Successful Application of an Ensemble Modeling Workflow for a Deepwater Field in US Gulf of Mexico

Abstract

A Deepwater field in the Gulf of Mexico is currently in its mid-production life and undergoing development optimization, which was underpinned by a robust evaluation using multiple geologic scenarios that were history matched to observed field data. This challenge was addressed using an ensemble modeling workflow that has been guided by BP's Top-Down Reservoir Modeling (TDRMTM) (1) principles. The ensemble contained 27 distinct geologic scenarios and was used to evaluate water injection, field expansion, and infills. For a wide range of geologic scenarios covering alternate structural and stratigraphic models, a workflow called "forced boxing" was executed along with the assisted history matching algorithm, PSO-MADS (2, 3), to fully history match historical rates and pressures for more than 27 distinct combinations of geologic scenarios. These scenarios covered a wide range of oil-in place, connectivity, aquifer strength, and relative permeability behavior. The final calibrated ensemble of 1000s of models was then down sampled to ~100 distinct models to use for the probabilistic evaluation of water injection, field expansion, and infills. The "forced boxing" technique was successful at finding high quality history matches for 25 of the 27 distinct geologic scenarios. The history matching workflow considered well-level production rates, including water cut, as well as reservoir pressure measurements (MDTs). A match quality acceptance workflow was set up to find acceptable models out of each geologic scenario. The resulting ensemble of models contained over 20,000 distinct simulation cases. A methodology was used to down sample those cases to a final ensemble of ~100 models covering the 25 geologic scenarios. This workflow is an improvement to conventional history matching and uncertainty workflows because this workflow ensures multiple geologic scenarios are matched and included in the ensemble and the final set of models gives a probabilistic view of the predictive outcomes. The ~100 model ensemble was then utilized to explore different field development opportunities and included the successful selection of an economic infill producer target. The "forced boxing" approach, which entailed history matching distinct static parameter combinations, was built to ensure diversity of scenarios/outcomes as opposed to traditional workflows that focus on finding the best history matches. Given the importance of cross-fault communication between different reservoir sands, the parameterization of fault throws as a variable controlled in the optimization process was also a novel addition to the history matching workflow.