Modern Deep Learning Framework Using Temporal Fusion Transformer to Optimize Sequestration Efficiency and Oil Production in CCUS Projects

Abstract Optimizing Water-Alternating-Gas (WAG) processes is critical for maximizing both oil recovery and carbon sequestration efficiency in CO2 Enhanced Oil Recovery (EOR) projects. Key design parameters include injection rate, the gas-to-water volume ratio (WAG ratio), and throughput—defined as the pore volume of CO2 injected per day. Conventional optimization using simulation models can be cumbersome due to the vast design space and high uncertainty. In this study, we apply a deep learning model—the Temporal Fusion Transformer (TFT)—for multivariate forecasting of oil production and CO2 sequestration efficiency across a range of WAG scenarios using field data from six legacy CO2 EOR projects. Digitized monthly production and injection data were collected from six U.S. fields representing diverse depositional environments. For each field, historical waterflood trends were isolated using exponential decline, and the incremental CO2 response was modeled across various WAG ratios. The TFT model was trained on this multivariate time-series data to forecast one year of production response for both oil and CO2 production. Hyper parameter optimization was used to optimize model inputs. The model achieved high predictive accuracy with R2 values exceeding 0.87 and robust MAE factors. The forecasts enabled actionable operational recommendations. For instance, adjusting the WAG ratio from 0.9 to 1.7 in the Denver Unit increased CO2 retention by 48% and utilization efficiency by 38%, while reducing the WAG ratio in East Vacuum from 0.7 to 0.2 yielded a retention improvement of 47%. These optimizations directly support enhanced carbon abatement by reducing CO2 recycling and improving sequestration permanence. To our knowledge, this is the first application of the TFT model for real-field CO2 EOR forecasting and WAG strategy optimization. The approach offers a scalable, data-driven alternative to simulation workflows, aligning EOR operations with climate and sustainability targets.