基于深度降噪自编码神经网络的中国大陆地壳厚度反演

程先琼; 蒋科植

doi:10.11939/jass.20200043

基于深度降噪自编码神经网络的中国大陆地壳厚度反演

程先琼^1, ,,
蒋科植^1,2

1.
中国成都 610059 成都理工大学地球物理学院
2.
中国四川乐山 614000 成都理工大学工程技术学院

基金项目: 国家自然科学基金（41774095、91755215）和成都理工大学工程技术学院院级基金项目（C122017008）共同资助

详细信息

通讯作者:
程先琼: e-mail：chxq@cdut.edu.cn

中图分类号: P313
计量
- 文章访问数: 799
- HTML全文浏览量: 441
- PDF下载量: 61
出版历程
- 收稿日期: 2020-03-30
- 修回日期: 2020-07-27
- 网络出版日期: 2021-03-19
- 发布日期: 2021-01-14

Inversion of crustal thickness in Chinese mainland based on deep denoising autoencoder neural network

Cheng Xianqiong^1, ,,
Jiang Kezhi^1,2

1.
College of Geophysics，Chengdu University of Technology，Chengdu 610059，China
2.
College of Engineering Technology，Chengdu University of Technology，Sichuan Leshan 614000，China

摘要

摘要: 本文采用基于数据驱动的深度降噪自编码网络构建了瑞雷面波群速度、相速度频散特性与地壳厚度的正反演函数关系，并利用最新频散模型反演了中国大陆的地壳厚度。对于神经网络架构体系的评价，除了考虑传统意义上的测试误差、训练误差之外，本文还用已知物理原理的正演结果与网络预测结果进行比较；在设计网络构架时，同时考虑地球模型和面波频散的正反演问题，即解码过程对应正演过程，编码过程对应反演过程。另外，针对观测频散数据包含噪声的特点，对训练样本加噪声，使解码器解码出无噪声输入，以达到对观测数据降噪的目的。对网络各种参数多次调试、分析再优化组合，最终获得稳健的神经网络，并据此反演出中国大陆的地壳厚度。本研究结果与已有的不同手段得到的地壳厚度模型的吻合度较高，表明深度降噪自编码神经网络能很好地揭示面波频散与地壳厚度之间的非线性关系，是利用面波频散反演地壳厚度的一种可行的和可信的方法。
- 深度学习 /
- 降噪 /
- 自编码神经网络 /
- 中国大陆 /
- 地壳厚度
Abstract: Chinese mainland is a composite continent composed of many micro-blocks, fold belts and orogenic belts after evolution over a long geological period. Chinese mainland crust is of complex crust-mantle structure, and crustal thickness is one of the most important parameters. At the same time, Rayleigh surface wave group velocity and phase velocity have strong constraints on the crust and upper mantle structure. In this paper, a data driven, named deep denoising autoencoder （DDAE） neural network is used to explore the relationship of forward and inversion between Rayleigh wave group velocity and phase velocity and crustal thickness, and we invert the crustal thickness of Chinese mainland by the latest dispersion model. In this paper, the evaluation of neural network architecture is put forward. In addition to the traditional test error and training error, it is compared with the result of network prediction with the forward process of the known physical principles. When designing the network architecture, the forward and inverse problems of the earth model and the surface wave dispersion are considered simultaneously, that is, the corresponding forward process is decoded, and the encoding process corresponds to the inversion process. At the same time, in view of the noise characteristics of the observed dispersion data, the training sample is polluted by noise, so the decoder decodes the noise-free input to denoising the observed data. After debugging, analyzing and optimizing the various parameters of the network, a robust neural network was finally obtained, and the crust thickness of the Chinese mainland was reproduced accordingly. The results of this study are in good agreement with the crustal thickness models obtained by different methods, suggesting that the deep denoising autoencoder neural network can well reveal the nonlinear relationship between surface wave dispersion and crustal thickness, and it is a feasible and credible method to solve the problem of surface wave dispersion inversion for crustal thickness.
- deep learning /
- denosing /
- autoencoder neural network /
- Chinese mainland /
- crustal thickness

HTML全文

下载: 全尺寸图片幻灯片

图 1 浅层降噪自编码网络结构图

Figure 1. Structure of shallow denoising autoencoder

下载: 全尺寸图片幻灯片

图 2 深度降噪自编码网络（DDAE）结构

Figure 2. Structure of DDAE

下载: 全尺寸图片幻灯片

图 3 深度降噪自编码的解码结果与基于频散、地壳厚度的物理关系的正演结果相关性

Figure 3. Comparison of DDAE decoder result with forward result based on physical relationship between crustal thickness and dispersion

