Closed-Loop CO2 Storage Control With History-Based Reinforcement Learning and Latent Model-Based Adaptation
arXiv:2605.02405v2 Announce Type: replace Abstract: Closed-loop management of geological CO2 storage requires control policies that adapt to uncertain reservoir behavior while relying on observations that are realistically available during operation. This work formulates CO2 injection and brine-production control as a partially observable sequential decision problem and studies deployable deep reinforcement-learning controllers trained with high-fidelity reservoir simulation. We first compare privileged-state, well-only, history-conditioned, masking-curriculum, and asymmetric teacher-student m
The increasing urgency of climate change mitigation and advancements in AI/ML are converging to enable more sophisticated and autonomous control systems for carbon capture and storage.
Efficient and reliable CO2 storage is critical for meeting net-zero targets, and AI-driven closed-loop control promises to significantly improve its economic viability and environmental safety.
The ability to autonomously manage complex subsurface CO2 injection in real-time, adapting to geological uncertainties, reduces operational costs and risks, potentially accelerating widespread adoption of carbon capture and storage.
- · Carbon capture and storage companies
- · AI/ML solution providers
- · Oil & Gas companies (repurposing infrastructure)
- · Environmental engineering firms
- · Fossil fuel industries without CCS integration
- · Legacy CO2 storage management manual processes
Improved efficiency and safety of geological CO2 storage operations through AI-driven adaptive control.
Reduced cost and increased scalability of CO2 removal technologies, accelerating decarbonization efforts globally.
Emergence of fully autonomous 'climate management' systems that integrate multiple geoengineering and carbon removal technologies controlled by advanced AI.
This signal links to a primary source. Continuum Brief monitors and indexes it as part of the live intelligence stream — we do not republish source content.
Read at arXiv cs.LG