Produção Científica

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 S√£o 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 ‚Äď 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 ‚ÄúRIOSS.‚ÄĚ 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

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.

Dissertação de Mestrado

Devido √† import√Ęncia econ√īmica dos reservat√≥rios carbon√°ticos da se√ß√£o Pr√©-sal da Bacia de Santos (Fm. Barra Velha), bastante esfor√ßo vem sendo empregado no
desenvolvimento de metodologias para a caracterização destas rochas. Diversos trabalhos têm confirmado a efetividade do uso de atributos sísmicos para a
estimativa propriedades de reservatório de tais rochas, como a porosidade. Além disso, técnicas de aprendizagem de máquina vêm ganhando notoriedade no setor de E&P, sendo ótimas ferramentas para estimar propriedades de rochas
heterog√™neas, como os carbonatos. O presente trabalho prop√Ķe uma metodologia de integra√ß√£o de dados s√≠smicos e de po√ßos, atrav√©s da modelagem 3D de porosidade na Fm. Barra Velha pelo m√©todo random forest. O objetivo foi analisar
como diferentes combina√ß√Ķes dos atributos de amplitude s√≠smica e imped√Ęncia ac√ļstica impactam os resultados. Os resultados foram considerados robustos, em conson√Ęncia com as m√©tricas de valida√ß√£o do modelo ‚ÄĒ correla√ß√£o de Pearson,
desvio padr√£o, Q-Q plots e crossplots. A √°rea de estudo consiste em uma extensa plataforma carbon√°tica e duas regi√Ķes de mounds carbon√°ticos. Ao longo da plataforma, a parte basal concentra porosidades altas, chegando a 15%. Os mounds apresentam porosidade de moderada a alta, chegando a 18%. A heterogeneidade da porosidade foi interpretada como resultado de varia√ß√Ķes relativas do n√≠vel do paleolago da Fm. Barra Velha, indicando uma tend√™ncia geral de afogamento da plataforma. As intrus√Ķes √≠gneas atuaram de forma geral como um fator redutor de porosidade nos carbonatos circundantes √†s mesmas. Por outro lado, as falhas associadas √†s intrus√Ķes podem ter atuado como conduto para fluidos hidrotermais, que teriam contribu√≠do para a gera√ß√£o de porosidade nos carbonatos da por√ß√£o
basal desta formação. Foi construído um catálogo de sismofácies da região de estudo na qual foram identificados mounds carbonáticos, plataformas carbonáticas, fácies debris e rochas ígneas. Portanto, a metodologia proposta para a modelagem 3D de porosidade, utilizando atributos sísmicos através do método random forest, foi efetiva em caracterizar potenciais reservatórios carbonáticos da Fm. Barra Velha na área de estudo, e esta metodologia pode ser aplicada em outras áreas, com características geológicas semelhantes.


Seismic attributes and machine learning techniques for identification and characterization of carbonate seismic facies of the Barra Velha Formation, in the Wildcat Prospect, Santos Basin.
Seismic data provides important information for prospect identification and reservoir characterization. For an effective seismic characterization, first we need to compute seismic attributes and combine them in a skillful way, to obtain as much information as possible to identify the different seismic patterns. Machine learning techniques applied to seismic interpretation have been successful in this regard and very useful in assisting with limitations involving data classification.

This work aims to apply a methodology for identification and characterization of carbonates facies from Barra Velha Formation, on the Wildcat Prospect in Santos Basin, using seismic attributes and a non-supervised facies classification that uses a machine learning method called Self-Growing Neural Network (SGNN). For the workflow, we performed the following steps: (i) Carbonate seismic patterns identification through seismic amplitude, where was possible to identify the build-up, characterized by chaotic seismic textures with a conical external geometry and internal fracturing; debris facies, exhibit prograde geometry with chaotic internal texture; carbonate platform facies showing a well defined flat parallel reflectors; and
the bottom lake facies, that does not have specific geometry and internally the reflectors are chaotic
(Neves et al., 2019). (ii) seismic attributes generation and analysis to characterize the carbonate seismic patterns, where the eigen coherence, dip steered enhancement, relief and relative acoustic impedance
assisted the seismic characterization. (iii) Principal component analysis (PCA), with amplitude filtered with
DSE, eigen coherence and relief as inputs, and (iv) non-supervised seismic classification with attribute clustering, using the seismic attributes and PCA results as input.

In the results, we could associate the seismic facies with the carbonate seismic patterns of Barra Velha Formation, previously identified. We obtained a clear differentiation between the carbonate platform and more fractured areas related to the build-ups. Similarly, the chaotic behavior of debris was well captured. However, the bottom lake patterns didn’t appear very evident. The PCA before the seismic classification brought better results. The attribute clustering method, has been an effective approach to differentiate the fractured zones mainly associated to build-ups and the facies layering of the carbonate platform in the Wildcat Prospect.

