Field Seismic Data Research Articles

Depthwise separable convolution Unet for 3D seismic data interpolation

In seismic exploration, dense and evenly spatial sampled seismic traces are crucial for successful implementation of most seismic data processing and interpretation algorithms. Recently, numerous seismic data reconstruction approaches based on deep learning have been presented. High dimension-based methods have the benefit of making full use of seismic signal at different perspectives. However, with the transformation of data dimension from low to high, the parameter capacity and computation cost of training deep neural network increase significantly. In this paper, we introduce depthwise separable convolution instead of standard convolution to reduce the operation cost of Unet for 3D seismic data missing trace interpolation. The structural similarity (SSIM), L1 hybrid loss function, and switchable normalization further improve the reconstruction performance of the network. The comparative experiments on the synthetic and field seismic data show that depthwise separable convolution can effectively reduce the number of network parameters and computation intensity with the interpolation results comparable to the standard convolution results.

3-D Seismic Fault Detection Using Recurrent Convolutional Neural Networks With Compound Loss

Fault detection is an essential component of seismic interpretation and plays a crucial role in industrial processes. However, it is also one of the main challenges, especially in delineating faults in 3D seismic data. Recently, the rapidly developing technology, deep learning, has proven to be a powerful tool for this task. A number of neural networks have been proposed for this purpose by regarding 3D fault detection as a semantic segmentation task. To further enhance the effectiveness of the deep learning methods, we propose a novel network architecture, named R2SE-Unet, to solve the 3D segmentation problem. In the neural network, we design a recurrent residual-SE convolution unit (RRCU-SE) that integrates the residual learning and Squeeze-Excitation module to store information in 3D seismic data. This component promotes the spread of 3D volumetric information and aids in learning spatial dependencies in 3D images. In addition, to reduce the impact of insufficient spatial resolution resulting from the base architecture of U-net, we add an attention unit between skip connection operations. These two new units enable our R2SE-Unet to exploit semantic information more accurately in the feature maps. After many experiments on region-based loss functions and distribution-based loss functions, we also propose a novel loss function, which takes the advantage of generalized dice (GDice) loss and balanced binary cross entropy (b-BCE) loss, named Gdice-bce, to effectively train R2SE-Unet. Although only synthetic seismic data samples are used to train the network parameters, our R2SE-Unet could produce more reliable fault feature maps on field seismic data than two other conventional fault detection neural networks. Thus, the proposed neural network is easy to train and reliably works for seismic fault interpretation on field seismic data.

Unpaired training: Optimize the seismic data denoising model without paired training data

With the development of seismic exploration technology, distributed acoustic sensing (DAS) has recently received attention in geophysics. However, owing to the complexity of the layout techniques in the DAS systems, and the unknown or harsh exploration environment, seismic data acquired by this technique usually contain the noise of diverse components, which increases the difficulty in subsequent data analysis and interpretation. This study has (1) trained a deep learning model that effectively suppressed noise with augmented noise data sets to obtain a high signal-to-noise ratio in the DAS vertical seismic profile (VSP) records, (2) introduced an attention module to enhance the extraction and recognition of signal features to recover effective signals under substantial noise interference, and (3) introduced adversarial loss and cycle-consistent loss to replace the commonly used L1 norm or L2 norm to train the network. The obtained hybrid training set containing unpaired synthetic and unpaired field data sets for model pretraining and fine tuning effectively improves the denoising performance of the seismic field data. In summary, this study develops an unpaired training-based DAS seismic data denoising method that transformed noisy DAS VSP data into noise-free data. By analyzing the noise suppression results of other methods, including qualitative and quantitative analyses, we demonstrate that our method successfully suppressed multiple types of noise in the DAS VSP data. The study has indicated clear and continuous signals in the denoising results and improved the denoising performance on the seismic field data.

