Seismic Velocity Inversion via Physical Embedding Recurrent Neural Networks (RNN)

Abstract

Seismic velocity inversion is one of the most critical issues in the field of seismic exploration and has long been the focus of numerous experts and scholars. In recent years, the advancement of machine learning technologies has infused new vitality into the research of seismic velocity inversion and yielded a wealth of research outcomes. Typically, seismic velocity inversion based on machine learning lacks control over physical processes and interpretability. Starting from wave theory and the physical processes of seismic data acquisition, this paper proposes a method for seismic velocity model inversion based on Physical Embedding Recurrent Neural Networks. Firstly, the wave equation is a mathematical representation of the physical process of acoustic waves propagating through a medium, and the finite difference method is an effective approach to solving the wave equation. With this in mind, we introduce the architecture of recurrent neural networks to describe the finite difference solution of the wave equation, realizing the embedding of physical processes into machine learning. Secondly, in seismic data acquisition, the propagation of acoustic waves from multiple sources through the medium represents a high-dimensional causal time series (wavefield snapshots), where the influential variable is the velocity model, and the received signals are the observations of the wavefield. This forms a forward modeling process as the forward simulation of the wavefield equation, and the use of error back-propagation between observations and calculations as the velocity inversion process. Through time-lapse inversion and by incorporating the causal information of wavefield propagation, the non-uniqueness issue in velocity inversion is mitigated. Through mathematical derivations and theoretical model analyses, the effectiveness and rationality of the method are demonstrated. In conjunction with simulation results for complex models, the method proposed in this paper can achieve velocity inversion in complex geological structures.

Supplementation of deep neural networks with simplified physics-based features to increase accuracy of plate fundamental frequency predictions

“…Therefore, benefit may be gained by successful integration of ML with modeling, potentially leading to improved designs from more efficient computation of possible solutions. Indeed, studies such as those undertaken by Zhu et al [14,15] for predicting total organic carbon (TOC) content of a reservoir, as well as Lu and Zhang [16] with the assessment of seismic velocity inversion, have successfully integrated physics models with ML, particularly neural networks. Nevertheless, available time and resources to develop a working predictive model may still be less than needed for integration of ML with numerical simulation.…”