Research Of 3D Geological Modeling Based On Parallel Kriging Algorithm

The interpolation algorithm is a fundamental tool for spatial analysis.The Kriging method in geostatistics satisfies the best linear unbiased estimation and is often applied in geological exploration,petroleum geology,hydrogeology,and other scenarios where high accuracy is required.Kriging interpolation is a complex process and there are two main issues to face when using it.First,spatial variability analysis involves more concepts,and the construction of a good variogram model requires the user to have a good understanding of the data distribution as well as the algorithm itself.The current relevant research is basically about the automatic fitting of a single theoretical variogram,and the automatic construction of a nested variogram is still to be explored.Second,the time complexity of the kriging algorithm is high,and its time consumption is too high when performing large-scale tasks.The common serial implementation of the program cannot give full play to the performance of hardware computing devices,so it needs to make full use of the computing power of computing devices through parallel computing technology to improve the running speed of the program.The current research on parallelization of the kriging algorithm basically uses traditional techniques and focuses on a single direction(CPU multi-core parallelism,GPU computing,distributed computing),lacking a more flexible scheme.For the spatial anisotropy analysis,this paper uses the artificial bee colony algorithm to fit the theoretical variogram,which is better than the classical least squares fitting;based on the vector space division to calculate the variogram in multiple directions,the nested variogram automatically constructed by this method can better reflect the spatial anisotropy.In order to improve the performance of the interpolation program,this paper proposes a parallelization scheme based on computational graphs by implementing the Kriging algorithm as a computational graph in TensorFlow.Compared with the methods proposed in other studies,our method is more flexible and easier to maintain,as it can achieve efficient utilization of multi-core CPU and GPU arithmetic power by coding only once.Moreover,our method shows significant performance advantages compared to the well-known serial implementations in the industry.In addition,this paper implements a distributed Kriging interpolation scheme through remote procedure call technology,which further reduces the running time of the program.The paper concludes with a visualization of the application of this method to 3D geological modeling,the flexible algorithm interface design makes the interpolation of unstructured meshes as easy as structured meshes.