CNN-LSTM Low-Resistivity Reservoir Fluid Identification Method Based on Attention Mechanism

Low-resistivity reservoirs are becoming increasingly significant in oil and gas exploration due to their extensive distribution and substantial reserves. However, their complex characteristics often render traditional logging interpretations insufficient. This study focuses on the low-resistivity reservoir fluids of the Hanjiang Formation and Zhujiang Formation in the Enping Sag of the Pearl River Mouth Basin. By analyzing gas logging data, quantitative fluorescence, and geochemical logging from 32 wells in the study area, we propose a hybrid model. This model combines the strengths of the Convolutional Neural Network (CNN) and Long Short-Term Memory Network (LSTM) and integrates an attention mechanism to enhance its ability to process small sample sizes of low-resistivity reservoir fluids, thereby improving recognition accuracy. The research findings show that the hybrid model significantly outperforms traditional single models in identifying low-resistivity reservoirs, achieving an accuracy rate of 88.43%, which surpasses single model performances by 2% to 8%. These results not only enhance the predictive accuracy of the model but also establish a new technical approach for identifying low-resistivity reservoirs.