Produção Científica
Dissertação de Mestrado
| Avaliação do modelamento da propagação de ondas sísmicas pelo método lowrank O modelamento da propagação da onda é uma parte fundamental para o imageamento sísmico, pelo qual se requer métodos precisos para representar o campo de onda. O método lowrank é aplicado à matriz propagadora determinada pela extrapolação da solução da equação da onda. Nesse trabalho, é avaliado o desempenho absoluto e relativo do método lowrank em meios homogêneos nos quais tem-se expressões teóricas, assim como em meios heterogêneos. Realizamos vários testes em modelos de meios homogêneos com diferentes discretizações espacias e temporais, e a resposta do modelamento foi comparada com a solução analítica, mostrando boa acurácia cinematicamente, enquanto o desempenho na dinâmica foi afetado pelo intervalo de amostragem temporal, embora para cada modelo foi possível determinar uma discretização temporal no qual conseguiu-se um erro dinâmico menor que 1% para modelar a onda em meios homogêneos. Para determinar o desempenho relativo foi considerado o método de diferenças finitas tradicional de segunda ordem em tempo e espaço. Os coeficientes de reflexão aproximados pelo método lowrank considerando um modelo com um refletor plano foram comparados com os coeficientes de reflexão de ondas planas para os quais tem-se expressões teóricas. Neste caso, observa-se que com o método lowrank obtém-se os coeficientes com erro menor que 1% quando comparada ao coeficiente teórico, mostrando melhor resultado do que obtido por diferenças finitas. O tempo computacional dos métodos foi avaliado, sendo que lowrank presenta um crescimento linear do tempo com o tamanho do modelo, isto sugere que para modelos de grande porte, lowrank economize o tempo de modelamento em comparação com diferenças finitas. |
|
|
Artigo em Revista
| Characterizing seismic facies in a carbonate reservoir, using machine learning offshore Brazil. The authors propose an approach that uses machine learning to characterize carbonate facies in a wildcat (Gato do Mato) prospect in the Santos basin, offshore Brazil. We analyzed different seismic attributes and selected those that best responded to the seismic patterns identified in the study area as input for an unsupervised classification. The classification method used is the self-growing neural network (SGNN) technique that consists of the following steps: 1. Seismic pattern identification in seismic amplitude. The main patterns identified are build-up, debris, carbonate platform and bottom lake facies. 2. Generation and analysis of seismic attributes to characterize seismic patterns. We chose Eigen coherence, dip-steered enhancement, relief and relative acoustic impedance to help in seismic characterization. 3. We performed principal component analysis (PCA) of the attributes: amplitude filtered from dip-steered enhancement (seismic-driven structural filtering), Eigen coherence and relief. 4. Unsupervised seismic classification from the PCA of seismic attributes (item 3 above). Using this approach, we associated the classified seismic facies with the patterns identified in the amplitude data. The seismic facies allowed us to differentiate the carbonate platforms from the build-up facies. However, the classification encountered difficulties in identifying the patterns associated with lake bottom facies and the chaotic seismic pattern of debris facies. |
|
|
Artigo em Revista
| Probabilistic Estimation of Seismically Thin-Layer Thicknesses with Application to Evaporite Formations The identification of potassium (K) and magnesium (Mg) salts prior to the well drilling is a key factor to avoid washouts, closing pipes, fluid loss damage, and borehole collapse. The Bayesian classification combines the outcomes from statistical rock physics and seismic inversion, providing the spatial occurrence of the most-probable salt types. It serves as a facies identifier of Mg–K-rich salts (bittern salts) before drilling. Nevertheless, the most-probable classification is limited to the seismic resolution which may underestimate seismically thin-layer thicknesses. Along with the most-probable facies, the Bayesian classification renders the facies probability volume. We demonstrate that the facies probability and facies-specific total thickness highly correlate to each other even under the threshold of seismic resolution. Thus, we employ the bittern-salts probability volume to predict thin-bed bittern-salts thickness in undrilled locations. To capture the variability of the seismic estimation, we resort to Monte Carlo-assisted simulations of wells that emulate the layering patterns of a site-specific deposition environment. These simulations are crucial to assist the estimation of the joint probability density function between the facies volume and the total thickness. Therefore, given the facies probability, the joint probability density function enables us to derive the conditional expectation and percentiles of thin-bed thicknesses. Furthermore, this paper proposes a method to quantify the negative influence of seismic noise in the estimation of thin-bed thicknesses. The blind well confirms the consistency of this technique to unfold the uncertainty in the seismic predictability of thin layers. We argue that this procedure is extendable to other facies. |
|
|
Artigo em Revista
