Post-stack Seismic Data Research Articles

Quadratic Unet for seismic random noise attenuation

In the field of seismic data processing, seismic denoising is essential to improve the quality of seismic data. Various deep learning methods have been proposed, showing promising performance for seismic denoising. However, most deep learning methods are based on lin ear neurons, resulting in limited expressive ability for complex seismic signals. To enhance denoising capability using non-linear neurons, we propose a supervised quadratic U-shaped network (QUnet) for seismic random noise attenuation. The quadratic neurons in QUnet are represented by a second-order polynomial of input data and weighted parameters. Nu merical synthetic seismic data experiments show that the proposed QUnet method achieves higher signal-to-noise ratio results and preserves the continuity of reflectors more effectively. We have also tested the effectiveness of QUnet on both prestack field data and post-stack seismic data. The detailed structures and weak signals of seismic data are better preserved by our proposed QUnet method.

The application of matching pursuit spectral blueing in post-stack seismic frequency enhancement

The calculation of the spectral blueing operator in the traditional spectral blueing method has singularity, which leads to poor performance in post-stack seimic frequency expansion. To this end, a frequency spreading technique based on matching pursuit (MP) and spectral blueing is proposed. Through time–frequency analysis processing, it is shown that the seismic signal extracted by matching tracking method has good stability and higher resolution. The specific process of the method in this paper firstly uses the matching tracking method to accurately divide the post-stack seismic data into multiple frequency-division seismic bodies; then, in the process of calculating the spectral blueing ization operators for each frequency band, the weighting idea is used to calculate the weights of the optimized spectral blueing ization operators for each frequency band based on the differences in energy in different frequency bands; finally, the optimized spectral blueing operator is convolved with seismic reflection coefficients to obtain high-resolution seismic data. The actual test results of poststack seismic data have proven that the frequency raising method proposed in this paper is superior to the traditional spectral blueing ization algorithm, greatly improving the high-frequency component information of poststack seismic data. After frequency extension, there are more seismic events and higher resolution. Finally, the practicability and rationality of the seismic data after frequency extraction are verified by a series of operations such as attribute extraction, well seismic calibration and inversion.

Seismic Facies-Controlled Porosity Prediction in a Tight Sandstone Reservoir Based on the XGBoost Algorithm

Porosity is one of the most important properties for evaluating hydrocarbon reservoirs. There is a strong nonlinearity between seismic data and rock physical properties. Machine learning methods possess the powerful ability to capture complex nonlinear relationships between these two parameters, holding significant potential for porosity prediction in tight sandstone reservoirs. In this study, we propose a seismic facies-controlled porosity prediction method based on the extreme gradient boosting algorithm. Through seismic meme inversion method,we obtain high-resolution P-impedance data. In the inversion process, lateral variations of seismic waveforms were utilized instead of the variogram function. We create labels for model training using porosity curves and borehole side P-impedance data, applying them to the extreme gradient boosting regressor. To demonstrate its feasibility, we apply the model to a real 3D case from an oil field in the Songliao Basin. Our application results demonstrate that this method can provide porosity predictions for tight sandstone reservoirs, maintaining consistency with seismic waveforms and adhering to seismic facies control patterns. The method uses only acoustic travel time, density, gamma rays, and spontaneous potential well-log data and post-stack 3D seismic data as inputs, allowing for quick porosity predictions in the field. Compared to the seismic facies-controlled inversion combined with support vector machine and multi-Layer perceptron methods, the method proposed in this study provides more reliable porosity predictions, offering insights and references for the exploration and development of tight sandstone reservoirs.

