Fred Aminzadeh

Three dimensional SEG/EAEG models; an update

This report updates the status of the 3-D SEG/EAEG Modeling Project (SEM). Previous updates appeared in TLE in February 1994, October 1994, and February 1995. The goal is to design two 3-D models, salt and overthrust, and simulate realistic 3-D surveys through numerical calculations.

Mining and fusion of petroleum data with fuzzy logic and neural network agents

Abstract Analyzing data from well logs and seismic is often a complex and laborious process because a physical relationship cannot be established to show how the data are correlated. In this study, we will develop the next generation of “intelligent” software that will identify the nonlinear relationship and mapping between well logs/rock properties and seismic information and extract rock properties, relevant reservoir information and rules (knowledge) from these databases. The software will use fuzzy logic techniques because the data and our requirements are imperfect. In addition, it will use neural network techniques, since the functional structure of the data is unknown. In particular, the software will be used to group data into important data sets; extract and classify dominant and interesting patterns that exist between these data sets; discover secondary, tertiary and higher-order data patterns; and discover expected and unexpected structural relationships between data sets.

Past, present and future intelligent reservoir characterization trends☆

Abstract As we approach the next millennium, and as our problems become too complex to rely only on one discipline to solve them more effectively, multi-disciplinary approaches in the petroleum industry become more of a necessity than professional curiosity. We will be forced to bring down the walls we have built around classical disciplines such as petroleum engineering, geology, geophysics and geochemistry, or at the very least, make them more permeable. Our data, methodologies and approaches to tackle problems will have to cut across various disciplines. As a result, todays “integration”, which is based on integration of results, will have to give way to a new form of integration, that is, integration of disciplines. In addition, to solve our complex problem, one needs to go beyond standard techniques and silicon hardware. The model needs to use several emerging methodologies and soft computing techniques: Expert Systems, Artificial Intelligence, Neural Network, Fuzzy Logic (GL), Genetic Algorithm (GA), Probabilistic Reasoning (PR), and Parallel Processing techniques. Soft computing differs from conventional (hard) computing in that, unlike hard computing, it is tolerant of imprecision, uncertainty, and partial truth. Soft Computing is also tractable, robust, efficient and inexpensive. In this paper, we reveal (explore) the role of Soft Computing techniques in intelligent reservoir characterization and exploration.

SEG/EAEG 3-D modeling project: 2nd update

The first update for the SEG/EAEG 3-D Modeling Project appeared in the February issue of TLE and the March issue of First Break. Our goal is to design salt and overthrust 3-D models and then simulate realistic 3-D surveys based on those models. Given the project’s significance and scope, we plan frequent progress reports. In conjunction with this goal and to solicit input, we have made recent presentations at SEG Cairo, the SEG/EAEG Summer Research Workshop in The Netherlands, the Stanford Exploration Project, and the Center for Wave Phenomena at the Colorado School of Mines.

3-D modeling project; 3rd report

The two geologic models (overthrust and salt dome) are now finalized, the acquisition phases are defined, and DOE laboratories have started the computations. Simulation for phase A is expected to be completed during the first quarter of 1995.

Applications of clustering in exploration seismology

Abstract Applying clustering techniques to seismic data is still at a nascent stage. This work represents an attempt in this direction. First, we give an overview of the cluster analysis and pattern recognition methods. A stacked seismic section is used to compute the features (amplitude, phase, frequency etc.) on a sample-by-sample basis. After factor analysis, for reducing the feature set to an orthogonal one, hierarchical clustering is performed on a small amount of data. The grouping information obtained from it is then passed onto a second phase in which a non-hierarchical clustering is done. The results were verified by model data.

Soft Computing for Reservoir Characterization and Modeling

The volume is the first comprehensive book in the area of intelligent reservoir characterization written by leading experts in academia and industry. It contains state-of-the-art techniques to be applied in reservoir geophysics, well logging, reservoir geology, and reservoir engineering. It introduces the basic concepts of soft computing techniques including neural networks, fuzzy logic and evolutionary computing applied to reservoir characterization. Some advanced statistical and hybrid models are also presented. The specific applications include different reservoir characterization topics such as prediction of petrophysical properties from well logs and seismic attributes.

