Flow Simulation

Key Takeaways

• Grid design for flow simulation requires discretizing the reservoir volume into cells small enough to capture the spatial heterogeneity of rock properties (permeability, porosity) that controls flow behavior, but not so small that the computational cost of solving the pressure-saturation equations at each time step becomes prohibitive: typical simulation grids for full-field models range from 100,000 to 10 million cells, with cell dimensions of 50 to 200 meters in the horizontal direction and 1 to 10 meters in the vertical direction; fine-scale geological models built from well log and seismic data (which may have cell dimensions of 5 to 25 meters horizontally and 0.5 to 1 meter vertically, totaling hundreds of millions of cells) must be upscaled to the coarser simulation grid by computing the effective properties (average porosity, upscaled permeability tensor) that best represent the fine-scale heterogeneity in the coarser cell; the upscaling step introduces uncertainty because no single set of coarse-cell properties can perfectly represent the effect of fine-scale heterogeneity on flow at the larger scale, and different upscaling algorithms (arithmetic, geometric, harmonic, or flow-based upscaling) produce different results that can significantly affect the predicted sweep efficiency and recovery factor.
• History matching is the iterative process of adjusting the simulation model parameters (primarily permeability distribution, fault transmissibility multipliers, and aquifer properties) until the simulated production rates, wellhead pressures, water cuts, and gas-oil ratios match the historical field measurements within acceptable tolerances: a good history match is necessary to validate the model before it is used for forecast predictions, because a model that cannot reproduce observed historical behavior has no credibility for predicting future behavior; however, history matching is an ill-posed inverse problem (many different combinations of model parameters can produce equally good matches to the limited production data), and a history-matched model that fits the data perfectly is not necessarily the correct geological model; ensemble-based history matching methods (including the Ensemble Kalman Filter and Ensemble Smoother with Multiple Data Assimilation) generate multiple model realizations that all match the historical data, preserving geological uncertainty and providing probabilistic production forecasts rather than the false precision of a single history-matched model.
• Black-oil simulation (the most common type) represents reservoir fluids as three components (stock-tank oil, solution gas, and water) characterized by empirical correlations (black-oil PVT tables) for the density, viscosity, and formation volume factor of each phase as functions of pressure, providing a computationally efficient representation of oil-gas-water phase behavior that is adequate for most primary and secondary recovery simulations but cannot accurately model complex compositional effects (miscible gas injection, retrograde condensation in gas condensate reservoirs, or surfactant flooding); compositional simulation uses an equation-of-state (EOS) model to predict phase equilibrium and fluid properties from the detailed molecular composition of the reservoir fluid at each cell and each time step, providing accurate modeling of multicomponent phase behavior for EOR processes (CO2 injection, rich gas cycling in gas condensate fields) at the cost of 5 to 20 times higher computation time than black-oil simulation; the choice between black-oil and compositional simulation is based on the recovery process being modeled, the importance of compositional effects for the specific fluid system, and the available computation budget.
• Dual-porosity and dual-permeability simulation models are required for naturally fractured reservoirs where fluid flow occurs through two distinct porosity systems (the rock matrix with low permeability and high storage capacity, and the fracture network with high permeability and low storage capacity) that exchange fluids through a matrix-fracture transfer function: in a dual-porosity model, each simulation cell contains both a matrix block (representing the bulk of the rock volume and the primary storage) and a fracture (representing the connected fracture network that conducts flow to the wellbore), with the matrix-fracture transfer controlled by the sugar cube geometry of the matrix blocks (characterized by the fracture spacing in each coordinate direction) and the imbibition-driven or viscous pressure-driven fluid exchange between matrix and fracture; dual-permeability models add the ability for fluid to flow directly between matrix cells (not just through fractures), providing a more realistic representation of tight carbonates and tight sands with microfractures; the characterization of fracture properties (orientation, spacing, aperture, and connectivity) for input to dual-porosity simulation models is one of the most challenging aspects of naturally fractured reservoir modeling, typically requiring integration of image log data, core fracture description, well test analysis, and seismic attribute analysis.
• Streamline simulation is a computationally efficient alternative to conventional finite-difference simulation for large-scale flood pattern optimization and heterogeneity characterization studies, using the concept of streamlines (lines tangent to the local velocity field of the flowing fluid) to solve the transport equations independently along each streamline rather than for all cells simultaneously: the pressure equation is solved on a conventional grid at each time step to determine the velocity field, then streamlines are traced through the velocity field and the saturation and composition equations are solved along each streamline, avoiding the need to invert the large coupled system of equations that conventional simulators require; streamline simulation can be 10 to 100 times faster than equivalent conventional simulation for certain problems (immiscible waterflood with simple fluid physics), enabling optimization studies with thousands of realizations that would be impractical with conventional simulators; the limitation is that streamline simulation is less accurate for problems with capillary pressure-dominated flow (near the flood front), gravity-dominated flow (in gas-oil systems with significant density difference), or complex compositional physics (EOR).