Abdollahfard Yaser, Mirabbasi Seyed Morteza, Ahmadi Mohammad, Hemmati-Sarapardeh Abdolhossein, Ashoorian Sefatallah
Petroleum Engineering Department, Amirkabir University of Technology, Tehran, Iran.
Department of Petroleum Engineering, Shahid Bahonar University of Kerman, Kerman, Iran.
Sci Rep. 2025 Apr 30;15(1):15251. doi: 10.1038/s41598-025-94617-7.
Permeability estimation plays an essential role in the assessment of reservoirs and hydrocarbon extraction. There are various methods to evaluate the formation and estimate the formation permeability, but in some cases, the evaluation may not be done or it may not be done correctly. This study focuses on a novel method to estimate the formation's permeability with appropriate accuracy using the mud loss data. Machine learning applications are becoming more popular nowadays and can succeed in many fields. This current research focuses on the application of mud loss data and deep learning to estimate the formation's permeability. To implement and validate our methodology, it is considered pilot cases including reservoir and drilling parameters values (depth, formation type, formation thickness, mud density, mud viscosity, and formation permeability). It is assumed that mud loss was occurred because of deferential pressure between formation pressure and bottom-hole pressure. The mud loss rate data were generated at different sets of reservoir and drilling data values using a reservoir simulator and then evaluated by calculating the correlation coefficients to ensure their validity and to check the fit under real conditions. This can be used to estimate the formation permeability values. One-dimensional convolutional neural networks(1D-CNN), a type of convolutional neural network, is utilized to be trained with data to perform a regression problem based on the contribution of flattening, dropout, and fully connected layers to estimate permeability with high accuracy (training data R = 0.970, testing data R = 0.964). Then the new deep learning method, Deep jointly informed neural network (DJINN), with the cooperation of neural networks and decision trees, provides a more accurate model than 1D-CNN (training data R = 0.978, testing data R = 0.972). These descriptions may provide new applications for mud loss data, where data while drilling can be used to predict formation permeability and provide insights for petroleum engineers to accurately measure design.
渗透率估算在油藏评价和油气开采中起着至关重要的作用。有多种方法可用于评估地层并估算地层渗透率，但在某些情况下，评估可能无法进行或无法正确进行。本研究聚焦于一种利用泥浆漏失数据以适当精度估算地层渗透率的新方法。如今机器学习应用越来越受欢迎，并且在许多领域都能取得成功。当前这项研究聚焦于利用泥浆漏失数据和深度学习来估算地层渗透率。为了实施和验证我们的方法，考虑了包括油藏和钻井参数值（深度、地层类型、地层厚度、泥浆密度、泥浆粘度和地层渗透率）的试点案例。假定泥浆漏失是由于地层压力与井底压力之间的压差所致。利用油藏模拟器在不同的油藏和钻井数据值集下生成泥浆漏失率数据，然后通过计算相关系数进行评估，以确保其有效性并检验在实际条件下的拟合情况。这可用于估算地层渗透率值。一维卷积神经网络（1D - CNN）作为卷积神经网络的一种类型，利用数据进行训练，基于展平、随机失活和全连接层的作用来执行回归问题，以高精度估算渗透率（训练数据R = 0.970，测试数据R = 0.964）。然后，新的深度学习方法——深度联合告知神经网络（DJINN），通过神经网络和决策树的协作，提供了一个比1D - CNN更精确的模型（训练数据R = 0.978，测试数据R = 0.972）。这些描述可能为泥浆漏失数据提供新的应用，即钻井过程中的数据可用于预测地层渗透率，并为石油工程师进行精确测量设计提供见解。