Facies-Guided Seismic Image Super-Resolution

Seismic field data usually suffer from low-resolution (LR) image quality. In recent years, high resolution (HR) seismic data acquisitions are either done in high-density seismic surveys, or by capturing high- and low-frequency bands in broadband seismic surveys. However, both methods increase the financial, computational, and memory overhead required for oil and gas exploration. Moreover, to improve the resolution of already acquired seismic field data in a certain area, another survey would be carried out in the same area with more advanced equipment. This is a time-consuming high-cost approach. In our previous work, we showcased the application of convolutional neural networks (CNN) for image super-resolution (SR) to increase the resolution of seismic images. While our application of SR for seismic images is optimized for better human perception, automatic seismic interpretation tasks may not benefit as much, such as facies classification considered in this work. We address this by proposing a novel loss function that incorporates facies information when training the SR model. The proposed loss function is applied to multiple SR models and results show that this facies-guided seismic image SR approach produces improved facies classification results while mitigating major sacrifices on the objective and subjective image quality of the reconstructed SR image.

Attention and Hybrid Loss Guided 2-D Network for Seismic Impedance Inversion

Deep learning (DL) methods, especially Convolutional Neural Networks (CNNs), achieve state-of-the-art performance on seismic impedance inversion. Most of the methods are based on one-dimensional (1D) convolution, tending to yield lateral discontinuities of impedance on field data applications. To alleviate this problem, we design a network equipped with two-dimensional (2D) convolutions and a coordinate attention (CA) block. The former can take the relationship between adjacent traces into consideration. The latter can capture the positional relationship of the geological structure, both horizontally and vertically. At the same time, we use a hybrid loss combined with edge operator and Mean Square Error (MSE) to further improve the stability of the designed network. Comparison experiments on synthetic SEAM model and field seismic data demonstrate the effectiveness of the adopted components, 2D convolution, CA and hybrid loss function, in improving the lateral continuity of inverted impedance. For field seismic data, the impedance predicted by proposed method shows improved lateral continuity and high resolution compared with the 1D network and Constrained Sparse Spike Inversion (CSSI) method using commercial software (InverTrace Plus module in Jason).

Coherent Noise Attenuation by Kurtosis-Guided Adaptive Dictionary Learning Based on Variational Sparse Representation

In seismic data denoising, coherent noise is usually challenging to remove because it has similar features with signal. Therefore, attenuating coherent noise is of great importance for seismic data. In order to obtain a good denoising result, it is necessary to fully consider the features of coherent noise and choose the appropriate denoising methods. Here, we propose a kurtosis-guided adaptive dictionary learning (KGADL) algorithm based on variational sparse representation model. First, the variational sparse representation is to construct initial dictionary that depends on the seismic data, so that it can accurately contain the features of the seismic data, so as to improve the accuracy of sparse representation. Then, we use the K-singular value decomposition (K-SVD) algorithm to update the dictionary and introduce kurtosis to measure each atom after updating. Atoms with high kurtosis values usually exhibit strong irregularities and are considered noise, while atoms with low kurtosis values have more feature of valid waves and are considered signal. The atoms with low kurtosis values characterizing valid signal are retained to get the special dictionary that adapts to the complexity of the input seismic data, and this dictionary is used to suppress the coherent noise in seismic data. Finally, the denoising performance of the proposed method is illustrated by examples of both synthetic and field seismic data.

Hilbert–Huang Transform Incorporating Analytical Mode Decomposition and Its Application in Fractured-Vuggy Carbonate Reservoir Prediction

High-precision time-frequency (TF) analysis (TFA) can accurately detect anomalies in oil and gas reservoirs. However, conventional TFA methods cannot effectively identify fractured-vuggy carbonate reservoirs, which are deep and strongly heterogeneous. In this paper, a TFA method based on analytical mode decomposition (AMD) and the Hilbert-Huang transform (HHT) is proposed. Specifically, seismic signals are decomposed by empirical mode decomposition (EMD) and then by AMD. Finally, the TF distribution of the seismic signals is obtained after Hilbert spectrum analysis (HSA). Synthetic seismic data demonstrate that the AMD-HHT method can effectively eliminate mode mixing and provide higher time and frequency resolutions and energy focus than conventional TFA methods. In addition, model and field seismic data also verify that the AMD-HHT method can effectively detect hydrocarbon-related energy anomalies hidden in broadband seismic data.

