Neural spline flow multi-constraint NURBS method for three-dimensional automatic geological modeling with multiple constraints

Abstract

The strike and the dip angle are vital for describing the geometry of the rock formations. However, in the interpolation and geological modeling, only the coordinates are considered; the strike and the dip angle are ignored. To this end, the neural spline flow (NSF) multi-constraint non-uniform rational B-splines (McNURBS) method is proposed in this study. Any complex high-dimensional joint distribution can be learned using the deep generative model, NSF; thus, the NSF model was used to perform the exact maximum likelihood estimation and joint sampling of three-dimensional (3D) geological point coordinates, strike, and dip angle, overcoming the shortcomings of conventional statistical models, which are difficult to extend to high-dimensional problems. In addition, the conventional single-constraint NURBS modeling method based on geological point coordinates was improved to obtain the McNURBS modeling method, which considers the geological point coordinates, strike, and dip angle during the modeling process. The practical application results show that by using the proposed method, a 3D geological model can be flexibly and automatically established considering both geological point coordinates and strike and dip angle constraints. Moreover, the fitting relative error (RE) of the proposed method was reduced by 52.4% compared to the conventional NURBS method. This study provides a convenient and efficient means to automatically build a reliable 3D geological model.