Full text loading...
-
A Reduced Order Surrogate Model for Optimal Reservoir Management
- Publisher: European Association of Geoscientists & Engineers
- Source: Conference Proceedings, ECMOR XIV - 14th European Conference on the Mathematics of Oil Recovery, Sep 2014, Volume 2014, p.1 - 12
Abstract
This work investigates the application in a Multifidelity Sequential Approximate Optmization (MSAO) framework of a proxy model that uses Trajectory Piecewise Linearization (TPWL) as a numerical complexity reduction technique and Proper Orthogonal Decomposition (POD) as the dimensional reduction technique. TPWL linearizes the reservoir simulation problem around previously converged states stored from a set of training simulations. Therefore, it is a semi-intrusive technique since it only needs simulation state maps and their derivatives to be exported. In this work TPWL stability is assured through the use of Petrov-Galerkin projection on the POD process.
The reservoir simulator used in this work is based on a fully implicit formulation in terms of mass fractions instead of saturations. The mass fraction formulation does not appear to affect the accuracy of TPWL when compared to the literature. Speedups in the order of five hundred and very good accuracy were obtained.
This work tests a new strategy for the number of required initial simulations. Instead of several simulations, it was used only one and considered the possibility of retraining if necessary. The need for retraining, the retraining process itself and how to assimilate new information into the TPWL model are studied in detail.
The problem in study is to determine optimum sequence of well controls to be used during the concession period in order to maximize NPV. The optimization algorithm proposed is based on a trust region framework. It uses TPWL with Kriging correction as a proxy model and decides when it must be retrained. A small number of simulations inside the trust region must be performed in order to build a Kriging correction model. Different schemes to perform these simulations were evaluated in order to increase the efficiency of the method. Different interpolation polynomials to be used in the Kriging correction also considered. The methodology was applied to synthetic reservoir models with good results.