基于机器学习的高含水油田剩余油预测方法

Prediction of remaining oil in high water cut oilfield based on machine learning

根据高含水油田剩余油分布特点,提出了含油饱和度等值线拟合样本预制作方法及基于人工神经网络的剩余油预测方法。利用数值模拟批量生成不同井距、物性、工作制度等条件下注采井组间剩余油分布场,编写模块自动提取少量含油饱和度等值线,使用多项式函数拟合不同时刻和层位含油饱和度等值线建立拟合参数样本数据集,实现机器学习样本参数量的大幅度降低。使用Tensorflow搭建神经网络模型,学习训练后形成注采井组间含油饱和度等值线预测模型,根据多个井组间等值线图叠加结果重构研究区含油饱和度场。基于高含水油田实际数据,与数值模拟相比,该方法对断层边界、层间剩余油富集区、井间局部零散剩余油均具一定预测能力;同时可将注采井生产动态数据快速转化为含油饱和度场数据,较传统方法的计算速度和定量化程度显著提高。

According to the distribution characteristics of remaining oil in high water cut oilfields,the pre-making method of fitting samples of oil saturation isolines and the remaining oil prediction method based on the artificial neural network are proposed.This paper applies the numerical simulation method to generate the remaining oil distribution fields between injection and production well groups under different well spacings,physical properties,working systems,and other conditions in batches.It programs a module to automatically extract a small amount of oil saturation isolines and constructs fitting parameter sample data set by polynomial functions to fit the oil saturation isolines at different times and horizon.This method can reduce the sample parameters of machine learning.Tensorflow is adopted to construct the neural network model.After the learning and training process,the oil saturation isoline prediction model between injection and production well groups is formed.The oil saturation field is reconstructed according to the superposition results of isoline maps between multiple well groups.Comparison between the actual data of high water cut oilfield and numerical simulation shows that the new method has prediction ability for fault boundary,interlayer residual oil enrichment area,and local scattered remaining oil between wells.This method can quickly convert dynamic data of oil and water wells into saturation field data.Compared with the traditional method,the proposed method significantly improves the calculation speed and quantification.