A Robust Method for Adaptive Model Uncertainty Reduction, 20-R9704

Printer Friendly Version

Principal Investigators
Alexander Sun
Alan Morris
Sitakanta Mohanty

Inclusive Dates:  04/01/07 – Current

Background -  Effective planning, exploitation, and management of oil reservoirs call for efficient and accurate reservoir simulation models. Many oil reservoirs are naturally fractured. Geologic faults and fractures form either preferential flow paths or barriers to fluid flow and, therefore, can greatly affect the performance of oil recovery technologies. Accurate representation of fault and fracture patterns in reservoir models is key to reliably estimating the reserve size and predicting reservoir production potential and safe production rates. However, knowledge of fault and fracture patterns in reservoirs can be highly uncertain because they are not directly observable. Technologies such as seismic imaging and horizontal drilling provide useful static data about fault patterns. On the other hand, dynamic production data (i.e., production rate as a function of time) can be used to infer additional information about geologic features that are not already accounted for in an existing model. An emerging need in the petroleum industry is how to integrate dynamic data to support real-time or near-real-time model updating and decision-making.

Approach - Traditional data assimilation techniques have practical limitations when applied to high-dimensional, nonlinear systems. The ensemble Kalman filter, introduced recently to handle high-dimensional, nonlinear models, can be a promising method for history-matching reservoir models. SwRI is developing a robust ensemble Kalman filter (REnKF) to circumvent issues associated with filter divergence and ensemble outliers, which are known to adversely affect the performance of the standard ensemble Kalman filter.

Accomplishments - The REnKF is currently being tested using simulated fractured reservoir problems in which randomly generated fracture networks are used. Initial tests show that the REnKF, when combined with synthetic production data, can be effective in the initial identification of geologic features that are unknown, such as faults or high permeability zones.

2007 Program Home