Abstract (
)
Download PDF (
)
HTML
(
)
Knowledge map
Save
Due to the advancements of seismic data acquisition and imaging with "wide band, wide azimuth, and high density", thin layers, small-scale (fractured-vuggy) anomalous bodies, and small-throw faults (strike-slip faults), etc. have become the main target geological bodies for high-fidelity and high-resolution imaging. In current seismic exploration, reasonable observation system design has received sufficient attention, but how to achieve customized high-resolution wavelets for the customers is rarely discussed. In theory, vibroseis is a method of obtaining the amplitude spectrum of expected seismic wavelet through the accumulation of single-frequency energy within time intervals in the frequency and time domain. Based on this, we propose the concept of customized high-resolution seismic wavelets and corresponding sweep signals for the customers. The first step is to generate (or customize) a zero-phase wavelet, which meets the requirements of high-fidelity and high-resolution on a specific target layer, and obtain its corresponding amplitude spectrum. The second step is to establish the mapping relationship between the amplitude spectrum of the customized wavelet and the vibrator sweep signals. Finally, the vibrator sweep signals are designed under the assumption of linear sweep frequency. We use formula derivation and model tests to demonstrate the feasibility of our method. Forward modeling and migration tests also show that the customized high-resolution wavelet could be used in high-resolution processing.
Abstract (
)
Download PDF (
)
HTML
(
)
Knowledge map
Save
Random noises contaminate seismic signals and reduce the signal-to-noise ratio of seismic data, which will affect subsequent seismic data processing. A denoising method based on supervised deep learning usually requires a large number of labels to train the network, but it is very challenging to make noise-free labels using observed seismic data. To attenuate random noises and extract useful signals from multi-dimensional seismic data, we propose an end-to-end neural network based on unsupervised deep learning, which consists of a fully connected module, an encoder module and a decoder module. A skip connection similar to a residual structure is added between the encoder and decoder to improve the performance of denoising. To strengthen the network further, a data enhancement method is used to segment large-scale multi-dimensional noisy seismic data into a large number of small-scale one-dimensional data for iteration. Appropriate slicing and sliding sizes for data enhancement could improve the calculation efficiency and denoising effect of the network. The application to synthetic data and actual data acquired in Bohai oilfield shows that the proposed method is better than traditional denoising techniques in random noise attenuation and signal extraction.
Abstract (
)
Download PDF (
)
HTML
(
)
Knowledge map
Save
Noise suppression is an important research topic in seismology and seismic signal processing. Accurately suppressing seismic noises and extracting effective signals is a key step in seismological research and seismic monitoring. Traditional denoising methods have some shortcomings, such as insufficient flexibility, difficulty in dealing with complex noises, information loss, and dependence on manual feature extraction. In order to overcome these shortcomings, this paper probes in a method of time-frequency domain transform combined with deep learning and its application to noise reduction. Five neural network models (including FCN, Unet, CBDNet, SwinUnet and TransUnet) are constructed for noise suppression after the time-frequency transformation of seismic data. This paper introduces three indicators: peak signal-to-noise ratio (PSNR), structural similarity index (SSIM) and root mean square error (RMSE) for quantitative evaluation of denoising performance. Numerical experiments show that the convolutional neural network(CNN) method based on time-frequency transform can effectively suppress common noise types (including random noises, swell noises and surface waves) and improve the signal-to-noise ratio of seismic data. The introduction of the Transformer module can further reduce above-mentioned noises and enhance the denoising performance of the CNN model. Further research will focus on an improved network structure for more complex seismic signals and the combination with other advanced techniques to improve denoising performance.
Abstract (
)
Download PDF (
)
HTML
(
)
Knowledge map
Save
The real earth media is characterized by the visco-elasticity, and it is of practical importance to study the full waveform inversion method in visco-elastic media. The full waveform inversion (FWI) is used to estimate the subsurface parameters by matching the simulated data obtained by the forward simulation with the actual recorded seismic data, and the optimization algorithm is generally used to solve the problem. To address the problems of complicated Hessian matrix calculation and slow convergence speed of conjugate gradient method in full waveform inversion, the modified FR conjugate gradient method is applied to the full waveform inversion. The modified FR conjugate gradient method uses more gradient information and has similar structure and properties as the quasi-newton algorithm, which makes the convergence more stable while speeding up the convergence speed and needs little more calculation. This paper use this method conduct a study of the time-domain visco-acoustic multiparameter full waveform inversion under the SLS model. The model test show that the modified method can obtain more accurate inversion results than the conventional FR conjugate gradient method, accelerate the convergence and resolution of deep layer. And the two-step inversion strategy is used to further improve the inversion effect of Q.
