Permeability is an essential parameter for the reservoir characteristics, which controls the flowing fluids in the reservoir hence the sweep efficiency and the ultimate recovery. The common practice in the petroleum industry is coring a limited number of wells, due to the expensive core recovery process, and measuring the permeability in the recovered cores then extend the concluded correlation to the un-cored wells. However, establishing a reliable permeability predictor is not an easy task in many heterogeneous formations due to the spatial variability of the permeability even at very close distances. Therefore, the conventional linear regression has often failed to address the formation’s heterogeneity, and an unsatisfied correlation coefficient has frequently obtained. Lower Qamchuqa formation, which is highly prolific producing formation in the Middle East, has been used as an example of highly heterogeneous carbonate systems. Well log measurements, which are available for most of the wells, in addition to the core data, were used to capture the high heterogeneity of the depositional environment. Core-log depth calibration was first performed to extract the accurate log measurements that correspond to the actual core data depth. Then, both core and log data were listed in a table for the statistical analysis using the neural network (NN) and multivariate regression approaches. A remarkable improvement in the correlation coefficient was obtained using the NN approach. The utilised training data and further verified by the validation data set have obtained a favourable accuracy compared with conventional linear regression or multivariate regression. The NN permeability predictor has proven its ability to overcome the complexity of the carbonate rock textures and the variety of the diagenesis alteration processes, which make the NN approach a superior method in obtaining an improved permeability predictor. Nevertheless, a regular update to estimate the permeability predictor would be necessary when new data acquired.

渗透率是油藏特性的一个重要参数，它控制着油藏中的流动流体，从而控制波及效率和最终采收率。由于昂贵的岩心回收过程，石油工业中的常见做法是对有限数量的井进行取心，然后测量回收岩心的渗透率，然后将得出的相关性扩展到未取心井。然而，由于渗透率的空间可变性，即使在非常近的距离，在许多非均质地层中建立可靠的渗透率预测器并非易事。因此，传统的线性回归往往无法解决地层的异质性，经常得到不满意的相关系数。下 Qamchuqa 组，这是中东地区产量很高的地层，已被用作高度非均质碳酸盐系统的一个例子。除了岩心数据外，大多数井都可以使用测井数据来捕捉沉积环境的高度异质性。首先进行岩心测井深度校准以提取与实际岩心数据深度相对应的准确测井测量值。然后，使用神经网络 (NN) 和多元回归方法将核心数据和日志数据都列在表中以进行统计分析。使用 NN 方法获得了相关系数的显着改善。与传统的线性回归或多元回归相比，利用的训练数据并通过验证数据集进一步验证获得了良好的准确性。NN 渗透率预测器已证明其能够克服碳酸盐岩结构的复杂性和成岩蚀变过程的多样性，这使得 NN 方法成为获得改进的渗透率预测器的优越方法。然而，当获得新数据时，需要定期更新以估计渗透率预测器。