This work focuses on the development of a new method for history matching problem that through a deterministic search finds a geologically feasible solution. Complex geology is taken into account evaluating multiple point statistics from earth model prototypes - training images. Further a function that measures similarity between statistics of a training image and statistics of any smooth model is introduced and its analytical gradient is computed. This allows us to apply any gradientbased method to history matching problem and guide a solution until it satisfies both production data and complexity of a prior model with desired accuracy. As a consequence of the approach, we sufficiently decrease the amount of forward simulations needed to resolve historical data and prior information.
Main Research Area:
EAGE Integrated Reservoir Modelling: Are we doing it right?, 2013