Given the critical importance of geological carbon storage in mitigating climate change, this study examines the influence of injection temperature on solubility and mineral trapping in saline aquifers using thermo-hydro-chemical (THC) simulations with the TOUGH code. To address the high computational cost of traditional simulations, we developed a lightweight deep learning model, GeoMTNet. Its shared encoder and dual-branch design enables unified extraction of global and local features and allows simultaneous prediction of CO2 storage amounts and spatial distribution fields—capabilities beyond those of conventional image-based networks. Using grid and formation properties as inputs, GeoMTNet rapidly predicts solubility and mineral trapping with high accuracy (R2 > 0.99 for amount; R2 > 0.97 and SSIM >0.85 for distribution) and achieves about 30,000 × speedup over traditional simulations. Both simulations and deep learning predictions show that injection temperature strongly affects a near-wellbore zone. Lower temperatures enhance early (<10 years) solubility trapping but inhibit late (>50 years) mineral trapping, whereas higher temperatures inhibit early solubility trapping but markedly promote late mineral trapping. The deep learning model effectively captures and explains the influence of injection temperature on CO2 storage dynamics. At 100 years, both dissolution storage and mineralization storage increase with higher injection temperatures, with mineralization storage exhibiting greater sensitivity. Therefore, higher injection temperatures are recommended in field operations to enhance safe storage amount, defined as the sum of dissolved and mineralized CO2 storage amounts.
扫码关注我们
求助内容:
应助结果提醒方式:
