Fast Closed-Loop Well Placement Optimization with Ensemble-Based Data-Space Inversion Framework

Closed-loop field development entails the (re-)optimization of field development decisions with ensembles of model realizations regularly updated based on the latest data gathered. Revising the subsequent decisions once new information becomes available leads to improvements in the quality of such decisions and, thereby, in the performance goals of the field development project. However, closed-loop workflows typically imply the use of computationally-intensive model-based optimization and data assimilation techniques requiring a large number of reservoir simulations, which makes large-scale applications inviable. Direct forecasting techniques such as data-space inversion (DSI) offer an alternative to conventional model-based approaches for reducing the computational load of workflows where conditioning of models (and model forecasts) to observation data must be repeated iteratively (e.g., closed-loop optimization and value of information workflows). Recent studies on DSI have focused either on the forecasting aspect only or on applications to closed-loop reservoir management, limited to well control optimization. In this work, we investigate how DSI can be applied to the closed-loop field development problem. We present our implementation of a generic DSI-based closed-loop framework using the ensemble-based methods available within mature open-source data assimilation and robust optimization tools. The developed workflow also utilizes machine learning techniques to better handle the presence of non-linearities in the system response which are typical of real-life applications. We apply the workflow to an infill drilling problem in a realistic synthetic oil-water reservoir. We optimize the placement of future infill wells to maximize economics of the project based on posterior production forecasts conditioned to the production data from existing wells. The infill drilling campaign is split in two phases. In each phase (separated by 5 years) a set of new wells is introduced. The DSI-based workflow is used to "close the loop" twice, with additional production data being incorporated between the two phases. The optimized well placement strategies obtained with DSI are verified against the response of the synthetic truth model. An increase of 300 million USD (10%) in terms of net present value on the truth model is achieved, attesting the validity of the approach. This study showcases the potential of DSI combined with ensemble-based optimization as an effective approach to accelerate closed-loop workflows involving a broader scope of decisions, such as well placement. This opens up opportunities to create practical tools that simplify the application of computer-assisted closed-loop and value of information workflows to assist practitioners in finding improved solutions to a wider range of real-life challenges. Copyright 2025, Society of Petroleum Engineers.