| 1950s, French scholars Cagniard and the former Soviet scholars Tikhonov proposed magnetotelluric method (magnetotelluric, MT). Magnetotelluric method using natural field source, has not shielded high resistance layer of high conductivity layer resolution ability, strong lateral resolution capability, exploration depth, exploration costs low, easy construction, data processing and interpretation technology is mature and so on. After half a century of research and development, has been widely used in oil and gas prospecting, exploration and metal ore, non-metallic mineral exploration, geothermal exploration, groundwater and cave exploration and other fields.Due to the two-dimensional interpretation often can not explain the existence of complex geological area, collecting field data showing important feature is the requirement for the ability of the conventional three-dimensional magnetotelluric inversion method for the further development of the future.Currently, the three-dimensional magnetotelluric sounding data are, the inverse problem has been a major research directions in the field of domestic and electromagnetic induction. Abroad in the mid-1970s, began to study the three-dimensional forward modeling. With the finite element method, finite difference method, boundary element method, integral equation method and other applications, MT 2D,3D modeling and inversion have made considerable progress. With the further development of three-dimensional forward, more and more people into the study of three-dimensional inversion in the past, so the birth of a number of inversion algorithms, mainly nonlinear conjugate gradient inversion, rapid relaxation inversion, conjugation Maximum Likelihood inversion gradient method, quasi-linear approximation inversion, Bayesian statistics and artificial neural network inversion inversion.In recent years, a lot of people in the development of three-dimensional magnetotelluric inversion algorithm made an effort to use a reasonable method for a wide range of approaches (eg:Mackie and Madden in 1993, Newman and Alumbaugh in the year 2000, Farquharon et al 2002). These methods have been able to recover reasonable conductivity changes, at least in the case of testing the theory has been verified data. However, the three-dimensional magnetotelluric inversion problem is far from solved. High-end workstations or parallel configuration of demand is still hampered dimensional computer program running applications, computer problems and authenticity of the actual data affects all conceived methods. Therefore, to improve the efficiency of the implementation of the three-dimensional inversion algorithm by a high degree of attention.For the conventional three-dimensional inversion method, M becomes larger so that the calculation time becomes longer, more importantly, increase your computer memory requirements may make the computer can not run. But unless there is a very strong structure Earth precedence constraints, as a result of the forward selection model strongly depends on the program, so the results may be misleading. For iterative inversion methods, this method can adapt to more conventional real geological model. However, this method is only general iterative approaches to calculate on the basis of the actual structure of the smallest three-dimensional inversion models have been restrictions on the large and growing importance M.Audio magnetotelluric sounding (Audio-frequency magnetotelluric, AMT) method is the application of natural field source, frequency domain electromagnetic exploration method based on plane wave Cargniard apparent resistivity defined. In the early 1960s, Kennecott started in the audio frequency band magnetotelluric mode observer test, proved to be feasible; then, Strangway, who use audio magnetotelluric sounding metal sulphide deposits in terms of finding a lot of work has led to many significance of results. This method instrument light, high exploration efficiency, operating frequency range 1Hz-20kHz, exploration depth range from a few meters to kilometers, especially for deep within kilometers of exploration and engineering resources. Due to the high frequency of observations, especially for shallow low resistivity layer has a higher resolution. Its shortcomings are not controllable field source, the signal is weak, vulnerable to the natural environment, particularly difficult to work in the mine, near the town. In terms of data interpretation, and conventional magnetotelluric sounding (MT) method is as vulnerable topography and local non-uniform body distortion caused by static offset impact, making two kinds of polarization of apparent resistivity curve serious differentiation, to data interpretation difficult.With further research work, the interpretation of magnetotelluric data from the early phase of the one-dimensional to two-dimensional and three-dimensional inversion inversion development.1987 by Constanble et al OCCAM inversion algorithm used successfully in a one-dimensional inversion of MT data, then further by deGroot-Hedlin, who studied for two-dimensional inversion of MT data. Compared with other algorithms, OCCAM inversion algorithm through several iterations to get less stable convergent solution. Dimensional magnetotelluric inversion data for less demanding of computer resources, the inversion algorithm is relatively mature, OCCAM algorithm fully meet the practical level. The algorithm is based on the model space, set the number of model parameters is M, MM-dimensional inversion need to calculate the sensitivity matrix, the computational workload is quite large when the grid parameters M model is large, so space-based three-dimensional MT inversion model is not with practicality.Space-based data inversion algorithm can overcome these difficulties. In general, the number of parameters to meet the N observed data is far less than the model change in the number of parameters M, N dimensional matrix calculation algorithm based on data space, when N<<M when calculating the amount is much smaller. Siripunvaraporn and Egbert in 2000 for the two-dimensional MT inversion algorithm successfully and to achieve the three-dimensional inversion in 2005. Greatly reduced due to the algorithm based on the data of the memory space requirements and speed of the computer technology, so the algorithm to achieve three-dimensional MT inversion practical as possible.Here, we have drawn a new three-dimensional magnetotelluric inversion algorithm, this algorithm is derived from the data space, the minimum penalty function and the formation of the N-dimensional data space. So the output will be canceled at the same time N x N equations M x M group replaces the conventional equations. In this case, the size of the N independent data will directly determine the size and the array of all the required number of calculations, and therefore three-dimensional geological simulation model will be much less than M. In fact, the data space method has been widely used in a variety of geological problems inversion (ie:Parker in 1994) and other physical field (Egbert et al.1994, Chua and bennett in 2001). If there are no other special restrictions, data space Approaching consider is the inversion algorithm instead of the conjugate gradient method. We believe that this type of approach the method is derived from the extension and development of two Vionnet Cam inversion algorithm.An efficient three dimensional inversion scheme for AMT data has been implemented based on data space inversion algorithm. A large set of observed data has been processed and inverted on a CPU/GPU workstation. The processing abilities of this package handling large scale model and large set of data has been proved to be effective, and computing time for inversion has been much reduced by using GPU parallel programming, and can be used for real data processing. Also inverted results have shown some advantages besides good agreement with known outcrop information to avoid distortion of static effect and improve on resolution to small targets compared with results of two dimensional inversion scheme. |
PDF Full Text Request