深度学习多地球物理场三维联合反演

目前深部资源勘探进入攻坚克难阶段，尤其是对深部矿产、隐伏矿产 资源储量的摸排，是当前应对拓展未来资源发展和提高国际竞争力的当务 之急。因此，开发针对深部资源勘探的技术，进行第二找矿空间资源高精 度勘查，在三维空间内获得资源分布，具有重大的战略意义。 随着地球物理探测技术和仪器的进步，日益增多的物探方法和地球物 理数据有效地降低了单一地球物理数据反演的多解性，为了更好地利用海 量地球物理数据所带来的信息，进行三维高精度矿产资源的勘探，快速发 展的机器学习方法提供了新的可能，这是因为其在降低数据处理过程中的 计算成本和人工成本的优势十分显著，采取新的手段有效地整合已知信息 并用于未知区域的预测是地球物理发展中值得探索的新方向。目前在地球 物理领域已有大量的工作应用了深度学习方法处理反演问题。然而，对于 多种数据不同维度、不同尺寸的输入，基于深度学习的联合反演仍然未被 实现，而实际应用中获得的地球物理数据常常具有多种维度、多种尺寸。 为了寻找解决此类问题的有效途径，本文借鉴了深度学习领域中多模 态学习的概念，将多维度、多尺寸的地球物理数据类比为计算机领域的声 音（一维）、图像（二维）、视频（三维）等多模态数据，互相补充信息联 合反演，以解决三维高精度找矿的具体问题。本文构建并开创性地提出了 三维联合反演的深度学习框架，将测井、重、磁、音频大地电磁方法视为 地下异常的多种数据的来源，将联合反演地下三维地质体作为示例，研究 了融合多种地球物理数据的人机交互算法。研究结果证明，我们所构建的 深度学习框架在解决多维多尺寸的地球物理数据联合反演问题上的可行性， 与单一地球物理数据反演或少量地球物理数据联合反演对比具有显著的优 势。最后将所提出的框架应用于实际场景中，构建了三种不同类型的几何 形状的数据集，根据反演结果不断调整数据集，证明了选择更匹配的训练 集和测试集对深度学习效果的重要性，并提出了一种基于此方向改善深度 学习效果的思路——在构建数据集时必要加入观测数据以及优化拟合差的 引导，有利于提高测试集和训练集的匹配程度从而提高预测准确率。

[1] LI Y, OLDENBURG D W J G. 3 -D inversion of magnetic data [J]. 1996, 61(2): 394-408.
[2] LI Y, OLDENBURG D W J G. Separation of regional and residual magnetic field data [J]. 1998, 63(2): 431-9.
[3] GALLARDO-DELGADO L A, PéREZ-FLORES M A, GóMEZ-TREVIñO E J G. A versatile algorithm for joint 3D inversion of gravity and magnetic data [J]. 2003, 68(3): 949-59.
[4] HOVERSTEN G M, CASSASSUCE F, GASPERIKOVA E, et al. Direct reservoir parameter estimation using joint inversion of marine seismic AVA and CSEM data [J]. 2006, 71(3): C1-C13.
[5] LIANG L, ABUBAKAR A, HABASHY T M J G. Estimating petrophysical parameters and average mud-filtrate invasion rates using joint inversi on of induction logging and pressure transient data [J]. 2011, 76(2): E21 -E34.
[6] GAO G, ABUBAKAR A, HABASHY T M J G. Joint petrophysical inversion of electromagnetic and full-waveform seismic dataJoint petrophysical EM and seismic inversion [J]. 2012, 77(3): WA3-WA18.
[7] GALLARDO L A, MEJU M A J J O G R S E. Joint two‐dimensional DC resistivity and seismic travel time inversion with cross‐gradients constraints [J]. 2004, 109(B3).
[8] HU W, ABUBAKAR A, HABASHY T M J G. Joint electromagnetic and seismic inversion using structural constraints [J]. 2009, 74(6): R99 -R109.
[9] LAN T, LIU N, HAN F, et al. Joint petrophysical and structural inversion of electromagnetic and seismic data based on volume integral equation method [J]. 2018, 57(4): 2075-86.
[10] DELL’AVERSANA P, BERNASCONI G, MIOTTI F, et al. Joint inversion of rock properties from sonic, resistivity and density well‐log measurements [J]. 2011, 59(6): 1144-54.
[11] HABER E, OLDENBURG D J I P. Joint inversion: a structural approach [J]. 1997, 13(1): 63.
