Nonlinear inversion for magnetotelluric sounding based on deep belief network

Abstract

To improve magnetotelluric (MT) nonlinear inversion accuracy and stabilitythis work introduces the deep belief network (DBN) algorithm. Firstlya network frame is set up for training in different 2D MT models. The network inputs are the apparent resistivities of known modelsand the outputs are the model parameters. The optimal network structure is achieved by determining the numbers of hidden layers and network nodes. Secondlythe learning process of the DBN is implemented to obtain the optimal solution of network connection weights for known geoelectric models. Finallythe trained DBN is verified through inversion testsin which the network inputs are the apparent resistivities of unknown modelsand the outputs are the corresponding model parameters. The experiment results show that the DBN can make full use of the global searching capability of the restricted Boltzmann machine (RBM) unsupervised learning and the local optimization of the back propagation (BP) neural network supervised learning. Comparing to the traditional neural network inversionthe calculation accuracy and stability of the DBN for MT data inversion are improved significantly. And the tests on synthetic data reveal that this method can be applied to MT data inversion and achieve good results compared with the least-square regularization inversion.

Keywords

magnetotellurics / nonlinear inversion / deep learning / deep belief network

References

[1] ConstableS C, ParkerR L, ConstableC G. Occam's inversion: A practical algorithm for generating smooth models from EM sounding data [J]. Geophysics, 1987, 52(3): 289-300

[2] SmithJ T, BookerJ R. Rapid inversion of two and three dimensional magnetotelluric data [J]. Journal of Geophysical Research, 1991, 96: 3905-3922

[3] RodiW, MackieR L. Nonlinear conjugate gradients algorithm for 2-D magnetotelluric inversion [J]. Geophysics, 2001, 66(1): 174-187

[4] MontahaeiM, OskooiB. Magnetotelluric inversion for azimuthally anisotropic resistivities employing artificial neural networks [J]. Acta Geophysica, 2014, 62(1): 12-43

[5] LiuB, LiS C, NieL C, WangJ L X, ZhangQ S. 3D resistivity inversion using an improved genetic algorithm based on control method of mutation direction [J]. Journal of Applied Geophysics, 2012, 87: 1-8

[6] DittmerJ K, SzymanskiJ E. The stochastic inversion of magnetics and resistivity data using the simulated annealing algorithm [J]. Geophysical Prospecting, 1995, 43(3): 397-416

[7] HuZ-z, HeZ-x, YangW-c, HuX-yun. Constrained inversion of magnetotelluric data with the artificial fish swarm optimization method [J]. Chinese Journal of Geophysics, 2015, 58(7): 2578-2587

[8] LiuJ-f, HuW-b, HuX-yun. Inversion of 2-D magnetotelluric data by differential ant colony algorithm [J]. Oil Geophysical Prospecting, 2015, 50(3): 548-555

[9] SujanG, RavineshC D, NathanJ D, NawinR. Self-adaptive differential evolutionary extreme learning machines for longterm solar radiation prediction with remotely-sensed MODIS satellite and Reanalysis atmospheric products in solar-rich cities [J]. Remote Sensing of Environment, 2018, 212: 176-198

[10] WangH, JiangH, WangL, XiZ-z, ZhangD-jun. Magnetotelluric inversion using artificial neural network [J]. Journal of Central South University: Science and Technology, 2015, 46(5): 1707-1714

[11] JiangF-b, DongL, DaiQ-wei. Electrical resistivity imaging inversion: An ISFLA trained kernel principal component wavelet neural network approach [J]. Neural Networks, 2018, 104: 114-123

[12] HintonG E, OsinderoS, TehY W. A fast learning algorithm for deep belief nets [J]. Neural Computer, 2006, 18(7): 1527-1554

[13] GoodfellowI, BengioY, CourvilleADeep learning [M], 2016, Boston, MIT Press

[14] LecunY, BengioY, HintonG E. Deep learning [J]. Nature, 2015, 521: 436-444

[15] ChengX-y, LiuH-p, XuX-y, SunF-chun. Denoising deep extreme learning machine for sparse representation [J]. Memetic Comp, 2017, 9: 199-212

[16] LeonardoB, ClaudiaC, LucianoF, GiorgioM. Integrating articulatory data in deep neural network-based acoustic modeling [J]. Computer Speech and Language, 2016, 36: 173-195

[17] ZeynettinA, AlfiiaG, AssafH, DanielL R, BradleyJ E. Deep learning for brain MRI segmentation: State of the art and future directions [J]. J Digit Imaging, 2017, 30: 449-459

[18] KasiprasadM, PanyamN S, MalojiS. A novel adaptive fractional deep belief networks for speaker emotion recognition [J]. Alexandria Engineering Journal, 2017, 56: 485-497

[19] TangB-b, LiuX, LeiJ, SongM-l, TaoD-p, SunS-f, DongF-min. Deep Chart: Combining deep convolutional networks and deep belief networks in chart classification [J]. Signal Processing, 2016, 124: 156-161

[20] DengL, FuS-s, ZhangR-xia. Application of deep belief network in polarimetric SAR image classification [J]. Journal of Image and Graphics, 2016, 21(7): 933-941

[21] LaloyE, HeraultR, LeeJ, JacquesD, LindeN. Inversion using a new low-dimensional representation of complex binary geological media based on a deep neural network [J]. Advances in Water Resources, 2017, 110: 387-405

[22] ZhuL-q, ZhangC, WeiY, ZhouX-q, HuangY-y, ZhangC-mo. Inversion of the permeability of a tight gas reservoir with the combination of a deep Boltzmann kernel extreme learning machine and nuclear magnetic resonance logging transverse relaxation time spectrum data [J]. Interpretation, 2017, 5(3): T341-T350

[23] CaoJ-x, WuS-kai. Deep Learning: Chance and challenge for deep gas reservoir identification [C]. International Geophysical Conference. QingdaoChina, 2017711712

[24] MauricioA P, JosphJ, AmirA, TaylorD. Deep-learning tomography [J]. The Leading Edge, 2018, 37(1): 58-66

[25] QianF, YinM, LiuX-y, WangY-j, LuC, HuG-min. Unsupervised seismic facies analysis via deep convolutional auto encoders [J]. Geophysics, 2018, 83(3): A39-A43

[26] LarochelleH, BengioY, LouradourJ, LamblinP. Exploring strategies for training deep neural networks [J]. Journal of Machine Learning Research, 2009, 10(1): 1-40