下载: 全尺寸图片幻灯片

图 4 DDAE预测地壳厚度与实际地壳厚度的相关性

Figure 4. Comparison of crustal thickness from DDAE with true crustal thickness

下载: 全尺寸图片幻灯片

图 5 基于深度降噪自编码神经网络反演的中国大陆地壳厚度

Figure 5. Crustal thickness of Chinese mainland from inversion by DDAE method

下载: 全尺寸图片幻灯片

图 6 CRUST2.0模型的地壳厚度图（Bassin et al，2000）（a）及其与DDAE反演地壳厚度的相关性（b）

Figure 6. Crustal thickness from the model CRUST2.0 （Bassin et al，2000）（a） and its correlativity with the crustal thickness from DDAE inversion （b）

下载: 全尺寸图片幻灯片

图 7 CUB2.0模型的地壳厚度图（Shapiro，Ritzwoller，2002）（a）及其与DDAE反演地壳厚度的相关性（b）

Figure 7. Crustal thickness from the model CUB2.0 （Shapiro，Ritzwoller，2002）（a） and its correlativity with the crustal thickness from DDAE inversion （b）

下载: 全尺寸图片幻灯片

图 8 Feng模型的地壳厚度图（Feng et al，2017）（a）及其与DDAE反演地壳厚度的相关性（b）

Figure 8. Crustal thickness from the model Feng （Feng et al，2017）（a） and its correlativity with the crustal thickness from DDAE inversion

下载: 全尺寸图片幻灯片

图 9 DDAE模型计算的理论群速度与Shen等（2016）所得观测群速度的相对误差

Figure 9. The relative error between group velocity calculated by our model DDAE and observed by Shen et al （2016）

下载: 全尺寸图片幻灯片

参考文献(69)

郭良辉,孟小红,石磊,陈召曦. 2012. 优化滤波方法及其在中国大陆布格重力异常数据处理中的应用[J]. 地球物理学报,55(12):4078–4088.

Guo L H,Meng X H,Shi L,Chen Z X. 2012. Preferential filtering method and its application to Bouguer gravity anomaly of Chinese continent[J]. Chinese Journal of Geophysics,55(12):4078–4088 (in Chinese).

胡卫剑,郝天珧,秦静欣,李志伟,江为为,姜迪迪,邢健,胡立天,徐亚,雷受旻. 2014. 中国海陆莫霍面及深部地壳结构特征:以阿尔泰—巴士海峡剖面为例[J]. 地球物理学报,57(12):3932–3943.

Hu W J,Hao T Y,Qin J X,Li Z W,Jiang W W,Jiang D D,Xing J,Hu L T,Xu Y,Lei S M. 2014. Moho depth and deep crustal structure in the land and seas of China and adjacent areas:An example of the Altay-Bashi Channel profile[J]. Chinese Journal of Geophysics,57(12):3932–3943 (in Chinese).

黄建平,傅容珊,许萍,黄建华,郑勇. 2006. 利用重力和地形观测反演中国及邻区地壳厚度[J]. 地震学报,28(3):250–258.

Huang J P,Fu R S,Xu P,Huang J H,Zheng Y. 2006. Inversion of gravity and topography data for the crust thickness of China and its adjacency[J]. Acta Seismologica Sinica,28(3):250–258 (in Chinese).

林彬华,金星,陈惠芳,廖诗荣,黄玲珠,张燕明,巫立华. 2019. 基于反向传播神经网络的闽台M_L震级偏差分析与修正[J]. 地震学报,41(6):723–734.

Lin B H,Jin X,Chen H F,Liao S R,Huang L Z,Zhang Y M,Wu L H. 2019. Analysis and revision of magnitude M_L deviation between Fujian and Taiwan based on BP neural network[J]. Acta Seismologica Sinica,41(6):723–734 (in Chinese).

刘光鼎. 2007. 中国大陆构造格架的动力学演化[J]. 地学前缘,14(3):39–46.

Liu G D. 2007. Geodynamical evolution and tectonic framework of China[J]. Earth Science Frontiers,14(3):39–46 (in Chinese). doi: 10.1016/S1872-5791(07)60021-9

刘启民,赵俊猛,卢芳,刘宏兵. 2014. 用接收函数方法反演青藏高原东北缘地壳结构[J]. 中国科学:地球科学,44(4):668–679.

