Inversion of Gravity Anomaly Based on AlexNet Deep Neural Network

doi:10.19657/j.geoscience.1000-8527.2022.079

Abstract

Abstract:

In order to solve the problems of traditional inversion methods, such as dependence of initial model and long time for calculation, this paper proposes a noval gravity anomaly inversion method based on AlexNet deep neural network. This method designs an Alex inversion network (AlexInvNet) for gravity anomaly inversion inspired by classical deep neural network AlexNet firstly; constructs labeled datasets by forward modeling using a large number of synthetic density models secondly; uses the dataset train the AlexInvNet thirdly; and finally inputs the gravity anomaly data to the trained AlexInvNet to obtain the inversion result directly. The inversion experimental results of synthetic models show that this method can invert the position and density of anomaly body accurately, with good generalization and anti-noise ability, better than the full connected network deep learning inversion method. The field data inversion result demonstrates that this method can solve gravity anomaly inversion problem effectively.

Key words: gravity anomaly, inversion, deep neural network, AlexInvNet

CLC Number:

P631

LIU Caiyun, LI Mengdi, XIONG Jie, WANG Rong. Inversion of Gravity Anomaly Based on AlexNet Deep Neural Network[J]. Geoscience, 2023, 37(01): 164-172.

Figures/Tables 11

Fig.1 Schematic diagram of underground grid division and gravity anomaly calculation of prism body

Fig.2 Flow chart of gravity anomaly inversion using AlexInvNet

Fig.3 AlexInvNet network structure diagram

Table 1 Parameters of AlexInvNet network structure

	类型	卷积/池化核大小	通道数	步长	输入		输出
第①层	卷积	1×11	96	4	1×101		45×96
第①层	池化	1×3	None	2	1×101		45×96
第②层	卷积	1×5	256	1		45×96		20×256
第②层	池化	1×3	None	2		45×96		20×256
第③层	卷积	1×3	384	1	20×256		18×384
第④层	卷积	1×3	384	1	18×384		16×384
第⑤层	卷积	1×3	256	1		16×384		6×256
第⑤层	池化	1×3	None	2		16×384		6×256
第⑥层	全连接	None	1024	None	6×256		1×1024
第⑦层	全连接	None	800	None	1×1024		1×800

Table 2 Related parameters for AlexInvNet network

分类	参数	值
数据集	训练集	4432
数据集	测试集	1109
网络参数	学习率	1×10^-3
	激活函数	ReLU
	优化器	Adam
	$λ$	1
	$μ$	0.01
训练过程	Epochs	40000
	Batch size	1000

Table 2 Related parameters for AlexInvNet network

分类	参数	值
数据集	训练集	4432
数据集	测试集	1109
网络参数	学习率	1×10^-3
	激活函数	ReLU
	优化器	Adam
	$λ$	1
	$μ$	0.01
训练过程	Epochs	40000
	Batch size	1000

Fig.4 Inversion results of typical validate dataset

Fig.5 Inversion results of generation tests of complex models

Fig.6 Inversion results of gravity anomaly with different density parameters

Fig.7 Inversion results of gravity anomaly added 5% and 10% Gaussian white noises

Fig.8 Gravity anomaly of San Nicolas massive sulfide copper-zinc deposit[1]

Fig.9 Inversion results of the gravity anomaly of San Nicolas massive sulfide copper-zinc deposit(N=600 m)

