Método paralelo de superiorização para problemas de ajuste de histórico usando priors sísmicos a suavidade por partes

Abstract

History Matching is a very important process used in managing oil and gas production since it aims to adjust a reservoir model until it closely reproduces the past behavior of a actual reservoir, so it can be used to predict future production. This work proposes the use of an iterative method called superiorization for the constrained optimization of a reservoir production model. The superiorization method is a bi-objective optimization approach, where the first objective is the production result and the second is a measure of the smoothness by parts of the reservoir. The second criterion seeks to optimize its functional without negatively affecting the optimization of the first criterion. The superiorization method is used in conjunction with some iterative algorithm. Our work used the tabu search algorithm in conjunction with superiorization. As a comparative approach for these algorithms, a non-superiorized genetic algorithm was developed, given that this technique is widely used in the literature to solve history matching. Tests were also performed with the non-superiorized tabu search to see how the superiorized algorithm differs. Both techniques are iterative and use population approaches. As the problem addressed is an inverse problem that is usually severely underdetermined, several possible solutions may exist for its resolution. Due to this, we also propose the use of seismic data from the reservoirs to use these data to verify the faults present in the reservoir. We can use piecewise smooth functions to reduce the number of possible results through regularization using the second optimization criterion of the superiorized version of the tabu search algorithm. Another critical factor in the history matching process is the simulation time, which is generally high. Thus, we also propose to investigate the use of parallelism of the solution using the CPU. The experiments were carried out in a 3D reservoir model to find correspondence for gas, oil, and water production values. The results obtained during the research show that the parallel approach decreases the execution time by up to 70%. The genetic approach obtained better mean values using statistical tests regarding the precision of the result. However, the tabu search and the superiorization method showed similar results but were more stable since the results of this algorithm have less variation.