Target-oriented time-lapse waveform inversion using deep learning-assisted regularization

Li, Yuanyuan; Alkhalifah, Tariq; Guo, Qiang

doi:10.1190/geo2020-0383.1

Cited by 20 publications

(6 citation statements)

References 52 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…In this way, to avoid solving the wave equation in the entire physical domain, we propose a new misfit function based on Re ´nyi statistics along with a technique for solving the wave equation only in a target region. Such a methodology inspired from the condensed matter physics named Patched Green's Function (PGF) [49][50][51], in which some of us have recently generalized the PGF technique for to apply it in problems of target-oriented modeling [52][53][54][55]. The PGF is a powerful methodology to reduce the computational cost in comparison to the classical modeling techniques.…”

Section: Plos Onementioning

confidence: 99%

Full-waveform inversion based on generalized Rényi entropy using patched Green’s function techniques

et al. 2022

View full text Add to dashboard Cite

The estimation of physical parameters from data analyses is a crucial process for the description and modeling of many complex systems. Based on Rényi α-Gaussian distribution and patched Green’s function (PGF) techniques, we propose a robust framework for data inversion using a wave-equation based methodology named full-waveform inversion (FWI). From the assumption that the residual seismic data (the difference between the modeled and observed data) obeys the Rényi α-Gaussian probability distribution, we introduce an outlier-resistant criterion to deal with erratic measures in the FWI context, in which the classical FWI based on l2-norm is a particular case. The new misfit function arises from the probabilistic maximum-likelihood method associated with the α-Gaussian distribution. The PGF technique works on the forward modeling process by dividing the computational domain into outside target area and target area, where the wave equation is solved only once on the outside target (before FWI). During the FWI processing, Green’s functions related only to the target area are computed instead of the entire computational domain, saving computational efforts. We show the effectiveness of our proposed approach by considering two distinct realistic P-wave velocity models, in which the first one is inspired in the Kwanza Basin in Angola and the second in a region of great economic interest in the Brazilian pre-salt field. We call our proposal by the abbreviation α-PGF-FWI. The results reveal that the α-PGF-FWI is robust against additive Gaussian noise and non-Gaussian noise with outliers in the limit α → 2/3, being α the Rényi entropic index.

show abstract

Section: Plos Onementioning

confidence: 99%

Full-waveform inversion based on generalized Rényi entropy using patched Green’s function techniques

et al. 2022

View full text Add to dashboard Cite

show abstract

“…Knowledge of the time-lapse changes is crucial for the characterization and management of the involved reservoir. Time-lapse full-waveform inversion has emerged as a promising approach for quantitatively inferring the property changes by using all the information in time-lapse seismic data, which are obtained by repeated seismic surveys over the same area [5]- [7].…”

Section: Introductionmentioning

confidence: 99%

“…Li et al [23] utilized the statistical distribution information in the wells to introduce high-resolution components into the EFWI result assisted by deep learning. To enhance the performance of time-lapse FWI, a prior model for the target monitor zone is constructed based on a DNN that relates an initial seismic estimation and well information [7].…”

Section: Introductionmentioning

confidence: 99%

“…Target-oriented inversion techniques have been developed to reduce the computational cost by focusing the inversion on a localized area [34]- [36]. Redatuming techniques retrieve a virtual data set for the target-oriented inversion by projecting sources and receivers to a datum level, preferably just above the target area [7], [37], [38]. To retain sufficient wave information in the virtual data set for interpreting the elastic properties, we exploit the multi-component data on the surface in our elastic redatuming, instead of only the PP reflections [39], [40], and generate the multi-component virtual data set at the datum [41].…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Target-Oriented Time-Lapse Elastic Full-Waveform Inversion Constrained by Deep Learning-Based Prior Model

Alkhalifah

2022

IEEE Trans. Geosci. Remote Sensing

Self Cite

View full text Add to dashboard Cite

Time-lapse (TL) seismic monitoring plays a vital role in reservoir characterization and management. Elastic fullwaveform inversion (EFWI) has been applied to time-lapse seismic data to allow for a quantitative estimation of time-varying elastic properties. However, the high-resolution inversion can be computationally intense and ill-posed. To estimate the highresolution time-lapse changes at a reasonable cost, we utilize two key techniques for the inversion: 1) we develop an elastic redatuming approach to retrieve the virtual elastic data for both base and monitor data at the target level using mainly a kinematically accurate velocity, thus, reducing the computational cost by focusing the high-resolution inversion on the target zone; 2) We integrate high-resolution well information and seismic data in the target-oriented inversion, where a high-resolution prior model is predicted by deep learning to regularize the inversion. A deep neural network (DNN) is capable of learning the mappings between the time-lapse seismic estimation and the facies interpreted from well information after the training process. Thus, we can derive a prior model for time-lapse changes by mapping the facies characterized by the property changes to the target inversion domain. We then implement the targetoriented TLEFWI regularized by the prior model, where the redatumed time-lapse elastic data and the prior model jointly contributes to the inversion result. The numerical examples validate that the proposed approach enables us to retrieve the time-lapse changes of elastic property in the target zone with improved resolution and well consistency.

show abstract

“…Recently, with the increased computational power and the revitalization of deep neural networks, significant research and development efforts have been put into data-driven ML methods for seismic imaging (Hale, 2013;Dahlke et al, 2016;Araya-Polo et al, 2017;Cao and Roy, 2017;Yuan et al, 2019;Li et al, 2021). For example, Um et al (2022) propose a network (U-net) to reconstruct CO 2 saturation and compare the results of the uncertainty analysis with the Monte-Carlo dropout and bagging methods.…”

Section: Introductionmentioning

confidence: 99%

Joint physics-based and data-driven time-lapse seismic inversion: Mitigating data scarcity

et al. 2022

View full text Add to dashboard Cite

In carbon capture and sequestration, developing rapid and effective imaging techniques is crucial for real-time monitoring of the spatial and temporal dynamics of CO2 propagation during/after injection. With continuing improvements in computational power and data storage, data-driven techniques based on machine learning (ML) have been effectively applied to seismic inverse problems. In particular, ML helps alleviate the ill-posedness and high computational cost of full-waveform inversion (FWI). However, such data-driven inversion techniques require massive high-quality training data sets to ensure prediction accuracy, which hinders their application to time-lapse monitoring of CO2 sequestration. We propose an efficient “hybrid” time-lapse workflow that combines physics-based FWI and data-driven ML inversion. The scarcity of the available training data is addressed by developing a new data-generation technique with physics constraints. The method is validated on a synthetic CO2-sequestration model based on the Kimberlina storage reservoir in California. The proposed approach is shown to synthesize a large volume of high-quality, physically realistic training data, which is critically important in accurately characterizing the CO2 movement in the reservoir. The developed hybrid methodology can also simultaneously predict the variations in velocity and saturation and achieve high spatial resolution in the presence of realistic noise in the data.

show abstract

Target-oriented time-lapse waveform inversion using deep learning-assisted regularization

Cited by 20 publications

References 52 publications

Full-waveform inversion based on generalized Rényi entropy using patched Green’s function techniques

Full-waveform inversion based on generalized Rényi entropy using patched Green’s function techniques

Target-Oriented Time-Lapse Elastic Full-Waveform Inversion Constrained by Deep Learning-Based Prior Model

Joint physics-based and data-driven time-lapse seismic inversion: Mitigating data scarcity

Contact Info

Product

Resources

About