A combined denoising method for Q-factor compensation of poststack seismic data

Attenuation is a main factor limiting the resolution of seismic data. Earth works as a low-pass filter, which has strong attenuation of the high-frequency data. The loss of high-frequency energy can be compensated by the inverse Q filtering strategy. However, this method will also increase the energy of random noise which limits its application. The inverse Q filtering algorithm also needs the Q-factor as the input parameter, which is not easy to obtain. In this paper, we proposed a three-stage process to correct the attenuation of poststack data. In the first stage, a robust structure-oriented filtering is applied to remove random noise while protecting the structure information to avoid high-frequency noise burst. In the second stage, the local centroid frequency shift (LCFS) method is used to estimate the Q factor along the seismic trace. This method combined shaping regularization and centroid frequency shift (CFS) method to improve the robustness and accuracy of Q estimation to some extent. The final stage is to apply a stable inverse Q-filtering. Synthetic and field data examples demonstrate that time-varying Q-value can be accurately estimated by using the local centroid frequency shift (LCFS) method, and the proposed workflow can compensate the attenuation without bursting of high-frequency random noise.

Harnessing 3D seismic data for mapping natural CO2 distribution: Paving the way for future CCS/CCUS implementation in the Lower Talang Akar Formation, Jabung Block, South Sumatra Basin

Hydrocarbon reservoirs saturated with CO2 are naturally widespread in the Lower Talang Akar Formation (LTAF) and deposited in the Jabung Block, South Sumatra Basin, Indonesia. The different CO2 saturation contents stored in the reservoirs pose a challenge to planning future carbon capture and storage (CCS) and carbon capture, utilization, and storage (CCUS) in the area. This study integrates poststack 3D seismic data and existing wells in the western part of the Jabung Block. It utilizes seismic attribute methods such as sum of negative amplitude, relative impedance, variance, and ant tracking to predict the origin, saturation, and distribution of CO2-saturated hydrocarbon accumulations in the sandstone reservoirs of the LTAF. The results show that CO2 originated from the Betara Deep source rock in the study area's eastern side and migrated to the reservoirs through major northwest–southeast-trending fault zones. The saturation of CO2 in the reservoirs is mainly controlled by the distance from the source and the reservoir depths, faulting, and fracturing intensity. The fields closer to the Betara source rock tend to be filled with higher contents of CO2 saturation. The results provide valuable guidance in CO2 identification and mapping for future CCS and CCUS implementation in the study area.

Evaluating machine learning predicted subsurface properties via seismic data reconstruction

In recent years, machine learning (ML) approaches have gained significant attention in seismic-based subsurface property estimation problems. However, because of the purely data-driven nature, it is challenging to evaluate the quality of the estimated properties in regions without ground truth data. In this article, we discuss evaluating the quality of ML predicted subsurface properties through ML-based seismic data reconstruction. We use a deep learning workflow to reconstruct the poststack seismic data, then use the misfit between the measured data and the reconstructed data as a proxy for the quality of ML predicted subsurface properties. We also use self-supervised learning to improve the model generalization when training the deep learning model for reconstruction. This proposed method is particularly valuable for subsurface properties without direct physical relation to seismic data. We provide both synthetic and field data examples to demonstrate the consistency of the proposed method.

Seismic data denoising using the 3D curvelet transform method

One of the major difficulties in processing and interpreting seismic data is the contamination of seismic signals by noise from numerous sources. Conventional denoising methods are mostly used for processing 2D seismic data, but noise also exists in the 3D data space, resulting in poor results using conventional denoising methods. Consequently, here, for seismic data denoising, a multiscale and multidirectional 3D curvelet transform was adopted. Our study combined the core principle of the 3D curvelet approach with the threshold iteration method to denoise simulated and actual seismic data with varying signal-to-noise ratios, and the results were quantitatively compared with the processing outcomes of existing denoising methods. The effectiveness of our method is illustrated using both genuine 2D and 3D post-stack seismic data, and synthetic 2D sections with added white and colored noise. Finally, we demonstrate how to prepare the data for frequency-domain full-waveform inversion using curvelet denoising. Despite the complexity of the procedure used to create the training samples, testing results using synthetic and real seismic data demonstrate that this method has mastered the capacity to suppress Gaussian and super-Gaussian noise from various training samples.

