Abstract
It has lately been reported several successful applications where the ensemble Kalman filter has been used to estimate reservoir properties such as permeability and porosity. However, a thorough investigation of robustness and performance is still missing for this approach. In this paper we aim at filling this gap by studying the robustness of the methodology. One aspect which is investigated is how the filter depends on the initial ensemble. As the initial ensemble is created in a stochastic way, one can not be certain that the results obtained from one run represent the filter performance. Another aspect of interest is how prior information can be used to obtain best possible initial fields. The influence of geostatistical information on the estimated solutions is studied. In addition, the quality of the estimated fields is investigated by evaluating if the estimated static fields are reasonable when treated as the solution of the history matching problem.
The estimation technique has been applied to the widely used PUNQ-S3 reservoir model, which is a small size synthetic 3-D reservoir engineering model. Both permeability and porosity are tuned, and measurements consist of well bottom-hole pressures, water cuts and gas-oil ratios. The initial fields are conditioned on the porosities in the gridblocks where the wells are located. By using a synthetic reservoir model it is possible to calculate the uncertainty of forecasts, and compare this with the true solution.