Abstract (
)
Download PDF (
)
HTML
(
)
Knowledge map
Save
Reverse time migration (RTM) based on the two-way wave equation may yield high-precision images of complex subsurface structures. However, RTM is sensitive to the velocity model. This paper proposes a new dip-constrained RTM method. The forward propagating wavefield from the source and backward propagating wavefield from the receiver are decomposed into different directional components, and then an automatic dip-constrained imaging condition is utilized to improve imaging quality in the presence of velocity errors. Meanwhile, our method may mitigate the artifacts caused by non-primary reflections, even when the velocity model is correct, by using the reflection signals within the Fresnel zone. To increase efficiency, we introduce the frequency-domain snapshot to compute the incident angle for source wavefield decomposition. Numerical tests on a complex velocity model show improved imaging in the case of inaccurate velocities and mitigated artifacts caused by multiple reflection waves in the case of the accurate model.
Abstract (
)
Download PDF (
)
HTML
(
)
Knowledge map
Save
The particle motion in a shear wave is perpendicular to the direction of propagation. An incident shear wave propagating in fracture-induced anisotropic media will split into a fast shear wave and a slow one, which is referred to as shear-wave splitting. The time delay between the fast and slow waves, which is generally proportional to fracture density, may be calculated using two methods: manual horizon picking and cross correlation. Due to the low signal-to-noise (S/N) ratio of shear-wave data, any single method cannot yield desired results. Therefore, we propose a horizon-based method for precise calculation. Large-scale time delays will be calculated first based on marker horizon tracking. For secondary horizons between two makers, a dynamic time warping (DTW) algorithm is used to calculate small-scale time delays. The time delays of both scales would then be combined for the time correction of slow shear waves. A case study confirms the effectiveness of the method for shear-wave data with low S/N ratio.
Abstract (
)
Download PDF (
)
HTML
(
)
Knowledge map
Save
Despite its extensive use in geophysical exploration, OBN data suffer from low signal-to-noise ratio caused by Z component contaminated by shear waves from horizontal components. Such leakage noises have a negative impact on dual-sensor merging based on Z component to obtain high-quality up-going and down-going waves for imaging, but it is hard to separate useful signals from leakage noises by using common filtering and matching attenuation algorithms. To suppress shear-wave leakage noises, we propose a matching attenuation method based on curvelet-domain extended filtering. The method constructs Hilbert transform records, time derivative records, and Hilbert transform followed by time derivative records from the X and Y component data of OBN to predict shear-wave leakage noises in the Z component, which enables the extended expression of Z component in the curvelet domain. Shear-wave leakage matching subtraction will then be performed using curvelet-domain least-squares extended filtering to separate effective signals from leakage noises. According to a model test and field data processing, our method has the advantage of leveraging curvelet transform for signal-noise separation and extended filtering for shear-wave leakage error prediction. Consequently, OBN data imaging will be improved because shear-wave leakage noises could be eliminated significantly without damaging effective signals.
Abstract (
)
Download PDF (
)
HTML
(
)
Knowledge map
Save
Noises in land seismic data, which can be classified into linear and nonlinear coherent noises, incoherent noises, and random noises, may come from near-surface formations, externally sourced wavefield, and other sources. The basic idea of noise suppression is to establish a model to predict signals or coherent noises and then remove coherent, random, and incoherent noises from seismic data. Some high-precision imaging methods, such as full waveform inversion and least-squares reverse time migration, should be accomplished using seismic data with high signal to noise ratios. This paper presents an overview of denoising theories and techniques, and develops a conceptual model of seismic data to show that signals or coherent noises with linear and/or nonlinear structures float in random noises that satisfy a certain probability distribution. Based on the conceptual model, various denoising methods are discussed, including AR model predictor, linear Radon transform, K-L transform, and Hankel matrix for the prediction of linear signals or coherent noises and Radon transform and polynomial fitting for the prediction of nonlinear (hyperbolic) signals or coherent noises. The fundamental point of denoising is optimal signal modeling, which is the basic idea for most denoising methods in commercial processing systems. The comparative analysis in this paper provides further insights on denoising theories to improve the effect of data processing.
Abstract (
)
Download PDF (
)
HTML
(
)
Knowledge map
Save
In China, seismic surveys deal with increasingly small and complex targets. Improved seismic resolution requires the downsizing of underground sampling grids. Conventional regular sampling methods are extremely costly. Compressed-sensing irregular sampling can design non-equally spaced shot and receiver points, without increasing investment, to obtain a uniform discrete distribution of CMP points and an irregular 3D data volume. The regularized reconstruction of irregular data with higher density has become a key issue in imaging. There are various reconstruction methods, most of which cannot balance accuracy and efficiency. Based on the compressed sensing theory, this paper uses a reconstruction method based on 3D curvelet transform, which can effectively capture the anisotropic and orientation features of seismic events for their optimal sparse representation. An algorithm of projection onto convex sets (POCS) is introduced to improve reconstruction accuracy. An optimization strategy with f-x domain conversion and OpenMP parallel acceleration is used to improve computational efficiency. This method realizes the reconstruction of irregularly acquired data with high density, high efficiency, and high precision based on compressed sensing. The application to the Guangli-Qingnantan shallow sea survey in Shengli Oilfield shows that the proposed method has high accuracy, high computational efficiency, and better imaging with improved resolution than a conventional regularly sampled high-density survey.
