CNN-GRU-ATT Method for Resistivity Logging Curve Reconstruction and Fluid Property Identification in Marine Carbonate Reservoirs

Geophysical logging curves are crucial for oil and gas field exploration and development, and curve reconstruction techniques are a key focus of research in this field. This study proposes an inversion model for deep resistivity curves in marine carbonate reservoirs, specifically the Mishrif Formation of the Halfaya Field, by integrating a deep learning model called CNN-GRU-ATT, which combines Convolutional Neural Networks (CNN), Gated Recurrent Units (GRU), and the Attention Mechanism (ATT). Using logging data from the marine carbonate oil layers, the reconstructed deep resistivity curve is compared with actual measurements to determine reservoir fluid properties. The results demonstrate the effectiveness of the CNN-GRU-ATT model in accurately reconstructing deep resistivity curves for carbonate reservoirs within the Mishrif Formation. Notably, the model outperforms alternative methods such as CNN-GRU, GRU, Long Short-Term Memory (LSTM), Multiple Regression, and Random Forest in new wells, exhibiting high accuracy and robust generalization capabilities. In practical applications, the response of the inverted deep resistivity curve can be utilized to identify the reservoir water cut. Specifically, when the model-inverted curve exhibits a higher response compared to the measured curve, it indicates the presence of reservoir water. Additionally, a stable relative position between the two curves suggests the presence of a water layer. Utilizing this method, the oil–water transition zone can be accurately delineated, achieving a fluid property identification accuracy of 93.14%. This study not only introduces a novel curve reconstruction method but also presents a precise approach to identifying reservoir fluid properties. These findings establish a solid technical foundation for decision-making support in oilfield development.