- Home
- Conferences
- Conference Proceedings
- Conferences
EAGE 2020 Annual Conference & Exhibition Online
- Conference date: December 8-11, 2020
- Location: Online
- Published: 08 December 2020
61 - 80 of 368 results
-
-
Frequency-Dependent Analysis of Q Factor Determination in Laboratory
More LessSummaryUltrasonic laboratory seismic physics model method can simulate attenuation data in a comprehensive, accurate and quantitative manner. Although there are many methods for measuring Q values in the laboratory, these various methods still duplicate the traditional measurement methods of the seismic data. Since the estimation of the inherent attenuation of the laboratory is extremely susceptible to factors such as scattering attenuation and diffraction effects, the traditional methods are directly applied to laboratory conditions and have large errors. The more applicable and accurate attenuation estimation methods are needed under laboratory conditions. In this paper, we compared several methods for estimating frequency-independent Q value (constant Q hypothesis) and a method for estimating frequency-dependent Q value via two-parameter regression (TPM) using a laboratory seismic physics model. Through the laboratory tests, we conclude that there is the frequency dependence of the Q value under laboratory conditions, and the frequency-dependent Q is more closely than frequency-independent Q value to match the Q value of the theoretical prediction and laboratory measurement. The frequency-dependent Q values can provide a reference for the accuracy and stability of the Q value measured by the seismic physical model.
-
-
-
Analysis of the 4D Signal at the Volve Field NCS - An Open Subsurface Dataset
Authors A. Hallam, C. MacBeth, H. Amini and R. ChassagneSummaryIn 2018, Equinor released to the public an open subsurface dataset comprising all the data of the Volve Oil field from the central Norwegian North Sea. The dataset includes 4D seismic that we interpret, calibrate and quantify to better understand the distribution of water sweep in the reservoir.
Calibration of a dynamic Petro-Elastic Model (PEM) is a key part of the 4D interpretation. Saturation changes in the reservoir are determined to be the primary reason for a 4D signal with a maximum impedance change of 5% expected. The PEM is then combined with 4D forward modelling of well logs to establish links between the the character of the 4D signal and water sweep in the three reservoir zones at Volve. Discrepancies between production logging results and those which best match the 4D seismic response can be combined with the qualitative understanding of the seismic uncertainty to suggest improvements to the reservoir simulation model.
-
-
-
Implementation of Large-Scale Integral Operators with Modern HPC Solutions
Authors M. Ravasi and I. VasconcelosSummaryNumerical integral operators of convolution type lie at the foundations of most wave-equation-based methods for processing and imaging of seismic data. Several of such methods require the solution of an inverse problem, which in turn calls for multiple forward and adjoint passes of the modelling operator. In this abstract we provide some insights into the numerical aspects of solving such systems of integral equations and present a framework that leverages open-source libraries for distributed storage and computing as well as for high-level symbolic representation of linear operators. To validate the effectiveness of our implementation, we evaluate the scalability of the forward and adjoint operations of the well-known time-domain multi-dimensional convolution (MDC) operator with respect to increasing size of the input data and number of computational resources. Finally, we use this implementation to solve the Marchenko equations by means of least-squares inversion for a 3D synthetic dataset composed of up to 9801 sources and receivers.
-
-
-
An Objective Function Based on q-Gaussian Distribution for Full-Waveform Inversion
Authors S.L. Da Silva, C.A. Da Costa, P. Carvalho, J. Araújo, L. Lucena and G. CorsoSummaryFull-waveform inversion (FWI) is a wave-equation-based inversion method to estimate the physical parameters of the geological structures by exploiting full-information of the seismograms. However, FWI is inherently an ill-posed problem that is sensitive to noise, especially to outliers in the dataset. Usually, this technique is formulated as a least-squares optimization problem that consists of to minimize the difference between the observed and the modelled seismic data (residuals). In this approach, the least-squares solution inversion problem determines the maximum likelihood for the residuals, where all residuals are assumed to follow a Gaussian distribution. However, the distribution of residuals is seldom Gaussian for non-linear problems. In this study, we propose an alternative objective function to mitigate FWI sensitivity to noise based on the q-Gaussian probability distribution. In contrast to Gaussian distribution, the q-Gaussian distribution has long-tails, being less sensitive to outliers. Application on acoustic synthetic noisy-data illustrates the performance between our proposal and FWI based on least-squares norm L2. In addition, we compare also with robust objective functions based on Huber criterion and least-absolute-values norm L1. Numerical experiments show that FWI based on the q-Gaussian probability distribution outperforms other approaches, especially in presence of outliers.