Abstract (
)
Download PDF (
)
HTML
(
)
Knowledge map
Save
Ringing and bubble noises common in shallow water seismic exploration may lead to strong low-frequency reverberations with periodic increase and decrease in envelope amplitude in seismic data. Conventional deterministic deconvolution using far-field wavelets and predictive deconvolution do not have good performance for suppressing such noises. The Ricker-compliant deconvolution method proposed by Claerbout is a paradigm for suppressing bubble effects in offshore seismic data. Based on the Claerbout's theory, this paper proposes a lag-log domain filter to suppress low-frequency reverberations in beat envelope amplitude, redefines the medium-time-delay signal in the lag-log domain, and divides the data into small-lag signals, medium-lag signals and large-lag signals. According to different signal distributions, a Kolmogorov filter is constructed to remove medium- and large-lag signals, or reverberations and bubbles. Testing results show that lag-log domain filtering gets a good result in suppressing low-frequency reverberations in beat envelope. A comparative study of real data processing shows that lag-log domain filtering performs better than several other methods in the suppression of low-frequency reverberations.
Abstract (
)
Download PDF (
)
HTML
(
)
Knowledge map
Save
Common horizon identification methods based on deep learning primarily focus on seismic amplitude without sufficient attention to the spatial relationship among horizons of different scales, resulting in discontinuous and even inaccurate interpretation. To address this problem,we propose a method based on the multi-scale attention UNet++, termed MR_CBAM_UNet++, which involves MultiResBlock to extract a broader spectrum of horizon scale features, CBAM to reduce the amplitude interference of non-target signals, and a UNet++. A joint loss function composed of Focal Loss and Dice Loss is utilized for network training, and the uniqueness constraint is incorporated to refine the results of horizon identification.According to its application to actual seismic data, the MR_CBAM_UNet++ model shows significantly improved capabilities compared to traditional models in suppressing non-horizon information and identifying horizons in complex subsurface conditions.A mean pixel accuracy rate of 86.19% is achieved for the test dataset,indicating more accurate horizon interpretation with better continuity. Additionally, the results of horizon identification are more geologically significant by using the uniqueness constraint.
Abstract (
)
Download PDF (
)
HTML
(
)
Knowledge map
Save
It is difficult to use a vertical transverse isotropy (VTI) and a horizontal transverse isotropy (HTI) model to characterize fractured shale gas reservoirs with horizontal bedding and several groups of parallel high-angle fractures. However, an orthorhombic anisotropy (OA) model is a feasible alternative. Based on the anisotropic characteristics of high-angle fractures described using the reflectivity equation, an indicator that represents the intensity of high-angle fractures is proposed. We derive an azimuthal elastic impedance equation containing a fracture indicator for OA media from the approximate equation of OA reflection coefficient and an equivalent OA parameter. We also establish a method to invert fracture indicator and fracture azimuthal elastic impedance. To improve the stability of the inversion, the derived azimuthal elastic impedance equation is rewritten in a two-parameter form. A model-based iterative least-squares inversion is used to achieve the robust estimation of fracture indicator and fracture azimuth. A model test demonstrates the reliable estimation of fracture indicator and fracture azimuth even at the signal-to-noise ratio of 2. A practical application also yields reliable predictions. This method is a useful tool to predict shale gas reservoirs with high-angle fractures.
Abstract (
)
Download PDF (
)
HTML
(
)
Knowledge map
Save
Subsurface formations with aligned dipping fractures can be equivalent to the media with tilted transverse isotropy (TTI). For TTI-type organic-rich shale gas reservoirs, the Young's modulus is not a sensitive brittleness indicator as a low value may emerge at a high degree of brittleness. Therefore, a hierarchical seismic inversion method is proposed to directly predict the brittleness and fracture parameters of TTI-type organic-rich shale gas reservoirs. We derive a linearized P-wave reflection coefficient equation for TTI media in terms of a new brittleness indicator formulated as the function of the Young's modulus and Lamé constant (λ), fracture density and fracture dip. The reflection coefficient equation is rewritten in the form of the Fourier series to separate fracture information from background matrix information. Based on the Bayesian framework, the hierarchical inversion method is established to predict brittleness indicator and fracture parameters. The process is composed of three steps. Firstly, we utilize the discrete Fourier series to decouple the isotropic (zero-order) and anisotropic (second-order) components of azimuthal seismic data. Secondly, we invert fracture density and fracture dip from the second-order component of azimuthal data. Thirdly, the brittleness indicator can be estimated using the zero-order component. The model test and practical application indicate the robustness of the method and credible prediction of high-quality brittle reservoirs.
