| Summary: | In this thesis two stochastic algorithms, the Ensemble Smoother (ES) and the Ensem-ble Kalman Filter (EnKF), have been utilized as automatic history matching meth-ods through the Statoil developed program the Ensemble based Reservoir Tool. Sta-toil, through the IO Center at NTNU, provided real historical production and pressuredata. These data was used to condition on a parametrized ECLIPSE reservoir modelof the Norne field. Parameters conditioned on includes the field parameters porosity,i-permeability, net-to-gross, and z-direction transmissibility multiplier, as well as fault multipliers, region transmissibility multipliers, minimum pore volumes and relative permeability endscaling options. The E-segment model and the Full-field model were both studied. With an ensemble size of 120 realizations for the E-Segment model and 80 realizations for the Full-field model, both algorithms performed well. In most of the results an initial high uncertainty in the prior provided the necessary coverage of the historical observed data. Overall the EnKF performed better than the ES, which is natural comparing time and computational power required. Some spurious correlations and one particular ensemble collapse were experienced. The EnKF-analysed ensemble production rates provides a lot less uncertainty than the initial ensemble, and for most plots a slightly better match than the reference case. The EnKF-algorithm is shown to be vulnerable to small ensemble sizes versus large amounts of conditioning parameters, as well as highly correlated historical observations.
|
|---|