新疆某地区音频大地电磁测深三维反演算法研究

发布时间：2018-04-22 10:34

本文选题：大地电磁 + 数据空间　；参考：《长江大学》2015年硕士论文

【摘要】：20世纪50年代,法国学者Cagniard和前苏联学者Tikhonov提出了大地电磁法(magnetotelluric, MT)。大地电磁法是利用自然界中本身存在的大地电磁场进行地球物理勘探,该方法避免高阻层屏蔽的影响、对高导层分辨能力强、横向分辨率高、勘探深度深大数十公里、勘探范围大、勘探费用低、野外施工简单方便、室内资料处理和后期地球物理解释方法成熟等优点。经过半个世纪的研究和发展,目前已被广泛应用于油气勘查、金属矿勘探、非金属矿勘探、地热勘探、地下水和溶洞勘查等领域。目前来说二维解释常常不能很好的说明存在于某些地质复杂区域,野外采集数据所呈现出的重要地质特征,因此对于常规的三维电磁反演的能力的研究是对大地电磁方法未来进一步发展的要求。目前,三维大地电磁测深数据正、反演问题,已经是国内外电磁感应领域的主要研究方向。国外在70年代中期,就开始了三维正演的研究。随着有限元法,有限差分法,边界元法,积分方程法等应用,MT二维、三维建模和反演都取得了长足的进步。随着三维正演的进一步发展,越来越多的人投入到三维反演的研究中来,因此诞生了很多不同的反演算法,主要有非线性共轭梯度反演、快速松弛反演、共轭梯度法极大似然反演、拟线性近似反演、贝叶斯统计反演和人工神经网络反演等。近年来,很多人在三维大地电磁反演算法的发展做出自己的努力,使用各种合理的大范围趋近方法(如：Mackie和Madden在1993年,]Newman和Alumbaugh在2000年,Farquharon等在2002年)。这些方法已被证明能够合理的去恢复电导率变化,至少在某些情况下,已被理论数据例子中得到验证。然而,该三维电磁反演问题远未得到解决。高端工作站配置或并行计算机需求仍然阻碍三维程序在实际应用中运行,计算机效率的改进和实际数据的真实性与准确性影响所有设想的方法。因此,三维反演算法执行效率的提高受到人们广泛的关注。对于常规三维反演方法,M的变大使得计算时间的变长,更重要的是计算机内存需求的增加可能使计算机不可能运行。但是除非地球结构有非常强的优先限制,由于正演的结果强烈的取决于模型方案的选择,这样的计算结果可能会误导我们。对于M×M阶系数矩阵,而且这种方法能够适应更多常规的真实地质模型。但是这种一般的迭代趋近方法只在最小三维反演结构模型已经有所限制的的基础上进行实际计算,而且M较大时逐渐引起重视。音频大地电磁测深(Audio-frequency magnetotelluric,AMT)法是应用天然场源、基于平面波Cargniard视电阻率定义的频率域电磁勘探方法。上世纪60年代初,Kennecott启动了在音频频段进行大地电磁方式观测的试验,证明是可行的；随后,Strangway等人应用音频大地电磁测深法寻找金属硫化矿床方面做了大量工作,取得了有意义的成果。该方法仪器轻便,勘探效率高,工作频率范围1Hz-20kHz,勘探深度从数米至千米范围,特别适用于深度在千米以内的资源和工程勘查。由于观测资料的频率较高,对浅部特别是低阻层具有较高的分辨率。其不足之处是场源不可控,信号微弱,易受自然环境的影响,尤其是在矿山、城区附近很难开展工作。在资料解释方面,与常规的大地电磁测深(MT)方法一样,容易受到地形起伏和局部非均匀体造成的静态偏移畸变影响,使得两种极化方式的视电阻率曲线严重分异,给资料解释带来困难。随着研究工作的深入,大地电磁资料的解释由早期的一维反演逐步向二维和三维反演发展。1987年由Constanble等人提出的OCCAM反演算法成功用于一维MT资料反演,后来由deGroot-Hedlin等人深入研究用于二维MT资料反演。与其他算法相比,OCCAM反演算法可以通过较少的几次迭代就得到稳定收敛的解。二维大地电磁资料反演对计算机资源的要求不太高,反演算法也比较成熟,OCCAM算法完全达到了实用化水平。该算法基于模型空间,设模型参数的个数为M,反演需要计算MxM维灵敏度矩阵,当模型的网格参数M很大时计算工作量相当大,所以基于模型空间的三维MT反演不具备实用性。采用基于数据空间的反演算法可以克服上述的困难。一般来讲能满足观测数据的参数个数N远小于模型换个参数的个数M,基于数据空间的算法计算N维矩阵, 当NM时计算量要小很多。Siripunvaraporn和Egbert于2000年将该算法成功用于二维MT反演,并于2005年实现了三维反演。由于基于数据空间的算法对计算机的内存和技术速度的要求度大大降低,所以该算法的实现使三维MT反演实用化成为可能。这里,我们引出一种新兴的大地电磁三维反演算法,这种算法源于数据空间,同时N×N方程组将取代M×M组常规方程组。这样的话,独立数据N的大小将直接决定所有计算数量的大小和所需要的数组,因此三维地质仿真模型将远远小于M。实际上,数据空间方法已经被广泛的应用于各种地质问题的反演(即：Parker在1994年)和其他的物理场(Egbert等在1994,Chua和bennett在2001年)。如果没有其他特殊的限制,数据空间趋近法考虑的是反演算法而不是共轭梯度法。我们认为这类趋近方法是源于二维奥卡姆反演算法的延伸和发展。本文采用基于数据空间的三维反演算法实现了大阵列三维音频大地电磁数据的反演,在CPU/GPU工作站上对一个观测实例进行了处理和计算。反演结果表明,该算法能完成大尺度模型和大阵列观测数据的三维反演,采用并行算法提高了反演速度,算法具有实用性。反演结果除与已知的露头或构造信息基本吻合外,还提供了丰富的地中电阻率参数变化和信息,避免了二维反演中静态偏移的影响,并大大提高了对小异常的分辨能力。
[Abstract]:In 1950s, the French scholar Cagniard and the former Soviet scholar Tikhonov proposed the magnetotelluric (MT). The magnetotelluric method is geophysical exploration using the magnetotelluric field in nature itself. This method avoids the influence of the shielding of the high resistivity layer, and is strong in high resolution, high horizontal resolution and depth of exploration. After half a century of research and development, it has been widely used in oil and gas exploration, metal prospecting, non-metallic ore exploration, geothermal exploration, groundwater and karst cave exploration. At present, the two dimensional interpretation is often not a good explanation for the important geological characteristics that exist in some complex geological regions and the field data collection. Therefore, the study of the ability of the conventional three-dimensional electromagnetic inversion is the need for the further development of the magnetotelluric method. The problem is the main research direction in the field of electromagnetic induction at home and abroad. In the middle of the 70s, the three