Statistical Synchrosqueezing Transform and Its Application to Seismic Thin Interbed Analysis

Synchrosqueezing transform (SST) benefits from an instantaneous frequency (IF) estimator in the time-frequency domain, providing an energy-concentrated time-frequency representation to describe the time-varying frequency of seismic signals. To enhance the concentration performance of SST, this paper theoretically proposes a statistical SST (SSST) by constructing a spectrum-weighted IF estimator in the short-time Fourier transform (STFT) domain. In this SSST, a linear chirp signal is introduced to better capture its chirp rate, and then the quadratic STFT spectra are weighted by a window-related weighting function. On this basis, two equations related to the IF and chirp rate are constructed to derive the spectrum-weighted IF estimator. Finally, the STFT coefficients are squeezed to the estimated IF trajectories by a frequency fixed-point iterative algorithm, thereby providing a more concentrated time-frequency representation for multi-component signals than existing advanced methods, while enabling to retrieve each component. Two synthetic examples and one field seismic data on thin interbeds are utilized to demonstrate the effectiveness of the proposed SSST and show its ability to highlight the time-varying frequency features of seismic signals, which is a promising seismic data analysis tool, such as characterizing thin interbed thickness variations.

AVO Inversion Based on Closed-Loop Multitask Conditional Wasserstein Generative Adversarial Network

Neural networks are commonly used for post-stack and pre-stack seismic inversion. With sufficient labelled data, the neural network-based seismic inversion results are more accurate than that use traditional seismic inversion methods. However, in the case of insufficient labeled data, the accuracy of neural networks-based seismic inversion results decreases and is even lower than those based on traditional inversion methods. In addition, the seismic inversion results based on neural networks generally suffer from lateral discontinuity. It further reduces the accuracy of the inversion results. To tackle these problems, we propose a pre-stack seismic amplitude variation with offset (AVO) inversion method based on Closed-Loop Multi-task conditional Wasserstein Generative Adversarial Network (CMcWGAN), which is a GAN-based AVO inversion method. CMcWGAN enables simultaneous and accurate inversion of P-wave velocity ( <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">V_p</i> ), S-wave velocity ( <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">V_s</i> ), and density ( ρ ). Moreover, it uses the low-frequency information of elastic parameters as a conditional input to alleviate the problem of lateral discontinuity in inversion results. Experimental results of simulated data show that the inversion results based on CMcWGAN have higher accuracy than those based on traditional AVO inversion methods. In addition, when the seismic angle gather is noisy, CMcWGAN has better robustness than the traditional methods. CMcWGAN can also obtain reasonable AVO inversion results in field seismic angle gather data.inversion results. Experimental results of simulated data show that the inversion results based on CMcWGAN have higher accuracy than those based on traditional AVO inversion method. In addition, when the seismic angle gather is noisy, CMcWGAN has better robustness than traditional method. CMcWGAN can also get reasonable AVO inversion results in field seismic angle gather data.

A Self-Supervised Denoising Method Based on Deep Noise Estimation

Seismic random noise attenuation is an essential procedure when processing seismic data. Due to various acquisition environments and complex geological conditions, random noise in seismic field data exhibits spatiotemporal levels, significantly increasing the difficulty of extracting seismic signals. The deep learning (DL) methods have shown excellent performances for seismic denoising. However, most existing discriminative DL methods cannot fit field data with noise levels and signal structures that differ from the training set. To tackle the unmatching challenge, we propose a self-supervised deep denoising model named noise estimation-based convolution neural network (NE-CNN), which contains a multiscale denoising module as a pretrained model and a dual-path noise estimation module that estimates the noise level of each data patch with a generalized Gaussian distribution and gray-level co-occurrence matrix (GLCM). With Stein’s unbiased risk estimate (SURE), we can fine-tune the pretrained denoising module according to the estimated noise levels in a self-supervised style solely on data to be processed, leading to a boost of robustness to nonstationary seismic noise. In synthetic and field data tests, NE-CNN performed satisfactorily compared with discriminative DL methods in denoising effects on seismic data with nonstationary noise.