EFFECTS OF TRANSVERSE ISOTROPY ON P-WAVE AVO FOR GAS SANDS

Velocity anisotropy should be taken into account when analyzing the amplitude variation with offset (AVO) response of gas sands encased in shales. The anisotropic effects on the AVO of gas sands in transversely isotropic (TI) media are reviewed. Reflection coefficients in TI media are computed using a planewave formula based on ray theory. We present results of modeling special cases of exploration interest having positive reflectivity, near‐zero reflectivity, and negative reflectivity. The AVO reflectivity in anisotropic media can be decomposed into two parts; one for isotropy and the other for anisotropy. Zero‐offset reflectivity and Poisson’s ratio contrast are the most significant parameters for the isotropic component while the δ difference (Δδ) between shale and gas sand is the most important factor for the anisotropic component. For typical values of Tl anisotropy in shale (positive δ and e), both δ difference (Δδ) and e difference (Δe) amplify AVO effects. For small angles of incidence, Δδ plays a...

Soft Computing for Intelligent Reservoir Characterization

This paper presents an overview of soft computing techniques for reservoir characterization. The key techniques include neurocomputing, fuzzy logic and evolutionary computing. A number of documented studies show that these intelligent techniques are good candidates for seismic data processing and characterization, well logging, reservoir mapping and engineering. Future research should focus on the integration of data and disciplinary knowledge for improving our understanding of reservoir data and reducing our prediction uncertainty. Introduction Accurate prediction of reservoir performance is a difficult problem. This is mainly due to the failure of our understanding of the spatial distribution of lithofacies and petrophysical properties. Because of this, the recovery factors in many reservoirs are unacceptably low. The current technologies based on conventional methodologies are inadequate and/or inefficient. In this paper, we propose the next generation of reservoir characterization tools for the new millennium – soft computing. Reservoir characterization plays a crucial role in modern reservoir management. It helps to make sound reservoir decisions and improves the asset value of the oil and gas companies. It maximizes integration of multi-disciplinary data and knowledge and improves the reliability of the reservoir predictions. The ultimate product is a reservoir model with realistic tolerance for imprecision and uncertainty. Soft computing aims to exploit such a tolerance for solving practical problems. Soft computing is an ensemble of various intelligent computing methodologies which include neurocomputing, fuzzy logic and evolutionary computing. Unlike the conventional or hard computing, it is tolerant of imprecision, uncertainty and partial truth. It is also tractable, robust, efficient and inexpensive. In reservoir characterization, these intelligent techniques can be used for uncertainty analysis, risk assessment, data fusion and data mining which are applicable to feature extraction from seismic attributes, well logging, reservoir mapping and engineering. Figure 1 shows schematically the flow of information and techniques to be used for intelligent reservoir characterization. The main goal is to integrate soft data such as geological data with hard data such as 3D seismic and production data to build a reservoir and stratigraphic model. While some individual methodologies (esp. neurocomputing) have gained much popularity during the past few years, the true benefit of soft computing lies on the integration of its constituent methodologies rather than use in isolation. This paper firstly outlines the unique roles of the three major methodologies of soft computing – neurocomputing, fuzzy logic and evolutionary computing. We will summarize a number of relevant and documented reservoir characterization applications. Lastly we will provide a list of recommendations for the future use of soft computing. This includes the hybrid of various methodologies (e.g. neural-fuzzy or neuro-fuzzy, neural-genetic, fuzzy-genetic and neural-fuzzy-genetic) and the latest tool of “computing with words” (CW). CW provides a completely new insight into computing with imprecise, qualitative and linguistic phrases and is a potential tool for geological modeling which is based on words rather than exact numbers. An appendix is also provided for introducing the basics in soft computing. Neurocomputing Neurocomputing represents general computation with the use of artificial neural networks. An artificial neural network is a computer model that attempts to mimic simple biological learning processes and simulate specific functions of human nervous system. It is an adaptive, parallel information processing system which is able to develop associations, transformations or mappings between objects or data. It is also SPE 59397 Soft Computing for Intelligent Reservoir Characterization D. Tamhane, SPE, P.M. Wong, SPE, University of New South Wales, F. Aminzadeh, SPE, FACT Inc. & dGB-USA, M. Nikravesh, SPE, Energy and Geoscience Institute (EGI)-University of Utah & Zadeh Institute of Information Technology. SPE 59397 SOFT COMPUTING FOR INTELLIGENT RESERVOIR CHARACTERIZATION 2 the most popular intelligent technique for pattern recognition to date. The basic elements of a neural network are the neurons and their connection strengths (weights). Given a topology of the network structure expressing how the neurons (the processing elements) are connected, a learning algorithm takes an initial model with some “prior” connection weights (usually random numbers) and produces a final model by numerical iterations. Hence “learning” implies the derivation of the “posterior” connection weights when a performance criterion is matched (e.g. the mean square error is below a certain tolerance value). Learning can be performed by “supervised” or “unsupervised” algorithm. The former requires a set of known input-output data patterns (or training patterns), while the latter requires only the input patterns. Figure 2 depicts a typical structure of a neural network, showing three layers of neurons. The lines represent how the neurons are connected. Each line is represented by a weight value. In this case, the inputs are passed to each layer and the results are obtained at the output layer. This is commonly known as the feedforward model, in which no lateral or backward connections are used. The full technical details can be found in Bishop. Applications. The major applications of neurocomputing are seismic data processing and interpretation, well logging and reservoir mapping and engineering. Good quality seismic data is essential for realistic delineation of reservoir structures. Seismic data quality depends largely on the efficiency of data processing. The processing step is time consuming and complex. The major applications include first arrival picking, noise elimination, structural mapping, horizon picking and event tracking. A detailed review can be found in Nikravesh and Aminzadeh. For interwell characterization, neural networks have been used to derive reservoir properties by crosswell seismic data. In Chawathé et al., the authors used a neural network to relate five seismic attributes (amplitude, reflection strength, phase, frequency and quadrature) to gamma ray (GR) logs obtained at two wells in the Sulimar Queen field (Chaves County). Then the GR response was predicted between the wells and was subsequently converted to porosity based on a field-specific porosity-GR transform. The results provided good delineation of various lithofacies. Feature extraction from 3D seismic attributes is an extremely important area. Most statistical methods are failed due to the inherent complexity and nonlinear information content. Figure 3 shows an example use of neural networks for segmenting seismic characters thus deducing information on the seismic facies and reservoir properties (lithology, porosity, fluid saturation and sand thickness). A display of the level of confidence (degree of match) between the seismic character at a given point versus the representative wavelets (centers of clusters) is also shown. Combining this information with the seismic model derived from the well logs while perturbing for different properties gives physical meaning of different clusters. Monson and Pita applied neural networks to find relationships between 3D seismic attributes and well logs. The study provided realistic prediction of log responses far away from the wellbore. Boadu also used similar technology to relate seismic attributes to rock properties for sandstones. In Nikravesh et al., the author applied a combination of k-means clustering, neural networks and fuzzy c-means (a clustering algorithm in which each data vector belongs to each of the clusters to a degree specified by a membership grade) techniques to characterize a field that produces from the Ellenburger Dolomite. The techniques were used to perform clustering of 3D seismic attributes and to establish relationships between the clusters and the production log. The production log was established away from wellbore. The production log and the clusters were then superimposed at each point of a 3D seismic cube. They also identified the optimum locations for new wells based on the connectivity, size and shape of the clusters related to the pay zones (see Figure 4). The use of neural networks in well logging has been popular for nearly one decade. Many successful applications have been documented. The most recent work by Bruce et al. presented a state-of-the-art review of the use of neural networks for predicting permeability from well logs. In this application, the network is used as a nonlinear regression tool to develop transformation between well logs and core permeability. Such a transformation can be used for estimating permeability in uncored intervals and wells. One example is shown in Figure 5. In this work, the permeability profile was predicted by a Bayesian neural network. The network was trained by a training set with four well logs (GR, NPHI, RHOB and RT) and core permeability. The network also provided a measure of confidence (the standard deviation of a Gaussian function): the higher the standard deviation (“sigma”), the lower the prediction reliability. This is very useful for understanding the risk of data extrapolation. The same tool can be applied to estimate porosity and fluid saturations. Another important application is the clustering of well logs for the recognition of lithofacies. This provides useful information for improved petrophysical estimates and well correlation. Neurocomputing has also been applied to reservoir mapping. In Wong et al. and Wang et al., the authors applied a radial basis function neural network to relate the conceptual distribution of geological facies (in the form of hand drawings) to reservoir porosity. It is able to incorporate the general property trend provided

Reservoir parameter estimation using a hybrid neural network

The accuracy of an artificial neural network (ANN) algorithm is a crucial issue in the estimation of an oil field’s reservoir properties from the log and seismic data. This paper demonstrates the use of the k-fold cross validation technique to obtain confidence bounds on an ANN’s accuracy statistic from a finite sample set. In addition, we also show that an ANN’s classification accuracy is dramatically improved by transforming the ANN’s input feature space to a dimensionally smaller, new input space. The new input space represents a feature space that maximizes the linear separation between classes. Thus, the ANN’s convergence time and accuracy are improved because the ANN must merely find nonlinear perturbations to the starting linear decision boundaries. These techniques for estimating ANN accuracy bounds and feature space transformations are demonstrated on the problem of estimating the sand thickness in an oil field reservoir based only on remotely sensed seismic data. 7 2000 Elsevier Science Ltd. All rights reserved.