Том: Computational Science and Its Applications - ICCSA 2024 Workshops. 24rd International Conference (Hanoi, Vietnam, July 1-4, 2024). Proceedings, Part III
Том: 14817
, Год издания: 2024
Многотомное издание: Computational Science and Its Applications - ICCSA 2024 Workshops. 24rd International Conference (Hanoi, Vietnam, July 1-4, 2024). Proceedings, Part III
Страницы: 276-289
Аннотация
The paper presents a novel method for 3D seismic modeling, combining classical grid-based methods such as finite difference methods with deep learning techniques to reduce numerical dispersion in the synthetic seismograms modeled on a coarse grid. The approach consists of three main steps. First, the inaccurate solution corresponding to the entire set of sources is computed on a coarse enough grid. Second, an accurate but computationally demanding fine-grid solution is simulated for a small number of sources to get a training dataset. Third, the neural network is trained to map coarse-grid solution to the fine-grid solution and then applied to the entire dataset. Through empirical investigation on a complex 3D SEG/EAGE Overthrust velocity model, the study highlights the significant reduction in computation time up to a factor of 15 for realistic 3D velocity models. This innovative approach offers promising advancements in seismic data processing and modeling.