[12] COLOMBO D, DE STEFANO M J T L E. Geophysical modeling via simultaneous joint inversion of seismic, gravity, and electromagnetic data: Application to prestack depth imaging [J]. 2007, 26(3): 326 -31.
[13] GALLARDO L A J G R L. Multiple cross‐gradient joint inversion for geospectral imaging [J]. 2007, 34(19).
[14] GALLARDO L, FONTES S, MEJU M, et al. Robust geophysical integration through structure-coupled joint inversion and multispectral fusion of seismic reflection, magnetotelluric, magnetic, and gravity images: Example from Santos Basin, offshore Brazil [J]. 201 2, 77(5): B237-B51.
[15] SUN J, LI Y. Joint inversion of multiple geophysical data using guided fuzzy c -means clustering [J]. 2016, 81(3): ID37 -ID57.
[16] ZEYEN H, POUS J J G J I. 3-D joint inversion of magnetic and gravimetric data with a priori information [J]. 1993, 112(2): 244-56.
[17] MOORKAMP M, HEINCKE B, JEGEN M, et al. A framework for 3 -D joint inversion of MT, gravity and seismic refraction data [J]. Geophysical Journal International, 2011, 184(1): 477 -93.
[18] ABDOLLAHI S, ZEYEN H, ARDESTANI V E, et al. 3D joint inversion of gravity data and Rayleigh wave group velocities to resolve shear-wave velocity and density structure in the Makran subduction zone, south -east Iran [J]. 2019, 173: 275 -90.
[19] KRIZHEVSKY A, SUTSKEVER I, HINTON G E J C O T A. Imagenet classification with deep convolutional neural networks [J]. 2017, 60(6): 84 -90.
[20] SZEGEDY C, LIU W, JIA Y, et al. Going deeper with convolutions; proceedings of the Proceedings of the IEEE conference on computer vision and pattern recognition, F, 2015 [C].
[21] WANG F, CASALINO L P, KHULLAR D J J I M. Deep learning in medicine—promise, progress, and challenges [J]. 2019, 179(3): 293 -4.
[22] SALMAN A G, KANIGORO B, HERYADI Y. Weather forecasting using deep learning techniques; proceedings of the 2015 international conference on advanced computer science and information systems (ICACSIS), F, 2015 [C]. Ieee.
[23] ZHU X X, TUIA D, MOU L, et al. Deep learning in remote sensing: A comprehensive review and list of resources [J]. 2017, 5(4): 8 -36.
[24] PUZYREV V J G J I. Deep learning electromagnetic inversion with convolutional neural networks [J]. 2019, 218(2): 817 -32.
[25] PUZYREV V, CELA J M J G J I. A review of block Krylov subspace methods for multisource electromagnetic modelling [J]. 2015, 202(2): 1241-52.
[26] ZHANG Y, PAULSON K J G P. Magnetotelluric inversion using regularized Hopfield neural networks [Link] [J]. 1997, 45(5): 725 -43.
[27] 韩卫雪, 周亚同, 石油物探 池 J. 基于深度学习卷积神经网络的地震数据随机噪声去除[J]. 2018, 57(6): 862-9.
[28] 奚先, 地球物理学进展 黄 J. 散射波场的深度学习反演成像法 [J]. 2018, 33(6): 2483-9.
[29] 李建国, 张卫东, 科技创新与应用 刘 J. 深度学习在测井岩性识别中的应用 [J]. 2015, (14): 21-2.
[30] 安鹏, 地球物理学进展 曹 J. 基于深度学习的测井岩性识别方法研究与应用 [J]. 2018, 33(3): 1029-34.
[31] 宋超, 郭智奇, 鹿琪, et al. 基于支持向量机法识别砂岩中流体类型 [J]. 2015, 30(2): 616-20.
[32] 单敬福, 陈欣欣, 赵忠军, et al. 利用 BP 神经网络法对致密砂岩气藏储集层复杂岩性的识别 [J]. 2015, (3): 1257-63.
[33] 张彦周, 刘叶玲, 勘探地球物理进展 谢 J. 支持向量机在储层厚度预测中的应用 [J]. 2005, 28(6): 422-4.
[34] LI Y, YANG D. Imaging of steel casing’s conductivity using surface electrical data and a deep learning approach [M]. SEG Technical Program Expanded Abstracts 2020. Society of Exploration Geophysicists. 2020: 636 -40.
[35] LI Y, YANG D J G. Electrical imaging of hydraulic fracturing fluid using steel -cased wells and a deep -learning methodElectrical hydraulic fracturing imaging [J]. 2021, 86(4): E315-E32.
