Research On Fast 3D Gravity Forward Modeling And Inversion Based On GPU Parallel Algorithm

Gravity exploration has the characteristics of low cost,high efficiency and wide coverage.At the same time,gravity exploration plays an important role in explaining the location,distribution,occurrence and geological meaning of geological bodies with different underground density.It is one of the important methods of geophysical exploration.With the development of computer technology,large-scale and high-precision three-dimensional inversion of gravity has been widely used.However,the realization of large-scale three-dimensional gravity interpretation needs to be supported by the three-dimensional inversion that can efficiently deal with mass data,and the problem of computational efficiency still exists at present.In this paper,the efficiency of gravity data is discussed.This paper first describes the current mainstream CUDA platform,analyzes its acceleration principle,hardware architecture,and introduces how to write parallel programs based on GPU.Then this paper begins with the gravity anomaly of the geological body,and obtains the discrete gravity anomaly and the gradient formula by meshing the underground space.Through the simulation calculation of the model forward modeling,the problem of the model size and calculation efficiency in the process of gravity forward modeling is analyzed,and the time-consuming part of the calculation is based on the analysis of the gravity forward performance.It is mainly the calculation of the positive matrix element.To solve this problem,the parallel computation of the forward matrix is used to verify the acceleration effect and accuracy by comparing the theoretical results.After that,the inversion theory is discussed in this paper.Focusing inversion and prior density constraint are used to solve the problem of multiple solutions of gravityinversion.The inversion results are improved by means of depth weighting and nonlinear step length.Then,the time-consuming part of the calculation is mainly based on the analysis of the inversion algorithm.The calculation of matrix related,the iteration in the inversion increases the calculation of these parts.At the same time,it also analyzes the problem of how to deal with the memory occupancy and the bandwidth transfer in the large-scale forward matrix calculation.Finally,this paper studies the parallel process of the inversion,carries on and performs the calculation,accelerates the inversion,and passes the theory.The results can better reflect the location,depth and density of the underground abnormal bodies,and also have a good acceleration effect.The research results in this paper have certain theoretical and practical application value for future large-scale and high-precision exploration data processing.