Inverse Modeling and Uncertainty Analysis
Our research in this area involves improving existing gradient-based optimization algorithms, optimal control theory and adjoint sensitivity computations, developing efficient reparameterization methods for reservoir models (Fig. 2) and data spaces and finding ways to efficiently integrate dynamic and static data. We are working to improve local search optimization methods such as the Quasi-Newton (particularly the LBFGS), the conjugate gradient and the Levenberg-Marquardt algorithms. Improvement of these algorithms would involve finding ways of speeding up the algorithms and reducing the storage requirements associated with them. Algorithms developed would be applied to reservoir parameter estimation. Improving reparameterization techniques is a key aspect of our research.
Fig. 2a:Fully-distributed reservoir model with high permeability streaks
Fig. 2a: Object-oriented reservoir model with four facies