Abstract
The Mixed Multiscale Finite Element method (MMsFE) is a promising alternative to traditional upscaling techniques in order to accelerate the simulation of flows in large heterogeneous porous media. Indeed, in this method, the calculation of the basis functions which encompass the fine-scale variations of the permeability field, can be performed in parallel and the size of the global linear system is reduced. However, we show in this work that a two-level MPI strategy should be used to adapt the calculation resources at these two steps of the algorithm and thus obtain a better scalability of the method. This strategy has been implemented for the resolution of the pressure equation which arises in two-phase flow models. Results of simulations performed on complex reservoir models show the benefits of this approach.
Highlights
Simulations of subsoil flows are based on geological models which provide a description of the geometry of the porous medium and its petrophysical properties such as porosity and permeability
Without the use of the ALEPH library, all MPI processes are used for the resolution of the coarse linear systems
The best configuration leads to a speedup of nearly 37 with 16 processes, which corresponds to solve coarse problem on one single node. These results tend to suggest that significant accelerations can be achieved during the resolution of the coarse system by limiting extra-node communications, which was carried out here by reducing the numbers of MPI processes and nodes at the same time
Summary
Simulations of subsoil flows are based on geological models which provide a description of the geometry of the porous medium and its petrophysical properties such as porosity and permeability. Like in the oil and gas industry for instance, the geological description of the studied area often leads to models where the permeability and porosity maps are defined on grids made of dozens or hundreds of millions of cells. This initial discretization will be referred to as finescale model in the following. Simulating multiphase flows, directly at this grid resolution, may lead to large computing times This problem gets even worse when several simulations have to be launched in order to perform a sensitivity analysis, a calibration of the model parameters, or test different flow scenarios. Various upscaling techniques and multiscale methods have been developed over the past years
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
