Synthetic seismic data generation for automated AI-based procedures with an example application to high-resolution interpretation

Deep learning is increasingly being used as a component of geoscience workflows for processing and interpreting seismic data. Training a supervised deep learning network is a data-hungry task: a significant number of data examples are needed and they must include labels. The data examples and their labels must have consistent patterns for the deep learning network to learn. Too few examples and/or poor-quality labels can lead to poor deep learning training results. One method to provide large quantities of training examples with high-quality labels is to create synthetic data. We discuss our techniques and experiences with our ongoing use of synthetic seismic data. We share our techniques as an open-source project concurrent with this paper at https://github.com/tpmerrifield/synthoseis. We hope that the geoscience community will share our enthusiasm for developing deep learning geoscience tools and for including synthetic seismic data in supervised deep learning training. We invite contributions from the geoscience community using the open-source model to collectively reduce the realism gap between synthetic data and field seismic data.

Deep learning has been applied to tackle the seismic inversion problem, bringing more efficiency and accuracy. However, bad spatial continuity and poor generalizability limit the practical application. To solve these problems, we propose a 2D end-to-end seismic inversion method based on domain adaption. Firstly, the proposed 2D network learns the inversion mapping of seismic data under the constraint of domain adaption layer, which can reduce the difference between the features of real seismic data and synthetic seismic data, improving the generalization ability on real seismic data. Then, the trained model is finetuned with well logging data. In the first process, the spatial continuity of the inversion result is guaranteed by the 2D training scheme. Meanwhile, due to the constraint of the domain adaption layer, our model not only performs well on the synthetic data but also has good generalization ability on the real seismic data. And we carefully discuss the mechanism of domain adaption layer. In the second process, finetuning introduces well logging information, which can further improve the ability to invert details. Moreover, in order to improve the inversion accuracy on real seismic data, we develop a new training data generation method that can generate the synthetic samples close to the real samples, and a 2.5D training strategy is adopted to improve the continuity of the 3D data. The experiments on both synthetic and real seismic data show that our method performs better than both the recursive inversion method and the 1D closed-loop CNN methods.

Historically, machine learning techniques and particularly neural networks have been applied in the domain of Geophysics to discern rock properties through a quantitative approach from seismic data and associated seismic attributes. However, the success or failure of neural network is dependent on amount of available data. In general, more labelled data during training of the neural network results in accurate predictions. Therefore, seismic based rock property prediction models and their calibration needs to be fed with labelled data coming from well logs in adequate amount. The fact that the number of wells are often limited can be a huge challenge in adopting neural networks predictions. Downton and Hampson (2019) developed a Hybrid Theory-guided Data Science (TGDS) technique to improve the training of the network by augmenting the amount of data used for training. This technique aims at overcoming the issue of scarcity of the well log data. By using the outputs of theory based components like Rock Physics theory as input to the data science component of augmenting the amount of the data. In this method, Rock Physics is applied to model the elastic response due to the changes in rock and fluid properties of the available well control to create a huge dataset of pseudo wells. Then, synthetic seismic gathers are modeled at the pseudo well locations to train a Deep Feed-Forwards Neural Network (DFNN). The parameters estimated during DFNN training with synthetic seismic gathers are then applied on a real seismic dataset. The final predicted P-impedance from this formerly described method was compared to conventional inverted Pre-Stack Elastic Inversion. This paper shows the application of the method of TGDS based synthetic seismic catalog generation followed by DFNN applied on real seismic dataset for a commercially oil producing field in North Sea. The reservoirs are injectite sands which are difficult to image but can be key pay zones with quite high porosity and permeability. Well log curves e.g. P-Impedance, Porosity and Water Saturation serve as guidelines to create the synthetic seismic data. Additionally, with the added advantage of increased synthetic data, pseudo wells were broken into training and validation datasets. So, the DFNN training is carried out on a portion of the training data subset. During the process of training the DFNN, the seismic data remains blind. The estimated DFNN non-linear operator is applied to the seismic data, however, it needs to be scaled to the real data as the training was performed on the synthetic seismic data. The final result of the data science and machine learning approach driven derived P-impedance showed higher frequency and better lateral continuity than the conventional theory based approach of Pre-Stack Elastic Inversion. Augmenting the amount of training data by generating synthetic data based on the statistics of from nearby well control and rock physics relationships is prime differentiator of this workflow. By using rock physics theory a large number of pseudo wells can be generated that gives a range of expected geologic scenarios. Key wells from nearby fields, that do not necessarily tie the seismic under consideration, can be incorporated into the analysis and also, if required additional wells could be incorporated into the analysis as drilled.