Artigo em Revista

Probabilistic Estimation of Seismically Thin‚ÄĎLayer Thicknesses with Application to Evaporite Formations
The identifcation of potassium (K) and magnesium (Mg) salts prior to the well drilling is a key factor to avoid washouts, closing pipes, fuid loss damage, and borehole collapse.
The Bayesian classifcation 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 identifer of Mg‚ÄďK-rich salts (bittern salts) before drilling. Nevertheless, the most-probable classifcation is limited to the seismic resolution which may underestimate seismically thin-layer thicknesses. Along with the most-probable facies, the Bayesian classifcation renders the facies probability volume. We demonstrate that the facies probability and facies-specifc total thickness highly correlate to each other even under the threshold
of seismic resolution. Thus, we employ the bittern-salts probability volume to predict thinbed 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-specifc 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 infuence of seismic noise in the estimation of thin-bed thicknesses. The blind well confrms 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.

Trabalho de Graduação

Processamento pós-estaqueamento e interpretação de dados sísmicos de reflexão aplicados na exploração de hidrocarbonetos
Os dados s√≠smicos de reflex√£o s√£o os produtos dos m√©todos geof√≠sicos mais amplamente empregados na explora√ß√£o de hidrocarbonetos. Isso se d√°, fundamentalmente, em fun√ß√£o da sua boa rela√ß√£o custo-benef√≠cio no que diz respeito √† profundidade de investiga√ß√£o do alvo a ser prospectado e das resolu√ß√Ķes horizontal e espacial associadas ao m√©todo. Para tanto, √© necess√°rio que o dado s√≠smico seja confi√°vel quanto √† imagem final, a qual, n√£o basta ter feito uma boa aquisi√ß√£o e um processamento adequado, pois ainda apresenta um certo n√≠vel de ru√≠do embutido, bem como, em diversos casos, ainda n√£o enfatiza de forma espec√≠fica o alvo. Neste contexto, o presente trabalho aborda o p√≥s-processamento, que normalmente melhora o car√°ter s√≠smico, e a utiliza√ß√£o de atributos s√≠smicos, que tem como principal objetivo enfatizar/real√ßar fei√ß√Ķes importantes para a interpreta√ß√£o. Os atributos utilizados foram: Similaridade, decomposi√ß√£o espectral, amplitude instant√Ęnea e frequ√™ncia instant√Ęnea. Os resultados obtidos confirmam a aplicabilidade de atributos s√≠smicos como uma poderosa ferramenta no aux√≠lio √† interpreta√ß√£o s√≠smica.

Dissertação de Mestrado

Plataforma interativa de análise de velocidade em dados sísmicos usando GPUs
Com o avan√ßo da explora√ß√£o de hidrocarbonetos, a ind√ļstria vem buscando continuamente meios de minimizar os riscos explorat√≥rios, onde um desses meios √© o aprimoramento das ferramentas utilizadas. Existem tr√™s etapas nessa explora√ß√£o: a aquisi√ß√£o de dados s√≠smicos, o processamento s√≠smico e a interpreta√ß√£o s√≠smica. O presente trabalho se situa no processamento s√≠smico, mais especificamente em uma de suas etapas, a an√°lise de velocidade s√≠smica, que tem como objetivo encontrar o campo de velocidade mais fidedigno da subsuperf√≠cie da terra atrav√©s de algoritmos conhecidos de an√°lise. Um dos objetivos desse trabalho √© a cria√ß√£o de meios para facilitar essa an√°lise de velocidade, atrav√©s da implementa√ß√£o desses algoritmos de forma que eles funcionem integrados em uma √ļnica plataforma. Outro ponto que o avan√ßo da explora√ß√£o s√≠smica trouxe foi o aumento consider√°vel do volume de dados s√≠smicos adquiridos e das tecnologias utilizadas, que elevaram consideravelmente a necessidade de computadores mais poderosos e tamb√©m √† busca de solu√ß√Ķes de alto poder computacional. Com base nessa necessidade, ser√° apresentada uma nova metodologia de an√°lise de dados s√≠smicos usando GPUs e os resultados obtidos da sua utiliza√ß√£o, mostrando sua viabilidade para acelerar algoritmos geof√≠sicos, em especial algoritmos voltados para √† an√°lise de velocidade. Ao final ser√£o discutidos os resultados e feita a compara√ß√£o de desempenho dos algoritmos paralelos e sequenciais.

<<  <   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  68   >  >>