Development and Field Application of a High Performance, Unstructured Simulator with Parallel Capability

Abstract

Introduction We present the development and field application of a high performance, unstructured simulator with parallel capability. The simulator integrates the subsurface with the surface network model. An earlier paper1 described the integrated surface model. This paper emphasizes the subsurface model and multi-field application. Object oriented concepts are used to gain flexibility, using efficient data structures to retain performance. The reservoir computational domain is composed of "Grids" and inter-grid connection structures that form the basis for parallelization. The Grids are ensembles of cells that are unstructured in general, and need not be contiguous. Besides unstructured (PEBI) grid simulation capability, this framework also forms the basis for simulating multiple, independently gridded reservoirs, with shared surface facilities, an important business problem. Earlier multi-reservoir workflows use LGR's in a fixed grid, making changes like adding a new reservoir difficult. Alternately, they use multiple processors with imperfect resolution of inter-reservoir coupling. In this paper, use of grid objects greatly facilitates the workflow; reservoir grids can be easily added or changed. Simultaneous solution of multi-reservoir and facilities system ensures consistent coupling throughout. Production stream mixing and discontinuous pressure drop correlations in the network are discussed. The mass-conservative implicit formulation, convergence criteria based on partial molar volume concepts, improved damping methods and implicit network coupling have resulted in significant performance gains. Benchmarking shows common speedup factors of 2–3 over industry simulators. This high performance enables simultaneous simulation of the large number of cells associated with multiple coupled reservoirs. Finally, we describe an application in a sub-sea development in deepwater West Africa. The development comprises six separate fields, which are to be tied back through subsea flowlines to a centrally positioned FPSO from where overall control of production and injection of water and gas will be maintained. The individual fields are spread out over many kilometers, so accurate representation of flowline performance is critical. Coupling of reservoir and surface networks was achieved using the new simulator, with good overall agreement in predicted performance from the previously used industry simulator. Greater stability and significant speedups were also achieved in the new runs. Grid "objects" The reservoir computational domain is composed of "Grids". The Grids are ensembles of cells that are unstructured in general, and need not even be contiguous. These are actually derived types, and not objects in the strict sense of the word; inheritance, polymorphism, and dynamic binding are not supported. However, we will occasionally refer to these as objects in this paper to simplify the conceptual discussion. All the necessary associated data is encapsulated within a Grid type and related Grid functions are stored in the same file. A Grid is a collection of cells that share certain common characteristics. For example, all the cells within a Grid share the same numerical treatment (IMPES or implicit), the same number of hydrocarbon components, and the same fluid characterization framework (e.g., compositional or black oil). Different Grids, however, can have different numerical treatment (IMPES or implicit), a different number of hydrocarbon components, and a different fluid characterization framework. As an example, it might be desirable to treat the near well areas in a fully implicit manner, while treating the rest of the reservoir as IMPES. Since there are no requirements on cell contiguity, all well blocks could be combined into a single implicit Grid, and all remaining cells could be combined into a second IMPES Grid.