dimensional forward modeling began. With the finite element method, the finite difference method, the boundary element method, the integral equation method and so on, the MT two-dimensional, three-dimensional modeling and inversion have made great progress. With the further development of three dimensional forward, the more The more people come into the study of three-dimensional inversion, so many different inversion algorithms are born, including nonlinear conjugate gradient inversion, fast relaxation inversion, conjugate gradient method maximum likelihood inversion, quasi linear approximation inversion, Bias statistical inversion and artificial neural network inversion. In recent years, a lot of people have been in three-dimensional magnetotelluric. The development of the inversion algorithm has made its own efforts to use a variety of reasonable large range approach methods (such as Mackie and Madden in 1993,]Newman and Alumbaugh in 2000, Farquharon in 2002). These methods have been proved to be able to restore electrical conductivity in a reasonable way, at least in some cases, have been tested in theoretical data examples. However, the three dimensional electromagnetic inversion problem is far from being solved. The high end workstation configuration or parallel computer demand still hinders the operation of the three-dimensional program in the actual application, the improvement of the computer efficiency and the real and accuracy of the actual data affect all the envisaged methods. Therefore, the efficiency of the 3D inversion algorithm is improved by people. There is wide concern. For the conventional three dimensional inversion, the change in the M's ambassador has to calculate the length of the time, and more importantly, the increase in the computer's memory demand may make the computer impossible. But unless the earth's structure has a very strong priority limit, the result is strongly dependent on the selection of the model scheme, such a calculation knot. The results may mislead us. For the M x M order coefficient matrix, and this method can adapt to more conventional real geological models. But this general iterative approach is only based on the minimum three-dimensional inversion structure model which has been limited, and it is gradually paid attention to when the M is larger. The Audio-frequency magnetotelluric (AMT) method is a frequency domain electromagnetic exploration method based on the natural field source, based on the plane wave Cargniard apparent resistivity. At the beginning of the 60s of last century, Kennecott started the experiment of magnetotelluric observation at the audio frequency band, which proved to be feasible; then, Strangway et al. Applied audio magnetotelluric. A lot of work has been done in the field of sounding to find metal sulfide deposits, which have made significant achievements. This method is portable, efficient in exploration, the range of work frequency 1Hz-20kHz, the depth of exploration from several meters to kilometer, especially for resources and engineering exploration within kilometers. The low resistivity layer has a high resolution. Its inadequacy is that the source is uncontrollable, the signal is weak, and it is easily affected by the natural environment. Especially in the mine, it is difficult to work in the vicinity of the city. As for the interpretation of the data, it is like the conventional magnetotelluric sounding (MT) method, which is easily affected by the topographic fluctuation and the static migration caused