Search

The estimation of the subsurface acoustic impedance (AI) model is an important step of seismic data processing for oil and gas exploration. The full waveform inversion (FWI) is a powerful way to invert the subsurface parameters with surface acquired seismic data. Nevertheless, the strong nonlinear relationship between the seismic data and the subsurface model will cause nonconvergence and unstable problems in practice. To divide the nonlinear inversion into some more linear steps, a 2D AI inversion imaging method is proposed to estimate the broadband AI model based on a broadband reflectivity. Firstly, a novel scheme based on Gaussian beam migration (GBM) is proposed to produce the point spread function (PSF) and conventional image of the subsurface. Then, the broadband reflectivity can be obtained by implementing deconvolution on the image with respect to the calculated PSF. Assuming that the low-wavenumber part of the AI model can be deduced by the background velocity, we implemented the AI inversion imaging scheme by merging the obtained broadband reflectivity as the high-wavenumber part of the AI model and produced a broadband AI result. The developed broadband migration based on GBM as the computational hotspot of the proposed 2D AI inversion imaging includes only two GBM and one Gaussian beam demigraton (Born modeling) processes. Hence, the developed broadband GBM is more efficient than the broadband imaging using the least-squares migrations (LSMs) that require multiple iterations (every iteration includes one Born modeling and one migration process) to minimize the objective function of data residuals. Numerical examples of both synthetic data and field data have demonstrated the validity and application potential of the proposed method.

Fault interpretation is an important part of seismic structural interpretation and reservoir characterization. In the conventional approach, faults are detected as reflection discontinuity or abruption and are manually tracked in post-stack seismic data, which is time-consuming. In order to improve efficiency, a variety of automatic fault detection methods have been proposed, among which widespread attention has been given to deep learning-based methods. However, deep learning techniques require a large amount of marked seismic samples as a training dataset. Although the amount of synthetic seismic data can be guaranteed and the labels are accurate, the difference between synthetic data and real data still exists. To overcome this drawback, we apply a transfer learning strategy to improve the performance of automatic fault detection by deep learning methods. We first pre-train a deep neural network with synthetic seismic data. Then we retrain the network with real seismic samples. We use a random sample consensus (RANSAC) method to obtain real seismic samples and generate corresponding labels automatically. Three real 3D examples are included to demonstrate that the fault detection accuracy of the pre-trained network models can be greatly improved by retraining the network with a few amount of real seismic samples.

Reservoir parameter estimation is one of the goals of amplitude-versus-angle (AVA) inversion and angle-domain common image gathers are the basis of AVA inversion. Therefore, the accuracy of kinematic and kinetic information on angle gathers is very important for reservoir characterization. Reverse time migration is one of the most physically accurate migration method. Generating angle gathers from reverse time migration with the Poynting vector method is very efficient. However, due to inaccurate angle measurement and uneven illumination, angle gathers calculated by the Poynting vector method are often not suitable for AVA inversion. In this paper, we propose an efficient method of angle gathers with accurate angular information and amplitude from reverse time migration. We firstly decompose source and receiver wavefield to their up-going and down-going parts by using analytic wavefield. We calculate propagation directions for source down-going wavefield and receiver up-going wavefield by the Poynting vector method and form the angle gathers with these angle information and decomposed wavefield. To reduce memory storage and improve computational efficiency, we decompose wavefield at excitation amplitude time by using a local spatial Fourier transform. We also use a spatial smoothed Poynting vector to improve the stability of angle measurement. We apply an illumination compensation image condition to recover the true amplitude. Numerical examples on Marmousi model and the SEAM two-dimensional (2D) model demonstrate the advantages of our proposed method. The angle gathers based on our method are cleaner with more focus on events energy and better continuity, suffering from less low-frequency noise in the shallow parts and with a distinct cutoff at large angle where reflection terminates. At last, we demonstrate the effectiveness of proposed method on a 2D marine field data example.