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。
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.
作者
邵广周
赵凯鹏
吴华
Shao Guangzhou;Zhao Kaipeng;Wu Hua(School of Geological and Surveying Engineering,Chang’an University,Xi’an 710054,China;School of Science,Chang’an University,Xi’an 710064,China)
出处
《吉林大学学报（地球科学版）》
EI
CAS
CSCD
北大核心
2020年第1期294-303,共10页
Journal of Jilin University：Earth Science Edition
基金
国家自然科学基金项目(41874123,41004043)
陕西省自然科学基金项目(2016JM4003)
长安大学中央高校基金项目(300102268402).
关键词
MPI
高阶交错网格
有限差分
镜像法
CPML
区域分解
MPI
high-order staggered-grid
finite-difference
image-method
CPML
region decomposition