Modeling of time-lapse seismic monitoring using CO2 leakage simulations for a model CO2 storage site with realistic geology: Application in assessment of early leak-detection capabilities

Deep learning is a data-driven technique that demands network models trained using big datasets. In seismic structure interpretation, it is very difficult, time-consuming, and relatively economic costs to prepare training datasets by directly annotating real seismic data. Seismic convolution method is an efficient way to synthesize seismic data, which can easily and quickly generate large amount of training datasets. However, there are large differences in feature space between synthetic seismic data and real seismic data. That results in the poor performance of network models trained using synthetic training datasets on real seismic data. In this paper, we propose to use Fourier domain adaptation (FDA) to achieve domain transfer. First, amplitude spectrum of synthetic seismic data is replaced with those of real seismic data to make feature space mapping. Then synthetic seismic data is used for transfer training of network models to improve its performance on real seismic data. The experimental results demonstrate that the FDA performs the domain transfer of synthetic seismic data to real seismic data, which improves the generalization of network models trained based on synthetic seismic data. Meanwhile, the FDA is a promising method for deep learning model transfer training in seismic structure interpretation.

Seismic data are always in low-resolution due to the limitations of seismic acquisition and processing technology, which bring challenges to subsequent seismic interpretation. Deep learning has been successfully applied to the seismic data super-resolution, all methods tend to directly learn the mapping relationship between low-resolution seismic images and super-resolution seismic images through some complex convolutional neural networks. But blindly increasing the depth of the network brings limited improvement to the super-resolution work. We propose a novel geophysical prior guided framework for seismic data super-resolution to solve this problem. Specifically, we select fault prior to guide the training of deep learning model: we use knowledge distillation technology to progressively propagate the fault prior from the teacher network (trained with the low-resolution synthetic seismic data/high-resolution fault prior and high-resolution synthetic seismic data pairs) to the student net-work (trained with the low-resolution synthetic seismic data and high-resolution synthetic seismic data pairs). To better propagate fault priors, we use feature space loss and soft ground truth loss in student network training. Finally, we use the trained student network to complete the super-resolution of synthetic validation seismic data and real seismic data. In addition, in our super-resolution framework, we directly process 3D seismic data instead of 2D seismic images, which further improves the effects of super-resolution and the subsequent seismic interpretation. Compared with the state-of-the-art seismic super-resolution method, the experimental results show that the super-resolution results of our method can depict the faults more clearly.

Interpretation of seismic data is still a process that involves manual work. By applying clustering algorithms to seismic attribute data, it is possible to automate interpretation to a certain degree. The first applications of clustering for seismic interpretation date from the early 1980s (Seber, 1984). In this work Seber applied K-means clustering, which remains one of the main clustering algorithms applied to seismic data. Sabeti and Javaherian (2009) applied K-means clustering to synthetic and real seismic data in an attempt to determine facies changes. Self-organizing maps (SOM) are another important clustering algorithm that has been applied to seismic data (Kohonen, 1990; 2001). Taner (2001) used SOM clustering to subdivide a seismic data set into four lithology classes. Strecker and Uden (2002) and Roy and Marfurt (2010) used SOM–based clustering to describe channel systems. Both Taner (2001) and Strecker (2002) emphasized the importance of well information for calibrating the results. Barnes and Laughlin (2002) applied both K-means and SOM clustering to seismic sections and to a 3D seismic data set. In their work they found a good correlation between the results of K-means and SOM clustering. Although almost no other clustering methods have been applied to seismic data, Paasche and Tronicke (2007) used Fuzzy K-means on 3D GPR data and assumed this method might also be applicable to 3D seismic data. In this work we test various clustering methods for their applicability to the segmentation of seismic data. The project is structured in two phases: firstly on synthetic data and secondly on real seismic. In the first phase we created synthetic structural models of reef bodies, salt structures, channels, karst features, fault zones, and volcanic structures. These models were then forward modelled to generate synthetic seismic data that were used as input for testing of clustering algorithms.

