A Physics-guided Generative AI Toolkit for Geophysical Monitoring

Junhuan Yang,Hanchen Wang,Yi Sheng,Youzuo Lin,Lei Yang

2024-01-06

Abstract:Full-waveform inversion (FWI) plays a vital role in geoscience to explore the subsurface. It utilizes the seismic wave to image the subsurface velocity map. As the machine learning (ML) technique evolves, the data-driven approaches using ML for FWI tasks have emerged, offering enhanced accuracy and reduced computational cost compared to traditional physics-based methods. However, a common challenge in geoscience, the unprivileged data, severely limits ML effectiveness. The issue becomes even worse during model pruning, a step essential in geoscience due to environmental complexities. To tackle this, we introduce the EdGeo toolkit, which employs a diffusion-based model guided by physics principles to generate high-fidelity velocity maps. The toolkit uses the acoustic wave equation to generate corresponding seismic waveform data, facilitating the fine-tuning of pruned ML models. Our results demonstrate significant improvements in SSIM scores and reduction in both MAE and MSE across various pruning ratios. Notably, the ML model fine-tuned using data generated by EdGeo yields superior quality of velocity maps, especially in representing unprivileged features, outperforming other existing methods.

Machine Learning,Artificial Intelligence,Computer Vision and Pattern Recognition,Signal Processing,Geophysics

What problem does this paper attempt to address?

This paper mainly discusses the problem of Full Waveform Inversion (FWI) in geophysics. FWI is an important technique for imaging subsurface velocity mapping using seismic waves, which is crucial for monitoring geological structures such as carbon sequestration. With the development of Machine Learning (ML), data-driven methods have shown advantages in improving accuracy and reducing computational costs. However, the lack of unprivileged data and the performance degradation after model pruning are the current challenges. The paper proposes a toolkit called EdGeo, which uses physics-guided diffusion models to generate high-fidelity velocity maps, solving the problems of data insufficiency and performance degradation after model pruning. Specifically, it uses the acoustic wave equation to generate corresponding seismic waveform data to fine-tune the pruned ML model. Experimental results show that EdGeo improves Structural Similarity Index (SSIM), reduces Mean Absolute Error (MAE) and Mean Square Error (MSE), especially when dealing with under-represented features, it outperforms other existing methods. In addition, EdGeo has been optimized for real-time performance and resource constraints, making it suitable for edge device deployment and ensuring local adaptability of the model in different environments. Through an end-to-end fine-tuning framework, it maintains efficient and accurate performance even when the model is heavily pruned.

A Physics-guided Generative AI Toolkit for Geophysical Monitoring

Revisit Geophysical Imaging in A New View of Physics-informed Generative Adversarial Learning

Domain Adaptation in Automatic Picking of Phase Velocity Dispersions Based on Deep Learning

FWIGAN: Full‐waveform inversion via a physics‐informed generative adversarial network

Controllable seismic velocity synthesis using generative diffusion models

Deep Velocity Generator: A Plug-In Network for FWI Enhancement.

Physics-Consistent Data-driven Waveform Inversion with Adaptive Data Augmentation

A prior regularized full waveform inversion using generative diffusion models

Physics-guided deep autoencoder to overcome the need for a starting model in full-waveform inversion

An Augmented Lagrangian Method-Based Deep Iterative Unrolling Network for Seismic Full-Waveform Inversion

Physics-guided Deep Learning-Based Inversion for Airborne Electromagnetic Data

Integrating Physics of the Problem into Data-Driven Methods to Enhance Elastic Full-Waveform Inversion with Uncertainty Quantification

Multimodal surface wave inversion with automatic differentiation

DiffusionVel: Multi-Information Integrated Velocity Inversion Using Generative Diffusion Models

Data-driven geophysics: from dictionary learning to deep learning

Deep-learning assisted regularized elastic full waveform inversion using the velocity distribution information from wells

Geological and Well prior assisted full waveform inversion using conditional diffusion models

GlobalTomo: A global dataset for physics-ML seismic wavefield modeling and FWI

Multiscale Data-driven Seismic Full-waveform Inversion with Field Data Study

Rapid seismic domain transfer: Seismic velocity inversion and modeling using deep generative neural networks