No abstract

No abstract

This paper discusses the parallel implementation of complex wells modeling in a large-scale simulation. Due to intensive computational requirements of giant reservoirs with hundreds of these complex wells, parallel implementation is essential for practical simulation studies. In this work, a distributed memory approach using message passing interface (MPI) is employed for parallelization where each processor is responsible for the computation of one or more complex wells. For inter-processor communication, a non-blocking technique is utilized to increase the parallel efficiency. Parallel implementation is not the only challenge for large-scale simulation. For instance, full-field simulation with hundreds of complex wells increases the probability of a nonconvergence solution for at least one or more complex wells during reservoir simulation. Robust algorithms are needed to guarantee convergence and improve performance. Therefore, in this paper, we propose treating complex wells as a subsurface network where it can be represented using graph theory. In addition, simulation results for a full-field with hundreds of intelligent complex wells are included. This will show the importance of well coupling with rigorous treatment of downhole controls and devices on accurately modeling large-scale and complex reservoir.

No abstract

Anomalous sub-diffusion processes governed by non-local operators have been used in various applications. These are recently adapted to model complex single-and multi-phase fluid flow in unconventional reservoirs to understand the long-time behavior of various quantities of interest. Such processes, requiring long-time simulation, are modeled by a non-local in-time fractional derivative based Darcy's law and a conservation of mass equation, leading to a coupled system of first-order fractional partial differential equations (FPDEs). Standard time-stepping discretization of these continuous models lead to computational bottleneck for long-time simulation. Memory limitations [in each node of high performance computing (HPC) clusters] dictate computational constraints for resolving fine scale spatial structures in the models. We describe and implement a parallel-in-time and parallel-in-space mixed finite element (FEM) based discretization computer model to simulate the FPDEs. Our hybrid parallel-in-time-and-space framework facilitates efficient long-time simulation of the first-order computer model to overcome the time-stepping computational bottleneck and, within the node memory constraints, approximate the continuous models with large degrees of freedom (DoF).