Using Ensemble Kalman Filter Techniques (EnKF) in Assisted History Matching

Wisam Shaker & Loran Taabbodi
36

Abstract

In general, a large computational effort is required, either in objective function evaluation in both gradient and gradient free optimization methods. Gradient method is highly dependent on the source code of the reservoir simulation. Gradient method does not work in a case of non-linearity. Free-gradient methods are not dependent on the simulator source code, but usually take thousands of simulation run to find the global minimal point. No guarantee that an optimal solution is found. A high dimensional response modelling may not represent a non-linear system well.

The Ensemble Kalam Filter (EnKf) has several major advantages for large scale history matching problems: It does not depend on the specific reservoir simulator; the computational cost is fairly low; it can account the uncertainty in the unknown parameters using a few Monte Carlo realizations.