Seismic Stratigraphic Interpretation Based on Deep Active Learning

Seismic stratigraphic interpretation plays an important role in geophysics and geosciences. Recently, deep learning has been explored for seismic stratigraphic interpretation. However, deep learning-based interpretation methods usually require sufficient labeled samples. This is often too hard to be satisfied in field seismic interpretation. In this paper, we propose a deep active learning-based method to address this issue. Active learning typically exploits prediction uncertainty to reduce labeling effort. We found that uncertainty of prediction is easily obtained in the field of seismic interpretation. Since adjacent seismic images are very similar, they should have similar predictions. When the model performs poorly, the predictions of adjacent images will differ significantly. Thus, the uncertainty can be easily obtained by measuring the similarity of the predictions of adjacent seismic images. Then, data with the highest uncertainty is annotated by geological expert and used for the next round of training. For few-shot active learning, initial models obtained by different initial training sets are quite different. We combine deep clustering and uncertainty sampling to select initial training datasets, with which a good initial model can be obtained. To improve generalization, we introduce a random thin plate spine transformation to simulate changes of terrain. We apply the proposed method to the F3 field seismic data. The results demonstrated that the proposed method can effectively improve performance of learned seismic interpretation network with very limited labeled samples.

Fault detection using a convolutional neural network trained with point-spread function-convolution-based samples

Automatic fault detection in seismic images using deep learning-based methods has attracted great interest in recent years. Detecting faults by deep learning methods is generally considered a supervised learning task, which requires numerous diverse training samples and corresponding accurate labels. Because field data with faults labeled by experienced interpreters are difficult to acquire, training samples are usually generated synthetically, in which the 1D seismic wavelets are convolved with reflectivity models. However, the 1D-convolution-based synthetic seismic images still differ from the real seismic images. Therefore, the network trained by such samples may fail to detect some faults in the field data. Full-wavefield approaches are the optimal seismic modeling methods, but they cannot be widely used at present due to the high cost. In this paper, we present an efficient approach to generate realistic synthetic seismic images by using a point-spread function (PSF)-based convolution. Because the size of a seismic sample used for deep learning training is a small range relative to the field seismic data, assuming the background velocity is close to homogeneous in such a small range, we can efficiently construct the PSF using the analytical Green’s function. With an Intel Core i9-10900K CPU, the PSF approach takes approximately 15 min to generate a seismic image of size 128 × 128 × 128, whereas it takes approximately 20 h to generate a seismic image of the same size using reverse time migration. The examples of one synthetic image and two field seismic images demonstrate that the network trained with the PSF convolution samples can predict more accurate and continuous faults than that trained with the 1D convolution samples.

A Genetic Algorithm Optimized Undersampling Method for Seismic Sparse Acquisition and Reconstruction

The irregular observation region poses challenges to seismic acquisition systems design. The commonly-used parallel-type acquisition system requires that the geophones are located at equally spaced positions and therefore is hard to implement in an irregular observation region. An acquisition system which allows implementation of sparse and irregular observation (e.g., the node-type geometry) followed by a reconstruction procedure is a solution. It can not only fit in irregular observation regions but also make a significant reduction in seismic acquisition costs. The seismic sparse acquisition can be mathematically modeled as a undersampling operator in the seismic reconstruction problem. A suboptimal undersampling pattern will lead to an inferior reconstruction result. To optimize the seismic sparse acquisition, I propose a undersampling method based on bionic intelligence in this study. In the proposed method, a Shannon entropy maximum model is proposed to improve the observed ergodicity and reduce the undersampling artifacts. To solve the maximum problem, an improved version of genetic algorithm is presented. The proposed method is applicable to irregular observation regions and can optimize the subsequent reconstruction performance. I provide the detailed algorithm framework and discuss the undersampling artifacts of different undersampling methods. The application to synthetic and field seismic data validates the effectiveness of the proposed method.

Weak Seismic Signal Enhancement Using Curvelet Transform and Compressive Sampling

Conventional curvelet-domain denoising methods suppress random noise by thresholding the amplitude of curvelet coefficients, which makes it hard to distinguish weak seismic signals from random noise because they share the same characteristic of weak amplitude in the curvelet domain. Here we put forward an innovative weak seismic signal enhancement method taht can distinguish weak seismic signals from random noise. After compressive sampling, the curvelet coefficients of weak seismic signals show significant amplitude reduction, whereas random noise does not. We take advantage of this characteristic and design a sensitivity coefficient, the absolute ratio of curvelet coefficients before and after compressive sampling. The sensitivity coefficient can distinguish weak seismic signals from random noise in the curvelet domain better than thresholding the amplitude of curvelet coefficients. The results of synthetic and field seismic data applications both indicate that our method outperforms the conventional curvelet-domain denoising method on weak seismic signal enhancement.

