Tight sandstone reservoirs are characterized by fine-grained rock particles, a high clay content, and a complex interplay between the electrical properties and gas content. These factors contribute to low-contrast reservoirs, where the logging responses of the gas and water layers are similar, resulting in traditional logging interpretation charts exhibiting a low accuracy in the fluid-type classification. This inadequacy fails to meet the fluid identification needs of the study area’s reservoirs and severely restricts the exploration and development of unconventional oil and gas resources. To address this challenge, this study proposes a fluid identification method based on Bayesian-optimized Support Vector Machine (SVM) to enhance the accuracy and efficiency of the fluid identification in low-contrast reservoirs. Firstly, through a sensitivity analysis of the logging responses, sensitive logging parameters such as the natural gamma, compensated density, compensated neutron, and compensated sonic logs are selected as input data for the model. Subsequently, Bayesian optimization is employed to automatically search for the optimal combination of hyperparameters for the SVM model. Finally, an SVM model is established using the optimized hyperparameters to classify and identify the following four fluid types: water layers, gas layers, gas–water layers, and dry layers. The proposed method is applied to fluid identification in the study area, and comparative experiments are conducted with the K-Nearest Neighbor (KNN), Random Forest (RF), and AdaBoost models. The classification performance of each model is systematically evaluated using metrics such as the accuracy, recall, and F1-score. The experimental results indicate that the SVM model outperforms the other models in fluid identification, achieving an average accuracy of 91.41%. This represents improvements of 16.94%, 4.39%, and 8.30% over the KNN, RF, and AdaBoost models, respectively. These findings validate the superiority of the SVM model for fluid identification in the study area and provide an efficient and feasible solution for fluid identification in tight sandstone reservoirs.
Read full abstract