Water-Alternating-Gas and CO2 Storage Optimization Using Time-Lapse Geophysical Monitoring and Deep Reinforcement Learning
Bibliographic record
Abstract
Abstract This study presents a new approach for optimizing water-alternating-gas (WAG) injection strategies and CO2 storage using deep reinforcement learning (DRL), supported by time-lapse gravity monitoring. We tested two reinforcement learning (RL) agents, Q-Learning (QL) and Double Deep Q-Network (DDQN), which interact with a high-fidelity reservoir simulation environment. The QL agent, despite its simplicity, demonstrates the fundamental concepts of optimal control of a WAG process using RL combined with gravity measurements. In contrast, the DDQN agent, combined with a convolutional neural network (CNN), outperforms other control methods by learning the spatial and temporal patterns of fluid movement within the subsurface. Comparisons with traditional, industry-standard WAG schedules reveal significant improvements in both Net Present Value (NPV) and CO2 storage efficiency using RL-optimized injection strategies. Time-lapse gravity data, simulated over 25 years, effectively capture and differentiate fluid displacement and accumulation under various injection regimes, making our proposed geophysical control approach possible. The industrial WAG schedule, involving regular switching between water and gas phases, results in suboptimal CO2 trapping due to early gas breakthrough and limited dissolution. Conversely, the DDQN-optimized policy enables longer CO2 retention in the reservoir, promoting greater dissolution into the formation brine and improving long-term geological storage. These findings underscore the importance of intelligent control strategies and advanced state representations in optimizing both economic and environmental objectives for CO2-enhanced oil recovery and geological carbon storage operations.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame machine prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.
Distilled classifier scores by category (both heads)
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.001 | 0.002 |
| Meta-epidemiology (narrow) | 0.001 | 0.000 |
| Meta-epidemiology (broad) | 0.001 | 0.000 |
| Bibliometrics | 0.000 | 0.000 |
| Science and technology studies | 0.000 | 0.001 |
| Scholarly communication | 0.001 | 0.001 |
| Open science | 0.001 | 0.001 |
| Research integrity | 0.001 | 0.001 |
| Insufficient payload (model declined to judge) | 0.001 | 0.000 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one source (direct Gemma or distilled Codex), not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".