Reinforcement Learning-Based Denoising Model for Seismic Random Noise Attenuation

The random noise attenuation is an essential step in seismic data processing. Due to complex geological conditions and acquisition environment, the intensity of effective signal and random noise varies in time and space. Additionally, the morphology of seismic event is complex and diverse, such as large dips and fast changes. These complex conditions necessitate that the denoiser adjust filtering policy dynamically. In this paper, we propose a reinforcement learning-based seismic denoising (RLSD) model with the framework of asynchronous advantage actor-critic (A3C). In A3C framework, the RLSD agent utilizes a policy network to learn a denoising policy for the state that is the sample of seismic data and selects a suitable filter from the preset action space composed of multiple simple and effective seismic filters with different parameters. Moreover, the RLSD agent utilizes a value network and a region-adaptive weighted reward function to accurately evaluate the denoising effect of nonstationary seismic signals. A curriculum learning approach is adopted to achieve convergence of the proposed RLSD model under complex seismic data by training from the stationary training data to nonstationary training data, and make the model more suitable to the data to be processed by using a local similarity-based reward function to fine-tune the model. The synthetic and field seismic data applications confirm that the proposed RLSD model achieves a significant performance in preserving nonstationary signals and suppressing noise by adaptively adjusting the denoising policy according to the complex structural features and noise levels. The source code is available on https://github.com/liangc-code/RLSD.

Frequency-Dependent AVO Inversion Using a Modified Window-Parameter-Optimized S-Transform for High Gas-Saturation Reservoir Delineation

The dispersion induced by fluid flow is a useful characteristic of hydrocarbon reservoirs and can be deduced from frequency-dependent amplitude versus offset (FAVO) inversion. However, FAVO inversion strongly depends on the spectral decomposition method, which directly affects the resolution of inversion results. In this letter, we propose a new time-frequency analysis method, named modified window-parameter optimized S-transform (MWPOST), for improving the resolution of FAVO inversion. In MWPOST, the optimized scale parameter of the window function can be adaptively obtained according to the amplitude spectra of actual signals. Synthetic signal analysis indicates that the method has excellent performance with higher time-frequency resolution. Then, we present a high-precision FAVO inversion method using MWPOST for high gas-saturation reservoir delineation. The field seismic data application illustrates that the dispersion factor inverted by the proposed method has a higher time resolution and exhibits better performance for delineating the reservoir boundary with high gas saturation. The proposed method can improve the precision of FAVO and it is feasible to detect the location and spatial distribution of high gas-saturation reservoirs.

A new hybrid optimization approach using PSO, Nelder-Mead Simplex and Kmeans clustering algorithms for 1D Full Waveform Inversion.

The FWI is formulated as a nonlinear optimization problem that traditionally uses local (derivative-based) minimization to find the scalar field of properties that best represents the field seismic data. This problem has a high computational cost and accuracy limited to local minima, in addition to suffering from a slow convergence rate (Cycle Skipping). Therefore, we developed a two-phase hybrid optimization algorithm based on DFO algorithms. The first use global minimization and clustering technique. The second use local minimization. In phase 1 we adopted the modified PSO and K-means algorithms and in phase 2, we adopted the ANMS. We call the hybrid algorithm of the PSO-Kmeans-ANMS. Where K-means is responsible for dividing swarms of particles into 2 clusters at every instant. This strategy aims to automatically balance the mechanisms of exploration and exploitation of the parameter search space by the hybrid algorithm, allowing one to find more precise solutions and consequently improving its convergence. The PSO-Kmeans-ANMS algorithm was validated on the set of 12 benchmark functions and applied to the FWI 1D problem. We compared PSO-Kmeans-ANMS with classic PSO, modified PSO, and ANMS algorithms. The metrics used were are the average execution time and the success rate (an error of ± 4% of the optimal solution). In all validation experiments and the FWI application, the PSO-Kmeans-ANMS performed well in terms of robustness and computational efficiency. In the case of FWI, there was a significant reduction in computational cost, thus presenting a relevant result.

