基于MPI的面波有限差分正演模拟

doi:10.13278/j.cnki.jjuese.20190021

摘要/Abstract

摘要： 近年来，瑞利波波形反演技术因其避开了常规频散曲线计算，直接进行波场计算和反演不再受水平层状介质理论假设的限制，得到广大学者的高度重视。但瑞利波波形反演过程中需要不断进行波场正演和逆推计算。另外，由于浅地表速度较小，模拟计算时需要较小的网格间距才能避免数值频散，这无疑大大增加了正演模拟的计算量。对于这一问题，通常采用并行化设计来提高正演模拟的计算效率。本文基于消息传递接口（MPI）并行有限差分算法，以区域分解思路将模型区间分解成若干子区域，各区域互相通信，共同完成对模型的正演计算。并详细给出了区域分解、坐标转换、区域通信、波场合并等并行方案中的具体实现方法和实现步骤。通过对弹性模型、Kelvin黏弹性模型和标准线弹性固体（SLS）黏弹性模型不同并行方案的计算结果进行分析，验证了本文并行方案的可行性和有效性。并行计算结果表明，与单处理器计算时间相比，增加处理器数目可以明显减少计算时间，但随着处理器数目的增加，不同处理器之间的通信时间也增大；因此，并行时需要选择合适的处理器数目。对于黏弹性介质模型，SLS黏弹性模型的并行计算效率优于Kelvin黏弹性模型。

关键词: MPI, 高阶交错网格, 有限差分, 镜像法, CPML, 区域分解

Abstract: In recent years, the technology of Rayleigh-waveform inversion is highly valued by scholars, because the wave field is calculated and inverted directly without the calculation of conventional dispersion curves. In other words, the waveform inversion method is no longer limited by the theoretical assumption of horizontal layered media. Rayleigh waveform inversion requires repeated forward and inverse calculations of wave field; in addition, due to the small velocity of the shallow surface, the simulation requires a small grid space to avoid numerical dispersion, which undoubtedly greatly increases the forward modeling calculation amount. Based on the idea of message passing interface (MPI) method, we applied a parallel finite-difference algorithm to wave field simulation to improve the computation efficiency of the forward modeling. Firstly, the whole calculation region was decomposed into several sub-regions; and then, the wave field was computed for each sub-region; finally, the whole wave field was completed together by communicating among sub-regions. In this paper, the detailed implementation methods and steps of the parallel schemes, such as region decomposition, coordinate transformation, region communication, and wave field combination and so on, are given. The analyzing results of the different parallel schemes in elastic model, Kelvin and standard linear solid (SLS) viscoelastic model show that our parallel scheme is feasible and effective. The parallel computation results indicate that multiple processors can significantly reduce the computing time compared with a single processor; however, the communication time between different processors also increases. It is necessary to select the appropriate number of processors in the parallel process. For viscoelastic medium model, the parallel computation efficiency of SLS viscoelastic model is better than that of Kelvin viscoelastic model.

Key words: MPI, high-order staggered-grid, finite-difference, image-method, CPML, region decomposition

中图分类号:

P631.4

邵广周, 赵凯鹏, 吴华. 基于MPI的面波有限差分正演模拟[J]. 吉林大学学报(地球科学版), 2020, 50(1): 294-303.

Shao Guangzhou, Zhao Kaipeng, Wu Hua. Finite Difference Forward Modeling of Surface Waves Based on MPI[J]. Journal of Jilin University(Earth Science Edition), 2020, 50(1): 294-303.

