基于广义回归神经网络的CO2驱最小混相压力预测

针对传统的最小混相压力预测方法应用不便或误差较大等问题,提出利用广义回归神经网络进行CO2驱最小混相压力预测。以油藏温度、C5+分子量、中间组分摩尔分数、挥发组分摩尔分数为输入变量,以最小混相压力为输出变量,建立广义回归神经网络预测模型,对CO2驱最小混相压力进行预测,将结果与其他预测方法进行对比,并做误差分析。实例计算结果表明,广义回归神经网络用于CO2驱最小混相压力预测是可行的,且具有精度高、收敛快、适用范围广、使用简便等特点。

Prediction of minimum miscibility pressure in CO2 flooding based on general regression neural network

To reduce the inconvenience and errors in traditional prediction methods for minimum miscibility pressure （MMP）, a general regression neural network （GRNN） model was established for MMP prediction in CO2 flooding. The main factors affecting CO2 MMP, such as reservoir temperature, mole percentage of oil components （volatile and intermediate） and molecular weight of C5＋, were employed as the input variables of the GRNN, and the CO2 MMP was used as the output variable. To evaluate the advantage of the new method, the predicted results were compared between the GRNN model and the traditional empirical formula. The results show that the GRNN model is feasible for CO2 MMP prediction, and also has the characteristics of good precision, fast convergence, wide applicability and simple application.