Development and Application of an LSTM-Based Surrogate Model for CO2 Flooding Reservoirs

Compositional simulation of CO2 water-alternating-gas (WAG) flooding in CCUS reservoirs is substantially more difficult than conventional water-flooding simulation because it involves strongly coupled multiphase flow, component transfer, phase behavior, pressure propagation, gas breakthrough and alternating injection–production controls. In the multi-block CCUS demonstration area of Shengli Oilfield, repeated compositional simulations for CO2 source allocation and injection–production optimization often suffer from poor convergence and long runtimes, making direct simulator-based optimization computationally impractical. Although conventional neural networks and response-surface-type surrogate models can accelerate single-scenario evaluation, they map a limited set of design variables to a limited set of outputs, and are therefore not suited to WAG processes whose response depends on the whole sequence of alternating injection and production controls. To address these limitations, this study developed a Long Short-Term Memory (LSTM) surrogate model, using WAG production schedules as input, to rapidly predict the oil and gas production under CO2 injection scenarios. The model takes seven synchronized monthly field-level variables, namely the liquid production rate, the water and gas injection rates, the field pressure and three phase inventories, and maps their 60-month profiles to the complete oil and gas production profiles in a single pass. Reservoir case studies were conducted for 12-month, 6-month and 3-month WAG cycles and continuous CO2 flooding schemes. On the five independent test schemes the surrogate reproduced the 5-year profiles with an RMSE of 0.32–1.47 m3/d for oil and 0.85–2.14 × 103 m3/d for gas. The average prediction time was reduced from 3791.38 s for numerical simulation to 2.91 s for the LSTM surrogate model, corresponding to an approximately 1302.9-fold speedup. These results indicate that the proposed modified LSTM surrogate provides an efficient tool for rapid scenario forecasting of CO2 WAG flooding strategies, with a potential use in scheme optimization.

Authors

Institutions

Publication Details

Journal
Energies
Published
2026-11-19
DOI
https://doi.org/10.3390/en19225594
Primary Topic
Reservoir Engineering and Simulation Methods
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

Development and Application of an LSTM-Based Surrogate Model for CO2 Flooding Reservoirs

Xiang Wang, Ji Qi, Huifang Hu, Peng Zhou et al.
Energies
Reservoir Engineering and Simulation Methods
article

Development and Application of an LSTM-Based Surrogate Model for CO2 Flooding Reservoirs

Xiang Wang, Ji Qi, Huifang Hu, Peng Zhou, Changfu Li
article en

Abstract

Compositional simulation of CO2 water-alternating-gas (WAG) flooding in CCUS reservoirs is substantially more difficult than conventional water-flooding simulation because it involves strongly coupled multiphase flow, component transfer, phase behavior, pressure propagation, gas breakthrough and alternating injection–production controls. In the multi-block CCUS demonstration area of Shengli Oilfield, repeated compositional simulations for CO2 source allocation and injection–production optimization often suffer from poor convergence and long runtimes, making direct simulator-based optimization computationally impractical. Although conventional neural networks and response-surface-type surrogate models can accelerate single-scenario evaluation, they map a limited set of design variables to a limited set of outputs, and are therefore not suited to WAG processes whose response depends on the whole sequence of alternating injection and production controls. To address these limitations, this study developed a Long Short-Term Memory (LSTM) surrogate model, using WAG production schedules as input, to rapidly predict the oil and gas production under CO2 injection scenarios. The model takes seven synchronized monthly field-level variables, namely the liquid production rate, the water and gas injection rates, the field pressure and three phase inventories, and maps their 60-month profiles to the complete oil and gas production profiles in a single pass. Reservoir case studies were conducted for 12-month, 6-month and 3-month WAG cycles and continuous CO2 flooding schemes. On the five independent test schemes the surrogate reproduced the 5-year profiles with an RMSE of 0.32–1.47 m3/d for oil and 0.85–2.14 × 103 m3/d for gas. The average prediction time was reduced from 3791.38 s for numerical simulation to 2.91 s for the LSTM surrogate model, corresponding to an approximately 1302.9-fold speedup. These results indicate that the proposed modified LSTM surrogate provides an efficient tool for rapid scenario forecasting of CO2 WAG flooding strategies, with a potential use in scheme optimization.

EnergiesVol. 19(22)
Sinopec (China) (CN), Changzhou University (CN)
Reservoir Engineering and Simulation Methods
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon