超深缝洞型油气藏孔隙度智能学习预测模型

A deep learning-based model for predicting porosity of ultradeep fractured-vuggy hydrocarbon reservoirs

康志江 邓紫妍 杨帆 周东升

Zhijiang KANG Ziyan DENG Fan YANG Dongsheng ZHOU

构建超深层碳酸盐岩油气藏孔隙度的预测模型在勘探开发中具有重要意义。超深缝洞型油气藏埋藏深度超过7 500 m，地震信噪比偏低，缝洞非均质性极为显著，地震波阻抗的缝洞孔隙度模型与测井解释结果相比偏差较大。为此提出了基于深度学习的多地震属性非线性关系孔隙度模型预测方法，选取与孔隙度相关的 8 类地震属性，构建与测井解释孔隙度相匹配的地震属性训练集。通过Box-Cox变换方式对数据集的偏态分布予以改善，运用地震相约束的深度学习模型以及贝叶斯算法加以优化，建立了多地震属性与孔隙度非线性关系数学模型。与常用的地震波阻抗体预测模型方法相比，新模型验证井吻合度由24 %提高至92 %，孔隙度模型预测精度大幅提升。

Constructing a porosity prediction model for ultradeep fractured-vuggy hydrocarbon reservoirs holds great significance for their exploration and exploitation. These reservoirs， with burial depths exceeding 7 500 m， exhibit relatively low signal-to-noise ratios （SNRs） and extremely significant heterogeneity in fractures and vugs. These factors lead to major deviations between the porosity predicted using seismic wave impedance-based models and well log interpretation results for fractured-vuggy reservoirs. In this study， we propose a deep learning-based model for predicting the reservoir porosity involving the nonlinear relationships among multiple seismic attributes. Specifically， eight seismic attributes related to reservoir porosity are selected to construct a seismic attribute training set that matches log-derived porosity. Then， the skewed distribution of seismic attributes in the training set is corrected using the Box-Cox transformation， and then optimized with the seismic facies-constrained deep learning model as well as the Bayesian algorithms. A mathematical model is thereby established illustrating the nonlinear relationships between multiple seismic attributes and reservoir porosity. Compared to the commonly used seismic wave impedance-based prediction model， the newly proposed model can increase the goodness of fit of validation wells from 24 % to 92 %， substantially enhancing the prediction accuracy.

10.11743/ogg20250215