Seismic coloured inversion to explore the hydrocarbon prospectivity of a field in the Upper Assam basin

This study demonstrates an integrated approach using a seismic-coloured inversion to derive the hydrocarbon potential of a field in the Upper Assam basin. The well-log analysis indicated the presence of hydrocarbon in the form of gas, oil, and water in sandstones with thin streaks of limestone and shale layers. The Acoustic Impedance (AI) model was generated from post-stack seismic data using a coloured inversion. Shale volume (Vshale), density, porosity and water saturation are the petrophysical parameters spatially populated using a multilayered feed-forward neural network on the inversion-based AI model. The estimated range of model-derived parameters varies as Velocity: 2136–5034 m/s, Vshale: 0–48%, Density: 2.14–2.72 gm/cc, Porosity: 6–29%, Water saturation: 9–28%, and demonstrates a reasonable correlation to well-log data. Seismic attribute for thin-fault likelihood was used to interpret the major faults. From the combined analysis based on interpreted faults and modelled parameters, the northwestern part of the study area displays a thick hydrocarbon-bearing zone (due to increasing thickness and reservoir quality). This interpretation is based on the consideration that faults are open. Thus, assuming continuity in the sequence, this northwestern region can be considered as one of the promising candidates for further exploration.

SeisResoDiff: Seismic resolution enhancement based on a diffusion model

High resolution of post-stack seismic data assists in better interpretation of subsurface structures as well as high accuracy of impedance inversion. Therefore, geophysicists consistently strive to acquire higher resolution seismic images in petroleum exploration. Although there have been successful applications of conventional signal processing and machine learning for post-stack seismic resolution enhancement, there is limited reference to the seismic applications of the recent emergence and rapid development of generative artificial intelligence. Hence, we propose to apply diffusion models, among the most popular generative models, to enhance seismic resolution. Specifically, we apply the classic diffusion model—denoising diffusion probabilistic model (DDPM), conditioned on the seismic data in low resolution, to reconstruct corresponding high-resolution images. Herein the entire scheme is referred to as SeisResoDiff. To provide a comprehensive and clear understanding of SeisResoDiff, we introduce the basic theories of diffusion models and detail the optimization objective's derivation with the aid of diagrams and algorithms. For implementation, we first propose a practical workflow to acquire abundant training data based on the generated pseudo-wells. Subsequently, we apply the trained model to both synthetic and field datasets, evaluating the results in three aspects: the appearance of seismic sections and slices in the time domain, frequency spectra, and comparisons with the synthetic data using real well-logging data at the well locations. The results demonstrate not only effective seismic resolution enhancement, but also additional denoising by the diffusion model. Experimental comparisons indicate that training the model on noisy data, which are more realistic, outperforms training on clean data. The proposed scheme demonstrates superiority over some conventional methods in high-resolution reconstruction and denoising ability, yielding more competitive results compared to our previous research.

Inversion with stratigraphy-guided deep learning

Property estimation in seismic exploration traditionally relies on seismic inversion, which is an ill-posed problem. However, recent advances in deep learning (DL), specifically supervised neural networks, indicate promise for accuracy improvements. Building upon this, stratigraphy-guided deep learning (SGDL) is a novel method that encodes stratigraphic units as discrete features within the DL model. Our primary objective is to evaluate SGDL in a scenario with available geologic data and field data calibration, such as well tops and horizons. We conduct a case study predicting porosity and acoustic impedance from poststack seismic data. Robustness evaluations demonstrate a 20% average enhancement in the correlation for acoustic impedance across 10 test wells from the Volve data set. We find that SGDL inversion outperforms traditional and other DL methods, reaching a 91% correlation for one benchmark blind well. These results offer compelling evidence that the incorporation of stratigraphic units as features in the DL model contributes to further enhancing the accuracy of property estimation. In summary, SGDL represents a novel approach that integrates DL with geologic data, offering significant enhancements in property estimation accuracy within the field of seismic inversion.

