Oil Geophysical Prospecting publishes new theories, methods, technologies, and new experiences within the field of exploration geophysics. The scope of the journal covers the seismic data collection, analysis, comprehensive interpretation, and nonseismic exploration. The goal of the journal is to provide valuable information for the scientific and product workers.
The perfectly matched layer (PML) is a very efficient method and has become widely used in seismic forward modeling. The development from split methods to various non-split methods greatly improves the computational efficiency of PML and reduces the required memory. The nearly perfectly matched layer (NPML) belongs to non-split PML, and it transforms directly the wave field. Therefore, it does not change the form of the governing equations and is easy to be implemented. However, the NPML suffers from large spurious reflections at grazing incidence and evanescent wave. Therefore, the convolution perfect matching layer (CPML) is still usually used. We implement the complex frequency-shifted NPML (CFS-NPML) to improve the capacity of absorbing the grazing-incidence wave. We introduce different damping profiles that are proportional to each other in the orthogonal direction to make CFS-NPML stable called multi-axial CFS-NPML (MCFS-NPML). Meanwhile, we design a new attenuation function which can improve the NPML absorption further. Eventually, the MCFS-NPML is able to absorb any incident wave more stably and efficiently.
The orderly thickness-graded thin interbed is a typical one in many thin interbed types. Thus, an accurate description of its time-frequency characteristics is of great significance for its identification. We first establish a geological model of orderly thickness-graded thin interbeds with different total thicknesses, different numbers of interbeds, and different thickness variation. Then we obtain synthetic data with the wave equation forward modeling, and the time-frequency spectrum with the affine smoothed pseudo-Wigner distribution. On this basis, we analyze the characteristics and variation laws of the time-domain waveforms, time–frequency spectra, and instantaneous frequency spectra of different thin interbeds. According to our research, for typical orderly thickness-graded thin interbeds, the frequency of thin beds increases while the spectrum of them is wide; the frequency of thick beds decreases while the spectrum of them is narrow. And the time–frequency variation law of thin bed is its main controlling mechanism. Besides, the time–frequency characteristics are different with thickness changes: when a single-bed thickness is less than 1/8 wavelength, the peak frequency decreases and the spectrum becomes narrower with its thickness decrease; when a single-bed thickness is greater than 1/2 wavelength, the first band is compressed and the bandwidth is broadened with its thickness increase. In addition, for a single-bed thickness between the above two, it has similar spectral characteristics as the single thin bed. Thus, single-bed thickness interval, numbers of interbeds and thickness variation are their main controlling factors. In this paper, we also propose the concept of transitional-graded thin interbeds and equivalent thin beds, so we can analyze thin beds, equivalent-thickness thin interbeds, and orderly thickness-graded thin interbeds under a unified characteristic frame.
The hydrocarbon seismic exploration in the areas with complex structure, complex lithology, and complex surface conditions has many difficulties, among which the severely undulating surface is one of the bottlenecks. The least-squares reverse time migration (LSRTM) has higher imaging resolution, amplitude fidelity, and balance than the conventional migration methods. However, the LSRTM based on rectangular grid cannot deal with complex surface and overcome severe undulating terrains such as piedmont zones. In addition, the seismic data in these areas often contain a lot of noises. Compared with Huber norm and mixed norm, the T distribution has a better performance in incomplete data, and it has no redundant parameters. So it is simple and practical. The results of Huber norm and mixed norm methods depend heavily on the parameter selection, which requires a lot of attempts. Therefore, a fully-staggered grid is introduced into the body-fitted grid, and the T distribution is extended to LSRTM with undulating surfaces. The linear Born forward equation of body-fitted is further deduced. On this basis, the LSRTM algorithm for undulating surface based on the body-fitted fully-staggered grids is proposed, which can better solve the influences of undulating surfaces. The model tests verify the effectiveness and complex-model suitability of the proposed algorithm.
The full waveform inversion (FWI) based on the prestack seismic wave field fitting is facing a few challenges in the application, for instances, low signal-to-noise ratio (SNR) land seismic data, missing low frequency information, complicated near-surface, and huge memory needed for massive 3D seismic data processing. In terms of low precision land seismic migrations, in this paper, we propose a velocity model building method by Gaussian beam tomography, which not only makes up the medium wave-number velocity components unachievable in the conventional methods based on ray tomography but also provides FWI with a high precise macroscopic velocity model that helps to effectively avoid cycle skips. Besides, in order to reduce the dependency of inversion on low frequencies, we replace the L
2 norm with a cross-correlation function based on phase matching, which figures out validly the problem of intensive computing of FWI. Furthermore, with a self-adjoint forward operator from pseudo-conservative wave equations, the gradient computing capacity based on the adjoint-state method is enhanced dramatically, which improves the inversion adaptation and realizes high-resolution velocity model building for land seismic data.
The natural gamma data of sandy braided river delta sedimentary environments in the Gasfield F, the East China Sea are selected as the training data to construct a deep convolutional neural network, which is used for log facies recognition for the first time. Four kinds of gamma ray (GR) curve shapes are selected as characteristics, and their values are converted into the image form. Several processing steps are carried out for these images such as normalization, noise addition, rotation, and grayscale turning, and then this image data set is enhanced and expanded. In this way, the training and test data sets are established. After that, a convolutional neural network is trained and used to establish the log facies recognition model. During the training process, dropout, local response normalization, and L
2 regularization are added to limit the complexity of the model and improve the generalization ability of the model. To automatically recognize the superposed deposition units of different grades in logging information, we use different scales of wavelet basis function and extreme value segmentation to classify the different-scale deposition units of logging data. The comparison with other algorithms demonstrates that the proposed method achieves better log facies recognition.
