3D variable-grid full-waveform inversion on GPU

Wang, Ziying; Huang, Jianping; Liu, Ding-Jin; Li, Zhenchun; Yong, Peng; Yang, Zhenjie

doi:10.1007/s12182-019-00368-2

Cited by 33 publications

(9 citation statements)

References 37 publications

(42 reference statements)

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Mitigating this problem, except the advance of computer hardware, optimization on inversion algorithms, numerical wavefield modeling and computer codes are important research directions. Among them, (Yao et al 2019c) improving the inversion and modeling algorithms are the key directions because computer code optimization, for example, load balance between processes, CPU vectorization, cache reuse and GPU acceleration (e.g., Wang et al 2019), can only speed up the inversion in tens of times or less generally. Insufficient information restricted by the technology and space of acquisition, the acquired seismic data are incomplete in spatial coverage and frequencies.…”

Section: Conclusion and Future Perspectivementioning

confidence: 99%

A review on reflection-waveform inversion

Yao

Wang

2020

Pet. Sci.

View full text Add to dashboard Cite

Full-waveform inversion (FWI) utilizes optimization methods to recover an optimal Earth model to best fit the observed seismic record in a sense of a predefined norm. Since FWI combines mathematic inversion and full-wave equations, it has been recognized as one of the key methods for seismic data imaging and Earth model building in the fields of global/regional and exploration seismology. Unfortunately, conventional FWI fixes background velocity mainly relying on refraction and turning waves that are commonly rich in large offsets. By contrast, reflections in the short offsets mainly contribute to the reconstruction of the high-resolution interfaces. Restricted by acquisition geometries, refractions and turning waves in the record usually have limited penetration depth, which may not reach oil/gas reservoirs. Thus, reflections in the record are the only source that carries the information of these reservoirs. Consequently, it is meaningful to develop reflection-waveform inversion (RWI) that utilizes reflections to recover background velocity including the deep part of the model. This review paper includes: analyzing the weaknesses of FWI when inverting reflections; overviewing the principles of RWI, including separation of the tomography and migration components, the objective functions, constraints; summarizing the current status of the technique of RWI; outlooking the future of RWI.

show abstract

Section: Conclusion and Future Perspectivementioning

confidence: 99%

A review on reflection-waveform inversion

Yao

Wang

2020

Pet. Sci.

View full text Add to dashboard Cite

show abstract

“…We found that it is not necessary to include all 25 channels in the input due to the redundancy among seismic data records, so our input only contains 8 channels (6,7,8,11,13,16,17,18). See Figure 11 for an explanation of these serial numbers.…”

Section: ) Training Configurationsmentioning

confidence: 99%

“…As 3D surveys become widely implemented, 3D FWI are required to solve complex geological cases [4], [5], [6]. However, both the computational cost of forward-modeling and the gradient calculation are much higher in the 3D cases than the 2D cases [7], [8]. Thus 3D FWI is still a challenging problem due to the high computational cost.…”

Section: Introductionmentioning

confidence: 99%

InversionNet3D: Efficient and Scalable Learning for 3D Full Waveform Inversion

Zeng,

Feng,

Wohlberg

et al. 2021

Preprint

View full text Add to dashboard Cite

Recent progress in the use of deep learning for Full Waveform Inversion (FWI) has demonstrated the advantage of data-driven methods over traditional physics-based approaches in terms of reconstruction accuracy and computational efficiency. However, due to high computational complexity and large memory consumption, the reconstruction of 3D high-resolution velocity maps via deep networks is still a great challenge. In this paper, we present InversionNet3D, an efficient and scalable encoder-decoder network for 3D FWI. The proposed method employs group convolution in the encoder to establish an effective hierarchy for learning information from multiple sources while cutting down unnecessary parameters and operations at the same time. The introduction of invertible layers further reduces the memory consumption of intermediate features during training and thus enables the development of deeper networks with more layers and higher capacity as required by different application scenarios. Experiments on the 3D Kimberlina dataset demonstrate that InversionNet3D achieves state-of-the-art reconstruction performance with lower computational cost and lower memory footprint compared to the baseline.

show abstract

“…In geophysics, previous studies have attempted to overcome these computational limitations through different approaches. Some of these studies have focused on the fact that, in FWI, most of the computational time is spent modelling wave propagation 5,6 , which requires the full numerical solution of the acoustic wave equation over discrete grids. The size of the grid is controlled by the acoustic properties of the imaging target and the temporal frequencies involved.…”

Section: Introductionmentioning

confidence: 99%

“…Because the grid spacing could be larger during the inversion of low frequencies, using a fixed grid results in oversampling and redundant computations. For this reason, previous studies have explored the use of adaptive grids, where the grid spacing changes with the background acoustic velocity 5,7 or the frequency of the propagated wave 8 , thus allowing the use of a lower number of grid points with respect to conventional FWI.…”

Section: Introductionmentioning

confidence: 99%

Computationally efficient full-waveform inversion of the brain using frequency-adaptive grids and lossy compression

Letizia¹,

Cueto²

2021

Preprint

View full text Add to dashboard Cite

A tomographic technique called full-waveform inversion has recently shown promise as a fast, affordable, and safe modality to image the brain using ultrasound. However, its high computational cost and memory footprint currently limit its clinical applicability. Here, we address these challenges through a frequency-adaptive discretisation of the imaging domain and lossy compression techniques. Because full-waveform inversion relies on the adjoint-state method, every iteration involves solving the wave equation over a discretised spatiotemporal grid and storing the numerical solution to calculate gradient updates. The computational cost depends on the grid size, which is controlled by the maximum frequency being modelled. Since the propagated frequency typically varies during the reconstruction, we reduce reconstruction time and memory use by allowing the grid size to change throughout the inversion. Moreover, we combine this approach with multiple lossy compression techniques that exploit the sparsity of the wavefield to further reduce its memory footprint. We explore applying these techniques in the spatial, wavelet, and wave atom domains. Numerical experiments using a human-head model show that our methods lead to a 30% reduction in reconstruction time and up to three orders of magnitude less memory, while negligibly affecting the accuracy of the reconstructions.

show abstract

3D variable-grid full-waveform inversion on GPU

Cited by 33 publications

References 37 publications

A review on reflection-waveform inversion

A review on reflection-waveform inversion

InversionNet3D: Efficient and Scalable Learning for 3D Full Waveform Inversion

Computationally efficient full-waveform inversion of the brain using frequency-adaptive grids and lossy compression

Contact Info

Product

Resources

About