by the local inhomogeneous body. The distortion effect makes the apparent resistivity curves of the two polarization ways seriously different, and brings difficulties to the interpretation of the data. With the deepening of the research work, the interpretation of magnetotelluric data from the early one-dimensional inversion to the two-dimensional and three-dimensional inversion is progressively developed by the OCCAM inversion algorithm proposed by Constanble et al. For the inversion of one dimension MT data in.1987. Later, deGroot-Hedlin and others have studied the inversion of two-dimensional MT data in depth. Compared with other algorithms, the OCCAM inversion algorithm can get the stable convergence through fewer iterations. The two-dimensional magnetotelluric data inversion is not very demanding for the computer resources, the inversion algorithm is more mature, and the OCCAM algorithm has been fully applied. The algorithm is based on the model space, the number of the model parameters is M, and the MxM dimension sensitivity matrix is calculated. When the model's grid parameter M is very large, the calculation of the 3D MT inversion based on the model space is not practical. The number of parameters that can meet the parameters of the observed data is far less than the number M of the model changing parameters, and the N dimensional matrix based on the data space algorithm is calculated. When the amount of computing in NM is less than that of.Siripunvaraporn and Egbert, the algorithm is successfully applied to the two-dimensional MT inversion in 2000, and the 3D inversion is realized in 2005. The algorithm based on data space has been realized. The requirements of the computer's memory and technical speed are greatly reduced, so the implementation of the algorithm makes it possible to apply the three-dimensional MT inversion. Here, we lead to a new three-dimensional magnetotelluric inversion algorithm. This algorithm is derived from the data space, and the N x N equations will replace the M x M group conventional equations. In this case, independent data N The size will directly determine the size of all calculated numbers and the array required, so the 3D geological simulation model will be far less than M. actually, and the data space method has been widely applied to the inversion of various geological problems (i.e., Parker in 1994) and other physical fields (Egbert and so on in 1994, Chua and Bennett in 2001). There are other special restrictions. The data space approach method considers the inversion algorithm rather than the conjugate gradient method. We think that this kind of approach is derived from the extension and development of the two-dimensional Occam inversion algorithm. In this paper, a three-dimensional inversion algorithm based on data space is used to retrieve the three dimensional audio magnetotelluric data of large array, in CPU/GPU The results show that the algorithm can complete the three-dimensional inversion of the large scale model and the large array observation data, and the parallel algorithm is used to improve the inversion speed, and the algorithm is practical. The inversion results also provide rich data in addition to the known outcrops or structural information. The variation and information of resistivity in the earth's surface can avoid the influence of static migration in two-dimensional inversion and greatly improve the resolving power of small anomalies.

【学位授予单位】：长江大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：P631.325

【参考文献】