基于机器学习的低渗透砂岩聚合物驱采收率预测

Predicting the Recovery Factor of Polymer Flooding in Low-permeability Sandstone Based on Machine Learning

在恶劣的油藏条件下,化学驱提高采收率方法的可行性主要在实验室进行,以探究化学驱方案在现场实施的可能效果,但此类实验通常昂贵且费时。为了提高筛选效率和研究变量关系,进行了3个聚合物驱油实验项目,其次通过构建14种机器学习基础模型来预测低渗透砂岩聚合物驱油实验的效率。结果表明:多层感知机(multi-layer perception,MLP)、随机树(random forest,RF)和极限梯度上升(extreme gradient boosting,XGB)模型表现最佳,它们在测试集的确定系数均为0.99,均方根误差分别为0.855、0.836和0.859。模型表明特征重要性由强至弱依次为含水率、累积注入孔隙体积、渗透率、非均质系数、孔隙度、聚合物注入量、聚合物浓度、注入压力。研究成果为室内物理低渗透砂岩聚合物驱提供了可靠的数据,给出了14种机器学习模型预测性能直接对比,建立了高拟合高泛化高稳定低误差的低渗透砂岩聚合物驱预测模型,有助于化学驱方案快速在低渗透储层应用,以及降低失败风险。

In harsh reservoir conditions,the feasibility of chemical flooding-enhanced oil recovery methods is mainly conducted in the laboratory to explore the possible effects of chemical flooding in the field,but such experiments are often expensive and time-consuming.To improve screening effectiveness and investigate the relationship between variables,three polymer flooding experiments were conducted.14 basic machine learning models were built to forecast the effectiveness of polymer flooding experiments in low-permeability sandstone.The results show that multi-layer perception(MLP),random forest(RF)and extreme gradient boosting(XGB)models have the best performance.Their coefficients of determination within the test set are all 0.99,and the mean square errors are 0.855,0.836 and 0.859,respectively.The model shows that the importance of characteristics from strong to weak is water content,cumulative injected pore volume,permeability,heterogeneity coefficient,porosity,polymer injection volume,polymer concentration,and injection pressure.The research results provide reliable data for indoor physical low-permeability sandstone polymer flooding,provides a direct comparison of the predictive performance of 14 machine learning models,and establishes a high fitting,high generalization,high stability,and low error prediction model for low-permeability sandstone polymer flooding,which will help chemical flooding schemes to be applied quickly in low-permeability reservoirs and reduce the risk of failure.