-
-
-
Seismic Attribute-Guided Automatic Fault Prediction by Deep Learning
Authors F. Jiang and P. NorlundSummaryFault identification in seismic data is a vital but time-consuming step in the seismic interpretation workflow. Recent studies demonstrate how deep-learning techniques, such as convolutional neural networks (CNN), can be used to automatically identify these faults with high accuracy. However, different levels of signal-to-noise ratios in seismic data can degrade prediction accuracy. A low resolution of predicted faults can cause multiple issues, such as failing to identify potential drilling hazards. In this abstract, a workflow is developed to combine the seismic data with multiple seismic attributes to train machine-learning models using a multichannel CNN architecture. A random forest is implemented to analyse the selection of each attribute in terms of a feature importance factor. Several attributes with a high-importance factor are selected as additional channels to feed into the multichannel CNN architecture. A comparison of fault predictions between a probability map generated from a model trained by seismic-only and a model trained using seismic-plus-attributes is presented. The results exhibit significant improvement on the continuity of fault segments and reveal missing fault planes not identified using a seismic-only model. Additionally, a modified generative adversarial network is implemented to reconstruct the fault probability map to help improve the resolution.
-
-
-
An Effective True-Amplitude Gaussian Beam Migration via Illumination Compensation
More LessSummaryConventional migrations are apt to structure imaging, but often incapable of generating true-amplitude subsurface image. We propose a true-amplitude Gaussian beam migration (GBM) method under the framework of seismic illumination compensation. A novel scheme based on the GBM is developed to estimate the point spread functions, with which the illumination compensation can be efficiently implemented in the local wave-number domain. The total computational cost of the proposed true-amplitude imaging includes one Born modelling process and two conventional GBM processes, which is more efficient than the true-amplitude imaging using least squares migration which requires multiple iterations. Numerical examples using synthetic data demonstrated the effectiveness and efficiency of the proposed method.
-
-
-
A Method for Predicting Mercury Injection Capillary Pressure Curves Based on NMR Echo Data
More LessSummaryIn this paper, a new method for predicting the mercury injection capillary pressure (MICP) curves based on the nuclear magnetic resonance (NMR) echo data is proposed. We calculate multiple characteristic parameters of NMR echo data: porosity (ϕ), the area enclosed by the NMR echo data (S_echo) and the decay time of NMR echo data (t_echo), establish the relationship between the aforementioned parameters of NMR echo data and the mercury saturation (Snw) at each capillary pressure point of the MICP curve, thereby the prediction model of the MICP curves based on the NMR echo data is obtained. We use the established model to predict the MICP curves of 14 tight sandstone core samples and the results show that the method for predicting the MICP curves based on NMR echo data has smaller calculation amount and higher precision. This method can be used to predict the MICP curves continuously with depth of well logging, which lays a foundation for the evaluation of reservoir pore structure.
-
-
-
Polymer Augmented Low Salinity Brine for Mixing Control in Low Salinity Waterflooding
Authors A. Darvish Sarvestani, B. Rostami and H. MahaniSummaryLow-salinity waterflooding (LSWF) is a promising IOR/EOR methodology which has been extensively investigated over the past ten years. However, mixing of the injected brine (low-salinity) with the in-situ high salinity brine (formation water) at the displacement front can decelerate oil recovery by LSWF. We propose that one promising approach to control in-situ mixing is to increase the viscosity of the low salinity injection brine by adding polymer which then improves the mobility ratio at the displacement front and subsequently suppresses dispersion of low salinity in high salinity. This, to our knowledge, has not been addressed before. To investigate systematically the mixing phenomenon in different salinity gradients and its sensitivity to HPAM polymer concentrations, sandpack experiments were designed and executed. Two sets of mixing experiments under single-phase condition were performed to capture the impact of absence or presence of polymer in the low-salinity brine. Our results show that polymer could be used as a mixing-control agent and the dispersivity of the system may be decreased to lower than 0.3 of its original value by adding 200 ppm of polymer.
-
-
-
The Discrete Orthonormal S-Transform for Seismic Data Reconstruction Based on Compressive Sensing
By Z. ZhaoSummaryIn recent years, compressive sensing has been widely used in seismic data reconstruction. According to the limitations of different sparse transform methods of compressive sensing, we propose to apply discrete orthogonal S transform (Dost) in compressive sensing as a new sparse transform method. Dost has the good time-frequency analysis capability, and has better sparsity while reducing the redundancy of S transform. We obtain seismic data reconstruction results through the highly convergent fast projection onto convex set (FPOCS) algorithm. The reconstruction results are evaluated with multiple parameters and other transformation methods. Testing results of synthetic and real data verify the correctness and effectiveness of this method, which reconstruction accuracy is higher than that of Fourier and Shearlet transform.
