基于二维散射变换的湖相碳酸盐岩储层厚度预测方法研究

济阳坳陷沙四段湖相碳酸盐岩受湖盆内沉积环境和构造运动等因素控制, 储层厚度变化大, 非均质性强, 油气开发难度大.为了准确预测湖相碳酸盐岩储层的厚度, 本文提出一种基于二维散射变换和随机森林的储层厚度预测方法.首先, 引入二维散射变换提取地震时频属性, 该变换是在二维小波变换的基础上, 通过迭代小波分解和非线性操作来实现的.与传统的二维小波变换对比, 散射变换提取的时频属性具有局部形变稳定性以及对噪声鲁棒性的优点, 有助于提高储层厚度预测的准确率.在此基础上, 在有限测井数据的条件下, 利用随机森林算法建立多尺度时频属性与测井解释厚度之间的非线性关系, 实现湖相碳酸盐岩储层预测.模型数据的预测结果表明, 与基于传统地震振幅属性的厚度预测和基于二维小波变换的储层厚度预测对比, 本文所提的厚度预测方法具有最优的性能.叠后三维地震数据的预测结果表明, 与基于传统地震振幅属性的厚度预测和基于二维小波变换的储层厚度预测对比, 本文所提方法的厚度预测结果与实际钻井数据误差更小, 提高了储层厚度预测的精度, 清晰刻画了灰礁、灰滩与灰泥等三种沉积亚相的空间展布, 有利于后续井位部署和优化.

Reservoir thickness prediction method of lacustrine carbonate based on two-dimension scattering transformation

The lacustrine carbonate rocks of the fourth member of Shahejie Formation in Jiyang Depression are controlled by the sedimentary environment and tectonic movement in the lacustrine basin. The thickness of the reservoirs varies greatly and the heterogeneity is strong, which makes it difficult to develop oil and gas. To accurately predict the thickness of lacustrine carbonate reservoirs, we propose a thickness prediction method for the lacustrine carbonate reservoirs based on the two-dimension scattering transform and random forest. First, the seismic time-frequency attributes are extracted by the two-dimensional scattering transform, which is generated by iterative wavelet decomposition and nonlinear operation based on two-dimensional wavelet transform. Compared with the traditional two-dimension wavelet, the scattering transform has local translation deformation stability, rotation deformation stability, and noise robustness, which are helpful to improve the accuracy of reservoir thickness prediction. Furthermore, due to the limitation of the well information, the random forest method is utilized to construct the nonlinear relationship between multi-scale scattering transform attributes and the interpreted thickness of the well. Then, the thickness prediction method of the lacustrine carbonate reservoir is proposed. Afterward, the synthetic seismic data is applied to the proposed method and the results reveal that the proposed method has the best performance in comparison with the traditional seismic amplitude attributes and the two-dimensional wavelet transform. Finally, the effectiveness of the proposed method is demonstrated by using the three-dimension post-stack seismic data. Compared with the prediction results based on traditional seismic amplitude attributes and the two-dimensional wavelet transform, the predicted result of the proposed method is completely consistent with the drilling well data, which can improve the accuracy of reservoir prediction. The proposed method can also identify the sedimentary sub-facies of the reef with limestone, the beach with limestone, and the plaster, which is beneficial for the subsequent well placement.