Subdivision of River Channel Sand Micro-scale Facies with Feature Attention Spatio-Temporal Network

Ruipu Zhao ¹ Lili Zeng ² Chendong Fu ³ Xiaoqing Zhao ^2*

• ¹ Tianjin University, Tianjin, China
• ² Northeast Petroleum University, Daqing, China
• ³ Daqing Branch, China National Logging Corporation, Daqing, China

Sedimentary micro-scale facies research plays a crucial role in characterizing the lateral and vertical evolutionary patterns and the contact relationships both within and between sedimentary micro-scale facies, which is vital for the redevelopment of high-water-cut oil reservoirs. However, challenges such as the heterogeneity of river channel sand in both horizontal and vertical dimensions, well connectivity, and the effectiveness of water injection require a more refined subdivision of channel sand sedimentary micro-scale facies. Traditional manual identification methods, which rely on effective thickness, formation thickness, and logging curve morphology, are labor-intensive and prone to subjectivity. To address these limitations, this paper proposes an innovative method that integrates well-logging sedimentology with statistical theory, selects multiple reservoir and logging parameters, and establishes novel classification standards for river channel sand sedimentary micro-scale facies. Based on deep learning techniques, the method introduces a feature attention module that dynamically assigns weights to logging parameters according to their linear relationships with recognition targets.Additionally, it incorporates spatial and temporal feature extraction modules to focus on key logging information, enabling precise boundary delineation and intelligent recognition of river channel sand sedimentary micro-scale facies. When applied to a real-world work area for residual oil development, this method establishes a classification standard that subdivides deltaic river channel sand sedimentary micro-scale facies into four distinct types, achieving over 94% accuracy in sedimentary micro-scale facies subdivision and identification. This approach not only provides a novel framework for analyzing connectivity between injection and production well groups but also demonstrates strong potential for application in other geological settings.