Towards Physics-based Machine Learning Framework of Seismic Wavefield Modeling and Full-waveform Inversion

李世乾 博士研究生

北京大学 人工智能研究院

2026.03.06（星期五）15:00，理科二号楼2821

摘要：

Global seismic tomography, taking advantage of seismic waves from natural sources, provides essential insights into the earth's internal dynamics. Advanced full-waveform inversion (FWI) techniques, whose aim is to meticulously interprete very detail in seismograms, confront formidable computational demands in forward modeling and adjoint simulations on a global scale. Recent advancements in Machine Learning (ML) offer a transformative potential for accelerating the computational efficiency of FWI and extending its applicability to global and regional scales.

This work presents a 3D global synthetic dataset tailored for seismic wavefield modeling, referred to as the Global Tomography (GlobalTomo) dataset. This dataset includes both surface seismograms and whole-domain wavefield data, based upon parameterization of long-wavelength velocity structures with spherical harmonics up to degree 8. Synthetics are simulated through AxiSEM3Doptimized for 3D global wavefield calculations.

By analyzing various ML baselines and establishing framework of FWI, we illustrate that ML approaches are suitable for global tomography, overcoming its limitations with rapid forward modeling up to 60000 times faster and achieving accurate inversion result. This work represents a cross-disciplinary effort to enhance our understanding of the earth's interior through physics-ML modeling.

报告人简介：

    李世乾，男，北京大学人工智能研究院22级博士生，研究领域为AI4S，Physical Reasoning，主要关注使用AI加速物理学模拟和反演问题，以第一作者在人工智能顶级会议NeurIPS，ICLR等发表论文6篇，研究成果曾被VALSE，AGU邀请做口头报告。