The Applicability Study Of Large-scale Set Magnetotelluric Data Three-dimensional Nonlinear Conjugate Gradient Inversion

Magnetotelluric sounding(MT) has been widely applied in many areas,because it has the merits with no artificial power, wide signal band, low cost, convenient working in field, no high resistivity layer shielding effect and high resolution in low-resistivity layer. Due to many factors, most of the research works are concentrated in 2D inversion currently, the existence of three-dimensional geological underground is one of the factors that influence the inversion effect, so three-dimensional magnetotelluric inversion has become a hot and difficult point for research scholars. Three-dimensional magnetotelluric inversion have not been entered the practical stage owing to the huge amount of data, calculate time-consuming and requires a large memory capacity and other factors. Therefore, this paper carried out conjugate gradient(NLCG) inversion algorithm research based on 3D large data sets.From the basic theory of three-dimensional nonlinear inversion objective function and its gradient calculation, step of model and linear search, and elaborates the basic strategy of the three-dimensional magnetotelluric inversion of large-scale data sets,listed the inversion flows. Firstly, the author applied the three-dimensional nonlinear conjugate gradient inversion algorithm to several typically geoelectric models,inversion results show that: the three-dimensional nonlinear conjugate gradient(NLCG) inversion algorithm has high lateral resolution and vertical resolution and confirms the validity and feasibility of the algorithm. Then this paper researched the large-scale data sets magnetotelluric data inversion strategy, including sub-block mosaic effect, fractional frequency multiple-step inversion strategy, measuring point thinning inversion strategy and thinning discrete frequency inversion strategy analysis,studies show that: the case of overlapping two survey lines and four measurement points, the mosaic effect of the sub-block inversion can give good results; discrete frequency inversion strategy in order to obtain more good inversion effect should be half or three-quarters of the cross thinning frequency points, and to ensure that there exists frequency near the depth of the target body; sub-block inversion strategy and fractional frequency multi-step inversion can better reflect the geo-body size,shape, and location; large-scale data sets measuring point thinning inversion show the measuring point cross-thinning inversion strategy should be selected without knowing the underground target body and thinning dot pitch should not be too large, otherwise poor inversion effect. In summary, the large-scale data sets MT inversion strategy can solve the problem of insufficient memory,time-consuming and low calculation precision.