Hydrogen production via in-situ combustion gasification: Insights from lab-scale modeling assisted by machine learning

Abstract

In-situ combustion gasification (ISCG) of heavy oil reservoirs has recently emerged as a promising approach for carbon-zero hydrogen (H₂) production. While combustion tube experiments report low H₂ production, field-scale projects recorded relatively high H₂ yields. To investigate this discrepancy, we conducted 2000 simulations via CMG-STARS assisted by machine learning and identified the combination of parameters that leads to the optimal scenarios. We found that, under the condition of co-injection of water and air, the temperature with the highest H₂ production is around 400 °C where gasification dominates H₂ generation, while cumulative CO₂ is lower than that of H₂. The optimized cases revealed up to 57.7 kg/m³ H₂ per unit of consumed oil and 0.0125 kg total H₂ produced. Our ML models achieved high predictive accuracy (training score >96 %, testing score >88 %), enabling fast evaluation of optimal input conditions. The limited availability of reactants and the inability to sustain high temperatures in combustion tube likely account for the low H₂ production. Machine learning promoted the preliminary validation of the feasibility of ISCG process. This paper also provides insights that can guide future research on H₂ generation via ISCG, thereby supporting the development of this promising technology.