Stochastic inversion of prestack seismic data using fractal-basedinitial models

摘要

In general, inversion algorithms rely on good startingmodels to produce realistic earth models. A new method,based on a fractional Gaussian distribution derived from thestatistical parameters of available well logs to generate realis-tic initial models, uses fractal theory to generate these mod-els. When such fractal-based initial models estimate P- and S-impedance profiles in a prestack stochastic inversion of seis-mic angle gathers, very fast simulated annealing — a globaloptimization method — finds the minimum of an objectivefunction that minimizes data misfit and honors the statisticsderived from well logs. The new stochastic inversion methodaddresses frequencies missing because of band limitation ofthe wavelet; it combines the low- and high-frequency varia-tion from well logs with seismic data. This method has beenimplemented successfully using real prestack seismic data,and results have been compared with deterministic inversion.Models derived by a deterministic inversion are devoid ofhigh-frequency variations in the well log; however, modelsderived by stochastic inversion reveal high-frequency varia-g.tions that are consistent with seismic and well-log data.