Inversion of low to medium frequency velocities and densities from AVO data using invertible neural networks

AVO (amplitude variation with offset) inversion and neural networks are widely used to invert elastic parameters. With more constraints from well log data, neural network-based inversion may estimate elastic parameters with greater precision and resolution than traditional AVO inversion, however, neural network approaches necessitate a massive number of reliable training samples. Furthermore, because the lack of low-frequency information in seismic gathers leads to multiple solutions of the inverse problem, both inversions rely heavily on proper low-frequency initial models. To mitigate the dependence of inversions on accurate training samples and initial models, we propose solving inverse problems with the recently developed invertible neural networks (INNs). Unlike conventional neural networks, which address the ambiguous inverse issues directly, INNs learn definite forward modeling and use additional latent variables to increase the uniqueness of solutions. Motivated by the newly developed neural networks, we propose an INN-based AVO inversion method, which can reliably invert low to medium frequency velocities and densities with randomly generated easy-to-access datasets rather than trustworthy training samples or well-prepared initial models. Tests on synthetic and field data show that our method is feasible, anti-noise capable, and practicable.

Direct Depth Domain Bayesian AVO Inversion

We present a new approach to perform Bayesian linearized amplitude-versus-offset (AVO) inversion directly in the depth domain using non-stationary wavelets. Bayesian linearized AVO inversion, which is a hybrid approach combining physics-based models with statistical learning, has gained immense popularity in the past decade because of its superior computational speed and its ability to estimate uncertainties in the inverted model parameters. Bayesian linearized AVO inversion is typically performed on time-domain seismic data; therefore, depth-imaged seismic cannot be inverted directly using this method, and would require depth-to-time conversion before AVO inversion can be done. Subsequently, time-to-depth conversion of the inverted volumes would be required for reservoir modeling and well-placement. Domain-conversions introduce additional sources of uncertainty in the geophysical workflows. Another drawback of conventional AVO inversion techniques is that the seismic wavelet is assumed to be stationary, and this assumption leads to a restriction in the length of the time-window over which the inversion can be performed. Therefore, AVO inversion is usually restricted to a narrow time window around the target of interest, and in case multiple targets are present at different depths, multiple inversions have to be run on the same seismic volume if we use traditional AVO inversion. AVO inversion in the depth-domain using non-stationary wavelets (or point-spread functions) is a fairly recent development, and has been previously presented in an iterative formulation that is computationally intensive compared to Bayesian linearized AVO inversion. Implementing linearized Bayesian inversion directly in the depth-domain using non-stationary wavelets is a convenient new approach that takes advantage of superior computational speed and uncertainty quantification without compromising the accurate spatial location that depth imaging provides. Bringing these two schools of thought together creates a novel, unique, and powerful seismic inversion technique that can be useful for quantitative interpretation and reservoir characterization.

Hydrocarbon Mapping on Reservoir Carbonate Using AVO Inversion Method

Amplitude Versus Offset (AVO) inversion has been applied for reservoir analysis focused on the horizon carbonate Peutu and Belumai. Simultaneous inversion analysis is used to determine gas anomaly inside carbonate-rocks and it’s spread laterally around target zones. It is based on the fact that small Vpand Vs value changes are going to show the better anomaly to identify reservoir fluid content. The AVO inversion method applies angle gather data as the input and then it is inverted to produce P impedance (Zp) and S impedance (Zs). Zp and Zs are derived to produce Lambda-Rho and Mu-Rho that are sensitive to fluid and lithology. Value of Mu-Rho between 44–65 Gpa gr/cc while value of Lambda-Rho smaller than 10 Gpa gr/cc (for carbonate-rock filled by fluid). This research found that Lambda-Rho is the best parameter to show the existence of hydrocarbon in the case of gas. While Mu-Rho is the best parameter to show the differences in lithology.

Rock Physical Modeling and Seismic Dispersion Attribute Inversion for the Characterization of a Tight Gas Sandstone Reservoir

Gas identification using seismic data is challenging for tight gas reservoirs with low porosity and permeability due to the complicated poroelastic behaviors of tight sandstone. In this study, the Chapman theory was used to simulate the dispersion and attenuation caused by the squirt flow of fluids in the complex pore spaces, which are assumed to consist of high aspect-ratio pores (stiff pores) and low aspect-ratio microcracks (soft pores). The rock physics modeling revealed that as the gas saturation varies, P-wave velocity dispersion and attenuation occurs at seismic frequencies, and it tends to move to high frequencies as the gas saturation increases. The velocity dispersion of the tight gas sandstone causes a frequency-dependent contrast in the P-wave impedance between the tight sandstone and the overlying mudstone, which consequently leads to frequency-dependent incidence reflection coefficients across the interface. In the synthetic seismic AVO modeling conducted by integrating the rock physics model and the propagator matrix method, the variations in the amplitudes and phases of the PP reflections can be observed for various gas saturations. The tests of the frequency-dependent AVO inversion of these synthetic data revealed that the magnitude of the inverted P-wave dispersion attribute can be used to indicate gas saturation in tight sandstone reservoirs. The applications of the frequency-dependent AVO inversion to the field pre-stacked seismic data revealed that the obtained P-wave dispersion attribute is positively correlated with the gas production from the pay zone at the well locations. Thus, the methods of the rock physics modeling and the frequency-dependent AVO inversion conducted in this study have good potential for the evaluation of the gas saturation in tight gas sandstone reservoirs.