科研之友
微信
新浪微博
Facebook
分享链接
摘要
Non-uniqueness presents challenges to seismic inverse problems, especially for time-lapse inversion where multiple inversions are needed for different vintages of seismic data. For time-lapse applications, the focus typically is to detect relatively small changes in seismic attributes at limited locations and to relate these differences to changes in the underlying physical properties. We propose a robust inversion workflow where the baseline inversion uses a starting model, which combines a high-frequency fractal component and a low-frequency component from well log data. This starting model provides an estimate of the null space based on fractal statistics of well data. To further focus on the localized changes, the inverted elastic parameters from the baseline model and the difference between two time-lapse data are summed together to produce the virtual time-lapse seismic data. This is known as double-difference inversion, which focuses primarily on the areas where time-lapse changes occur. The misfit function uses both data and model norms so that the ill-posedness of the inverse problem can be regularized. We pre-process the seismic data using a local correlation-based warping algorithm to register the time-lapse datasets. Finally, very fast simulated annealing, a nonlinear global search method, is used to minimize the misfit function. We demonstrate the effectiveness of our method with synthetic data and field data from Cranfield site used for CO2 sequestration studies.
- 出版日期2013-6
