【目的】为了深入探究页岩储层内部的孔裂隙分布规律以优化页岩油气的开发方案和提高产能。【方法】提出一种基于卷积神经网络的页岩CT图片分形维数预测方法,自主搭建适用于油页岩CT图片的卷积神经网络模型,分别将不同温度热解下的油页岩样品CT切片和其对应的分形维数作为数据集和标签,对搭建好的卷积神经网络进行训练并预测,实现对页岩CT图片分形维数的提取。【结果】结果证明,通过卷积神经网络预测的页岩CT图片分形维数与盒子计数法计算得到的分形维数十分接近,大约差0.01,且在计算速度更快的情况下还可以很大程度地忽略CT图片的噪声和伪影。新方法有效地捕捉到了图像的结构特征,能够对图片的分形维数进行可靠的预估并具有较好的抗干扰能力。

【Purposes】The development of shale oil and gas often requires a thorough understand⁃ ing of the internal pore-fracture distribution patterns within shale reservoirs to optimize development  strategies and enhance production capacity. In this context, the fractal dimension holds significant im⁃ portance for reflecting the distribution patterns of pores and fractures within shale formations.【Meth⁃ ods】In this study, a convolutional neural network-based method for predicting the fractal dimension  of shale Computed Tomography (CT) images is proposed. An independent convolutional neural net⁃ work model is constructed, specifically designed for oil shale CT images. CT slices of oil shale  samples treated with different temperatures, along with their corresponding fractal dimensions, are  employed as the dataset and labels. The constructed convolutional neural network is trained and uti⁃ lized for prediction to realize, effectively extracting fractal dimensions from shale CT images.【Find⁃ ings】The trained model is applied to various practical scenarios and compared with the box-counting  method. The results demonstrate a high degree of similarity between the predicted fractal dimensions  of shale CT images by using the convolutional neural network and those computed through the box- counting method, with a difference of approximately 0.01. Additionally, the convolutional neural net⁃ work method exhibits robustness against interference while also significantly accelerating the computa⁃ tion process compared with the box-counting method. Therefore, it can be concluded that the pro⁃ posed method effectively captures the structural characteristics of images, allowing for reliable estima⁃ tion of image fractal dimensions with notable resilience to noise and artifacts.