Liu Q M,Zhao J M,Lu F,Liu H B. 2014. Crustal structure of northeastern margin of the Tibetan Plateau by receiver function inversion[J]. Science China Earth Sciences,57(4):741–750. doi: 10.1007/s11430-013-4772-5

孙伟家,符力耘,魏伟,林羿,唐清雅. 2018. 中国东部地区的壳-幔过渡带结构[J]. 地球物理学报,61(3):845–855.

Sun W J,Fu L Y,Wei W,Li Y,Tang Q Y. 2018. The crust-mantle transition structures beneath eastern China[J]. Chinese Journal of Geophysics,61(3):845–855 (in Chinese).

滕吉文,曾融生,闫雅芬,张慧. 2002. 东亚大陆及周边海域Moho界面深度分布和基本构造格局[J]. 中国科学:D辑,32(2):89–100.

Teng J W,Zeng R S,Yan Y F,Zhang H. 2002. Depth distribution of Moho and tectonic framework in eastern Asian continent and its adjacent ocean areas[J]. Science in China:Series D,46(5):428–446.

隗永刚,杨千里,王婷婷,蒋长胜,边银菊. 2019. 基于深度学习残差网络模型的地震和爆破识别[J]. 地震学报,41(5):646–657.

Wei Y G,Yang Q L,Wang T T,Jiang C S,Bian Y J. 2019. Earthquake and explosion identification based on Deep Learning residual network model[J]. Acta Seismologica Sinica,41(5):646–657 (in Chinese).

危自根,储日升,杨小林,谢军,田忠华,凌媛,董非非. 2019. 汉中盆地及邻区地壳结构和地震活动性研究[J]. 地震学报,41(4):445–458.

Wei Z G,Chu R S,Yang X L,Xie J,Tian Z H,Ling Y,Dong F F. 2019. Crustal structure and seismic activity in the Hanzhong basin and its adjacent areas[J]. Acta Seismologica Sinica,41(4):445–458 (in Chinese).

熊小松. 2010. 中国大陆莫霍面深度与变化特征及其地球动力学意义[D]. 北京: 中国地质科学院: 91–93.

Xiong X S. 2010. Moho Depth and Variation of the Continent in China and Its Geodynamic Implications[D]. Beijing: Chinese Academy of Geological Sciences: 91–93 （in Chinese）.

杨宝俊,刘万崧,王喜臣,李勤学,王建民,赵雪平,李瑞磊. 2005. 中国东部大兴安岭重力梯级带域地球物理场特征及其成因[J]. 地球物理学报,48(1):86–97.

Yang B J,Liu W S,Wang X C,Li Q X,Wang J M,Zhao X P,Li R L. 2005. Geophysical characteristics of Daxinganling gravitational gradient zone in the east China and its geodynamic mechanism[J]. Chinese Journal of Geophysics,48(1):86–97 (in Chinese).

杨文采,瞿辰,侯遵泽,颜苹,于常青. 2017. 中国大陆造山带地壳密度结构特征[J]. 地质论评,63(3):539–548.

Yang W C,Qu C,Hou Z Z,Yan P,Yu C Q. 2017. Crustal density structures beneath orogenic belts of Chinese continent[J]. Geological Review,63(3):539–548 (in Chinese).

易桂喜,姚华建,朱介寿,van der Hilst R D. 2008. 中国大陆及邻区Rayleigh面波相速度分布特征[J]. 地球物理学报,51(2):402–411.

Yi G X,Yao H J,Zhu J S,van der Hilst R D. 2008. Rayleigh-wave phase velocity distribution in China continent and its adjacent regions[J]. Chinese Journal of Geophysics,51(2):402–411 (in Chinese).

易桂喜,姚华建,朱介寿,van der Hilst R D. 2010. 用Rayleigh面波方位各向异性研究中国大陆岩石圈形变特征[J]. 地球物理学报,53(2):256–268.

Yi G X,Yao H J,Zhu J S,van der Hilst R D. 2010. Lithospheric deformation of continental China from Rayleigh wave azimuthal anisotropy[J]. Chinese Journal of Geophysics,53(2):256–268 (in Chinese).

张广成,吴庆举,潘佳铁,张风雪,余大新. 2013. 利用H-K叠加方法和CCP叠加方法研究中国东北地区地壳结构与泊松比[J]. 地球物理学报,56(12):4084–4094.

