基于门控循环单元神经网络的储层孔渗饱参数预测

Prediction of reservoir porosity,permeability,and saturation based on a gated recurrent unit neural network

孔隙度、渗透率和饱和度等物性参数是表征储层质量的重要参数,也是储层评价的重要依据。根据测井数据估算岩石的孔隙度、渗透率和饱和度参数,进而评价储层,是测井解释的基本内容。作为一种适于解决非线性和时序性问题的新型深度学习算法,门控循环单元(gated recurrent unit,GRU)神经网络算法能较好地反映出孔渗饱参数与测井数据之间的非线性映射关系以及不同深度历史数据之间的关联。基于GRU神经网络的储层孔渗饱参数预测方法首先采用基于Copula函数的相关性测度法筛选与孔渗饱参数关联度较高的测井参数,而后利用GRU神经网络建立测井数据与孔渗饱参数之间的非线性映射关系。对四川盆地某探区实际测井数据进行了GRU神经网络储层孔渗饱参数预测的模型训练和预测试验,最后将预测结果与多元回归分析、循环神经网络等方法的预测结果进行比较,结果表明,以均方根误差和Pearson相关系数为评价指标,基于门控循环单元神经网络的储层孔渗饱参数预测方法效果优于其它方法。

Porosity,permeability,and saturation are important parameters to characterize and evaluate reservoirs.The interpretation of well logging data can be used to evaluate the porosity,permeability,and saturation parameters of the rock,and thus discover reservoirs.A method for predicting reservoir porosity,permeability,and saturation based on a gated recurrent unit(GRU)neural network was proposed in this study.The GRU neural network is a novel deep learning algorithm suitable for solving nonlinear and time-dependent problems.As such,it may also be able to identify the non-linear mapping relationship between porosity,permeability,and saturation parameters and logging data,as well as the correlations among historical data at different depths.First,a correlation measurement method based on a copula function was employed in this work to select the logging parameters that best correlate with the porosity,permeability,and saturation parameters.Then,the GRU neural network was used to identify the non-linear mapping relationship between logging data and porosity,permeability,and saturation parameters.The results of the application to an exploration area in the Sichuan basin showed that the proposed method was superior to multiple regression analysis and recurrent neural network methods with respect to two evaluation indexes,namely the root mean square error and the correlation coefficient.