Comparação de métodos de otimização para o problema de ajuste de histórico em ambientes paralelos

Abstract

The process of history matching aims on the determination of the models' parameters from a petroleum reservoir. Once adjusted, the models can be used for the prediction of the reservoir behavior. This work presents a comparsion of different optimization methods for this problem's solution. Derivative based methods are compared to a genetic algorithm. In particular, the following methods are compared: Levenberg-Marquadt, Quasi-Newton, Non Linear Conjugate Gradient, steepest descent and genetic algorithm. Due to the great computational demand of this problem, the parallel computing has been widely used. The comparsions among the optimization algorithms were performed in an heterogeneous parallel computing environment and the preliminar results are presented and discussed.