Application of transformers based generative adversarial neural networks as an algorithm for seismic amplitude inversion

Introduction. Seismic amplitude inversion enables the transition from wavefield amplitudes to the distributions of physical properties and the prediction of their propagation. However, the inversion problem is often ill-posed due to solution non-uniqueness and instability, as well as the limited frequency range of seismic data, which typically necessitates iterative optimization.

Aim. The objective of this study is to develop and test a method for impedance inversion using Generative Adversarial Networks (GAN) without the need for constructing a low-frequency model or extracting the impulse from well log data. This method aims to minimize the degrees of freedom of the inversion algorithm, relying on neural networks’ ability to extract all necessary information directly from the input data, and signifi cantly reducing the time required for predicting the propagation of acoustic properties.

Materials and methods. The study examines three model architecture scenarios for predicting pseudo-acoustic logging, each differing in the number of neural networks. Each scenario is adapted to the specifics of real data, where the number of wells available for analysis is often limited. The proposed training approach used in GANs is applied for solve model regularization and overfitting tasks. The developed method is tested on real data from a site in Eastern Siberia.

Results. The results of the new method are compared with those of the standard model-based acoustic inversion algorithm. Our method demonstrates comparable metrics in quantitative evaluation and qualitative analysis, requiring significantly less time to obtain the forecast.

Conclusion. Our method demonstrates high efficiency in using deep learning technologies to solve seismic data inversion problems and can be considered as a rapid method for acoustic inversion.

1. Zhdanov M.S. Theory of Inverse Problems and Regularization in Geophysics. Moscow: Nauchny Mir, 2007, 712 p.

2. Danko D.A. Development of principles for studying unconventional clay reservoirs based on petroelastic modeling and amplitude inversion of seismic data. 2018, 273 p.

3. Kim Y., Nakata N. Geophysical inversion versus machine learning in inverse problems, Lead. Edge, 2018, vol. 37, no. 12, pp. 894–901.

4. Wu B., Meng D., Wang L., Liu N., Wang Y. Seismic impedance inversion using fully convolutional residual network and transfer learning, IEEE Geosci. Remote Sens. Lett., 2020, vol. 17, no. 12, pp. 2140–2144.

5. Sheng H., Wu1 X., Si X., Li J., Zhang S., Duan X. Seismic Foundation Model (SFM): All-Purpose Feature Extraction from Seismic Data for Diverse Geophysical Applications, 85th EAGE Annual Conference & Exhibition, 2024, p. 1–5.

6. Meng D., Wu B., Wang Z., Zhu Z. Seismic impedance inversion using conditional generative adversarial network, IEEE Geoscience and Remote Sensing Letters, 2021, vol. 19, p. 1–5.

7. Vaswani A., Shazeer N., Parmar N., Uszkoreit J., Jones L., Gomez A.N., Kaiser Ł., Polosukhin I. “Attention is all you need,” Advances in neural information processing systems, vol. 30, 2017.