High-Dimensional Seismic Data Reconstruction Based on Linear Radon Transform–Constrained Tensor CANDECOM/PARAFAC Decomposition

Random noise and missing seismic traces are common in field seismic data, which seriously affects the subsequent seismic processing flow. The complete noise-free high-dimensional seismic dataset in the frequency–space (f-x) domain under the local linear assumption are regarded as a low-rank tensor, and each high dimensional seismic dataset containing only one linear event is a rank-1 tensor. The tensor CANDECOM/PARAFAC decomposition (CPD) method estimates complete noise-free seismic signals by characterizing high-dimensional seismic signals as the sum of several rank-1 tensors. In order to improve the stability and effect of the tensor CPD algorithm, this paper proposes a linear Radon transform–constrained tensor CPD method (RCPD) by using the sparsity of factor matrix in the Radon domain after high-dimensional seismic signal tensor CPD and uses alternating direction multiplier method (ADMM) to solve the established optimization problem. This proposed method is an essential realization of the high-dimensional linear Radon transform, and the results of synthetic and field data reconstruction prove the effectiveness of the proposed method.

Deep learning for characterizing CO2 migration in time-lapse seismic images

Time-lapse (or 4-D) seismic data play an important role in monitoring the spatial CO2 distribution during and after the injection period. However, traditional interpretation or prediction of CO2 distribution is time-consuming and might be sensitive to the quality of 4D seismic data. To solve these problems, we propose a deep-learning-based method to efficiently and accurately characterize CO2 plumes in time-lapse seismic data. We first introduce a workflow to build 3-D realistic impedance models containing CO2 plumes with various shapes, sizes, and locations. From the impedance models, we then simulate synthetic seismic datasets and automatically obtain the corresponding CO2 label volumes. We extract real noise from field seismic datasets and add the noise to the synthetic ones to make them more realistic. We further construct a diverse and realistic training dataset with the combination of synthetic data containing CO2 plumes and real data without CO2 plumes that are randomly cropped from field seismic data before CO2 injection. We finally utilize the training datasets without any human labeling to train a 3-D deep U-shape convolutional neural network for detecting CO2 plumes in the Sleipner time-lapse seismic images. Compared to traditional interpretation methods that take several days or even weeks, our method takes only 29 s using one graphics processing unit (GPU) to predict CO2 plumes in a 512*512*256 seismic volume. Besides, our CO2 prediction can achieve 95.8% accuracy (compared to the manual interpretation) and could distinguish reflections of CO2 plumes from the ones of pre-existing fluids, thin layers, and noise. To more accurately characterize the CO2 plumes migration, we use dynamic image warping to compute relative shifts that register the time-lapse seismic volumes before and after CO2 injection and then apply the same shifts to the predicted CO2 plumes. By doing this, we are able to reduce the inconsistencies that may be introduced by acquisition, processing, push-down effect (velocity decrease by injected CO2), and pull-up effect (wavelet distortion), which is helpful to more accurately characterize the CO2 plumes migration.

Inclusion of seismic attributes in reservoir ensemble coverage analysis

There is a lack of quality-control (QC) methods to ensure measured seismic data are within the span of modeled seismic data in the context of ensemble-based seismic history matching of reservoir models. The dimensionality of seismic data makes it difficult to visualize the data and further compare them to the large number of ensembles in an efficient manner. Two attributes called coverage and importance are introduced to incorporate the key elements of reviewing an ensemble. The coverage attribute delineates where the set of models replicates the measured data, and the importance attribute identifies where it is important to fit the data above the noise threshold. The two attributes are then combined to highlight in which spatial area our reservoir model ensemble appropriately models the data and where a significant discrepancy exists between our ensemble of models and the measured data. The attributes are closely connected to noise, as coverage always must be analyzed in terms of the noise level. Although noise may not be explicitly corrected for, the methodology corrects the attributes for the noise assessed. The method is applied on two data examples from field seismic data: a 4D absolute difference amplitude map and a 4D relative impedance difference cube. The first example shows how changing the oil-water contact of the ensemble can improve the coverage without any history matching, and the second shows how it is more difficult to get a good coverage using 3D seismic attributes rather than using 2D maps of seismic data. The proposed QC attributes provide tools to better manage coverage of seismic data in the ensemble.