| dc.description.abstract | Ensemble-based reservoir management procedures have become the standard when it comes to uncertainty quantification; this in turn has resulted in increased processing requirements, which are not only time-consuming but can also negatively impact the economics of the projects. To mitigate the problems posed by conventional methods in such cases, researchers have shifted their focus to data-driven approaches that could prove to be robust, computationally inexpensive, and effective while dealing with closed-loop reservoir management.
These direct forecasting techniques have shown enormous potential, and data space inversion (DSI) is one such technique. DSI conditions its predictions using historically observed data. This study presents the implementation of DSI using Ensemble Smoother with Multiple Data Assimilation (ES-MDA) and an optimization tool using a stochastic simplex approximate gradient. Both tools used for ES-MDA and optimization are mature tools with large-scale commercial applications. The optimization and sensitivity are performed on a synthetic field, Drogon, selected specifically due to its realistic characteristics.
The entire workflow was implemented based on two different sets of observations. The predictions are conditioned on their respective observations and are then compared to the synthetic truths. These results are then optimized along with the synthetic truths. In both cases, the DSI ensemble prediction and optimization results have an acceptable match with their respective truths, thus establishing the viability and robustness of the DSI technique. In the DSI framework, one can choose specific properties to predict, which eliminates the time-consuming calculation of redundant properties.
The results confirm the huge potential of this technique, especially when used in closed-loop reservoir management, providing us with efficient, inexpensive, and accurate predictions for decision-making. | |