Zhang G C,Wu Q J,Pan J T,Zhang F X,Yu D X. 2013. Study of crustal structure and Poisson ratio of NE China by H-K stack and CCP stack methods[J]. Chinese Journal of Geophysics,56(12):4084–4094 (in Chinese).

张攀,朱良保,陈浩朋,王清东,杨颖航. 2014. 用接收函数方法研究中国境内地壳结构[J]. 地震学报,36(5):850–861.

Zhang P,Zhu L B,Chen H P,Wang Q D,Yang Y H. 2014. Crustal structure in China from teleseismic receiver function[J]. Acta Seismologica Sinica,36(5):850–861 (in Chinese).

曾融生,孙为国,毛桐恩,林中洋,胡鸿翔,陈光英. 1995. 中国大陆莫霍界面深度图[J]. 地震学报,17(3):322–327.

Zeng R S,Sun W G,Mao T E,Lin Z Y,Hu H X,Chen G Y. 1995. The depth of Moho in the mainland of China[J]. Acta Seismologica Sinica,17(3):322–327 (in Chinese).

朱介寿,曹家敏,蔡学林,严忠琼,曹小林. 2002. 东亚及西太平洋边缘海高分辨率面波层析成像[J]. 地球物理学报,45(5):646–664.

Zhu J S,Cao J M,Cai X L,Yan Z Q,Cao X L. 2002. High resolution surface wave tomography in East Asia and West Pacific marginal seas[J]. Chinese Journal of Geophysics,45(5):646–664 (in Chinese).

Bassin C G L, Laske G, Masters G. 2000. The current limits of resolution for surface wave tomography in North America[J]. EOS Trans Am Geophys Union, 81（48）: F897.

Chen Y L,Niu F L,Liu R F,Huang Z B,Tkalčić H,Sun H,Chan W. 2010. Crustal structure beneath China from receiver function analysis[J]. J Geophys Res:Solid Earth,115(B3):B03307.

Cheng X Q,Liu Q H,Li P P,Liu Y. 2019. Inverting Rayleigh surface wave velocities for crustal thickness in eastern Tibet and the western Yangtze craton based on deep learning neural networks[J]. Nonlin Process Geophys Discuss,26:61–71. doi: 10.5194/npg-26-61-2019

de Wit R W L,Valentine A P,Trampert J. 2013. Bayesian inference of Earth’s radial seismic structure from body-wave travel times using neural networks[J]. Geophys J Int,195(1):408–422. doi: 10.1093/gji/ggt220

de Wit R W L,Käufl P J,Valentine A P,Trampert J. 2014. Bayesian inversion of free oscillations for earth’s radial (an)elastic structure[J]. Phys Earth Planet Int,237:1–17. doi: 10.1016/j.pepi.2014.09.004

Devilee R J R,Curtis A,Roy-Chowdhury K. 1999. An efficient,probabilistic neural network approach to solving inverse problems:Inverting surface wave velocities for Eurasian crustal thickness[J]. J Geophys Res:Solid Earth,104(B12):28841–28857. doi: 10.1029/1999JB900273

Feng M,An M J,Dong S W. 2017. Tectonic history of the Ordos block and Qinling orogen inferred from crustal thickness[J]. Geophys J Int,210(1):303–320. doi: 10.1093/gji/ggx163

He R Z,Shang X F,Yu C Q,Zhang H J,van der Hilst R D. 2014. A unified map of Moho depth and v_P/v_S ratio of continental China by receiver function analysis[J]. Geophys J Int,199:1910–1918. doi: 10.1093/gji/ggu365

Hinton G E,Salakhutdinov R R. 2006. Reducing the dimensionality of data with neural networks[J]. Science,313(5786):504–507. doi: 10.1126/science.1127647

Huang Z X,Su W,Peng Y J,Zheng Y J,Li H Y. 2003. Rayleigh wave tomography of China and adjacent regions[J]. J Geophys Res:Solid Earth,108(B2):ES4-1–4-10.

Lebedev S,Adam J M C,Meier T. 2013. Mapping the Moho with seismic surface waves:A review resolution analysis and recommended inversion strategies[J]. Tectonophysics,609(1):377–394.

Legendre C P,Deschamps F,Zhao L,Chen Q F. 2015. Rayleigh-wave dispersion reveals crust-mantle decoupling beneath eastern Tibet[J]. Sci Rep,5(16644):1–7.

