The Study Of Three-dimensional Magnetotelluric Occam Inversion Based On MPI Parallel Algorithm

Magnetotelluric sounding (MT) widely applied in the earth’s interior structure, deepgeological research and deep prospecting areas. Restricted by many factors, its dataprocessing technologies mainly use1D and2D method currently, and the inversionresults is easy affected by underground3D geological body. Therefore3D MT forwardand inverse becomes the research hotspot at home and abroad. Due to the large amountof calculation, the huge amounts of memory requirements, etc.,3D MT inversion has notyet entered the process of the practicality. This paper uses MT data space Occaminversion of the serial and parallel algorithm that based on the staggered grid finitedifference and analyze the problems on the adaptability of inversion, to provide referenceto the practicality of the3D MT data processing.Based on the basic principle of magnetotelluric theory, by adopting Yee gridstaggered sampling under the condition of the MAXWELL’s equations in quasi staticdiscrete, combining with the corresponding boundary conditions, the large sparse systemof linear equations was generated. Then use the PCG iteration method solved the linearsimultaneous equations. And via computing a fault model and other typical Geoelectricmodel, the reliability and accuracy of3D magnetotelluric forward modeling based onstaggered grid was verified.Then, based on the brief introduction of the basic principle of model space Occaminversion and data space Occam inversion, the expressions of the sensitivity matrix aboutimpedance component were deduced. By studying the one dimensional search methodapplied in searching the Lagrange multiplier, a new search method that combination ofNewton method and dichotomy was present. This study shows that the new method hasless forward compute and less run time compared to the traditional search algorithm suchas forward-back search method and reducing the range search algorithm. Then with a single low resistivity body model calculation this paper showed that the inversionmethod is effective. And on this basis, through the studying of the parallel computingtheory and the relationship between the individual parts of Occam inversion algorithm,the paper analyzed what parallel computing schema the inversion algorithm can take andthe theoretical parallel computing speed. And, with the help of the MPI parallelcomputing packages, we set up a cluster based on PC machines. According to the test ofcommunications speed between nodes, we optimized the parallel computing algorithm,and speeded up the computation speed.At last, with parallel computing platform and date space MT inverse algorithm, wedid the theoretical tests on the typical geoelectric model. And via the analysis of theimpact of noise immunity, the initial background resistivity, the number of measuringpoint, the measuring point distribution, the selection of frequency, the influence of theselection of data sets, the paper showed that: the program has a good anti-noise property,and initial background resistivity should be chosen close to the target volume resistivityvalue, the frequency should be appropriately increased near the target body, themeasuring point should be homogeneously and symmetrically distributed above thetarget body relatively, and select the total impedance tensor for inversion has a bettereffect.