General Parallel Reservoir Simulation

Abstract

Abstract Faster reservoir simulation turnaround time continues to be a major industry priority. Simultaneously, model sizes are reaching a billion cells and the recovery mechanisms and reservoir management processes to be modeled are rapidly changing and are becoming computationally more expensive. A new reservoir modeling solution has been developed to quickly solve these largest and most complex modeling studies within ExxonMobil. This latest generation reservoir simulator has been designed from the ground up with 60 years of internal reservoir simulator development experience. Some of the key learnings incorporated into the new reservoir simulator include a requirement for a flexible and modular software framework, a general flow formulation that is decoupled from a very fast and highly accurate phase behavior engine, and unstructured architecture for unstructured grids. This simulator is optimized for massive distributed memory parallelism to take advantage of ExxonMobil's world class supercomputer, Discovery. A new fluid agnostic formulation has been developed based on general material balance. A highly optimized and highly accurate fluid library supports liquid-liquid-vapor calculations and is the only differentiation between black-oil and compositional options. These considerations are critical for maintaining computationally efficient software under heavy software development with a large development team. The distributed memory parallelism has been tested and has shown to be very efficient on the Discovery platform. Strong scalability tests have been run to 16,000 cores with good parallel performance to 6000 unknowns per core. These are some of the largest core counts for parallel reservoir simulation with unstructured grids seen in the industry and reduce model run times from days to minutes. These drastically reduced run times have allowed the new simulator to include heavy computational methods, such a nonlinear finite volumes or implicit reactions, for practical use as well as supporting very large models in excess of 100 million cells. Flexible well management control is provided through Python scripts. This allows users to customize asset-specific control strategies via a well-known and straight-forward scripting language. Alternatively, an internal optimization engine can manage the field subject to high level constraints provided by the user.