Research On Geological Voxel Modeling Method Based On Transition Probability

How to integrate multi-source geological information to establish a geological model that conforms to complex geological reality has always been a difficult point in earth science.In scenarios such as oil and gas exploration,mine evaluation and hydrological environments where the randomness and uncertainty of geological variables are concerned,geological voxel models that are easy to integrate multi-source information are often used as carriers for visualization,interpretation and analysis of underground geological objects.Using geostatistics to integrate multi-source geological information to predict geological variables in voxels is a key step in the construction of geological voxel models.However,the current geological reality is the result of multiple stages and multiple geological actions.The spatial distribution of geological variables often presents high-order Markov and local anisotropy,and geological sampling is sparse and uneven.High-quality geological voxel models that conform to complex geological reality require reasonable spatial correlation calculation modelsn,high-accuracy geological variable prediction methods,and appropriate modeling methods.It has always been a difficult problem in geomodeling and urgently needs to solve.Transiogram can describe the high-order Markov property of geological variables and can be used to accurately describe the spatial correlation between two geological variables.Artificial neural networks are good at learning complex nonlinear relationships between variables and can be applied to the probabilistic fusion of multiple two-point transition probabilities.Focusing on the problem of low accuracy modeling caused by the above characteristics of geological variables,this thesis studies the construction method of geological voxel model based on transition probability.The calculation model of two-point transition probability considering locally varying anisotropy in folded structures and the high-accuracy probability fusion method are the key research objects of the thesis.A method for constructing a 3D geological voxel model in folds and faults is formed,and the validity of the method is verified by case study.It provides method support and case reference for realizing the construction of geological voxel model containing folds and faults.Specific work includes:1)A method for constructing 3D locally varying anisotropy direction field of fold structure is proposed.The occurrence vector field obtained by the previous method using the measured occurrence interpolation is inaccurate,and the further obtained locally varying anisotropy direction field has little structural meaning.A method of estimating the occurrence at the tangent of the rock layer is studied to supplement a large number of occurrences that are consistent with the geological structure.Further study combined with vector analysis and a 3D local anisotropic directional field with clear structural meaning is studied.2)The calculation model of two-point transition probability in fold structure is proposed.The calculation model of two-point transfer probability in fold structure is studied based on the 3D local anisotropic direction field and referring to the transiogram.The algorithm for calculating the two-point locally varying anisotropic distance with structural meaning in fold structure is studied to solve the problems of slow efficiency and lack of structural meaning of traditional local anisotropic distance.The locally varying anisotropy transiogram calculation model suitable for describing the spatial correlation of two points in fold structure is studied to obtain the accurate transition probability between two points in fold structure.This method is not only intuitive,but also can deduce the relationship between layer thickness,sequence,and locally varying anisotropic transiogram of horizontal stratum.3)A self-learning and high accuracy geological variable prediction method based on artificial neural network fusion two-point transition probability is proposed.For the problem that the weight coefficients of most probability fusion models are difficult to determine,the sample features and sample images are composed of two-point transition probabilities from the transiogram.A probability fusion method with self-learning and high accuracy is proposed based on feedforward neural network and convolution neural network.The probability fusion method based on convolution neural network is not limited by the number of known points.Both synthetic model experiments and practical cases verify that these two methods have more advanced probability fusion methods with higher classification accuracy.4)The 3D voxel geological modeling method in folds and faults is proposed.For the construction of geological voxel model in fold structure,a two-step modeling method of locally varying anisotropic directional field simulation geological voxel model simulation is proposed.In a study area with combined fold structure,only one plane geological map is used to successfully simulate the 3D stratigraphic voxel model consistent with the actual geological structure.For the sliding displacement of geological objects on both sides of the fault surface caused by fracture,the voxel coordinates before fracturing are restored by using the fault sliding field.The 3D voxel geological modeling method in the fault system is formed,and the 3D geological voxel model is constructed in a complex fault system.The main innovations of the thesis are as follows:1)The objective function of the local anisotropic direction field and its solution method with the strike and dip of the tangent line of the geological surface as the variable are proposed,and the high-precision 3D reconstruction of the anisotropic direction field of the fold structure is realized.2)Based on the 3D local anisotropy direction field and the high-order Markov characteristics of geological objects,a two-point transition probability calculation model for local anisotropy is constructed,which can more effectively describe the spatial variation characteristics of geological objects.3)A 3D geological voxel model building method integrating two-point transition probabilities is proposed,which improves the accuracy of building the complex geological structure voxel model.The above research work provides new theoretical and methodological support for the construction of 3D geological voxel model.Based on the exploration of spatial correlation calculation model of geological variables,geological information fusion and construction method of geological voxel model in this thesis,this thesis hope to improve the model quality in the construction of 3D geological voxel model containing folds and faults.