Reservoir Model Updating by Ensemble Kalman Filter – Practical Approaches Using Grid Computing Technology

Inversion of reservoir flow models using dynamic production observations and realistic a priori geologic models quantifies a posteriori model parameter and forecast uncertainty. However, geologic model complexity, the requirement of ensembles or repetitions of many simulations, and the computational difficulty of realistic problems have slowed broad application of inverse methods. <br><br>This paper describes a method for reservoir model updating with Ensemble Kalman Filters (EnKF) using grid computing. EnKF uses a randomized ensemble of reservoir models for uncertainty estimation, and requires no gradient calculations. A computational grid combines the resources of many networked computers to solve large-scale data and computation problems. <br><br>An ensemble manager (ResGrid) drives the EnKF on a computational grid of high performance computers. ResGrid provides a web interface, or portal, to obtain grid resources and control execution without requiring detailed knowledge of grid computing architecture, security certification, protocols, or commands. A new queue mechanism in ResGrid reduces queue waits and increases resource utilization in multicluster environments, which is especially useful for a sequential, synchronized process like EnKF. <br><br>A synthetic case indicates that ResGrid efficiently performs EnKF inversions to obtain accurate, uncertainty-aware predictions of reservoir production. The ResGrid EnKF is open-source and available for downloading.<br>