-
-
-
Full Waveform Inversion Based on the Acoustic-Elastic Coupled Equation
More LessSummaryCurrent full waveform inversion (FWI) does not make full use of all components data: using only the pressure component (by acoustic FWI) or velocity component (by elastic FWI) data. Based on the acoustic-elastic coupled equation (AECE), we propose a multiple component FWI method to get P-wave and S-wave velocities simultaneously. We use both pressure and velocity component data to form the objective function. Using the adjoint-state method, we deduce the adjoint equation of the AECE and the gradient formula in time domain. Numerical experiments show that results of our method using both pressure and velocity component data are the best, using only velocity component data are the second, and using only pressure component data are the worst. Meanwhile, FWI based on the AECE achieves better results than traditional elastic FWI. This indicates that this approach is an effective way for model building in OBC/OBS applications.
-
-
-
Microseismic Hypocenter Location Using an Artificial Neural Network
Authors Q. Hao, U.B. Waheed, M. Babatunde and L. EisnerSummaryThe sharp increase in the occurrence of human induced earthquakes globally requires real-time source location capabilities, particularly in areas where no prior seismic activity occurred. Recent advances in the field of machine learning coupled with available computational resources provide a great opportunity to address the challenge. Researchers have started looking into using convolutional neural networks (CNNs) for hypocenter determination by training on already located seismic events. We propose an alternate approach to the problem. We train a feed-forward neural network on synthetic P-wave arrival time data (based on a velocity model or empirical data). Once trained, the neural network can be deployed for real-time location of seismic events using observed P-wave arrival times. The use of a feed-forward neural network allows fast training compared to CNNs. We show sensitivity of the proposed method to the training dataset (density and distribution of the training sources), noise in the arrival times of the detected events, and size of the monitoring network.
-
-
-
Wavefield Solutions from Machine Learned Functions that Approximately Satisfy the Wave Equation
Authors T. Alkhalifah, C. Song, U. Bin Waheed and Q. HaoSummarySolving the Helmholtz wave equation provides wavefield solutions that are dimensionally compressed, per frequency, compared to the time domain, which is useful for many applications, like full waveform inversion (FWI). However, the efficiency in attaining such wavefield solutions depends often on the size of the model, which tends to be large at high frequencies and for 3D problems. Thus, we use a recently introduced framework based on predicting such functional solutions through setting the underlying physical equation as a cost function to optimize a neural network for such a task. We specifically seek the solution of the functional scattered wavefield in the frequency domain through a neural network considering a simple homogeneous background model. Feeding the network a reasonable number random points from the model space will ultimately train a fully connected 8-layer deep neural network with each layer having a dimension of 20, to predict the scattered wavefield function. Initial tests on a two-box-shaped scatterer model with a source in the middle, as well as, a layered model with a source on the surface demonstrate the successful training of the NN for this application and provide us with a peek into the potential of such an approach.
-
-
-
Down/down deconvolution
Authors G. Hampson and G. SzumskiSummaryImages constructed from up-coming wavefields can be very effectively deconvolved using up/down deconvolution, however, they have geometry related drawbacks that reduce the quality of the shallower zones. In contrast, the down-going wavefield, which is imaged using mirror imaging, does not suffer from such geometry related disadvantages, however, it lacks a powerful deconvolution technique akin to up/down deconvolution. We use a modified Delft feedback model to describe the up- and down-going scattered wavefields. Using these results, we illustrate how up/down deconvolution works and then go on to introduce a new idea called down/down deconvolution. This new technique inverts the down-going wavefield for the Earth’s response in the absence of a free-surface. The free-surface multiples are removed and the 3D source wavefield is deconvolved to produce a result that is theoretically the same as up/down deconvolution. As a result we can combine the geometrical advantages of the down-going wavefield and the benefits of a powerful deconvolution technique. We illustrate this new idea using a synthetic dataset and a real 3D OBN dataset from the North Sea.
-
-
-
Processing and imaging of a multi-petabyte OBN survey in the North Sea
Authors T. Rayment, G. Hampson and L. LetkiSummaryThe Utsira ocean-bottom node (OBN) survey was acquired during 2018/19 covering 1510 km2 over the Utsira high, which is ∼90 nautical miles west of Stavanger. As it was acquired using several asynchronous triple-source vessels, deblending was an essential step to recover signal at target depths. Deblending also removed very strong nearby seismic interference, by using the interfering survey’s shot times, with as many as 14 active sources firing.
Up-down deconvolution is a key step to address the challenges observed in the area as it achieves both free-surface multiple elimination and 3D signature deconvolution. A similar approach was developed to tackle the same issues for the down-going wavefield resulting in a more effective and efficient processing flow for imaging shallow targets than conventional techniques.
The long offsets and rich azimuthal sampling of the wavefield meant FWI could solve model building and imaging challenges over a range of depths. Such challenges included shallow channels and deeper cemented sand injectites.
