Research On 3-D Forward Modeling Parallel Algorithm Of DC Resistivity Data

The direct current (DC) resistivity is based on crustal rocks or minerals in the difference of the conductivity and carries out the geological exploration through research and observation of artificial establishment of the distribution of current field. Three-dimensional (3-D) forward modeling algorithm of DC resistivity method is used to find out the rules of the geoelectric field by geoelectric modeling and sources distribution computing. 3-D forward modeling algorithm, especially on how to improve its efficiency, is always the hot issue in this field.This paper summarizes the fundamental property, boundary conditions and the finite difference method of steady current field. This paper also gives the connection coefficients for some special points based on the finite difference method .We gain a large sparse linear system Ax=b when we use the 3-D forward modeling algorithm of DC resistivity method based on the finite difference method. Matrix A is a big sparse definite matrix and has non-zero elements only on the six lines near the main diagonal line and the main diagonal line. The line system computing takes most time of the processing, so it is the key point for improving the efficiency of the line system computing. There are many methods to do with the linear system. The conjugate gradient method (CG) is one iteration method which suits for the symmetrical and positive sparse linear system. It is one of the most popular methods and has excellent parallelism.GPU is different from CPU on the hardware and it is especially fit for the float point computing and parallel programming. CUDA is one of the most successful and popular platform for the GPU parallel programming. CUDA gives plentiful API for the programmer to make thread parallel programs and these APIs can take full advantage of the computing capability and memory bandwidth. MPI is a traditional parallel programming based on the message passing interface model. It has a great development during those years. The MPI parallel programs are based on the communication between the processes. MPICH is an important realization of MPI. This paper introduces the parallelism of CG method and gives the 3-D forward modeling parallel algorithm of DC resistivity programs using CUDA and MPI based on the previous research achievements. And last this paper verifies the correctness and efficiency improvement of the programs by the general geophysics model computing.