Answers are grounded in this paper and cite their sources.
Accurate oil and gas production-rate forecasting is essential for reservoir management and field-development planning. Capturing the relationships between production history, reservoir conditions, and operational variables remains challenging because of their nonlinear and time-dependent behavior. This study proposes a Reservoir-Aware Dual-Attention Recurrent Network (RA-DARN) for field-level production-rate forecasting using historical production and reservoir data. The proposed architecture integrates bidirectional gated recurrent learning with dual-attention mechanisms. Two feature-aggregation pathways named aggregated encoding features (AEF) and aggregated original features (AOF) are used to preserve both encoded temporal patterns and the original input information, improving prediction reliability. Using an eight-observation input window, RA-DARN was trained for one-step-ahead prediction and evaluated through recursive multistep forecasting. The model achieved R2 values of 0.998 for oil and 0.980 for gas production rates, with corresponding RMSE values of 11.06 STB/day and 172.15 MSCF/day. It also achieved lower prediction errors than the evaluated conventional machine-learning models. Comparison with the history-matched PETREL–ECLIPSE realization showed closer statistical agreement between the RA-DARN predictions and observed oil and gas production rates for the investigated field case. This comparison reflects the specific simulation calibration achieved in this study and does not establish general superiority over reservoir simulation. RA-DARN therefore offers a complementary tool for production-rate forecasting, alongside the physical interpretation and scenario evaluation provided by reservoir simulation. Validation remains limited to one oilfield, and further assessment is required to establish applicability to other reservoirs.
No takes yet. Share an insight, caveat, or question.
Dawood et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: