The Bot That Wants Your Oil and Your Carbon Too

New ML framework optimizes CO2 injection in shale to boost oil recovery and maximize carbon storage simultaneously

27 Mar 2026

A study published in Energy & Fuels in 2026 proposes a machine learning framework that could help US shale operators lift oil recovery rates while simultaneously locking more carbon dioxide underground, resolving a long-standing tension between the two objectives.

The research, led by Lei and colleagues, applies a surrogate model to rapidly test a wide range of CO₂ injection parameters, identifying configurations that perform well on both measures. Conventional reservoir simulation achieves the same task at far greater computational cost, making field-by-field design iteration difficult in practice.

The commercial case for the work is clear. Shale wells in the Permian Basin and Bakken formations typically recover less than 10 per cent of oil in place. CO₂ injection is one of the few credible tools for improving that figure, but the way injection is designed has historically favoured either production or storage, not both.

Federal 45Q tax credits, which reward operators that demonstrate verifiable, permanent CO₂ storage as part of enhanced oil recovery programmes, add a financial dimension to the optimisation problem. The ability to maximise both outcomes simultaneously carries measurable compliance and revenue implications.

Researchers caution that shale formations vary considerably across US basins. Any model will need validation and recalibration before deployment in new geological settings, limiting near-term applicability at scale.

Whether the framework can be standardised across the geological diversity of US shale plays, and how it will fare under intensifying regulatory scrutiny of corporate carbon claims, remains to be seen.