![]() Nearly all geophysical inverse problems are notoriously non-unique (many models fit the data equally well) and non-linear (small variations in the measured data may induce large variations in the model). The difficulties in solving them come from several constraints imposed by real-world scenarios. Inverse problems are not restricted to geophysics and arise in many fields including medical imaging and non-destructive testing. isotropic or anisotropic resistivity) is progressively updated by minimizing the discrepancies between the observations and the simulated data until they fall below a chosen tolerance. The interpretation of EM measurements collected in complex geological settings typically relies on inversion when a subsurface model described by a set of parameters (e.g. These methods utilize low- and medium-frequency EM energy to map variations in the subsurface resistivity and characterize the electrical structure of the geological formations at depths ranging from a few metres to tens of kilometres. Electromagnetic (EM) methods are widely applied in various areas of geophysics including hydrocarbon and mineral exploration, CO 2 storage monitoring, geothermal reservoir characterization and hydrogeological studies. Examples with different survey geometry and noise levels confirm the feasibility of the DL inversion, opening the possibility to estimate the subsurface resistivity distribution in real time.Ĭontrolled source electromagnetics (CSEM), Image processing, Inverse theory, Neural networks, Numerical modelling 1 INTRODUCTIONĮlectrical resistivity is an important property of geological formations with high sensitivity to porosity and fluid saturation. Several fully convolutional network architectures are compared in terms of their accuracy, generalization and cost of training. The pre-trained networks can reliably estimate the position and lateral dimensions of the anomalies, as well as their resistivity properties. The performance of the method is demonstrated on models of strong practical relevance representing an onshore controlled source electromagnetic CO 2 monitoring scenario. ![]() Deep neural networks based on fully convolutional architecture are trained on large synthetic data sets obtained by full 3-D simulations. This approach does not require calculation of the gradient and, once the network is trained, provides results instantaneously. In this paper, I explore the potential of deep learning (DL) methods for electromagnetic (EM) inversion. Probabilistic inversion methods, despite their great potential in uncertainty quantification, still remain a formidable computational task. Existing approaches are largely based on deterministic gradient-based methods, which are limited by non-linearity and non-uniqueness of the inverse problem. Inverse problems are commonly posed as least-squares optimization problems in high-dimensional parameter spaces. Geophysical inversion attempts to estimate the distribution of physical properties in the Earth’s interior from observations collected at or above the surface.
0 Comments
Leave a Reply. |
AuthorWrite something about yourself. No need to be fancy, just an overview. ArchivesCategories |