Stochastic Modeling Of Reservoir With Seismic Information Constraints

Lithofacies and reservoir properties are significant information for reservoir description and reservoir evaluation,and they are also very vital for exploration and development of oil and gas reservoirs.Reservoir characterization and modeling is the key method and tool for obtaining lithofacies and reservoir properties.With the development of oil and gas exploration and development,reservoir models with more detailed and fine-scaled are demanded.In general,seismic data contain abundant information about lithofacies and reservoir properties.However,the relationship between these information and seismic attributes is extremely complicated,which leads to some challenges for using seismic data to gain corresponding lithofacies and reservoir properties.Therefore,it is urgent to study how to use seismic data as a constraint to attain information of reservoir properties and lithofacies.Stochastic inversion,which is usually based on geostatistics,plays an important role in reservoir characterization and modeling.Compared with deterministic inversion,stochastic inversion is superior in improving resolution and reducing uncertainty of reservoir modeling results.In geostatistics,seismic data can be directly used as a constraint to implement stochastic inversion.Besides,it can be considered as conditional information to constrain reservoir characterization and modeling after it is transformed into probabilistic information by various means.At first,this dissertation studies the methods that can discriminate facies and predict reservoir properties exploiting seismic data.These methods are based on the Bayes theorem.They can not only provide the identified lithofacies or predicted reservoir properties,but also obtain confidence probability of the results that is very important for uncertainty evaluation.In addition,it can be used as probabilistic constraint information in geostatistics.The proposed method takes the various characteristics of the data into account and also incorporates with the kernel function and Fisher transform,which contributes to more accurate results.Considering the disadvantages of the existing geostatistical method of simulation and inversion,we study on stochastic simulation and stochastic inversion integrating different information from different resources.In the framework of probability perturbation inversion strategy,the dissertation combines multi-point geostatistics,which incorporates the advantages of pixel-based and object-based simulation algorithm,statistical rock physics and forward modeling with quantum annealing algorithm to achieve lithofacies and reservoir properties inversion.The inversion results with relatively high resolution are generated in a small number of iterations,enhancing the utility of stochastic inversion based on multi-point geostatistics.Moreover,the advantages and disadvantages of two-point and multi-point geostatistics are analyzed.Although the calculation speed of two-point geostatistical approaches that are widely applied is relatively fast,its capacity to reproduce the geometries of geological bodies is insufficient.On the other hand,multi-point geostatistics considers the spatial relationship between multiple points,and it can describe the features of the target more effectively,while multi-point geostatistics is mainly applied to discrete variables.Thus,multi-point geostatistics and sequential Gaussian simulation,which is the commonly used two-point geostatistical method,are combined with probability perturbation method.Facies and reservoir properties are simultaneously produced through this inversion approach.The present method integrates the benefits of multi-point and two-point geostatistical methods.Sequential Gaussian simulation is controlled by facies generated by conditional simulation based on multi-point geostatistics.As a result,the differences of reservoir properties in different lithofacies can be better reflected.Inverted reservoir models of facies and reservoir properties with fine-scaled are outputted.The results can not only be used for uncertainty evaluation,but also guide permeability modeling and other follow-up researches.The model tests and successful applications of field data fully verify the feasibility and effectiveness of these methods.