石油行业中应用的替代模型的介绍

INTRODUCTION OF THE PROXY MODELS IN PETROLEUM INDUSTRY

油田开发优化是一种计算密集型任务,需要大量的油藏模拟,这种模拟非常昂贵,尤其对于大宗复杂的油藏模型。替代模型在历史拟合、生产优化和预测过程计算量少、成本低,能有效地评估新方案的目标函数值和大幅度减少计算过程中反复调用模拟模型所引起的庞大的计算负荷,从而加速油田开发优化过程。本文主要介绍常见的四种替代模型:多项式回归模型(PRM)、泛克里金模型(MKG)、薄板样条插值(TPSM)、人工神经网络(ANN)。

Oilfield development optimization is a kind of computing intensive task, which needs a large number of reservoir simulations and is expensive, especially for the large complex reservoir model. The proxy model has advantages of less computation and low cost in the history matching. It can evaluate the objective function value of the new project effectively and reduce the huge computational load caused by using simulation model directly and repeatedly during the optimization process. The paper presents four kinds of proxy models: polynomial regression model, multivariate Kriging model, thin-plate spline interpolation and artificial neural network.