Abstract (
)
Download PDF (
)
HTML
(
)
Knowledge map
Save
In order to comprehensively promote the application of deep-water reservoir prediction technology in the Kaiping sag, the Pearl River Mouth Basin, we conduct a study on stratigraphic division and depositional system, together with lithologic and petrophysical properties in the Enping Formation, based on 3D seismic, drilling, and logging data. The log-constrained sparse spike inversion is utilized to determine the spatial distribution of favorable sandstone reservoirs. Combined with the seismic multi-attribute analysis method, multiple attributes with varying sensitivities to seismic information were extracted along the layer. Three attributes with high correlation coefficients were selected for weight fusion. Drilling data reveal that deeply buried Eocene formations in the Kaiping sag consist of interbedded sandstones and mudstones. Sandstones in the Enping Formation have better reservoir properties, i.e. higher porosity and higher permeability, than in the Wenchang Formation. According to sedimentation, reservoir properties, seismic inversion results, and seismic attribute analysis, deltaic sandstones with good continuity and high impedance in the Enping Formation may function as promising reservoir rocks. The northwest of KP11 tectonic zone is predicted to be a promising area for drilling. A favorable drilling well location is proposed aiming to establish a theoretical direction for future oil and gas exploration.
Abstract (
)
Download PDF (
)
HTML
(
)
Knowledge map
Save
Offset and azimuth attributes have obvious advantages in detecting complex geological anomalies such as faults and channels. To improve information mining and result presentation, we carry out the study of Locally Linear Embedding(LLE)-based dimensionality reduction and technical process for prestack seismic attributes based on the fundamentals and technical features of the classic non-linear dimensionality reduction algorithm, LLE. Using the LLE algorithm, prestack seismic attributes in a high-dimension space extracted from partial stacks will be mapped into a dimension-reduction attribute volume in 3D space. As per a case study in Bakken field, the LLE non-linear technique is superior to Principal Component Analysis (PCA), a linear algorithm, in dimensionality reduction and feature preservation. Another practical application shows that the LLE technique can realize dimensionality reduction and fusion of prestack seismic attributes for accurate identification and characterization of geological anomalies.
Abstract (
)
Download PDF (
)
HTML
(
)
Knowledge map
Save
Thin coal seams in the Pinghu Formation, Pingbei slope zone of the Xihu sag, cause strong seismic reflections and geologic misinterpretation. These strong reflections have a remarkable negative impact on reservoir prediction, including lateral extension and vertical thickness prediction, through seismic inversion. Based on different reflection coefficients of coal-mud-sand associations, we propose an angle-gather conditioning technique to identify and eliminate strong reflections and thus weaken the effect of coal seams. Gather conditioning suppresses strong reflections of coal seams and side lobes and highlights effective signals of sandstone reservoirs. A prestack inversion shows improved reservoir prediction with decreased uncertainties, which offers support to development well drilling and post-drilling evaluation of the Pinghu Formation in the slope zone.
Abstract (
)
Download PDF (
)
HTML
(
)
Knowledge map
Save
In the lake-delta depositional system, tight sandstone reservoirs are characterized by complex pore structures, diverse pore types, and low permeability, and log interpretation and formation evaluation is facing the challenges. Permeability is a key parameter for reservoir evaluation and productivity prediction, and traditional calculation methods from log interpretation are not accurate and cannot meet production requirements. Aiming at this problem, two closely related controls on reservoir permeability are analyzed: microscopic pore structure and macroscopic flow unit, and a new permeability prediction method based on rock type and flow zone indicator (FZI) are proposed. First, core experimental results are analyzed, rock types are determined, core FZIs are calculated, rock types are classified using the cumulative frequency method, and the permeability model for each rock type is constructed. Then, sensitive well logs are selected to form labels, and a deep neural network (DNN) is used to predict reservoir flow unit index (FZI). Finally, log porosities and FZIs are input into the model for each type to calculate permeability. The application in low-porosity low-permeability reservoirs in the HG Formation, the XH sag, China, shows good results with logarithmic error of 0.18, which is smaller than of other DNN methods. The new method includes both data-driven machine learning methods and mechanism-based or knowledge-driven physical model construction, which embodies the idea of data and model jointly driven intelligence, and significantly improved the accuracy of permeability evaluation of tight sandstone reservoirs. Furthermore, it is also referential to permeability prediction for tight sandstone reservoirs in other lake-delta sedimentary systems.
Office Online Journal Online
Online Submission