Advances in Tightly Coupled Reservoir/ Wellbore/Surface-Network Simulation
Summary One of the most perplexing and difficult challenges in the industry is deciding how to develop a new oil or gas field. It is necessary to estimate recoverable reserves, design the most efficient exploitation strategy, decide where and when to drill wells and install surface facilities, and predict the rate of production. This requires a clear understanding of energy distribution and fluid movements throughout the entire system, under any given operational scenario or market-demand situation. Even after a reservoir-development plan is selected, there are many possible facility designs, each with different investment and operating costs. An important, but not always considered, fact is that each facility scheme could result in different future production rates owing to various types, sizes, and configurations of fluid-flow facilities. Selecting the best design for the asset requires the most accurate production forecasts possible over the forecast life cycle. No other single technology has the ability to provide this insight, as well as tightly coupled reservoir and facility simulation, because it combines all pertinent geological and engineering data into a single, comprehensive, dynamic model of the entire oilfield flow system. An integrated oilfield simulation system accounts for all dynamic flow effects and provides a test environment for quickly and accurately comparing alternative designs. This paper provides a brief background of this technology and gives a review of a major development project where it is currently being applied. Finally, we describe some recent significant advances in the technology that make it more stable, accurate, and rigorous. Introduction Finite-difference reservoir simulation is widely used to predict production performance of oil and gas fields. This is usually done in a "stand-alone" mode, where individual well performance is commonly calculated from pregenerated multiphase wellbore flow tables that cover various ranges of wellhead and bottomhole pressures, gas/oil ratios (GOR's) and water/oil ratios (WOR's). The reservoir simulator determines the predicted production rate from these tables, normally assuming a fixed wellhead pressure and using a flowing bottomhole pressure calculated by the reservoir simulator. With this scheme it is not possible to consider the changing flow-resistance effects of the piping system as various fluids merge or split in the surface network. Neglecting this interaction of the surface network can, in many cases, introduce substantial errors into predicted performance. Basing multimillion- (in some cases, billion-) dollar exploitation designs on performance predictions that are suboptimal can be very detrimental to the asset's long-range profitability. To help eliminate this problem, considerable attention is being given to coupling reservoir simulators and multiphase facility network simulators to improve the accuracy of forecasting. Landscape Surface-network simulation technology was first introduced in 1976.1 Although successfully applied in selected cases, the concept was not widely adopted because of the excessive additional computing demands on computers of that era. In those earlier applications, the time consumed by the facility calculations could actually exceed the reservoir calculations.2,3 As computer performance has increased by orders of magnitude, this has become less of an issue. Reservoir model sizes have increased dramatically with much finer grids that take advantage of the increased computer power, but there was no need for a corresponding increase in the size of the facility models. Today, with tightly coupled reservoir/wellbore/surface models, the facility calculations are a fairly small part of the overall computing time and there is considerable effort in the industry to build these types of systems.4,5 Chevron's current tightly coupled oilfield simulation system is CHEARS®***/PIPESOFT-2™****. CHEARS® is a fully implicit 3D reservoir simulator with black-oil, compositional, thermal, miscible, and polymer formulations. It has fully implicit dual porosity, dual permeability options, and unlimited multiple-level local grid refinement. PIPESOFT-2™ is a comprehensive multiphase wellbore/surface-network simulator. It has black-oil, compositional, CO2, steam, and non-Newtonian fluid capabilities. It can solve any type of complex nested looping, both surface and subsurface. The coupling is done at the wellbore completion interval, which is the natural domain boundary between the flow systems. We refer to our implementation as "tightly coupled" because information is dynamically exchanged directly between the simulators without any intermediate intervention. A simple representation of the interaction between the simulators is shown in Fig. 1. Gorgon Field Example The following is an example of how this technology is currently being used. The Gorgon field is a Triassic gas accumulation estimated to contain over 20 Tscf of gas, located 130 km offshore northwest Australia in 300 m of water (Fig. 2). It is currently undergoing development studies for an LNG project. Field and Model Description. The field is 45 km long and 9 km wide, and it comprises more than 2000 m of Triassic fluvial Mungaroo formation in angular discordance with a Jurassic-age unconformity. It has been subdivided into 11 vertical intervals (or zones) on the basis of regional sequence boundaries and depositional systems. These 11 zones were first modeled individually with an object-based modeling technique before being stacked into a 715-layer full-field geologic model. This model was subsequently scaled up to a 46-layer reservoir simulation model, reducing the size of the model from 4.5 million cells to 290,000 cells. While the scaleup process preserved the original 11 zone boundaries, the majority of the layers were located in regions identified as key flow units. In addition to vertical subdivision, seismic and appraisal well data suggest structural compartmentalization, resulting in six major fault blocks. After deactivating appropriate cells, the final simulation model contained 50,000 active cells and was initialized with 35 independent pressure regions. Each of these regions corresponds to a single zone in a single fault block.