History matching a field's performance to understand the reservoir behavior and characterize its static and dynamic properties has been a key activity for reservoir engineers for a long time. Recently, significant effort has been made to devise techniques that allow this process to be automated. These techniques suffer with a number of limitations and often yield History Match (HM) solution points in the uncertainty space that carry certain bias inherent to the algorithm used. Common limitations include:

  • Techniques more often failing to yield any realization with acceptable HM quality at the well levels,

  • Large number of iterations and simulation runs required to minimize the HM errors to acceptable values and,

  • Significant volume of information generated during this iterative process, which becomes unmanageable to interpret and reconcile.

Consequently, the assisted HM unfortunately becomes just an iterative process of minimizing the objective function, and the main goal of understanding the reservoir performance gets diluted. When these techniques are applied for Enhanced Oil Recovery (EOR) studies, the list of uncertainty parameters becomes very extensive due to the addition of parameters defining the EOR process itself. As a result, the number of scenarios increases in addition simulations tend to become slower due to incorporation of EOR module and model requirements for space and time resolution to simulate physio-chemical phenomena. Traditional HM techniques then become cumbersome which might lead to inadequate characterization of the potential upside and downside scenarios. This in turn could adversely impact business decisions involving huge capital investments for any field (re-)development opportunities.

This paper will discuss a technique that evolved during the HM exercise for an EOR pilot area in the Middle East. The methodology preserves the idea of Assisted History Match (AHM) to generate multiple HM realizations, however, the solution points are found much quicker eliminating large number of iterations, thereby minimizing the computational expense.

The devised methodology is based on the combination of Design of Experiment (DoE) based Stochastic Uncertainty Management (SUM) workflow with the gradient based calculation (Adjoint) approach to find multiple HM realizations. First, a complete stochastic uncertainty management workflow is applied sampling the entire uncertainty space and multiple realizations are screened ensuring enough variability in parameters and acceptable HM error. The gradient based calculations are then applied on each of these selected realizations to further minimize the HM error.

The Adjoint method yields final set of improved HM realizations with different starting points in the solution space that were obtained via DoE. This helps in two ways – firstly, it covers the entire uncertainty space for better characterization of upside and downside cases and secondly, provides a focused view of the reservoir characteristics.

Additionally, the paper would also offer insights gained from the HM exercise into water coning behavior in a viscous oil reservoir and illustrate the reservoir parameters and their significance that need to be addressed to adequately capture the coning phenomenon.

You can access this article if you purchase or spend a download.