A deep learning-based direct forecasting of CO2 plume migration☆
Section snippets
Introduction and motivation
Geologic carbon capture and storage technology has been considered as one of the promising technologies to mitigate the long-term CO2 concentration in the atmosphere as well as to accomplish the objective of decreasing greenhouse gas emission and limiting the global temperature exceeding to a threshold of 2° by 2100 (Alcalde et al., 2018; Viebahn et al., 2015; IEA, 2012). This technology involves collecting CO2 from large fossil fuel usage emitting sources such as power plants and subsequent
Materials and methods
In this section, we first describe the data generation. Next, we introduce two key aspects of the developed framework in effectively predicting the CO2 plume migration. The deep convolutional autoencoder is adopted to perform data parameterization by mapping the high-dimensional plume extent onto a low-dimensional latent space. Then, the non-linear relationship between observation variables and dimension-reduced prediction variables is established using a neural network. Lastly, we define four
Results
In this section, we first use the four defined metrics to evaluate the accuracy, effectiveness, and efficiency of the developed framework to forecast the high-dimensional CO2 plume distribution with streaming observation variables in three dramatically different scenarios. Then we investigate the influence of the parameterization level on the forecasting performance of the developed framework. We compare with PCA-based parameterization to evaluate whether the convolutional autoencoder can map
Discussion
To accelerate real-time decision making for carbon storage, we propose an inversion-free direct forecasting framework to reduce computational costs in traditional inversion-based model calibration (history matching). In this framework, we learn the observation-prediction relationship in their low-dimensional latent space and then infer the prediction from the observation data based on the learned ML model. We train the ML model using physics-based model simulation samples, so the observation
Conclusions and future work
Because of the high heterogeneity in the CO2 deposition environment, the CO2 plume shapes are complex and it is difficult to delineate their movement over space and time. In this work, we develop a deep learning-based framework to fast and accurately forecast the CO2 plume migration directly from the streaming observation data. The framework is tested against three dramatically different scenarios to demonstrate its robustness and effective-ness in quantifying the shape and spatio-temporal
Credit author statement
Ming Fan implemented the numerical experiments, prepared the figures and analyzed the results. Dan Lu formulated the problem, contributed to the research plan, and interpreted the results. Siyan Liu interpreted the results. All the three authors contributed to the manuscript preparation.
Declaration of competing interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
Acknowledgments
Primary funding support for this work is provided by the Science-informed Machine Learning to Accelerate Real Time Decision Making for Carbon Storage (SMART-CS) Initiative, funded by the US Department of Energy (DOE), Office of Fossil Energy and Carbon Management. Additional support is from the Artificial Intelligence Initiative as part of the Laboratory Directed Research and Development Program of Oak Ridge National Lab- oratory, managed by UT-Battelle, LLC, for the US DOE under contract
References (53)
- et al.
A review of developments in carbon dioxide storage
Appl. Energy
(2017) - et al.
Quantifying the effects of depositional environment on deep saline formation co2 storage efficiency and rate
Int. J. Greenh. Gas Control
(2018) - et al.
Towards a robust parameterization for conditioning facies models using deep variational autoencoders and ensemble smoother
Comput. Geosci.
(2019) - et al.
Continuum-scale convective mixing in geolog- ical co2 sequestration in anisotropic and heterogeneous saline aquifers
Adv. Water Resour.
(2013) - et al.
Geologic co2 sequestration monitoring design: a machine learning and uncertainty quantification based approach
Appl. Energy
(2018) - et al.
Reducing uncertainty in geologic co2 sequestration risk assessment by assimilating monitoring data
Int. J. Greenh. Gas Control
(2020) - et al.
A comparative anal- ysis of risk assessment methodologies for the geologic storage of carbon dioxide
Energy Proc.
(2011) - et al.
Co2 plume evolution in a depleted natural gas reservoir: modeling of conformance uncertainty reduction over time
Int. J. Greenh. Gas Control
(2020) - et al.
Comprehensive study of the interactions between the critical dimensionless numbers associated with multiphase flow in 3d porous media
Fuel
(2019) - et al.
The role of the spatial heterogeneity and correlation length of surface wettability on two-phase flow in a co2-water-rock system
Adv. Water Resour.
(2020)
Multilayer feedforward networks are universal approximators
Neural Network.
Inversion using a new low-dimensional representation of complex binary geological media based on a deep neural network
Adv. Water Resour.
An efficient bayesian data-worth analysis using a multilevel Monte Carlo method
Adv. Water Resour.
Geological storage of co2 in saline aquifers—a review of the experience from existing storage operations
Int. J. Greenh. Gas Control
Quantification of risk profiles and impacts of uncertainties as part of us doe's national risk assessment partnership (nrap)
Energy Proc.
Recent advances in risk assessment and risk management of geologic co2 storage
Int. J. Greenh. Gas Control
A deep learning-accelerated data assimilation and forecasting workflow for commercial-scale geologic carbon storage
Int. J. Greenh. Gas Control
Prospects of carbon capture and storage (ccs) in China's power sector–an integrated assessment
Appl. Energy
Deep-learning-based workflow for boundary and small target segmen- tation in digital rock images using unet++ and Ik-ebm
J. Petrol. Sci. Eng.
Tensorflow: Large-Scale Machine Learning on Heterogeneous Systems
Estimating geological co2 storage security to deliver on climate mitigation
Nat. Commun.
On the dispersion of a solute in a fluid flowing through a tube
Proc. R. Soc. London, A
Representation learning: a review and new perspectives
IEEE Trans. Pattern Anal. Mach. Intell.
A real field application
Optimization and Monitoring of Geological Carbon Stor- Age Operations
Interaction between prop- pant compaction and single-/multiphase flows in a hydraulic fracture
SPE J.
Cited by (4)
Advancing spatiotemporal forecasts of CO<inf>2</inf> plume migration using deep learning networks with transfer learning and interpretation analysis
2024, International Journal of Greenhouse Gas ControlResearch on Short-Term Prediction Method of Liquefied Gas Concentration based on Mixed Intelligence
2024, International Journal of Multiphysics
- ☆
This manuscript has been authored by UT-Battelle LLC, under contract DE-AC05- 00OR22725 with the US Department of Energy (DOE). The US government retains and the publisher, by accepting the article for publication, acknowledges that the US government retains a nonexclusive, paid-up, irrevocable, worldwide license to publish or repro-duce the published form of this manuscript, or allow others to do so, for US government purposes. DOE will provide public access to these results of federally sponsored research in accordance with the DOE Public Access Plan (http://energy.gov/downloads/doe-public- access-plan).