基于深度学习的远震P波接收函数与面波频散数据联合反演方法研究

大量研究表明深度学习技术在地球物理反演问题上具有巨大的价值，但目前基于深度学习的地球物理联合反演研究较少。本申请将针对远震P波接收函数与面波频散数据联合反演开展研究，拟发展一种基于深度学习的联合反演新方法，快速可靠地获取地壳上地幔速度结构。相比于传统的联合反演方法，新方法的优点是不依赖初始模型的选取，同时能快速地联合反演两种数据，特别适合大量的接收函数与面波频散数据反演，如密集台阵得到的地震数据。另外，新方法计划采用基于物理信息引导以及数据增强的训练策略，在很大程度上可以提高网络的泛化性能。本项目计划将新方法应用到华南地区的实际远震P波接收函数和面波频散数据中，反演得到华南地区地壳和上地幔三维横波速度结构，并对反演结果进行评估以及结合其他研究进行综合地质解释。

Numerous studies have shown that deep learning techniques have great potentials in geophysical inversion problems. However, there are very few studies on application of deep learning on geophysical joint inversion problems. This proposal aims at developing joint inversion of teleseismic P-wave receiver function and surface wave dispersion data via deep learning to efficiently and reliably determine seismic velocity structure for the crust and upper mantle. Compared to conventional joint inversion methods of receiver function and surface wave dispersion data, the proposed method does not depend on the selection of initial models, and meanwhile it is more efficient, which is extremely suitable for inverting large amounts of receiver function and surface wave dispersion data such as for the case of dense seismic arrays. In addition, the new method adopts strategies of physical model constraints and data augmentation during the neural network training process, which can greatly improve the generalization ability of the neural network. The new joint inversion method will be applied to real teleseismic P-wave receiver function and surface wave dispersion data in South China to obtain S-wave velocity structure of crust and upper mantle, which will be evaluated and interpreted jointly with other geophysical and geological results in this region.