基于众核架构的地震波场延拓方法

地震波场延拓方法与高性能计算技术的有效结合是保证波场延拓精度、提高波场延拓效率的重要途径。为此，本研究拟基于众核架构计算机建立高性能地震波场延拓理论和方法。兼顾地震正演、采集、处理、反演等领域中的地震波场延拓算法，构建面向用户的界面友好的地震波场延拓框架；针对快速傅里叶变换、矩阵分解、稀疏矩阵求解、隐式有限差分等关键基础算法需要大量节点或卡间数据通讯的特点，研发基于多种众核硬件的分布式计算技术，形成基于众核架构的地震波场延拓基础算法库；建立基于众核架构超级计算机的多层次并行优化策略，形成面向地震勘探多领域应用的高度可拓展性的地震波场延拓平台；研发基于众核架构的高保真数据压缩与解压缩技术，实现在众核架构下使用不同类型的处理器进行波场延拓与数据压缩时的高效并发运行。本项目力图解决地震勘探在高性能计算方面的基础科学问题，研究成果将会促进高性能计算在地震勘探领域中的有效应用。

Seismic wavefield extrapolation is a critical foundation of seismic exploration. The deep involvement of high performance computing can effectively enhance the efficiency while maintaining the accuracy of wavefield extrapolation. Therefore, we plan to develop a high performance seismic wavefield extrapolation theory and method based on many-core structure. The framework of the platform with a user-friendly interface will be constructed by giving overall consideration to wavefield extrapolation involved in the areas of forward modeling, acquisition, processing, and inversion. Then we will develop a many-core-based library of fundamental algorithms including FFT, matrix decomposition, sparse matrix operation, and implicit finite difference, which require a large amount of computing resources and/or involve intensive node communications. We will also develop an optimization strategy for multi-layer parallel computing on many-core structure super computer. The library and the optimization strategy are then incorporated into the framework to build the wavefield extrapolation platform that is highly feasible and can fulfill the demand from different areas of seismic exploration. In order to achieve efficient concurrent implementations of seismic wavefield extrapolation and data compression utilizing different processors on the same node, a data compression method featured with high fidelity and many-core computation will be developed as well. This project endeavors to solve the fundamental scientific issues of seismic exploration in the aspect of high performance computing. The research achievements will promote the application of high performance computing in the area of seismic exploration.