参考文献

[1] 吴华,李庆春,邵广周. 瑞利波波形反演的发展现状及展望[J]. 物探与化探, 2018, 42(6):1103-1111. Wu Hua, Li Qingchun, Shao Guangzhou. Development Status and Prospect of Rayleigh Waveform Inversion[J]. Geophysical and Geochemical Exploration, 2018, 42(6):1103-1111.
[2] Tarantola A. Inversion of Seismic Reflection Data in the Acoustic Approximation[J]. Geophysics, 1984(49):1259-1266.
[3] Haskell N A. The Dispersion of Surface Waves on Multilayered Media[J]. Bulletin of the Seismological Society of America, 1953, 43(1):17-34.
[4] Schwab F. Surface Wave Dispersion Computation:Knopoff's Method[J]. Bulletin of the Seismological Society of America, 1970, 60(5):1491-1520.
[5] Carcione J M, Herman J C, Kroode A P E. Seismic Modeling[J]. Geophysics, 2002, 67(4):1304-1325.
[6] 裴正林,牟永光. 地震波传播数值模拟[J]. 地球物理学进展, 2004, 19(4):933-941. Pei Zhenglin, Mou Yongguang. Numerical Simulation of Seismic Wave Propagation[J]. Progress in Geophysics, 2004, 19(4):933-941.
[7] Virieux J. P-SV Wave Propagation in Heterogeneous Media:Velocity-Stress Finite-Difference Method[J]. Geophysics, 1986, 51(4):889-901.
[8] Levander A R. Fourth-Order Finite-Difference P-SV Seismograms[J]. Geophysics, 1988, 53(11):1425-1436.
[9] Berenger J P. A Perfectly Matched Layer for the Absorption of Electromagnetic Waves[J]. Journal of Computational Physics, 1994, 114(2):185-200.
[10] Collino F, Tsogka C. Application of the Perfectly Matched Absorbing Layer Model to the Linear Elastodynamic Problem in Anisotropic Heterogeneous Media[J]. Geophysics, 2001, 66(1):294-307.
[11] Johan O A R, Joakim O B, William W S. Viscoelastic Finite-Difference Modeling[J]. Geophysics, 1994, 59(9):1444-1456.
[12] Zahradnık J, Priolo E. Heterogeneous Formulations of Elastodynamic Equations and Finite Difference Schemes[J]. Geophys J Int, 1995, 120(3):663-676.
[13] Bohlen T, Saenger E H. Accuracy of Heterogeneous Staggered-Grid Finite-Difference Modeling of Rayleigh Waves[J]. Geophysics, 2006, 71(4):T109-T115.
[14] Robertsson O A J. A Numerical Free-Surface Condition for Elastic/Viscoelastic Finite-Difference Modeling in the Presence of Topography[J]. Geophysics, 1996, 61(6):1921-1934.
[15] Rune M. Free-Surface Boundary Conditions for Elastic Staggered-Grid Modeling Schemes[J]. Geophysics, 2002, 67(5):1616-1623.
[16] 周竹生,刘喜亮,熊孝雨. 弹性介质中瑞雷面波有限差分法正演模拟[J]. 地球物理学报, 2007, 50(2):567-573. Zhou Zhusheng, Liu Xiliang, Xiong Xiaoyu. Finite Difference Modeling of Rayleigh Surface Wave in Elastic Media[J]. Chinese Journal of Geophysics, 2007, 50(2):567-573.
[17] Thomas B, Tobias M M, Bernd M. Parallel Finite-Difference Modeling of Seismic Wave Scattering in 3-D Elastic Random Media[C]//SEG Technical Program Expanded Abstracts 2001. Tulsa:SEG, 2001:1147-1150.
[18] Felix R, Mauricio H, Albert F, et al. Generalized Elastic Staggered Grids on Multi-GPU Platforms[C]//SEG Technical Program Expanded Abstracts 2012. Tulsa:SEG, 2012:1-5.
[19] Liu Xiaobo, Chen Jingyi, Lan Haiqiang, et al. Numerical Modeling of Wave Propagation with an Irregular Free Surface and Graphic Processing Unit (GPU) Implementation[C]//SEG Technical Program Expanded Abstracts 2015. Tulsa:SEG, 2015:3704-3709.
[20] 周洲. 基于多线程并行计算的有限差分法弹性波数值模拟[J]. 硅谷, 2015, 4:76. Zhou Zhou. Numerical Simulation of Elastic Waves Using Finite Difference Method Based on Multithread Parallel Computation[J]. Silicon Valley, 2015, 4:76.
[21] 张明财,熊章强,张大洲. 基于MPI的三维瑞雷面波有限差分并行模拟[J]. 石油物探, 2013, 52(4):354-362. Zhang Mingcai, Xiong Zhangqiang, Zhang Dazhou. Finite Difference Parallel Simulation of 3D Rayleigh Surface Wave Based on MPI[J]. Geophysical Prospecting for Petroleum, 2013, 52(4):354-362.
[22] 钟飞, 张伟, 焦标强,等. 可控震源粘弹性波动方程有限差分模拟[J]. 煤田地质与勘探, 2011, 39(2):57-65. Zhong Fei, Zhang Wei, Jiao Biaoqiang, et al. Finite Difference Simulation of Viscoelastic Wave Equation in Vibroseis[J]. Coal Geology & Exploration, 2011, 39(2):57-65.
[23] Foster I. Designing and Building Parallel Programs:Concepts and Tools for Parallel Software Engineering[M]. New Jersey:Addison-Wesley, 1995.
[24] 杨庆节,刘财,耿美霞,等.交错网格任意阶导数有限差分格式及差分系数推导[J]. 吉林大学学报(地球科学版), 2014, 44(1):375-385. Yang Qingjie, Liu Cai, Geng Meixia, et al. Staggered Grid Finite Difference Scheme and Coefficients Deduction of Any Number of Derivatives[J]. Journal of Jilin University (Earth Science Edition), 2014, 44(1):375-385.
[25] 崔永福, 李国发, 吴国忱, 等. 基于面波模拟和曲波变换的去噪技术[J]. 吉林大学学报(地球科学版), 2016, 46(3):911-919. Cui Yongfu, Li Guofa, Wu Guochen, et al. Seismic Denoising Technique Based on Surface Wave Modeling and Curvelet Transform[J]. Journal of Jilin University (Earth Science Edition), 2016, 46(3):911-919.
[26] Roden J A, Gedney S D. Convolutional PML(CPML):An Efficient FDTD Implementation of the CFS_PML for Arbitrary Media[J]. Microwave and Optical Technology Letters, 2000, 27(5):334-339.
[27] Damir P, Ray M. Convolutional Perfectly Matched Layer for Isotropic and Anisotropic Acoustic Wave Equations[C]//SEG Technical Program Expanded Abstracts 2010. Tulsa:SEG, 2010:2925-2929.

Metrics

Viewed

Full text

Abstract

Cited

Shared

Discussed