Reservoir classification and evaluation of tight sandstone formations based on electric imaging image processing and XGBoost algorithm

Imaging logging serves as a crucial technique for acquiring reservoir information. Extracting characteristic reservoir parameters, such as fractures and porosity, from log images is of significant importance for reservoir classification and evaluation. The tight sandstone reservoirs in the second member of Xujiahe Formation in the Western Sichuan Depression exhibit poor physical properties, low porosity and permeability, and diverse lithological types. Collectively, these factors contribute to complex fracture development patterns and pose challenges for characterizing their heterogeneity. To address these challenges, this study proposes an integrated evaluation method for tight sandstone reservoirs that combines electrical imaging log image processing with machine learning. The proposed approach initially employs the Otsu method combined with the watershed segmentation algorithm to segment electrical imaging log images and extract fracture features. Subsequently, the ensemble tree model eXtreme Gradient Boosting (XGBoost) is utilized to process the feature parameters, leveraging the advantages of machine learning in parameter processing and feature extraction. This ultimately enables reservoir classification and evaluation. Particularly under small-sample conditions, this method effectively enhances identification performance and facilitates the selection of high-quality reservoirs, thereby laying the groundwork for improved recovery efficiency. In practical application, this method successfully extracted characteristic parameters from fractures in tight sandstone reservoir samples and achieved reservoir classification and preferential selection. Validation against production data confirmed a classification accuracy of 88% (95% CI: 84% - 96%), effectively ensuring identification precision. This study provides more accurate geological interpretation and technical support for the drilling and development of tight sandstone reservoirs in the second member of the Xujiahe Formation in the Western Sichuan Depression. • Hybrid Otsu-watershed method enhances fracture detection in tight sandstone reservoirs, improving segmentation accuracy. • XGBoost algorithm achieves 88% accuracy in reservoir classification using fracture parameters. • Integration of imaging logging and machine learning enables precise identification of high-quality reservoir zones. • Fracture aperture is the most critical parameter for distinguishing reservoir types in tight sandstone formations. • Field application validation confirms the model's reliability in optimizing exploration and development strategies.