The joint inversion of seismic data for elastic and petrophysical properties is an inverse problem with a nonunique solution. There are several factors that impact the accuracy of the results, such as the statistical rock-physics relations and observation errors. We have developed a general methodology to incorporate a linearized rock-physics model in a multivariate multimodal prior distribution for Bayesian seismic linearized inversion. The prior distribution is used to define a mixture model for elastic and petrophysical properties and introduce physics-based correlations between the properties. Using the rock-physics prior model and a convolutional seismic forward model in the Bayesian inversion framework, we obtain an analytical expression of the spatially independent conditional distributions to be used as a proposal distribution in a Gibbs sampling algorithm. We then combine the sampling algorithm with geostatistical simulation methods to compute the spatially correlated posterior distribution of the model parameters. We apply our method to a real angle-stack seismic data set to generate multiple geostatistical realizations of facies, P-velocity, S-velocity, density, porosity, and water saturation. The method is validated through a blind well test and a comparison with the standard Bayesian linearized inversion.
One of the main objectives in the reservoir characterization is estimating the rock properties based on seismic measurements. We have developed a stochastic sampling method for the joint prediction of facies and petrophysical properties, assuming a nonparametric mixture prior distribution and a nonlinear forward model. The proposed methodology is based on a Markov chain Monte Carlo (MCMC) method specifically designed for multimodal distributions for nonlinear problems. The vector of model parameters includes the facies sequence along the seismic trace as well as the continuous petrophysical properties, such as porosity, mineral fractions, and fluid saturations. At each location, the distribution of petrophysical properties is assumed to be multimodal and nonparametric with as many modes as the number of facies; therefore, along the seismic trace, the distribution is multimodal with the number of modes being equal to the number of facies power the number of samples. Because of the nonlinear forward model, the large number of modes and as a consequence the large dimension of the model space, the analytical computation of the full posterior distribution is not feasible. We then numerically evaluate the posterior distribution by using an MCMC method in which we iteratively sample the facies, by moving from one mode to another, and the petrophysical properties, by sampling within the same mode. The method is extended to multiple seismic traces by applying a first-order Markov chain that accounts for the lateral continuity of the model properties. We first validate the method using a synthetic 2D reservoir model and then we apply the method to a real data set acquired in a carbonate field.
5 February 2021 in IEEE Geoscience and Remote Sensing Letters
Determination of lithofacies is one of the most important steps for reservoir characterization. Well log curves are not always sufficient to determine lithology as some times the signals are similar for different lithologies. We believe that the sequence of sedimentary patterns that follows the general geologic rules can be essential to help this disambiguation. This work aims to present a computational system based on deep recurrent neural networks (RNNs) as an effective method to automatically identify lithofacies patterns from well logs. We show that bidirectional long-short-term memory (BiLSTM) RNNs can learn long-term dependencies between time steps of sequence data improving the context available to facies classification. We validated our method by applying it to a real case study from Rio Bonito Formation (Paraná Basin, Brazil) and the proposed method is compared with XGBoost, Random Forest, Naïve Bayes, and support vector machine (SVM) learning approaches. The results indicate that the performance of the proposed method for lithology identification is higher when compared with these other methods.
1 September 2016 in Expert Systems with Applications
Neural networks have been applied to seismic inversion problems since the 1990s. More recently, many publications have reported the use of Deep Learning (DL) neural networks capable of performing seismic inversion with promising results. However, when solving a seismic inversion problem with DL, each author uses, in addition to different DL models, different datasets and different metrics for performance evaluation, which makes it difficult to compare performances. Depending on the data used for training and the metrics used for evaluation, one model may be better or worse than another. Thus, it is quite challenging to choose the appropriate model to meet the requirements of a new problem. This work aims to review some of the proposed DL methodologies, propose appropriate performance evaluation metrics, compare the performances, and observe the advantages and disadvantages of each model implementation when applied to the chosen datasets. The publication of this benchmark environment will allow fair and uniform evaluations of newly proposed models and comparisons with currently available implementations.
4 October 2018 in IEEE Geoscience and Remote Sensing Letters
Domain-specific methods for deblurring particular sorts of objects have gained increasing attention due to the ineffectiveness of generic methods. We present a simple and effective convolutional neural network that deblurs postinversion acoustic impedance images. The architecture of our model consists of a convolutional layer that highlights edges and contours related to interfaces between rock layers; a locally connected layer that performs a convolutional step with unshared weights; and, finally, two fully connected layers that perform a nonlinear estimation of acoustic impedance values. We use the updated Standford VI reservoir model as training data set, which is composed of 150 acoustic impedance sections, each section with 200 traces. In this letter, we adopt a strong supervised learning that exploit, trace by trace, the data set of the inverted and ground truth impedance images. We also present an analysis comparing the frequency bandwidth among the latent, blurry, and deblurred images. Furthermore, the peak signal-to-noise ratio is calculated and compared with a classical deblurring method. We additionally address the requirement of deep learning for the huge amount of training examples by inserting rectified linear units and keeping the network architecture simple. The experimental results demonstrate the efficacy of the proposed method.