Font Size: a A A

The Parallel Algorithm Research Of Three-dimensional Staggered-grid Finite Difference Numerical Simulation For CSAMT

Posted on:2013-01-26Degree:DoctorType:Dissertation
Country:ChinaCandidate:R ChenFull Text:PDF
GTID:1110330371482233Subject:Earth Exploration and Information Technology
Abstract/Summary:PDF Full Text Request
The controllable source audio magnetotelluric method (referred to as CSAMT method)is a kind of electromagnetic sounding technology which was gradually developed in1960s and 1970s. The method adopts the artificial field source method. Whencompared with the natural source magnetotelluric sounding method, this method ischaracterized in being high in signal to noise ratio, fast and efficient. Since the 1980s,the method has been widely used in exploration of energy, metals and nonmetallicmineral resources and in many fields of hydrology, engineering, environment, disastergeological survey in China. However, the forward-inversion study of CSAMT athome and abroad has been limited to 2.5-dimension. The study of the electromagneticsimulation forward-inversion method in the three-dimensional geological body forCSAMT is still the focus of the study of electromagnetic method at home and abroad.This paper starts from Maxwell equations that CSAMT satisfies, and adopts theseparation manner of the total field, which is divided into the primary field(background field) and the secondary field. According to the knowledge of the basictheory and the potential function of the electromagnetic field, the primary electricfield and the primary magnetic field are solved in a one-dimensional numericalsimulation manner. The secondary field uses the staggered-grid finite differencemethod, and derives the expression of the magnetic field component and electric fieldcomponent. On this basis, the comparative analysis is carried out to the method andthe serial integral equation method, which verifies the correctness of staggered-gridfinite difference method, thus transforming CSAMT secondary field calculation to thesolving problems of the large complex coefficient sparse linear equation set.Then, according to the characteristics of the coefficient matrix formed by CSAMT, incombination of the preconditioning methods and solution methods in the iterativesolution process, a different preconditioning technology combination scheme is proposed. After a comparative analysis, the BlcokJacobi and incomplete LUpreconditioning method combined with the Biconjugate Gradient Stabilized solutionfor solving equations are adopted in the serial process, with good numerical accuracyand stability. Meanwhile, in the CSAMT three-dimensional numerical simulationprocess, the forward simulation operation is required for different frequencies. Thereis a good parallel feature between different frequencies. This paper assigns thecomputational tasks of the primary field and secondary field with differentfrequencies to each node through the MPI message-passing mechanism for calculation,so as to improve the computational efficiency of the forward simulation. In thesolution of the secondary field equations, the CUDA programming model is used andthe advanced sparse matrix compression storage manner is applied, which carries outthe parallel processing to Sparse Matrix-Vector Multiplication, vector product, vectorinner product and other operations with a longer time and a higher parallel degree. Itimproves the efficiency so that the efficiency of the iterative solution is increased.Finally, four three-dimensional geological models are established, the correctness ofCSAMT three-dimensional forward simulation parallel algorithm is analyzed andverified, and the parallel algorithm performance is evaluated. The experiment showsthat the application of MPI + CUDA parallel algorithm will improve the performanceof CSAMT three-dimensional numerical simulation. The computational efficiencywill be increased by nearly three to five times with the use of the four-node MPI +CUDA parallel manner.
Keywords/Search Tags:controllable source audio magnetotelluric method, three-dimensional numerical simulation, iterative algorithm, MPI, CUDA
PDF Full Text Request
Related items