| Mineralogy based classification of carbonate rocks using elastic parameters: A case study from Buzios Field Brazilian presalt carbonate reservoirs are highly heterogeneous. This feature is mostly justified by the nature of the original depositional system and subsequently diagenetic processes. Consequently, reserve estimates and production forecasting are under large uncertainties. In this geologic context, it is of great relevance to develop techniques that helps to obtain a detailed description on the spatial distribution of these different rocks. In doing so, it contributes to the understanding of presalt carbonate sedimentary deposits, providing inputs for more predictive reservoir models. Traditionally, these carbonates are grouped into three classes, from which only one exhibits reservoir properties. Using a dataset from Buzios Field, this work proposes a characterization of presalt carbonate reservoir rocks by grouping them in terms of their mineral composition. Taking advantage of rock physics concepts, we aim to potentialize the use of elastic parameters for multiple rock type discrimination. We explored several attempts for rock classification by using a Bayesian approach. Among all the tested propositions, a two-step workflow for five lithotypes classification, emerges as the most appropriate for the Buzios Field. In this scheme, three lithotypes represent good-quality reservoirs and the other two are low-porosity and Mg-clay-rich carbonates. The average root-mean-square error of the most likely a posteriori rock proportions is around 8.4%, only approximately 1% higher than the conventional three lithotypes configuration. To support that, we compared different methodologies for Bayesian classification at well-log scale through acoustic impedance and compressional to shear velocity ratio. Potential applicability of the proposed methodology at field scale is reinforced by similar results achieved using well-logs filtered to the seismic bandwidth. |
|
|
Artigo em Revista
| The influence of subseismic-scale fracture interconnectivity on fluid flow in fracture corridors of the Brejões carbonate karst system, Brazil The present study used a multitool approach to characterize fractures of several orders of magnitude in large fracture corridors, caves, and canyons to investigate their impact on fluid flow in carbonate units. The study area is the Brejões carbonate karst system that is located in the Neoproterozoic Salitre Formation in the Irecê Basin, São Francisco Craton, Brazil. The approach included satellite imagery, used for interpreting the regional structural context, Unmanned Aerial Vehicle (UAV) and ground-based Light Detection And Ranging (LiDAR) imagery, used for detailed structural interpretation. Regional interpretation revealed that fracture corridors, caves and canyons occur along a N–S-oriented anticline hinge. An advanced stage of karstification caused fracture enlargement and intrabed dissolution, and the formation of caves and canyons. A river captured by the highly fractured zone along the anticline hinge played an important role as an erosive agent. Detailed characterization of fracture corridors comprised structural analysis, topological studies, persistence estimations, power-law fitting of fracture trace length distributions, and identification of network backbones. Our results indicate that fracture corridors comprise four subvertical fracture sets: N–S and E-W and a conjugate pair, NNE-SSW and NW-SE. Fractures observed in the caves show the same dominant directions. Fracture directions are consistent with a common origin associated with the anticline folding. Fracture traces range from 1.0 m to 300 m, comprising both subseismic (<50 m) and seismic scale fractures (>50 m). Networks have dominance of node terminations Y and X (notably Y), CB values higher than 1.8, high P20 and P21 persistence values, and highly interconnected backbones. Fracture network connectivity is associated with power-law exponents greater than 2.5 for the fracture trace distributions, indicating large influence of subseismic-scale fractures on fluid flow. As the final result of folding and karstification, large volumes of secondary macroporosity were created, particularly in the zone of maximum fracture intensity around the hinge zone of the anticline. This scenario can be used to understand better oil reservoirs formed in similar structural controls in near-surface conditions. |
|
|
Artigo em Revista
| A constrained version for the stereology inverse problem: Honoring power law and persistences of the fracture traces exposed on arbitrary surfaces We present a stereological study in a cave setting that is part of a karstic carbonate system located in the São Francisco Craton, Brazil. Using a Lagrangian approach, a constrained version of the nonlinear inverse problem of stereology is solved. Besides the classical demand of fitting the histogram of fracture traces measured on arbitrary exposed surface, it is imposed that the solution honors also measures of surface intensity (p21) and power law exponent obtained from fracture traces on the same exposed surfaces. Estimates of volumetric intensities (p32) of conjugate fracture pairs might be also imposed to be close values. The resulting cost functional is minimized using the Particle Swarm Optimization (PSO) method. The implemented version of PSO furnishes the best solution and a set of suboptimal quasisolutions, from which the solution uncertainty is evaluated. A key aspect of the implemented approach is that all terms composing the cost functional are normalized, obtaining as a result, robustness for the weighting parameters to changes in the input data. The resulting stochastically simulated discrete fracture networks honor all statistical observations but, in general, do not reproduce the positions of the observed fracture traces. The methodology is applied to synthetic and field data examples. All obtained solutions are stable and geologically reliable. |
|
|
Artigo em Revista
