Study Of Full-Waveform Inversion For Subsurface Q Structure

The flow of reservior prediction based on seismic velocity has been well developed, including rock physics, well data interpretation, seismic data processing, velocity model building, pre- or post-stack inversion. But the study of quality factor (Q) in these fields falls behind. Following the way of velocity research work, we can deal with the theoretical and practical problems about Q.First, I investigate the mechanisms, the parameterization, and the observation characterizations of intrinsic attenuation. Then I use three methods, full-waveform sonic data analysis, 1D full-waveform inversion(FWI) using genetic algorithm and 2D FWI using local decent algorithm, to study the dissipation properties of subsurface rocks. Also, I investigate the problems of parameterization, inversion strategies, initial model building, in FWI for Q and evaluate the inversion results using rock physics and the known reservior distribution.In the process of velocity inversion, well data plays an important role in building the initial model, evaluating the results and geological anaysis. So it can do the same work in the inversion for Q. I use the P head wave of full-waveform sonic data to analyse the dissipation properties of land facies reservior in Well YBX qualitatively and quantitively and compare the result with well data interpretation. It shows that, the sand reserviors correspond to strong attenuation zone, but not the fractured shale reservior; sandstone shows stronger attenuation than calystone in sonic frequancy range, and the magnitude decrease with clay content when it is greater than 30%; there is no significant relationships between the attenuation and porocity, water saturation; when considering both Q and velocity, the anomaly of reservior parameters can be extracted easily.1D FWI based on Kennett's method and genetic algorithm has high computation efficiency and reserch ability, and is less dependent on initial model. These algorithms are adapted for Q inversion and then applied to synthetic and real data set. The synthetic case shows that the inversion results of velocity and Q is consistent with real model in trend and the high/low velocity anomalies and high attenuation anomalies are obvious. But there is still some disagreement between the inverted model and the real one. The error of Q inversion is higher than the velocity. The real data case shows that the inverted 1/Qp model is consistent with full-waveform sonic result in trend, and the main anomalies can be distinguished. But there are some uninterpretable oscillations in the inverted model. So the accuracy of inversion for Q does not satisfy the quantitive or semi-quntitive analysis. The quality of seismic data is the main reason for bad inversion result.2D FWI based on wave equation in frequency domain and local decent algorithm has high accuracy and convergence speed. To be used for Q inversion, its theoretical form, model parameterization and inversion strategies must be adapted. I made some discussion about the relationship between velocity and Q in FWI, as well as the inversion strategies and initial model building for Q inversion. It shows that using the complex velocity as the model parameter will introduce much more uncertainties to the inversion. It is better to treat the Q as real parameter and simultaneously update the velocity and Q with subspace method. Because the coupling of velocity and Q varies in different inversion stages, to update the model parameter in order of low frequency velocity→high frequency velocity→high frequency velocity+Q can reduce the uncertainty of inversion to a maximum extent.