A stochastic inversion method integrating multi-point geostatistics and sequential Gaussian simulation

Facies and reservoir properties are important parameters to characterize reservoirs. Seismic inversion has become one of the most widely used techniques in reservoir modeling, characterization and prediction. Stochastic inversion based on geostatistical theory can integrate different information and data and permits to establish reservoir models with high resolution. Of it, the probability perturbation method, which is an optimized process by iteration, is an effective strategy to solve inverse problems. It can integrate different probabilistic information into a joint probability by the Tau model and can produce invertion results in a small number of iterations. Based on the probability perturbation method, we integrate multi-point statistics simulation and sequential Gaussian simulation to form a new strategy for inverting facies and reservoir properties. A series of models of facies are generated by conditional simulation with multi-point geostatistics at first. Multi-point geostatistics allows considering the spatial correlation among multiple points and can produce fine-scaled facies models. Then, perturbing the probability distribution gains a new facies model. Next, several models of reservoir properties are established by using facies-controlled sequential Gaussian simulation. Compared with the direct sequential Gaussian simulation, facies-controlled sequential Gaussian simulation can better characterize the spatial continuity of reservoir properties in different facies because it analyzes the variograms for different lithofacies. Finally, elastic attributes are computed by statistical rock physics, synthetic seismograms are computed by forward modeling and these seismograms are matched with actual seismic data. This method allows obtaining models of facies and reservoir properties simultaneously in a relatively small number of iterations. To demonstrate the feasibility and effectiveness of the method, we test it on the Stanford VI model. Then, the method is applied to a seismic section from a region of China.