| Well log analysis for lithology and fluid contacts in Rovuma Basin – Mozambique: Application of cluster and discriminant analyses This study applies the cluster and discriminant analyses in geophysical well log data from the Rovuma sedimentary Basin - Mozambique. The main objective was to determine the lithological profile and fluid contacts in reservoirs. Well log data from five wells drilled on the same basin were used. For the discrimination, a reference well was chosen for training, and the obtained functions from it were then applied to the remaining wells. The classification process comprehended three main phases, namely, the separation of shale/non-shale layers along the entire logged section, separation of water/hydrocarbon within reservoirs and the separation of oil/gas within hydrocarbon bearing zones. The two methods, cluster analysis and discriminant analysis, were applied in parallel and the results are compared in each classification phase. The quality of reservoirs was also assessed by applying cutoffs in relation to shale content and effective porosity, delineating net reservoirs. In general, both methods converged to the same lithological model and fluidtypes in reservoirs. Gas has been indicated as the most predominant hydrocarbon in the basin. |
|
|
Artigo em Revista
| Comparison between oil spill images and look-alikes: an evaluation of SAR-derived observations of the 2019 oil spill incident along Brazilian waters Three SAR-derived observations of dark surface patches along the Northeastern Brazilian coastline by the end of 2019 were misreported in the Brazilian media as oil spill-related. Unfortunately, these observations were misled by false positives or look-alikes. Therefore, this paper aims to technically evaluate these look-alike classes by analyzing image attributes found to be helpful to the identification of ocean targets, including oil spills, rain cells, biofilms, and low wind conditions. We use image augmentation to extend our dataset size and create the probability density function curves. The processing includes image segmentation, optimal attribute extraction, and classification with random forest classifiers. Our results contrast with the open-source oil spill detection system and patch classifier methodology called “RIOSS.” Analysis of the feature probability density functions based on optimal attributes is promising since we could capture most of the false positive targets in the three SAR-reported images in 2019. The only exception was the biofilm slick observed on October 28th, where the RIOSS mistakenly classified this organic patch as a low wind region with oil spots. This pitfall is acceptable at this project stage since we had only five biogenic film samples to train the algorithm. |
|
|
Artigo em Revista
| Low-rank seismic data reconstruction and denoising by CUR matrix decompositions. Low-rank reconstruction methods assume that noiseless and complete seismic data can be represented as low-rank matrices or tensors. Therefore, denoising and recovery of missing traces require a reduced-rank approximation of the data matrix/tensor. To calculate such approximation, we explore the CUR matrix decompositions, which use actual columns and rows of the data matrix, instead of the costly singular vectors derived from singular value decomposition. By allowing oversampling columns and rows, CUR decompositions obviate the need for the exact rank. We evaluate three different procedures for randomly selecting columns and rows to obtain the CUR. Once the low-rank approximation is estimated, data reconstruction is achieved by an iterative optimization scheme. To demonstrate the effectiveness of CUR matrix decompositions for multidimensional seismic data recovery, we present examples of 3D and 4D synthetic and field data. Results derived by CUR compare well to conventional eigenimage-family methods. |
|
|
Tese de Doutorado
| ADVANCED TECHNIQUES FOR 3D RESERVOIR CHARACTERIZATION: MODELS FOR THE BUZIOS FIELD, SANTOS BASIN This thesis aims to build workflows using innovative 3D reservoir characterization techniques that have the capacity of provide robust, automatized, and with fair predictability lithological facies and petrophysical properties models from well data and secondary data can honor the heterogeneities for complex reservoirs, such as the presalt Aptian carbonates in the Brazilian marginal basins. First, I applied for the first-time an integrated approach between 4D sedimentary modeling and geostatistical modeling for facies reconstruction for presalt reservoirs from Barra Velha Formation within the Buzios Field located in the Santos Basin. The results from this first study demonstrate that this methodology makes it possible for the geomodeler to incorporate its conceptual geological knowledge about the facies paleoenvironment to be modeled in a more precise manner and the integration with geostatistics allows proper matching of the facies model with well data. For the second study, I used of a neural network algorithm for a multi-attribute unsupervised seismic facies classification. My results from this study showed that this technique allows the automatized creation of a 3D seismic facies model that can be further associated with the porosity and permeability distributions from well data for a qualitative inference of the reservoir properties. The third study present in this thesis approaches also for the first-time the use of a pre-commercial and innovative machine learning methodology that implements geostatistical concepts for supervised estimation of petrophysical reservoir properties from multiple secondary variables called EMBER. The results from this study allowed the quantitative evaluation of the reservoirs from Barra Velha Formation through the creation of effective porosity and permeability 3D models also helping to address the uncertainties related to effective porosity distribution. Before this research, two of those methodologies had never been applied for carbonate rocks characterization or presalt Aptian reservoir modeling. Finally, I hope that those standardized methodologies and the results discussions from this thesis can help the presalt studies to build a better understanding of these reservoirs origins diminishing complexity and provide alternatives for the classical reservoir characterization approaches which not always can properly provide robust results impacting the future of oil and gas exploration and production. |
|
|
<< < 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 > >>
