Parallel Computation Research Of 3-D Magnetotelluric Forward Modeling And Inversion Based On MPI

Magnetotelluric Sounding is a method of natural alternating electromagnetic field on field source electrical structure about the Earth geophysical, which is an indispensable means of deep mineral geophysical exploration, it widely applied in resources and mineral research. Nowadays, the magnetotelluric data main use one dimensional and two-dimensional inversion interpretation, but actually, geoelectric section electrical distribution is often complex, it's difficult to meet the assumptions conditions of one dimensional and two-dimensional. To deal with that one dimensional and two-dimensional inversion measured data, cannot receive reliable results. To solve this problem, it has to take three-dimensional magnetotelluric inversion. The biggest difference between three-dimensional magnetotelluric and one dimensional two dimensional magnetotelluric is the different ground model assumptions, three-dimensional magnetotelluric think resistivity are different along the horizontal, vertical, vertical changes, so they are fully with the true underground space state.The difficulty of three-dimensional magnetotelluric inversion lies in a long time forward calculation, long time of inversion Jacobian partial derivative matrix computation, non-uniqueness of inversion are too serious. Therefore, a major problem of the three-dimensional magnetotelluric inversion is the calculation amount is too large, especially the frequency points and measuring points are too much, and mesh generation model are dense, using the whole tensor impedance elements to calculate the sensitivity matrix, common PC hard to bear, speed operation is unacceptable, which is biggest obstacles to restricted the use of three-dimensional magnetotelluric inversion. The emergence of parallel computing makes the entire three-dimensional inversion into practical become true. Parallel computing is the process of solving calculation by use multiple computing resources. It can quickly solve large and complex computational problems, to overcome the existence single computer memory limitations, not only improve calculation speed, but also save memory costs. Therefore, take parallel processing technology into the three-dimensional magnetotelluric inversion field, in order to solve the three-dimensional magnetotelluric inversion problem provides an effective way fundamentally. The main contents include the following six parts. The first section general introduce the research status of magnetotelluric inversion, magnetotelluric one-dimensional, two-dimensional, three-dimensional inversion method and the parallel computing applications in field of electromagneticThe second section briefly describes MPI, including the MPI features, two modes, communication means and the basic design of MPI programs.The third section describes the basic theory of three-dimensional magnetotelluric inversion. Three-dimensional forward modeling use staggered-grid finite difference numerical simulation, the theoretical formula given in detail, including the boundary conditions, equations, and tensor impedance, apparent resistivity, strike phase. Three-dimensional inversion using a data space algorithm is based on the spatial version of Occam method. By the mathematical derivation to sensitivity matrix of the model from the original (M×M) computation and storage space in the form of data conversion to the data space (N×N) computation and storage, usually the amount of model parameters M is much larger than the amount N, so the method can reduce the computational load and required memory. Finally, this paper gives the process of calculation of the data space in version.The fourth section presents three-dimensional magnetotelluric forward parallel algorithm, for different frequencies values corresponding independent electromagnetic, we used frequency parallel thinking. In order to test the correctness and validity of three dimensional forward parallel algorithms, we designed the four models and calculation respectively. Four models are presented in detail the response of apparent resistivity cross-section diagram phase and frequency corresponding slices, and to analysis. Finally, to compare with the serial algorithm, proved that three-dimensional forward parallel program is correct and efficient.Part five gives the parallel algorithm of magnetotelluric inversion respectively. Through calculated three-dimensional inversion program analysis and test, we found that three-dimensional inversion for a long time, store more is a sensitivity matrix, multiple forward and cross product matrix. For this feature, we have developed three-dimensional magnetotelluric parallel programs, inversion and give the calculation flow chart. To validate the three-dimensional inversion the parallel program, we test the model I and IV in fourth chapter respectively, the results show that the three-dimensional inversion of parallel programs is correct and efficient. In addition, we also developed a three-dimensional inversion procedure visual program interface, by direct operation of the program can easily prepare the required data and inversion results into a map.Finally, Part six summarizes the main achievements of this paper and pointed out certain deficiencies and improvement in the future.