[36] GUO R, YAO H M, LI M, et al. Joint inversion of audio-magnetotelluric and seismic travel time data with deep learning constraint [J]. 2020, 59(9): 7982 -95.
[37] OH S, NOH K, YOON D, et al. Salt delineation from electromagnetic data using convolutional neural networks [J]. 2018, 16(4): 519-23.
[38] OH S, NOH K, SEOL S J, et al. Cooperative deep learning inversion of controlled -source electromagnetic data for salt delineation [J]. Geophysics, 2020, 85(4): E121-E37.
[39] 张志厚, 廖晓龙, 曹云勇, et al. 基于深度学习的重力异常与重力梯度异常联合反演 [J]. 2021, 64(4): 1435-52.
[40] GOODFELLOW I J, BULATOV Y, IBARZ J, et al. Multi-digit number recognition from street view imagery using deep convolutional neural networks [J]. 2013.
[41] LE CUN Y, BOSER B, DENKER J, et al. Handwritten digit recognition width a backpropagation network. Advances in Neural Information Systems, 2 [Z]. Morgan Kaufman, San Mateo, CA. 1990.
[42] BISHOP C M, NASRABADI N M. Pattern recognition and machine learning [M]. Springer, 2006.
[43] ZHANG Y, LING C J N C M. A strategy to apply machine learn ing to small datasets in materials science [J]. 2018, 4(1): 25.
[44] RUDIN C J N M I. Stop explaining black box machine learning models for high stakes decisions and use interpretable models instead [J]. 2019, 1(5): 206 -15.
[45] LEIJNEN S, VEEN F V. The neural netwo rk zoo; proceedings of the Proceedings, F, 2020 [C]. MDPI.
[46] POUYANFAR S, SADIQ S, YAN Y, et al. A survey on deep learning: Algorithms, techniques, and applications [J]. 2018, 51(5): 1 -36.
[47] YOUNG T, HAZARIKA D, PORIA S, et al. Recent trends in deep learning b ased natural language processing [J]. 2018, 13(3): 55 -75.
[48] BENGIO Y, SIMARD P, FRASCONI P J I T O N N. Learning long -term dependencies with gradient descent is difficult [J]. 1994, 5(2): 157-66.
[49] SUTSKEVER I, MARTENS J, HINTON G E. Generating text with recurrent neural networks; proceedings of the Proceedings of the 28th international conference on machine learning (ICML-11), F, 2011 [C].
[50] MIKOLOV T, JOULIN A, CHOPRA S, et al. Learning longer memory in recurrent neural networks [J]. 2014.
[51] SALEHINEJAD H, SANKAR S, BARFETT J, et al. Recent advances in recurrent neural networks [J]. 2017.
[52] BRYANT G A J F I P. Animal signals and emotion in music: Coordinating affect across groups [J]. 2013, 4: 990.
[53] HSU C-C, ZHUANG Y-X, LEE C-Y J A S. Deep fake image detection based on pairwise learning [J]. 2020, 10(1): 370.
[54] MARRA F, GRAGNANIELLO D, COZZOLINO D, et al. Detection of gan -generated fake images over social networks; proceedings of the 2018 IEEE conference on multimedia information processing and retrieval (MIPR), F, 2018[C]. IEEE.
[55] AGGARWAL A, MITTAL M, BATTINENI G J I J O I M D I. Generative adversarial network: An overview of theory and applications [J]. 2021, 1(1): 100004.
[56] LONG J, SHELHAMER E, DARRELL T. Fully convolutional networks for semantic segmentation; proceedings of the Proceedings of the IEEE conference on computer vision and pattern recognition, F, 2015 [C].
[57] NIE D, WANG L, GAO Y, et al. Fully convolutional networks for multi-modality isointense infant brain image segmentation; proceedings of the 2016 IEEE 13Thinternational symposium on biomedical imaging (ISBI), F, 2016 [C]. IEEE.
[58] RONNEBERGER O, FISCHER P, BROX T. U-net: Convolutional networks for biomedical image segmentation; proceedings of the Medical Image Computing and Computer-Assisted Intervention –MICCAI 2015: 18th International Conference, Munich, Germany, October 5 -9, 2015, Proceedings, Part III 18, F, 2015 [C]. Springer.
[59] WENG Y, ZHOU T, LI Y, et al. Nas-unet: Neural architecture search for medical image segmentation [J]. 2019, 7: 44247 -57.
[60] BALTRUŠAITIS T, AHUJA C, MORENCY L-P J I T O P A, et al. Multimodal machine learning: A survey and taxonomy [J]. 2018, 41(2): 423 -43.