Li Y H,Gao M T,Wu Q J. 2014. Crustal thickness map of the Chinese mainland from teleseismic receiver functions[J]. Tectonophysics,611:51–60. doi: 10.1016/j.tecto.2013.11.019

Liu Q H,Li X,Ye M,Ge S S,Du X S. 2014. Drift compensation for electronic noise by semi-supervised domain adaption[J]. IEEE Sensors Journal,14(3):657–665. doi: 10.1109/JSEN.2013.2285919

Liu Q H,Hu X N,Ye M,Cheng X Q,Li F. 2015. Gas recognition under sensor drift by using deep learning[J]. Inter J Intelli Syst,30(8):907–922. doi: 10.1002/int.21731

Liu Q Y,van der Hilst R D,Li Y,Yao H J,Chen J H,Guo B,Qi S H,Wang J,Huang H,Li S C. 2014. Eastward expansion of the Tibetan Plateau by crustal flow and strain partitioning across faults[J]. Nat Geosci,7(5):361–365. doi: 10.1038/ngeo2130

Marone C. 2018. Training machines in earthly ways[J]. Nat Geosci,11(5):301–302. doi: 10.1038/s41561-018-0117-5

Masters G, Barmine M P, Pasadena K S. 2007. Mineos User’s Manual in Computational Infrastructure for Geodynamics[M]. Pasadena: California Institute of Technology: 9–89.

Meier U,Curtis A,Trampert J. 2007. Global crustal thickness from neural network inversion of surface wave data[J]. Geophys J Int,169:706–722. doi: 10.1111/j.1365-246X.2007.03373.x

Meier U,Trampert J,Curtis A. 2009. Global variations of temperature and water content in the mantle transition zone from higher mode surface waves[J]. Earth Planet Sci Lett,282(1/2/3/4):91–101.

Rifai S, Vincent P, Muller X, Glorot X, Bengio Y. 2011. Contractive autoencoders: Explicit invariance during feature extraction[C]//Proceedings of the 28th International Conference on Machine Learning. Madison: Omnipress: 833–840.

Rumelhart D E,Hinton G E,Williams R J. 1986. Learning representations by back-propagating errors[J]. Nature,323(6088):533–536. doi: 10.1038/323533a0

Shapiro N M,Ritzwoller M H. 2002. Monte-Carlo inversion for a global shear-velocity model of the crust and upper mantle[J]. Geophys J Int,151:88–105. doi: 10.1046/j.1365-246X.2002.01742.x

Shen W S,Ritzwoller M H,Kang D,Kim Y,Lin F C,Ning J Y,Wang W T,Zheng Y,Zhou L Q. 2016. A seismic reference model for the crust and uppermost mantle beneath China from surface wave dispersion[J]. Geophys J Int,206(2):954–979. doi: 10.1093/gji/ggw175

Stolk W,Kaban M,Beekman F,Tesauro M,Mooney W D,Cloetingh S. 2013. High resolution regional crustal models from irregularly distributed data:Application to Asia and adjacent areas[J]. Tectonophysics,602:55–68. doi: 10.1016/j.tecto.2013.01.022

van der Baan M,Jutten C. 2000. Neural networks in geophysical applications[J]. Geophysics,65(4):1032–1047. doi: 10.1190/1.1444797

Vincent P, Larochelle H, Bengio Y, Manzagol P A. 2008. Extracting and composing robust features with denoising autoencoders[C]//Proceedings of the 25th International Conference on Machine Learning. New York, NY: ACM: 1096–1103.

Wang W L,Wu J P,Fang L H,Lai G J,Cai Y. 2017. Crustal thickness and Poisson’s ratio in southwest China based on data from dense seismic arrays[J]. J Geophys Res:Solid Earth,122(9):7219–7235. doi: 10.1002/2017JB013978

Wei Z G,Chen L,Li Z W,Ling Y,Li J. 2016. Regional variation in Moho depth and Poisson’s ratio beneath eastern China and its tectonic implications[J]. J Asia Earth Sci,115:308–320. doi: 10.1016/j.jseaes.2015.10.010

Zhang P,Yao H J,Chen L,Fang L H,Wu Y,Feng J K. 2019. Moho depth variations from receiver function imaging in the northeastern North China Craton and its tectonic implications[J]. J Geophys Res:Solid Earth,124(2):1–19.

Zhou Y,Nolet G,Dahlen F A,Laske G. 2006. Global upper-mantle structure from finite-frequency surface-wave tomography[J]. J Geophys Res:Solid Earth,111:B04304.