PDF(Pubmed)
Abstract：
The most crucial aspect in determining field development plans is the oil recovery factor (RF). However, RF has a complex relationship with the reservoir rock and fluid properties. The application of artificial neural networks is able to produce complex correlations between reservoir parameters that affect the recovery factor. This research provides a new approach to improve the accuracy of the ANN model in the form of steps including removing outlier data, selecting input parameters, selecting transferring functions, selecting the number of neurons, and determining hidden layers. By applying these steps, an ANN model was selected with nine input parameters consisting of oil viscosity, water saturation, initial oil formation volume factor, formation thickness, initial pressure, permeability, specific gravity of oil, porosity, and original oil in place. Furthermore, based on the correlation coefficient, a tangent sigmoid transferring function, 30 neurons, and two hidden layers were determined. The proposed ANN correlation gives the best accuracy compared to the previous correlations. This is proved by the highest correlation coefficient of 0.91657.
摘要：
确定油田开发计划的最关键方面是石油采收率（RF）。然而,RF与储层岩石和流体性质有着复杂的关系。人工神经网络的应用能够在影响采收率的储层参数之间产生复杂的相关性。本研究提供了一种新的方法来提高神经网络模型的准确性，包括去除离群数据，选择输入参数，选择传递函数，选择神经元的数量，并确定隐藏层。通过应用这些步骤，选择了一个ANN模型，该模型具有9个输入参数，包括油粘度，含水饱和度,初始油形成体积因子，地层厚度,初始压力,渗透性,石油的比重,孔隙度,原油到位。此外，根据相关系数，正切S形传递函数，30个神经元，并确定了两个隐藏层。与先前的相关性相比，所提出的ANN相关性给出了最好的准确性。这由0.91657的最高相关系数证明。
