Author: João Lucas Braga Da Silva

Surrogate models are techniques to approximate the objective functions of expensive optimization problems. Recently, Random Forests have been studied as a surrogate model technique for combinatorial optimization problems. Nonetheless, Random Forests contain several hyper-parameters that are used to control the prediction process. Despite their importance, research on the effects of these hyper-parameters is scarce. Therefore, this paper performs a systematic investigation of the effects of different combinations of values for the Random Forest hyper-parameters on the approximation of well-known multi-objective combinatorial benchmark problems. The results show that the number of samples to consider when building each tree and the minimum number of samples to be at the leaf node are the two most important hyper-parameters in this context.

Three-dimensional seismic data provides valuable insights into structural interpretation when dealing with naturally fractured reservoirs. The faults or fractures may control the direction of fluid flow in the reservoirs, especially during the sensitive analysis in projects of carbon sequestration and reservoir development, since these features may impact the rock-fluid interaction and oil field production. However, characterizing faults and fractures in pre-salt reservoirs using conventional approaches is challenging due to a complex and heterogeneous overlying evaporitic unit which reduces the seismic-to-noise ratio. Therefore, seismic data conditioning targeted at improving the seismic image is necessary for quality subsurface investigations. This work provides a systematic series of filters to enhance and detect discontinuities in post-stack seismic volume, primarily by reducing migration artifacts and salt multiples, thereby maintaining the true geological characteristics of the data. We emphasize the delineation of the fault and potential fracture zones by computing seismic attributes in a post-stack time-migrated 3D seismic in the Santos Basin’s deep waters. Initially, the seismic signal-to-noise ratio was analyzed, and a sequence of structural-oriented filters and spectral balancing was applied to generate vertically refined images with enhanced seismic amplitudes. To assess the quality of the conditioned data, we employed the Similarity and Curvature attributes to identify the distribution of discontinuities and compare the before-and-after denoising approach. The data conditioning and analysis of seismic attributes improved our understanding of the distribution and compartmentalization of the reservoirs, highlighting predominantly NE-trending discontinuities. The systematic processing significantly improved the signal-to-noise ratio, allowing us to identify the relative direction of structural elements and insights into potential fractured zones over the reservoir.

Machine learning techniques have been widely used in the oil and gas industry to improve the qualitative and quantitative characterization of subsurface reservoirs. Because rock properties are strongly influenced by lithological and sedimentological information, lithofacies classification is an important step in 3D reservoir modeling. The aim of this study is to use supervised classification algorithms to predict the spatial distribution pattern of lithofacies classes using borehole and seismic data. In this study, lithofacies classes are distributed away from the wells using a machine-learning classifier. Seismic data attributes extracted from well locations are utilized as training data features in various supervised classification algorithms. Machine learning classifiers trained and evaluated for lithofacies classification include K-nearest neighbors, support vector machine, Gaussian naive Bayes, decision tree, Gradient Boosting, and Random Forests. A number of parameters are optimally determined in order to achieve the highest value of classification accuracy in the model. Comparing machine learning classifiers based on evaluation metrics reveals that ensemble-based decision tree approaches such as Random Forests and Gradient Boosting are the most effective for supervised classification. The results are validated using testing data and have an 80% classification accuracy. The predicted volume of lithofacies classes contributes to improved 3D reservoir modeling for the pre-salt carbonate reservoir.

Exact finite element de Rham subcomplexes relate conforming subspaces in H1(Ômega), H(curl; Ômega), H(div; Ômega), and L2(Ômega) in a simple way by means of differential operators (gradient, curl, and divergence). The characteristics of such strong couplings are crucial for the design of stable and conservative discretizations of mixed formulations for a variety of multiphysics systems. This work explores these aspects for the construction of divergence-free vector shape functions in a robust fashion allowing stable and faster simulations of mixed formulations of incompressible porous media flows. The resulting schemes are verified by means of numerical tests with known smooth solutions and applied to a benchmark problem to confirm the expected theoretical and computational performance results.

This article presents a computable and efficient procedure for a posteriori error estimations of approximate solutions of Darcy’s flows given by a Multiscale Hybrid Mixed method. Based on a partition of the domain by polytopal macro subregions, this is a strategy for efficient simulations of problems with strongly varying solutions. The flux approximations interacting with the skeleton (subregion boundaries) are strongly constrained by a given trace space. Whilst the dimension of the piecewise polynomial trace space is expected to be as low as possible, the small-scale features of the solution are supposed to be accurately resolved by completely independent local stable mixed solvers, the trace variable playing the role of Neumann boundary data. As the method already gives an equilibrated global H(div)-conforming flux approximation, the methodology for the error estimation only requires a potential reconstruction. In addition to usual residual errors and indicators associated with the potential reconstruction, the estimation also takes into account the effect of discretizations of practical nonhomogeneous Dirichlet and Neumann boundary conditions. Based on the proposed a posteriori error estimator, ℎ-adaptive algorithms are constructed to guide a proper choice of the trace space. The performance of the error estimator and the adaptive scheme is numerically investigated through a set of illustrating test problems.

Lower Cretaceous carbonates in the pre-salt succession in the Santos Basin, eastern Brazil, are highly heterogeneous in terms of their reservoir characteristics as a result of depositional and diagenetic factors. Electrofacies have widely been used for reservoir zonation and, when allied with computer-based methods such as neural networks, may help with the study of such complex reservoir rocks and with the identification of high-quality reservoir zones. In this work, an unsupervised artificial neural network known as a self-organizing map (SOM) was used to carry out a zonation of the pre-salt carbonates in the Aptian Barra Velha Formation, the main reservoir unit in the Santos Basin. Available data included gramma-ray, neutron porosity, resistivity deep, sonic, density, photoelectric factor, total porosity and effective porosity profiles from 21 wells together with mineralogical models. Core descriptions and thin section images were used as additional data for the lithological characterization of the electrofacies and consequently for reservoir zonation. A total of four electrofacies were defined from the SOM application, and five reservoir zones were identified.

The characterization of the reservoir zones also considered the structural locations of the wells based on the relative depth to top- Barra Velha Formation; well locations were classified as structurally high, intermediate or low. Based on the reservoir zone characteristics, the results could be correlated with zonations in previous studies. A general tendency was noted for there to be an increase of finer-grained sediments in the formation in wells located in structural lows; packstone and mudstone facies were prevalent in these wells and were in general characterized as poor-quality reservoir rocks. By contrast, the shrubstones and grainstones which were more frequent in structurally high wells comprised higher quality reservoir rocks.

The basal reservoir zone showed wide lithological variation compared to the overlying reservoir zones. Grainstone-dominated facies were identified in the middle of the formation, and the uppermost reservoir zones were characterized by an upward increase in shrubstones and reworked grainstones which in general pointed to better quality reservoirs.