|Theme||Visible||Selectable||Appearance||Zoom Range (now: 0)|
Seismic inversion consists in finding an earth model that best fits the data. For this purpose, we have to minimize a least square function that measures the amplitude of the misfits according to norm to be chosen in the date space. In general the chosen norm is Euclidean or L2. Unfortunately, such a norm is not adapted to data corrupted by a high amplitude correlated noise : the noise is, in this case, inverted as a part of the signal, and the inversion results are unacceptable. Our paper is a quest for some norms in the data space that allow to reject a high amplitude correlated noise in the residuals so that the model does not match the data but matches the signal instead.
Due to the advantages of greater robustness and higher computational efficiency, the acoustic impedance inversion remains one of the most popular reservoir prediction tools in the industry. But either the conventional two-step sparse-spike inversion or the one-step model-based inversion suffers from the problem of the subjective determination of the constant damping factor, because different damping factors will cause obvious different inversion results. Therefore, we propose a Bayesian adaptive impedance inversion method based on the Taylor series expansion and Bayesian theory. The damping factor can be automatically adjusted according to the noise level of seismic data and obtain the best compromise between resolution and stability of the inversion result. Numerical and real data examples validate the adaptive ability and high-resolution advantages of the proposed method, which indicates a good application prospect in the thin-layer exploration. Presentation Date: Tuesday, October 13, 2020 Session Start Time: 9:20 AM Presentation Time: 9:45 AM Location: Poster Station 8 Presentation Type: Poster
We explore the reliability of three-term AVO inversion under conditions of varying random noise and varying far offset source-receiver reflection angle. We use stochastic forward modeling with noise followed by AVO inversion which allows calculation of Bayesian probabilities for AVO ‘true positives’. Our results quantify how increases in fold - or signal to noise- compare to increases in source-receiver offset and can be used to guide seismic acquisition design. We have used a single set of rock properties in this example, but have ‘normalized’ our measurement of seismic signal in a way that attempts to make our results applicable to all reservoirs.
Hare, Jennifer L. (The University of Texas at Dallas) | Ferguson, John F. (The University of Texas at Dallas) | Aiken, Carlos L.V. (The University of Texas at Dallas) | Brady, Jerry L. (Arco Alaska Inc.)
The gravity signal of interest is the observed gravity changes in the reservoir fluid densities.
Seismic inversion in different domain is utilized to get reflectivity information by considering the fitting degree between real seismic response and synthetic seismogram. Conventional seismic inversion methods including time domain inversion and spectral inversion only considered the difference of amplitudes in time domain or the comparison of their spectrum. However, some small modifications of underground reflectivity or the noise contamination can change mainly or only the seismic amplitudes in the time domain, some other the frequency responses of the signal. Considering the time domain inversion’s quality of anti-noise and the superior resolution of frequency domain inversion, we present a new robust inversion method based on Bayesian scheme which can enhance the resolution of seismic inversion using partial spectrum of seismic data. We referred to our method as time-frequency domain joint inversion (TF-joint inversion). Finally, applications on theoretical model and field seismic data demonstrate that our method could achieve high-resolution results and be stable in the presence of severe noise level.
Different deconvolution and seismic inversion methods were proposed to eliminate the influence of the bandlimited wavelet and yield high resolution reflection coefficient sequence. Considering the classification of sparse seismic inversion in different transform domain, it can be divided into two main theoretical frameworks including time domain and spectrum domain algorithms. Spectral inversion (Castagna, 2004; Portniaguine, 2005; Chopra, 2006) is utilized to estimate the sparse reflectivity or layer thickness based on spectral decomposition under the constraint of objective function in the frequency domain. At present many researches indicate that the spectral inversion can determine the thin layer thickness within tuning thickness. Yuan (2009) discussed the ill-posed property of spectral inversion theoretically mainly including the ambiguity and stability. Yuan (2013) developed sparse reflectivity inversion method in frequency domain based on Bayesian learning theory.