4D Seismic Monitoring and Forecasting of CO2 Sequestration with Neural Networks
Abstract
Monitoring during and after CO2 injection is essential for the safety and effectiveness of CO2 sequestration projects. Among all the geophysical monitoring techniques, 4D Seismic imaging is one of the most useful methods for monitoring CO2 injection and storage in the geological subsurface (Ma, et al., App. Geo., 2016). 3D seismic surveys are implemented in the time-lapse mode to image the subsurface changes over time (Lumley, Lead. Edge., 2010). However, the high cost of 3D seismic acquisition, availability of equipment, and exclusion zones around surface structures limit the frequency of the data acquisition (Koster et.al, Lead. Edge, 2000), which make in-situ 4D seismic monitoring of the reservoir changes technically challenging or even infeasible. Hence, it is important to fill the gaps between time-lapse seismic images (interpolation) concerning the spatio-temporal characteristics of the existing sparse measurement. Moreover, it is also a big challenge to forecast new seismic images from the past measurements (extrapolation) without violating the physical law. In this study, we develop a spatio-temporal neural network-based model that incorporates physical knowledge to interpolate/extrapolate high-fidelity seismic images effectively and efficiently. This model is built based on an autoencoder that obtains the latent spaces of the input sequential images. The long short-term memory (LSTM) structure is incorporated in the model to recognize the sequential and temporal dynamic. Moreover, the optical flow is used as a regularization term in the loss function. It can describe the spatial movement of the images over time. We have implemented this method to the worlds first industrial-scale CO2 storage operation project, Sleipner. Via the numerical and expert evaluations, our work produces high-quality 2D/3D seismic images at a reasonable cost and it demonstrates its potential in real-time monitoring and near-future forecasting of the CO2 storage reservoir.
- Publication:
-
AGU Fall Meeting Abstracts
- Pub Date:
- December 2021
- Bibcode:
- 2021AGUFM.S25A0215F