Event-Targeting Model Calibration Used for History Matching Large Simulation Cases

摘要

Large simulation models with excessive simulation time are traditionally a challenge for the reservoir engineer. This paper introduces a new concept of Event Targeting Model Calibration used for History Matching and Uncertainty Quantification in Reservoir Simulation. It is shown that the history matching process of large reservoir simulation models can be significantly improved by coupling different experimental design, optimization and analysis techniques. Assisted history matching techniques using stochastic and direct search methods have already proven to outperform manual workflows on small and moderate size simulation models. The application of the same techniques and methodology when confronted with the largest simulation models (defined by long simulation times) have proven challenging. A more sophisticated use of the assisted history matching tool box is necessary to utilize CPUs for multiple simulation models as efficient as possible even if distributed computing capabilities are available. The paper describes a workflow employing assisted history matching techniques to handle ‘monster’ simulation models. An Event Targeting Model Calibration work process is introduced which focuses on key historical events and divides the production history into main time periods. The optimization algorithms usually used in assisted history matching studies are replaced by experimental design methods to investigate the different time periods. Analysis techniques like a newly implemented cluster analysis is used to identify alternative history matched models in a multi-objective optimization formulation. Optimization algorithms are finally used for fine-tuning purposes. In this framework, a remarkable improvement of both the history matching process and uncertainty quantification is possible. This is a significant break through, improving the capability of understanding reservoir uncertainties for large field developments. This paper summarizes experiences from several different complex history matching studies and outlines guidelines to apply state-of-the-art optimization techniques in combination with experimental design methods to the problem of History Matching and Uncertainty Quantification of large simulation cases.