For existing wells in unconventional fields to remain profitable, regularly analyzing and forecasting well performance under various operational strategies is essential. There is also a growing need to create models that are scalable across the entire field. Data-driven methodologies, like decline curve analysis (DCA) and machine learning based methods, fall short in capturing crucial physical components. This is made worse by the absence of key parameters, including PVT and downhole pressures across several wells. Scalability issues make it difficult for models like numerical simulations and RTA to be applied to large well counts. The goal of this study is to develop physics-constrained, data-driven reservoir model for pressure and PVT aware production forecasting from routinely collected data.
To find rate-pressure relationships in unconventionals, we propose a hybrid data-driven and physics-informed model based on sparse nonlinear regression (SNR). Utilizing recent developments in machine learning and sparsity techniques, hybrid SNR is a novel framework for determining governing equations underlying fluid flow in unconventionals. Complex, non-uniform fractures, and multi-phase flow of fluids do not follow the same diagnostics behavior as planar fractures and single-phase flow, but exhibit more complex behavior not explained by analytical equations. To forecast the well for various flowing pressures and operating conditions, the hybrid SNR approach identifies these complexities by combining the most important pressure and time features that explain behavior of the phase rates.
The technique was validated using a benchmark model with varying and constant bottom hole pressures (BHP). It is then applied for complex real field cases on multi-phase wells. The findings show that the model is accurate enough to identify the necessary features that control rate-pressure behavior and to predict production for new BHPs. The method is robust because it can be used with any well, regardless of fluid type or flow conditions, and can be applied to any reservoir complexity because it does not depend on mechanistic fractures or simulation model assumptions. The method also enables the identification of dominant flow regimes by using terms with the highest contributions instead of the usual line fitting procedure.
Traditional methods like RTA tend to be interpretive and not suitable for field-scale applications because it necessitates identification of flow regimes and uses mechanistic model assumptions. The proposed technique makes use of a library of data-driven functions as well as features from common flow-regime equations, which serve as the foundation for traditional RTA. The model, however, is not constrained to known fixed relationships between pressure and rates that are only applicable under specific hypotheses (planar fractures, single-phase flowing conditions etc.). The method is novel as it allows for proposal of multiple functional forms for pressure and time, and hybrid SNR can resolve the proper functions based on the dominating multiple complex flow regimes.