Seismic resolution improving by a sequential convolutional neural network.

Thin-bed soft rock is one of the main factors causing large deformations of tunnels. In addition to relying on some innovative construction techniques, detecting thin beds early during surface geological exploration and advanced geological prediction can provide a basis for planning and implementing effective coping measures. The commonly used seismic methods cannot meet the requirement for thin beds detection accuracy. A high-resolution (HR) seismic signal processing method is proposed by introducing a sequential convolutional neural network (SCNN). The deep learning dataset including low-resolution (LR) and HR seismic is firstly prepared through forward modeling. Then, a one-dimension (1D) SCNN architecture is proposed to establish the mapping relationship between LR and HR sequences. Training on the prepared dataset, the HR seismic processing model with high accuracy is achieved and applied to some practical seismic data. The applications on both poststack and prestack seismic data demonstrate that the trained HR processing model can effectively improve the seismic resolution and restore the high-frequency seismic energy so that to recognize the thin-bed rocks.

Composite loss function for 3-D poststack seismic data compression

This work introduces composite functions to compute distortion in volumetric seismic data. Several loss functions, such as those based on Lp-functions, ignore the structure of 3-D seismic data, treating it as a unidimensional vector. Alternatively, applying distinct functions in each axis, properly designed for dimension reduction, can evaluate seismic data error according to its unity and magnitude. We thus propose a novel multidimensional composite loss function to evaluate through dimension reduction, suitable for seismic data compression within a method named 3DSC-GAN. It replaces the usual peak signal-to-noise ratio (PSNR) metric as the distortion function. An extensive study is conducted to analyze potential combinations of functions for the 3-D poststack seismic data compression problem. Results indicate that the new function contributes to improve the neural network learning step. Our method provides superior reconstructions, both quantitatively and qualitatively, when compared to the PSNR metric.

Reservoir characterisation using hybrid optimisation of genetic algorithm and pattern search to estimate porosity and impedance volume from post-stack seismic data: A case study

Reservoir characterisation using hybrid optimisation of genetic algorithm and pattern search to estimate porosity and impedance volume from post-stack seismic data: A case study

Posterior sampling with convolutional neural network-based plug-and-play regularization with applications to poststack seismic inversion

Uncertainty quantification is a crucial component in any geophysical inverse problem, as it provides decision makers with valuable information about the inversion results. Seismic inversion is a notoriously ill-posed inverse problem, due to the band-limited and noisy nature of seismic data; as such, quantifying the uncertainties associated with the ill-posed nature of this inversion process is essential for qualifying the subsequent interpretation and decision-making processes. Selecting appropriate prior information is a crucial — yet nontrivial — step in probabilistic inversion because it influences the ability of sampling-based inference algorithms to provide geologically plausible posterior samples. However, the necessity to encapsulate prior knowledge into a probability distribution can greatly limit our ability to define expressive priors. To address this limitation and following in the footsteps of the plug-and-play (PnP) methodology for deterministic inversion, we develop a regularized variational inference framework that performs posterior inference by implicitly regularizing the Kullback-Leibler divergence loss — a measure of the distance between the approximated and target probabilistic distributions — with a convolutional neural network-based denoiser. We call this new algorithm PnP Stein variational gradient descent and determine its ability to produce high-resolution trustworthy samples that realistically represent subsurface structures. Our method is validated on synthetic and field poststack seismic data.

