(_2) dissolution in water in the low-temperature high-pressure deep-ocean environment. This model comprises a chemistry module responsible for all chemical reactions among different phases and a flow module responsible for mass and energy transfer. The chemistry module is based on the mass action law that considers the inhibition effect of salt on CO(_2) hydration and dissolution. We implement this model in an open-source Matlab-based code MRST-HYD using a sequential iteration approach. The code has been validated through tests against one theoretical solution and one numerical code, Geoxim, developed at IFP Energies nouvelles. Finally, we apply this code to simulate CO(_2) injection into one-dimensional and two-dimensional deep-ocean sediments.

Consistent discretization methods are a natural fit for the novel cut-cell gridding technique for reservoir simulation, which preserves the orthogonality characteristic in the lateral direction. Both uniform (global) and novel hybrid (local) variants of consistent discretization methods are implemented and tested in the vicinity of fault representations on cut-cell grids. Novel consistent discretization methods, which do not require major intrusive changes to the solver structure of industrial-grade reservoir simulators, are investigated in this work. Cell-centered methods such as multi-point flux approximation (MPFA), average multi-point flux approximation (AvgMPFA), and nonlinear two-point flux approximation (NTPFA) methods fit naturally into the framework of existing industrial-grade simulators. Thus, cut-cell compatible variants of AvgMPFA and NTPFA and their novel hybridizations with TPFA are implemented and tested. An implementation of the relatively more computationally expensive MPFA is also made to serve as accuracy reference to AvgMPFA and NTPFA. AvgMPFA and NTPFA multiphase simulation results are compared in terms of accuracy and computational performance against the ones computed with reference MPFA and TPFA methods on a set of synthetic cut-cell grid models of varying complexity including conceptual models and a field-scale model. It is observed that AvgMPFA consistently yields more accurate and computationally efficient simulations than NTPFA on cut-cell grids. Moreover, AvgMPFA-TPFA hybrids run faster than NTPFA-TPFA hybrids when compared on the same problem for the same hybridization strategy. On the other hand, the computational performance of AvgMPFA degrades more rapidly compared to NTPFA with increasing “rings” of orthogonal blocks around cut-cells. Auspiciously, only one or two “rings” of orthogonal blocks around cut cells are sufficient for AvgMPFA to deliver high accuracy.

The study of thin sections provides crucial information about the structure of sedimentary rocks. Different properties, such as mineral composition, texture, grain morphology, presence of clay minerals, and porosity level, can be derived from thin section analysis. These features directly determine the quality of crude reservoirs. In this context, manual grain identification from petrographic thin sections usually demands considerable time and effort, so machine learning and image processing techniques have become more frequent in the last few years. Obtaining large and reliable labeled data sets for supervised learning workflows is a complex and critical process. We devise a completely unsupervised approach for granulometric classification using thin section images. The introduced workflow first pre-processes the thin section image by denoising and dividing it into different image patches. In the second stage, the image patches are used to train an unsupervised convolutional neural network. Then, the trained network segments the grains in each patch of the pre-processed image. The training strategy uses transfer learning to guarantee the same initialization parameters of the neural network while processing the image patches. Next, a watershed transform is applied to recover the borders of the segmented grains. Finally, a granulometric calculation and classification process is performed by considering the grain contours restored through the implemented methodology. The results obtained with the proposed algorithm are concordant with those obtained from the analysis of sieved thin sections derived from controlled experiments in the laboratory.

Clarifying methane’s adsorption and diffusion properties in kerogen contributes to efficiently exploiting shale gas reservoirs. We refined the kerogen III-series model to construct the kerogen III-B and kerogen III-C molecular structures. In contrast to the traditional simplified slit model containing small organic matter, a mixed kerogen-quartz slit model was further proposed. These inorganic-organic models, including kerogen II-D, kerogen III-B, and kerogen III-C, provide a realistic reservoir environment for the study of the shale gas adsorption and diffusion characteristics in shale. Based on these models, we investigated the competitive adsorption of methane and carbon dioxide using the grand canonical Monte Carlo (GCMC) method. We then studied the diffusion characteristics of methane molecules throughout the model area and in the different adsorption blocks classified as the inner slit zone, surface zone, and matrix zone using the molecular dynamics (MD) method. The results showed that carbon dioxide gradually replaces methane molecules as the injection pressure of carbon dioxide increases, causing desorption and diffusion of methane. The order of the overall diffusion capability of methane in the kerogen slit models is kerogen II-D >kerogen III-C >kerogen III-B. In addition, the diffusion capability of methane molecules in the different zones is ordered as inner slit zone >surface zone >matrix zone. This work is a step towards more effective exploitation of shale gas.

Results of an application of a dispersive wave hydro-sediment-morphodynamic model in the western circulation cell of the Ria Formosa lagoon located in the Algarve region of the southern Portugal are presented. This area of interest has a couple of features that complicate the application of the dispersive wave model: (1) the area has a complex irregular geometry with a number of barrier islands that separate the lagoon from the Atlantic Ocean, artificial and naturally occurring tidal inlets, and a number of curling channels inside the lagoon that interconnect the inlets and serve as waterways between the lagoon settlements; (2) the tidal range in the area can reach up to 3.5 m; therefore, the terrain inside the lagoon is characterized by vast salt marshes and tidal flats, and the wetting-drying process is a key component of any hydrodynamic simulation in this area. A model representation of the area has been developed by generating an unstructured finite element mesh of the circulation cell, and collecting data on parameters that characterize the tidal waves in the area, and bottom friction and sediment transport models used in the simulations. The results of the simulations indicate that the dispersive wave model can be applied in coastal areas with nontrivial underlying physical processes, and complex irregular geometries. Moreover, the dispersive term of the model is capable of capturing additional flow characteristics that are otherwise not present in hydrodynamic simulations that involve the nonlinear shallow water equations; and these additional flow features can, in their turn, affect the resulting sediment transport and bed morphodynamic process simulations.

Being a fully Lagrangian particle method, the material point method (MPM) discretizes a material domain by using a collection of material points. The momentum equations in MPM are solved on a predefined regular background grid, so that the grid distortion and entanglement are completely avoided in MPM. The contact algorithm of MPM is developed via the background grid and the impenetrability condition between bodies. The contact algorithm of MPM is applied to solve some impact and perforation problems. This study concerns the validation of the contact algorithm of MPM. Solutions from MPM with the contact algorithm are compared to those from the finite element method (FEM) with the penalty method. For two impact problems, the results from MPM with the contact algorithm are in good agreement with those obtained with the FEM penalty method. For the perforation problem of aluminum plate, the results obtained using MPM with the contact algorithm are better than those from the FEM penalty method. We think that for impact problems without extreme large deformations, it is better to use the FEM penalty method. For impact problems with extreme large deformations, it is better to use the contact algorithm of MPM.