[61] D'MELLO S K, KORY J J A C S. A review and meta -analysis of multimodal affect detection systems [J]. 2015, 47(3): 1 -36.
[62] SNOEK C G, WORRING M, SMEULDERS A W. Early versus late fusion in semantic video analysis; proceedings of the Proceedings of the 13th annual ACM international conference on Multimedia, F, 2005 [C].
[63] PIGNATELLI A, NICOLOSI I, CARLUCCIO R, et al. Graphical interactive generation of gravity and magnetic fields [J]. 2011, 37(4): 567 -72.
[64] CHEN T, ZHANG G J C, GEOSCIENCES. Forward modeling of gravity anomalies based on cell mergence and parallel computing [J]. 2018, 120: 1 -9.
[65] EFREMOVA A, BULYCHEV A J M U G B. The accuracy of d etermining the effective layer density at gravity logging [J]. 2008, 63(6): 405.
[66] LINDSEY N J, NEWMAN G A J G. Improved workflow for 3D inverse modeling of magnetotelluric data: Examples from five geothermal systems [J]. 2015, 53: 527 -32.
[67] CHERKOSE B A, SAIBI H J G. Investigation of the Ayrobera geothermal field using 3D magnetotelluric data inversion, Afar depression, NE Ethiopia [J]. 2021, 94: 102114.
[68] HANNESON C, UNSWORTH M. Magnetotelluric Exploration at Mount Meager, Southwestern Canada: Imaging the Magma Body Beneath a Holocene Volcanic Center; proceedings of the AGU Fall Meeting Abstracts, F, 2021 [C].
[69] KARLSDóTTIR R, VILHJáLMSSON A M, GUðNASON E Á J J O V, et al. Three dimensional inversion of magnetotelluric (MT) resistivity data from Reykjanes high temperature field in SW Iceland [J]. 2020, 391: 106498.
[70] GHALATI F H, CRAVEN J A, MOTAZEDIAN D, et al. Modeling a fractured geothermal reservoir using 3-D AMT data inversion: insights from Garibaldi Volcanic Belt, British Columbia, Canada [J]. 2022, 105: 102528.
[71] KELBERT A, MEQBEL N, EGBERT G D, et al. ModEM: A modular system for inversion of electromagnetic geophysical data [J]. 2014, 66: 40 -53.
[72] HAO Z, WANG Z, BAI D, et al. Intelligent detectio n of steel defects based on improved split attention networks [J]. 2022, 9: 1478.
[73] ELKINS A, FREITAS F F, SANZ V J A P A. Developing an app to interpret chest X-rays to support the diagnosis of respiratory pathology with artificial intelligence [J]. 2019.
[74] HARDER M, SMITH B S, HETMAN C, et al. The evolution of geological models for the DO-27 kimberlite, NWT, Canada: Implications for evaluation [J]. 2009, 112: 61-72.
[75] MACNAE J J J O G E. Applications of geophysics for the detection and exploration of kimberlites and lamproites [J]. 1995, 53(1 -3): 213-43.
[76] KEATING P, SAILHAC P J G. Use of the analytic signal to identify magnetic anomalies due to kimberlite pipes [J]. 2004, 69(1): 180 -90.
[77] POWER M, HILDES D. Geophysical strategies for kimberlite exploration in northern Canada; proceedings of the Proceedings of Exploration, F, 2007 [C].
[78] JANSEN J, WITHERLY K. The Tli Kwi Cho kimberlite complex, Northwest Territories, Canada: A geophysical case study [M]. SEG Technical Program Expanded Abstracts 2004. Society of Explora tion Geophysicists. 2004: 1147 -50.
[79] DEVRIESE S G, DAVIS K, OLDENBURG D W J I. Inversion of airborne geophysics over the DO-27/DO-18 kimberlites—Part 1: Potential fields [J]. 2017, 5(3): T299-T311.
[80] FOURNIER D, KANG S, MCMILLAN M S, et al. Inversion of airbor ne geophysics over the DO-27/DO-18 kimberlites—Part 2: Electromagnetics [J]. 2017, 5(3): T313-T25.
[81] KANG S, FOURNIER D, OLDENBURG D W J I. Inversion of airborne geophysics over the DO-27/DO-18 kimberlites—Part 3: Induced polarization [J]. 2017, 5(3): T327-T40.
[82] ASTIC T. A framework for joint petrophysically and geologically guided geophysical inversion [D]; University of British Columbia, 2020.