Diving-wave time-lapse delay for CO2 thin layer detection

SUMMARY We have derived an analytical approximate expression to estimate the delay in diving seismic waves due to thin layers of CO2. The expression is valid for high frequencies and can be used to estimate the delay in diving waves at seismic frequencies for large separations between the source and receiver (offset). The approximation may be used to assess CO2 detection limits using diving waves and to support survey planning for CO2 monitoring and full-waveform inversion (FWI) cycle skipping analysis. In this study, we analyse the diving-wave response to a thin layer of CO2 for band-limited data using acoustic finite-difference modelling, and compare the results against the analytical calculations. We find that the responses are offset-dependent and related to double- and single-leg interactions between the diving waves and the CO2. To test the methods, we created a synthetic representation of the 2010 subsurface conditions for the top CO2 layer at the Sleipner storage complex in the North Sea, by combining base and monitor post-stack seismic data with field velocity trends. Using the acoustic finite-difference method, we model pre-stack data that captures the complexity of field data and demonstrate the use of the diving-wave delay for CO2 migration monitoring and CO2 thin layer detection.

Application of Post Stack Acoustic Impedance Inversion to Lateral Rock Property Prediction: A Case Study Eti-field Offshore Niger Delta

This study presents the result of a Model-based seismic inversion technique which was used to invert an acoustic impedance structure within a reservoir interval by intergrating well logs and 3D post stack seismic data obtained from Eti-field offshore Niger Delta. The purpose was to delineate lateral and vertical alternations in subsurface rock properties which is caused by difference in lithofacies within the reservoir interval. This would help to define hydrocarbon fairways better and constrain the range of hydrocarbon zones for field development. The inversion workflow used in this study includes forward modelling of reflection coefficients from a low frequency impedance model driven from well logs and convolution of the reflection coeffiecients with a source wavelet derived from the seismic data. P-impedance inversion analysis at the control well location gave a near perfect correlation of 0.993686 (correlation coefficient of ≈ 99.4 %) between the original P-impedance log, initial guess P-impedance model log and inverted P-impedance log. The estimated error observed was 5533.1 which corresponds to (0.112308) about 11.23 %. Seismic inversion analysis realized an acoustic impedance striucture with P-impedance values ranging from 7000 to about 50000 ft/s*g/cc and having a general increase with depth trend. Impedance slice extracted from the impedance volume at top of the reservoir predicted lateral variations in P-impedance at well control and away from well control. This information can be invaluable in delineating more prospective reservoir zones in the field and thereby enhancing optimum field development which aids in reservoir management decisions.

Lateral Rock Characteristic Evidence of Hydrocarbon Restoration from Post-Stack Impedance Inversion: A Case Study of ZED-field Offshore Niger Delta

Lateral predictions of rock properties that are descriptive of a reservoir have been studied using a model-established seismic inversion approach that was used to convert input seismic data into an acoustic impedance structure to optimize hydrocarbon restoration in the field. This was actualized by integrating wireline logs and 3D post stack seismic data procured from ZED-Field offshore Niger Delta. The inversion system employed in this study comprises of forward modelling of reflection coefficients starting with a low-frequency impedance model induced from well logs and convolution of the reflection coefficients with the source wavelet extracted from the seismic input. P-impedance analysis at reservoir C5000 delineated from the control well gave a near perfect correlation of 0.998109 (≈ 99.8 %) between the original P-impedance log, initial P-impedance model log and inverted P-impedance log with an estimated error of 0.0616932 which is about (6.16%). Seismic inversion analysis realized an acoustic impedance structure with P-impedance values ranging from 3801 to 11073.0m/s*g/cc and having an overall increase with depth tendency. Specifically, a low-impedance structure was observed at the reservoir window that can be profiled laterally elsewhere from the existing wells. Impedance slices extracted from the impedance volume at the top and base of reservoir C5000 clearly showed low impedance values away from the existing wells which are evidence of new hydrocarbon prospects (hydrocarbon-charged sands) in the field that can be explored for improved hydrocarbon recovery and field development. Therefore, the realized attribute could supply litho-fluids knowledge inside the current reservoirs, and likewise aid in defining probable hydrocarbon regions of low impedance that can be tested with the input seismic to boost the interpretation of reservoir attributes with feasible integration to seismic stratigraphy for improved reservoir prediction away from the extant wells. This information can be invaluable in delineating more prospective reservoir zones in the field, and thereby enhancing optimum field development which aids in reservoir management decisions.

