An efficient deep learning-based workflow for real-time CO2 plume visualization in saline aquifer using distributed pressure and temperature measurements

Geoenergy Science and Engineering, Journal Year: 2024, Volume and Issue: 239, P. 212990 - 212990

Published: May 31, 2024

Language: Английский

---

Exploring Sparsity-Promoting Dynamic Mode Decomposition for Data-Driven Reduced Order Modeling of Geological CO2 Storage

SPE Annual Technical Conference and Exhibition, Journal Year: 2024, Volume and Issue: 113

Published: Sept. 20, 2024

Abstract In the context of addressing climate change, advanced computational methods are crucial for enhancing efficiency and efficacy carbon capture storage (CCS) technologies. This study explores application Sparsity-Promoting Dynamic Mode Decomposition (Sp-DMD) developing reduced-order models (ROMs) that effectively manage complexity subsurface CO2 simulations. By focusing on critical state variables—reservoir pressure plume saturation—this research aims to investigate whether Sp-DMD can accurately dynamic characteristics behavior in geological formations over extended periods. is centered Illinois Basin Decatur Project (IBDP), a CCS initiative targeting injection into deep saline reservoir. The pre-existing IBDP Eclipse300 open-source simulation model, originally calibrated with both permeability porosity modifications, required 8 hours simulate 3-year history. revising history-matching process focus solely modification, re-calibrating using multi-level readings from monitoring well bottom-hole data well, we reduced runtime 3 historical period. Additionally, include 9-year post-injection phase, culminating total 4 12 years. From this full-order model (FOM), developed ROM by extracting 3D saturation at various timesteps, which were then flattened vectors form comprehensive snapshot matrix. matrix was segmented 109 months (9 years) training 27 validation employs data-driven techniques such as Singular Value (SVD) eigen decomposition, enhanced L1 norm regularization, coherent fluid dynamics within complex settings. achieved mean absolute errors (MAE) 4.78E-05 0.63 psi during its phase. When tested remaining months, it yielded an MAE 1.17 0.009 saturation. Remarkably, reconstructed 12-year dataset just 1 minute 43 seconds—significantly faster than model's 4-hour requirement. Moreover, demonstrated capability forecast up 500 years only hour 40 minutes, establishing itself potent tool long-term monitoring. performance not significantly reduces demands but also facilitates efficient rapid forecasting pressure, thus streamlining transition data-intensive FOM expedient ROM. project utilized workstation 12th Gen Intel(R) Core(TM) i9-12900H, 2.50 GHz, 16 GB RAM, supporting extensive demands. Our approach provides meaningful balance between speed accuracy modeling systems. Although still exploratory paves way further investigations scalability robustness applications simulation.

Language: Английский

---

Time series predictions in unmonitored sites: a survey of machine learning techniques in water resources

Environmental Data Science, Journal Year: 2025, Volume and Issue: 4

Published: Jan. 1, 2025

Abstract Prediction of dynamic environmental variables in unmonitored sites remains a long-standing challenge for water resources science. The majority the world’s freshwater have inadequate monitoring critical needed management. Yet, need to widespread predictions hydrological such as river flow and quality has become increasingly urgent due climate land use change over past decades, their associated impacts on resources. Modern machine learning methods outperform process-based empirical model counterparts hydrologic time series prediction with ability extract information from large, diverse data sets. We review relevant state-of-the art applications streamflow, quality, other discuss opportunities improve emerging incorporating watershed characteristics process knowledge into classical, deep learning, transfer methodologies. analysis here suggests most prior efforts been focused frameworks built many at daily scales United States, but that comparisons between different classes are few inadequate. identify several open questions include inputs site characteristics, mechanistic understanding spatial context, explainable AI techniques modern frameworks.

Language: Английский

---

Hybrid Newton method for the acceleration of well event handling in the simulation of CO2 storage using supervised learning

Computers & Geosciences, Journal Year: 2025, Volume and Issue: unknown, P. 105872 - 105872

Published: Feb. 1, 2025

Language: Английский

---

Deep learning framework for history matching CO2 storage with 4D seismic and monitoring well data

Geoenergy Science and Engineering, Journal Year: 2025, Volume and Issue: unknown, P. 213736 - 213736

Published: Feb. 1, 2025

Language: Английский

---

Deep Learning for Subsurface Flow: A Comparative Study of U‐Net, Fourier Neural Operators, and Transformers in Underground Hydrogen Storage

Journal of Geophysical Research Machine Learning and Computation, Journal Year: 2025, Volume and Issue: 2(1)

Published: March 1, 2025

Abstract Subsurface flow research is essential for the sustainable management of natural resources and environment. Deep learning (DL) has significantly advanced this field by developing efficient accurate surrogate models to replace computationally expensive physics‐based simulations. These are commonly used predict spatiotemporal evolution state variables, such as gas saturation reservoir pressure, in heterogeneous geological formations. Despite various DL applied task, there a lack studies systematically comparing their performance. This absence comparative analysis leads somewhat arbitrary model selection subsurface research, resulting suboptimal performance potentially inaccurate predictions. To bridge gap, we conduct systematic comparison study three popular architectures—U‐Net, Fourier Neural Operators (FNO), Segmentation Transformer (SETR)—in modeling underground hydrogen storage (UHS). We focus on UHS due its promise enhancing clean energy resilience cyclic operational conditions that represent common scenarios applications. evaluate based accuracy, training cost, inference speed. The shows U‐Net achieves highest followed SETR FNO. lower FNO offers competitive accuracy with least memory usage, demonstrating potential transformers flow. Our results provide guidance selecting wide range problems.

Language: Английский

---

Reservoir Surrogate Modeling Using U-Net with Vision Transformer and Time Embedding

Processes, Journal Year: 2025, Volume and Issue: 13(4), P. 958 - 958

Published: March 24, 2025

Accurate and efficient modeling of subsurface flow in reservoir simulations is essential for optimizing hydrocarbon recovery, enhancing water management strategies, informing critical decision-making processes. However, traditional numerical simulation methods face significant challenges due to their high computational cost limited scalability handling large-scale models with uncertain geological parameters, such as permeability distributions. To address these limitations, we propose a novel deep learning-based framework leveraging conditional U-Net architecture time embedding improve the efficiency accuracy data assimilation. The designed train on maps, which encode uncertainty properties, trained predict high-resolution saturation pressure maps at each step. By utilizing from previous step inputs, model dynamically captures spatiotemporal dependencies governing multiphase processes reservoirs. incorporation embeddings enables maintain temporal consistency adapt sequential nature evolution over periods. proposed can be integrated into assimilation loop, enabling generation forecasts reduced overhead while maintaining accuracy. bridging gap between physical accuracy, this study contributes advancing state art simulation. model’s ability generalize across diverse scenarios its potential real-time applications, production strategies history matching, underscores practical relevance oil gas industry.

Language: Английский

---

Efficient super-resolution of pipeline transient process modeling using the Fourier Neural Operator

Energy, Journal Year: 2024, Volume and Issue: 302, P. 131676 - 131676

Published: May 20, 2024

Language: Английский

---

Enhancing subsurface multiphase flow simulation with Fourier neural operator

Heliyon, Journal Year: 2024, Volume and Issue: 10(18), P. e38103 - e38103

Published: Sept. 1, 2024

Language: Английский

---

Gradient-Guided Convolutional Autoencoder for Predicting Co2 Storage in Saline Aquifers with Multiple Geological Scenarios and Well Placements

Published: Jan. 1, 2024

Download This Paper Open PDF in Browser Add to My Library Share: Permalink Using these links will ensure access this page indefinitely Copy URL DOI

Language: Английский

---