This survey illustrates how advances in acquisition and processing technologies enable large-scale, high-density OBN surveys to be acquired and processed in accelerated time frames leading to new insights even in relatively well-explored areas.
-
-
-
P-S Separation from Multi-Component Seismic Data Using Deep Convolutional Neural Networks
More LessSummaryThe accurate separation of single-mode waves from multi-component seismic data is of great significance for elastic imaging and inversion. Traditional separation methods require accurate velocity information and do not perform well on the far offset. In this paper, we propose to use deep convolutional neural networks for P-S separation tasks. We design a training and testing workflow that can handle arbitrary seismic data size. We train the model on one synthetic dataset and directly evaluate the trained model on another without re-training or fine-tuning process. Our results indicate that the proposed method can capture polarization information from the data and perform well on both near and far offset, without providing near surface velocity model. The proposed method can easily extend to 3D or anisotropic media.
-
-
-
SSGST-Based Prestack Fluid Mobility Calculation Method and Its Application
Authors J. Liao, H.D. Huang, F. Xu, J. Zeng and X.D. TianSummaryFluid mobility attributes extracted from poststack low-frequency seismic data have exhibited some potential for reservoir characterization and fluid identification. Compared to poststack seismic data, prestack seismic data contains more information about reservoirs and fluids. In order to extract fluid mobility information from prestack seismic data, we establish the relationship between fluid mobility and incidence angle based on frequency-dependent AVO analysis. We define the relationship as prestack fluid mobility which means that fluid mobility varies with angle/offset (FVA/FVO). By establishing models of reservoirs with different fluids, the corresponding prestack fluid mobility is estimated. The results show that prestack fluid mobility can distinguish the oil-bearing reservoir (class III AVO) from the water-bearing reservoir (class IV AVO). In order to estimate prestack fluid mobility, we further derive a Synchrosqueezed Generalized S-transform-based prestack fluid mobility calculation method. This method estimates the FVO by the difference in fluid mobility of the different partially stack gathers. The method is applied to the seismic data. Compared with poststack fluid mobility, FVO can better identify fluid properties and reduce the uncertainty of hydrocarbon-bearing reservoir prediction in the absence of wells.
-
-
-
Elastic Anisotropy of Transversely Isotropic Rocks Containing Aligned Cracks and Applications to Experiment and Field Data
More LessSummaryRock physics models provide the basis for evaluating the elastic properties of cracked/fractured media. A sphere-equivalency method of elastic wave scattering was developed to accurately calculate the elastic properties of a transversely isotropic solid containing aligned cracks. To validate the validity and accuracy, the theory was applied to a recent experiment made with a VTI medium containing cracks and shows significantly better agreement with the data. For a more realistic situation, the method was furtherly applied to interpret the borehole acoustic anisotropy measurement, showing that the theory can adequately explain the anisotropic characteristics of the field data.
-
-
-
Multichannel Blind Acoustic Impedance Inversion Based on 2D-TV Regularization
More LessSummaryWe presented a multichannel blind acoustic impedance inversion method. The method is an extension of the principle of Euclid deconvolution, which can make full use of the information of multi-trace seismic data to obtain acoustic impedance without estimating seismic wavelet. Significantly, the 2D total variation (TV) constraint is added to the cost function to suppress the random noise and keep the boundary characteristics of the stratum. Also, to obtain the absolute impedance, the low-frequency information extracted from the well logs is used to supplement low-frequency component of the inversion result. Finally, to demonstrate the effectiveness of the proposed method, we apply the method to synthetic data and field data, and confirm that the proposed method can achieve credible acoustic impedance.
-
-
-
An Adaptive Demultiple Method Based on Inversion of Two-Dimensional Nonstationary Filter
More LessSummaryThis study uses the separability of the primary and multiple waves in the Radon domain to invert the filter coefficients at each point in the space-time profile, thereby suppressing the multiples using a two-dimensional nonstationary filtering technique. Compared with the parabolic Radon transform, it does not need to perform the inverse Radon transform, and alleviates the truncation effect caused by clearing the data in the Radon domain. Compared with two-dimensional nonstationary filtering, the uncertainty and subjectivity of filter design are avoided. Therefore, it is not just a simple combination of parabolic Radon transform and two-dimensional non-stationary filtering. Synthetic and field data examples show that this method has better ability of demultiple and amplitude preserving.
-
-
-
Seismic Processing with Deep Convolutional Neural Networks: Opportunities and Challenges
More LessSummaryDeep convolutional neural networks (DCNNs) are growing in popularity in seismic data processing and inversion due to their achievements in signal and image processing. In this paper we explore the link between DCNN and seismic processing. We demonstrate the potential of the application of DCNNs to seismic processing by analysing its performance with data deblending as an example. We discuss challenges and issues to solve before deploying DCNNs to production, and suggest some directions of study.
-