GNGTS 2015 - Atti del 34° Convegno Nazionale

130 GNGTS 2015 S essione 3.3 compressive sensing approach such that invalid solutions are preliminarily discharged. This procedure abandons the global optimization in exchange for lower computational costs and a more robust estimation. Interval velocity consistency constraint. Interval velocity is directly related to the geological formation, rock properties, stack, and migration (Buland et al. , 2011; Claerbout, 1985). Therefore a prior rough model is usually available. Moreover it can be limited between boundaries and this is intrinsically translated in a corresponding smooth RMS velocity. If we assume the V RMS at location t 0 to be known, the V RMS at location t 0 + ∆t has to be ranged as: (5) An analogues condition can be written also for a passed time location ( t 0 - ∆t ). By means of a matching pursuit approach, outliers can be preliminarily discharged and a valid solution can be obtained iteratively. By using this scheme a new double constraint is added at each iteration so that convergence to the solution is accelerated. Synthetic example. We tested V RMS builder on the synthetic Pluto data set. Pluto data set is one of several test sets released by the Subsalt Multiples Attenuation and Reduction Technology Joint Venture (SMAART JV). It is designed to emulate deep water subsalt prospects as found in the Gulf of Mexico. Depth true model and corresponding V RMS are represented in Fig. 1a and Fig. 1b respectively. The processing flow is set up by the following steps: 1) at the first iteration no a-priory model is used. Data is binned, NMO and stacked using a set of constant velocities profiles ranging from 4500 to 13500 ft/s; 2) coherency panels are produced and local maxima time locations are extracted starting from the water bottom position (assumed to be known); 3) for each location, RMS velocity corresponding to the maxima is computed; 4) V RMS builder uses both monotonic and interval velocity consistent constraints to redefine valid picked time locations. A greedy approach starts from the global maxima of the entire panel. The following time location is chosen by a maximum energy criteria and, if constraints are fulfilled, the corresponding element of the solution is stored. This algorithm produces more and more constraints at each step leading to a stable convergence. The process ends when all the time locations selected at point 3) are processed; 5) RMS velocities at valid time locations are interpolated in time to get the final velocity profile; 6) once estimation has been done for all CMP locations a smoothing spatial filter is applied; 7) data is time migrated with the updated V RMS model with about 10% of perturbation; 8) steps 2 to 6 are iterated to refine the model. Test shows that two iterations are enough to reach convergence. Final V RMS model is represented in Fig. 1c. This result shows that, even with such a complex model, automatic V RMS builder performs a good job and the low-medium frequencies of the velocities are well retrieved. In Fig. 2 we show a sub-sampled set of CRPs obtained by migrating the data using: a) uniform mean velocity (9000 ft/s); b) final resulting V RMS model; c) V RMS model obtained from true depth model. We can see that CRP are correctly aligned and they considerably differ from the ones computed using the true model just under the salts. In Fig. 3 we show time migrated panels using: a) uniform mean velocity (9000 ft/s); b) final resulting V RMS model; c) V RMS model obtained from true depth model. Once again the panel obtained with the automatic V RMS builder looks extremely coherent and very similar to the one computed using the true model.

RkJQdWJsaXNoZXIy MjQ4NzI=