A reliable 4D seismic attribute is required for its quantitative use along with production history data to update the reservoir model. Because of the noisy, less reliable 4D attributes computed from field seismic data, 4D seismic has historically been used for qualitative analysis only. We present the computation of an appropriate 4D seismic attribute and the analysis of 4D seismic data to estimate a less noisy real seismic 4D attribute, which can be compared with a synthetic 4D seismic attribute. The difference of the two 4D attributes computed from real and synthetic seismic data can then be minimized to update the reservoir model. The analysis is demonstrated on a real data example from Alba field. We investigate two types of 4D attributes summed over the reservoir: 1) the difference of squared seismic amplitudes, and 2) the square of the difference of seismic amplitudes. We prefer the second attribute, because this attribute is positive, the real data attribute matches better with the synthetic data attribute, and it is more appropriate for quantitative analysis. Although the preferred attribute is less sensitive to noise, preconditioning of the real data is must. For the Alba data set, we apply time shifts to align two real seismic datasets and FK filtering to attenuate crosscutting noises from 4D data.

Deep learning (DL) applications of seismic reservoir characterization often require the generation of synthetic data to augment available sparse labeled data. An approach for generating synthetic training data consists of specifying probability distributions modeling prior geologic uncertainty on reservoir properties and forward modeling the seismic data. A prior falsification approach is critical to establish the consistency of the synthetic training data distribution with real seismic data. With the help of a real case study of facies classification with convolutional neural networks (CNNs) from an offshore deltaic reservoir, we have highlighted several practical nuances associated with training DL models on synthetic seismic data. We highlight the issue of overfitting of CNNs to the synthetic training data distribution and propose regularization strategies to address it. We demonstrate the efficacy of our proposed strategies by training the CNN on synthetic data and making robust predictions with real 3D partial stack seismic data.

Partial CRS stack is one of alternative seismic data processing methods, which does not depend on macro velocity model. One of the advantages of this method compared with conventional CMP stack is the summation process in each CMPs in order to obtain better signal quality. CRS travel time formula can be applied to obtain partial stacked CRS supergathers. The CRS gathers obtained by this method are regularized on offset domain and have better S/N ratio than conventional gathers. This method is very helpful to improve the quality of land seismic data subject to the random noise and lack of fold coverage. The purpose of this research is to examine the efficiency of the partial CRS stack methods to highlight AVO/A (Amplitude Versus Offset/Angle) anomalies by using synthetic seismic data. The tested AVO anomaly is class 3 where the negative amplitude increases with increasing distance/offset. Seismic modeling is performed on a simple anticline model composed of shale and sand layers. The synthetic seismic data are provided to the processing using partial CRS stack to test the offset aperture that is considered to influence the seismic amplitude. AVO analysis was conducted on the pre-stack time migrated (PSTM) synthetic data with and without partial CRS stack for three offset apertures. The seismic amplitude responses change with the apertures. Dipping reflectors also give different amplitude response from flat reflectors. It can be inferred from the analysis that although the class of AVO anomalies on the partial CRS stack gathers are almost same, there should be an optimum offset aperture to avoid ambiguous result of the AVO attributes.

