Mapping 3D Overthrust Structures by a Hybrid Modeling Method

Abstract

A rational three-dimensional (3D) geological model with complex characteristics generated on a small amount of data is a crucial data infrastructure for scientific research and many applications. However, reconstructing structures with multi-Z values on a single point caused by folding or overthrusting is still one of the bottlenecks in 3D geological modeling. Combined with the multi-point statistics (MPS) method and fully connected neural networks (FCNs), this study presented a hybrid framework for 3D geological modeling. The loss functions of FCN and the conventional MPS method jointly form the kernel function of the proposed method, which is constrained by stratigraphic sequence and stratum thickness. The input and output parameters of the FCN are the coordinates and corresponding elevations of geological contacts, respectively. To solve the kernel function, the initial model, in which geological surfaces are generated by the FCNs, is generated using a sequential process. An iterative MPS process with an Expectation Maximization-like (EM-like) algorithm is carried out to illuminate the artifacts in the initial model. Ten orthogonal cross-sections are extracted from the overthrust model created by SEG/EAGE as the modeling data source. The results illustrated that the geometry and spatial relationships of strata and faults are retained well with the geological constraints. The comparison of virtual boreholes from the results and the real model shows that the accuracy of the geological object reaches 75%. The presented method provides a new idea for simulating 3D structures with multi-Z values, which overcomes the limitations of the conventional MPS-based 3D modeling method.