Deriving Unconventional Reservoir Predictive Models from Historic Data using Case-Based Reasoning (CBR)

Saputelli, Luigi (Frontender Corporation) | Verde, Alexander (Frontender Corporation) | Haris, Zameel (Frontender Corporation) | Díaz, Daniel (Frontender Corporation) | Diaz, Daniel (now with Geopark Colombia)



Decline curve analysis models are adequate for unconventional field production forecasting as a function of wells scheduling and high-level screening of reservoir capacity, depletion scenarios and market needs, however those models do not consider rigorous physics since they assume constant production conditions. For complex field development and production operations optimization, integrated reservoir performance models constrained with well and surface network pressure must be considered. Alternatively, numeric simulation-driven forecasting methods provide an advanced level of subsurface response however they require intensive model tuning effort which may not be practical for a large number of wells with limited reservoir data.

The objective of this effort was to develop an automated workflow to generate production forecasts in the context of integrated reservoir to facilities production modeling. We leveraged on reservoir analytic models and gradient based optimization to identify associated model parameters. Since gradient-based optimization required the specification of an initial guess, we used case-based reasoning to focus on the most relevant parameters and to select better initial guesses a smaller solution range to narrow possible solutions.

The proposed solution was successfully tested in a large tight-gas field with approximately 250 wells and a long production history. For this field, rate-transient-analysis models provided certain advantages as they captured field performance response using pressure vs. rate fundamental modeling with the tuning of few parameters. Decline curve models were used as inputs to derive a full physics-based reservoir and well performance models therefore translating time-dependent models into fully pressure and time dependent models. The identified parameters, such as reservoir pressure, well permeability and well drainage radius provided the means to generate full physics well responsive models. The calibrated reservoir and well models were able to reproduce production history with minimum error as well as to provide a means to optimize production using an integrated reservoir, well and facilities production model, which were not possible using decline curve models alone.