ABSTRACTSignificant advances have been made towards fault detection using deep learning. However, the fault labelling of seismic data requires great human effort. The resulting small sample problem makes traditional deep learning methods difficult to achieve desired results. Existing research proposes to train a deep learning model with labelled synthetic seismic data to get good fault detection results. However, due to the complexity of the actual geological situation, there are inevitable differences between synthetic seismic data and real seismic data in many aspects such as seismic signal frequency, frequency of fault distribution and degree of noise disturbance, which lead to the fact that the deep learning model trained by synthetic seismic data is difficult to get good fault detection result in field data applications. We propose to use transfer learning to reduce the impact of data differences to solve this problem: part of the deep transfer learning model is used to learn fault‐related features. And the other part of the deep transfer learning model is used to mine common features between the real seismic data and the synthetic seismic data, which makes the deep transfer learning model more suitable for real seismic data. Compared with the latest research progress, our method can greatly improve the effect of fault detection without real data label, which can significantly save the cost of manual label processing.

Fueled by the continuing development of computer technology, traditional techniques for reservoir characterization have evolved into a multi-disciplinary process. In particular, geostatistical techniques have provided a framework for integrating all available engineering and geoscience data. In this paper, we will demonstrate a seismic-guided geostatistical method on a realistic seismic data set. A cross section from a three-dimensional reservoir facies model was used to test the capabilities of this method. Synthetic seismic data were generated from the geologic cross section and then input to a seismic inversion process to estimate acoustic impedance. Five porosity logs derived from the known model were used as hand data. Using this hard data, and utilizing varying correlation coefficients between the hard data and soft data, a set of gridded maps were generated using a geostatistical method. Comparisons of the results with the known model clearly indicate the potential of this type of approach to provide meaningful reservoir characterizations consistent with the available data. The integration procedure illustrated in this paper can be used as a basis by engineers and geoscientists to develop specific procedures for their applications. The demonstrations in characterization quality improvement can be used to justify the need for integrating seismic data into the reservoir description process.

Noise attenuation has been a long-standing but still active topic in seismic data processing. The deep convolutional neural networks (CNNs) have been recently adopted to remove the learned random noise from noisy seismic data, but it is still difficult to improve the generalization ability of learned denoisers due to the limited diversity of training data sets. In this letter, we investigate an end-to-end deep denoising CNNs (DCNNs) with a novel data generation method involving multidimensional geological structure features for seismic denoising. To learn an optimized network denoiser, seismic amplitude data are extracted from 3-D synthetic seismic data along three directions (i.e., two spatial directions and one temporal direction) to prepare a training data set. Compared with using seismic data from only a certain single direction to generate all training samples, this strategy enables DCNNs to learn abundant geological structural information from three directions, and helps DCNNs have a better performance on noise reduction. Another 3-D synthetic seismic data and 3-D real land data examples with plentiful faults and fluvial channels are used to illustrate that the optimized network denoiser can be directly extended to attenuate random noise. The denoising results demonstrate that DCNNs learned from the multidimensional geological structures can accomplish the self-adaptive random noise attenuation, and meanwhile preserve spatial geological structures.

Robust low frequency seismic bandwidth extension with a U-net and synthetic training data

Deep-learning-based seismic data interpretation has received extensive attention and focus in recent years. Research has shown that training data play a key role in the process of intelligent seismic interpretation. At present, the main methods used to obtain training data are synthesizing seismic data and manually labeling the real data. However, synthetic data have certain feature differences from real data, and the manual labeling of data is time-consuming and subjective. These factors limit the application of deep learning algorithms in seismic data interpretation. To obtain realistic seismic training data, we propose label-to-data networks based on cycle-consistent adversarial networks in this work. These networks take random labels and unlabeled real seismic data as input and generate synthetic seismic data that match the random labels and have similar features to the real seismic data. Quantitative analysis of the generated data demonstrate the effectiveness of the proposed methods. Meanwhile, test results on different data indicate that the generated data are reliable and can be applied for seismic fault detection.