Use of External Parallelization to Improve History Matching

Abstract Although automatic history matching is a difficult task due to production and reservoir characterization uncertainties, it is possible to develop methodologies and computer programs to improve the matching process. An important tool in the development of this type of methodology is the external parallelization of reservoir simulators where simulations can be divided into a network of workstations, reducing significantly the total time of the process, especially for some network types. This work shows some applications of the use of external parallelization applied to the history matching process. Due to the high the number of simulations which are necessary, a routine was created to distribute the runs over a network of workstation. The software PVM (Parallel Virtual Machine) was used and an algorithm was developed to make the best use the target network. There is a great number of applications of this technique in reservoir simulation. Every application which requires several simulations can be improved with external parallelization. A routine to manage all simulations is very important to make the best use of a network. Introduction The modeling of hydrocarbon reservoirs can be so complex that many times the only possible and reliable technique to be used is making adjustments in the reservoir model until numerical simulators results match observed production data. Such a process, known as history matching, is usually very slow due to the high number of variables and degree of uncertainty and heterogeneity of reservoir parameters. The complexity of the process leads most people to believe that automatic history matching is neither viable nor possible. For this reason, "trial and error" procedures based on personal experience is a common practice. However, recent hardware and software developments can be used to accelerate a history matching process, making it faster and more reliable than "trial and error" procedures. Although it is not yet possible to have the whole process automatic, the automation of several steps is both possible and necessary. This work deals with one important tool which can be used to accelerate history matching which is external parallelization, i.e., the use of a software to efficiently distribute simulation runs that can be performed in parallel in an existing network of workstations. The software developed in this work (MPS - Multiple Parallel Simulations) is based on PVM (Parallel Virtual Machine) libraries. The applications cited here are specifically applied to history matching but they can be used to any process that requires several simulation runs. The ultimate application of MPS is to manage all simulations on a network.