Comparative history matching for subsurface hydraulic characterizations.

Abstract:

Two history matching models are proposed for subsurface characterization based on the simulated annealing method with unconditional geostatistical simulation and a radial basis function network as random walk transition kernels. For verification, the developed models and ensemble Kalman filter (EnKF) based history matching are comparatively applied to two synthetic cases. In Case 1, history curves at 16 observation wells and the statistical information on the targeted properties are fully informed while, in Case 2, only the hydraulic measurements are available. Through in-depth comparisons, it is concluded that the developed models as well as the EnKF based history matching model show reasonable prediction accuracy with full information on spatial statistics of Case 1, whereas the prediction accuracies of the models are severely deteriorated with reduced information in Case 2. For both cases, the developed simulated annealing (SA) based history matching model with a radial basis function network has the highest predictability and the EnKF model shows similar prediction accuracy. However, from the perspective of the prediction stability, the EnKF model is found to be inferior to the developed models. This inferiority of the EnKF model is attributed to the sensitivity of the model to the composition of the initial ensemble.