Simulação de reservatórios de petróleo em paralelo utilizando malhas não-estruturadas 2D e 3D

Abstract

Large - scale simulation s involving hundreds of thousands of grid blocks require application of parallelization techniques in order to achieve practical computational times. There are essentially three types of parallelization: distributed memory, shared memory, and a combination of the two mentioned. This work is based on the first approach, where the domain is divided among processes and each one is responsible only for its portion of the reservoir . This a pproach has two main advantages: it reduces the required memory per process and allows the simulations to be carried out using clusters with large number of processors. In this work, open source libraries were used to partition the computational domain, manage the grid information between the process e s, and solve the linear syste m of equations generated from the discretization of partial differential equation modeling fluid flow in the reservoir. ParMetis (Parallel Graph Partitioning and Fill - reducing Matrix Ordering) is used to partition the computational domains, FMDB (Flexible Distributed Mesh Database) is responsible to manage the grid information between the process e s, and PETSc (Portable, Extensible Toolkit for Scientific Comput ation) solves the linear system of equations. The numerical approach is based on the Element based Finite Volume Method (EbFVM) in conjunction with unstructured meshes . The main challenge was to manage the grid, fluids, and reservoir data set in such a way that the comm unications between the processe s were reduced. It was used an in - house compositional, multicomponent/multiphase simulator called UTCOMP, which was developed at The University of Texas at Austin, in order to perform this implementation. It is show n that the EbFVM is suited for modeling reservoirs with complex geometries , and its performance in parallel mode is presented . The results are evaluated in terms of oil and gas production curves, speedup curves and CPU times for various case studies.