Abstract
Current unconventional well forecasting relies primarily on decline curve analysis or data-driven methods, which lack predictive power for new well designs and changing operating conditions. While full-physics reservoir simulation provides more reliable predictions, its computational demands make it impractical for large-scale application. This study presents a novel methodology using reduced physics-based 1D simulation models that balance computational efficiency with physical accuracy, enabling rapid forecasting of thousands of wells while maintaining the ability to model design changes and operating constraints.
The workflow employs a computationally efficient single-well reservoir simulation incorporating essential physics for robust prediction. For each well, an automated two-stage history matching process is implemented. The first stage estimates fundamental parameters (fracture area and oil in place) while the second performs gradient-based calibration of detailed parameters. The workflow leverages a fully differentiable simulator with adjoint capability, enabling efficient sensitivity calculations and massive parallelization. This architecture allows for rapid processing of large well counts while maintaining physical consistency.
Application across multiple field cases demonstrates that while multiple parameter configurations can match production data equally well, even in 1D models, this non-uniqueness can be effectively managed. By anchoring the detailed calibration process to rate transient analysis results, the parameter space is appropriately constrained while preserving key degrees of freedom. This approach enables quantification of prediction uncertainties while maintaining physical consistency. The methodology successfully delivers pressure-aware forecasts that capture both reservoir and operational effects at computational speeds suitable for full-field implementation.
This work introduces a practical synthesis of established technical elements to enable physics-based forecasting at scale in unconventional reservoirs. The key innovation lies in combining adjoint-based calibration with rate transient analysis constraints in an automated workflow. Beyond forecasting, the calibrated models provide valuable insights into subsurface characteristics and well behavior, offering a dual-purpose tool for both prediction and diagnostics. The methodology bridges the gap between simplified decline curves and full-physics simulation, providing a scalable solution for modern field development challenges.