References 37

[1]	李泽林, 姚长利, 郑元满. 基于Lp范数稀疏优化算法的重力三维反演[J]. 地球物理学报, 2019, 62(10):3699-3709.
[2]	HU Z L, LIU S, HU X Y, et al. Inversion of magnetic data using deep neural networks[J]. Physics of the Earth and Planetary Interiors, 2021, 311:106653. DOI URL
[3]	WEI C, LI X F, ZHEN X D. The group search based parallel algorithm for the serial Monte Carlo inversion method[J]. Applied Geophysics, 2010, 7(2): 127-134. DOI URL
[4]	于鹏, 王家林, 吴健生. 二度半长方体组合模型的重力模拟退火反演[J]. 地球物理学报, 2007, 50(3):882-889.
[5]	万玲, 林婷婷, 林君, 等. 基于自适应遗传算法的MRS-TEM联合反演方法研究[J]. 地球物理学报, 2013, 56(11):3728-3740.
[6]	朱凯光, 林君, 韩悦慧, 等. 基于神经网络的时间域直升机电磁数据电导率深度成像[J]. 地球物理学报, 2010, 53(3):743-750.
[7]	熊杰, 刘彩云, 邹长春. 基于粒子群优化算法的感应测井反演[J]. 物探与化探, 2013, 37(6):1141-1145.
[8]	熊杰, 孟小红, 刘彩云, 等. 基于差分进化的大地电磁反演[J]. 物探与化探, 2012, 36(3):448-451.
[9]	刘双, 刘天佑, 冯杰, 等. 蚁群算法在磁测资料反演解释中的应用[J]. 物探与化探, 2013, 37(1):150-154.
[10]	XIONG J, ZHANG T. Multiobjective particle swarm inversion algorithm for two-dimensional magnetic data[J]. Applied Geophysics, 2015, 12(2):127-136. DOI URL
[11]	张荣, 李伟平, 莫同. 深度学习研究综述[J]. 信息与控制, 2018, 47(4):385-397,410.
[12]	王昊, 严加永, 付光明, 等. 深度学习在地球物理中的应用现状与前景[J]. 地球物理学进展, 2020, 35(2): 642-655.
[13]	LI S C, LIU B, REN Y X, et al. Deep-learning inversion of seismic data[J]. IEEE Transactions on Geoscience and Remote Sensing, 2020, 58(3): 2135-2149. DOI URL
[14]	WU B, MENG D, WANG L, et al. Seismic impedance inversion using fully convolutional residual network and transfer learning[J]. IEEE Geoscience and Remote Sensing Letters, 2020, 17(12): 2140-2144. DOI URL
[15]	杨庭威, 曹丹平, 杜南樵, 等. 基于深度学习的接收函数横波速度预测[J]. 地球物理学报, 2022, 65(1):214-226.
[16]	伊小蝶, 吴帮玉, 孟德林, 等. 数据增广和主动学习在波阻抗反演中的应用[J]. 石油地球物理勘探, 2021, 56(4):707-715.
[17]	王德涛, 陈国雄. 基于时间卷积网络的地震波阻抗反演[J]. 地球科学, 2022, 47(4):1492-1506.
[18]	奚先, 黄江清. 基于卷积神经网络的地震偏移剖面中散射体的定位和成像[J]. 地球物理学报, 2020, 63(2):684-714.
[19]	WANG W Q, WANG Q, LU W K. Physics-constrained seismic impedance inversion based on deep learning[J]. IEEE Geoscience and Remote Sensing Letters, 2022, 19:7503305.
[20]	WU B Y, MENG D L, ZHAO H X. Semi-supervised learning for seismic impedance inversion using generative adversarial networks[J]. Remote Sensing, 2021, 13(5):909. DOI URL
[21]	HUANG J Y, NOWACK R L. Machine learning using U-net convolutional neural networks for the imaging of sparse seismic data[J]. Pure and Applied Geophysics, 2020, 177(6):2685-2700. DOI
[22]	ALFARRAJ M, ALREGIB G. Semisupervised sequence modeling for elastic impedance inversion[J]. Interpretation, 2019, 7(3):SE237-SE249. DOI URL
[23]	王鹤, 刘威, 席振铢. 基于深度置信网络的大地电磁非线性反演[J]. 中南大学学报, 2019, 26(9):2482-2494.
[24]	WU S H, HUANG Q H, ZHAO L. Convolutional neural network inversion of airborne transient electromagnetic data[J]. Geophysical Prospecting, 2021, 69(8):1761-1772. DOI URL
[25]	LIU Z G, CHEN H, REN Z Y, et al. Deep learning audiomagnetotellurics inversion using residual-based deep convolution neural network[J]. Journal of Applied Geophysics, 2021, 188:104309. DOI URL
[26]	MOQHADAS D. One-dimensional deep learning inversion of electromagnetic induction data using convolutional neural network[J]. Geophysical Journal International, 2020, 222(1):247-259. DOI URL
[27]	NOH K, YOON D, BYUN B. Imaging subsurface resistivity structure from airborne electromagnetic induction data using deep neural network[J]. Exploration Geophysics, 2019, 51(2):214-220. DOI URL
[28]	WANG H, LIU Y H, YIN C C, et al. Stochastic inversion of magnetotelluric data using deep reinforcement learning[J]. Geophysics, 2021, 87(1):E49-E61. DOI URL
[29]	GUO R, YAO H M, LI M K, et al. Joint inversion of audio-magnetotelluric and seismic travel time data with deep learning constraint[J]. IEEE Transactions on Geoscience and Remote Sensing, 2021, 59(9):7982-7995. DOI URL
[30]	LIU W, XI Z Z, WANG H, et al. Two dimensional deep learning inversion of magnetotelluric sounding data[J]. Journal of Geophysics and Engineering, 2021, 18(5):627-641. DOI URL
[31]	PUZYREV V, SWIDINSKY A. Inversion of 1D frequency- and time-domain electromagnetic data with convolutional neural networks[J]. Computer and Geosciences, 2021, 149:104681. DOI URL
[32]	张志厚, 廖晓龙, 曹云勇, 等. 基于深度学习的重力异常与重力梯度异常联合反演[J]. 地球物理学报, 2021, 64(4):1435-1452.
[33]	ZHANG L Z, ZHANG G B, LIU Y, et al. Deep learning for 3-D inversion of gravity data[J]. IEEE Transactions on Geoscience and Remote Sensing, 2022, 60:5905919.
[34]	YANG Q G, HU X Y, LIU S, et al. 3-D gravity inversion based on deep convolution neural networks[J]. IEEE Geoscience and Remote Sensing Letters, 2021, 19:3001305.
[35]	薛瑞洁, 熊杰, 张月, 等. 基于卷积神经网络的磁异常反演[J]. 现代地质, 2023, 37(1):194-209.
[36]	王蓉, 熊杰, 刘倩, 等. 基于深度神经网络的重力异常反演[J]. 物探与化探, 2022, 46(2):451-458.
[37]	KRIZHEVSKY A, SUTSKEVER I, HINTON G E. Imagenet classification with deep convolutional neural networks[M]//Advances in Neural Information Processing Systems 25.NIPS, 2012:1097-1105.