Regularized deep learning for unsupervised random noise attenuation in poststack seismic data

Abstract Deep learning methods achieve excellent noise reduction performances in seismic data processing compared with traditional methods. However, deep learning usually requires a large number of pairwise noisy-clean training data, which is an extremely challenging task. In this paper, an unsupervised approach without clean seismic data is proposed to suppress random noise. Seismic data is divided into odd and even traces, which serve as the input and output of the depth network, so that the proposed algorithm can be trained directly on the original data. What is more, the proposed method introduces two regularization terms to solve the over-smoothing problem caused by reconstruction of adjacent traces. The first term considers an ideal denoising network that does not cause oversmooth as a constraint, while the second term considers the structural information existing in seismic data. Experiments on synthetic post-stack data illustrate that the proposed method obtain a higher signal-to-noise ratio than the comparison methods. In the application of field post-stack seismic data, the proposed method can effectively maintain the seismic amplitude and generate good spectral characteristics.

Machine learning for carbonate formation drilling: Mud loss prediction using seismic attributes and mud loss records

Due to the complexity and variability of carbonate formation leakage zones, lost circulation prediction and control is one of the major challenges of carbonate drilling. It raises well-control risks and production expenses. This research utilizes the H oilfield as an example, employs seismic features to analyze mud loss prediction, and produces a complete set of pre-drilling mud loss prediction solutions. Firstly, 16 seismic attributes are calculated based on the post-stack seismic data, and the mud loss rate per unit footage is specified. The sample set is constructed by extracting each attribute from the seismic trace surrounding 15 typical wells, with a ratio of 8:2 between the training set and the test set. With the calibration results for mud loss rate per unit footage, the nonlinear mapping relationship between seismic attributes and mud loss rate per unit size is established using the mixed density network model. Then, the influence of the number of sub-Gausses and the uncertainty coefficient on the model's prediction is evaluated. Finally, the model is used in conjunction with downhole drilling conditions to assess the risk of mud loss in various layers and along the wellbore trajectory. The study demonstrates that the mean relative errors of the model for training data and test data are 6.9% and 7.5%, respectively, and that R2 is 90% and 88%, respectively, for training data and test data. The accuracy and efficacy of mud loss prediction may be greatly enhanced by combining 16 seismic attributes with the mud loss rate per unit footage and applying machine learning methods. The mud loss prediction model based on the MDN model can not only predict the mud loss rate but also objectively evaluate the prediction based on the quality of the data and the model.

Unsupervised seismic random noise attenuation by a recursive deep image prior

The presence of random noise in field data significantly reduces the precision of subsequent seismic processing steps. As a result, random noise suppression is essential to improve the quality of field data. Because most traditional algorithms characterize seismic data linearly, the denoising accuracy is still open to be improved. As an unsupervised deep-learning method, the deep image prior (DIP) algorithm can characterize seismic data nonlinearly. The DIP uses randomly generated noise as input and noisy seismic data as desired output for random noise attenuation over several rounds of training epochs. However, determining the optimal training epoch for obtaining the final denoised result of unlabeled noisy data remains a challenge. To terminate the DIP training in time and obtain the denoised result, we design an improved quality control criterion (IQCC) based on adjacent estimations of seismic signal. To further improve the denoising accuracy, a recursive strategy is developed that uses the previous desired output as the new input and the previous denoised result as the new desired output. To obtain the optimal denoised results using the suggested recursive algorithm, a convergence condition also is established. Numerous examples of synthetic prestack and poststack data demonstrate the effectiveness of the designed IQCC and our recursive strategy with a convergence condition in protecting the effective signal, especially when compared with the curvelet thresholding algorithm and the original DIP. Furthermore, the denoising accuracy is on par with that of the supervised learning algorithm, demonstrating the adaptability of our recursive DIP under the convergence condition. Its superiority is further supported by field poststack seismic data processing, which uses the local similarity for performance assessments.