Research And Application Of Stochastic Inversion Based On Geostatistics

As the difficulty of exploration increases,it is a new challenge for us to change from the initial structural reservoirs to find lithologic reservoirs and improve the accuracy of reservoir prediction.Inversion is the main method of obtaining lithology in geophysics.Compared with deterministic inversion,stochastic inversion based on geostatistics can make full use of geological information,seismic and logging information,so that the results not only meet the framework of geological structure,the reflection characteristics of seismic,but also meet the wellbore.Logging constraints thus inherit the resolution of the logging data in the longitudinal direction and inherit the advantages of the continuity of the seismic data in the lateral direction.However,the characteristics of random inversion lead to relatively low computational efficiency.The quantum annealing algorithm introduced in this paper further increases the computational efficiency compared to traditional simulated annealing.Sequential Gaussian simulation can obtain multiple equal-probability simulations based on Kriging estimation.The inversion and optimization algorithms under the constraints of seismic data can screen and evaluate the results to provide the most reliable lithologic and physical parameters,which has practical significance in seismic interpretation.Aiming at the problem that the conventional geostatistical inversion has great dependence on the well and the Kriging estimation result obtained when the well data is less is often different from the real underground structure,the method of plane wave decomposition filter is introduced in this paper.Seismic records to obtain the local dip of the formation,using geologic structure-oriented logging velocity interpolation to provide an accurate and stable result instead of Kriging estimation,to make efforts to improve the accuracy and speed of random inversion.The feasibility and effectiveness of the method are further proved by the test of model data and real data.