关键词: 珠江口盆地, 文昌组, 成岩相, 图像处理, 机器学习, 测井

Abstract: The characteristics of sandstone diagenetic facies are complex and diverse. Taking the Eocene sandstones in the Lufeng Sag of the Pearl River Mouth Basin as the research object, it is of great practical significance to establish an accurate method for the classification and identification of sandstone diagenetic facies. To accurately classify the diagenetic facies, diagenetic parameters and image processing techniques were employed. The intensities of compaction, cementation, and dissolution were characterized by the content of each component in cast thin sections. Based on intensity difference, the diagenetic facies were classified into Class I, Class II, and Class III. Subsequently, a multilayer perceptron (MLP) neural network model was established by combining logging curve data including natural gamma, acoustic time difference, compensated neutron, compensated density, and resistivity, along with the characteristics of three diagenetic facies classes. Model parameters were optimized by controlling variables, ultimately achieving the classification of diagenetic facies and logging identification. Compared with the actual thin section diagenetic facies, the MLP model achieved an identification accuracy of 82.11%. When applied to well LF13-9-1, the model identified 23 samples with an accuracy of 82.61%, demonstrating good identification performance. The sandstone diagenetic facies classification and identification method integrating image processing technology and an MLP model provides a reference for future sandstone diagenetic facies identification.

Key words: Pearl River Mouth Basin, Wenchang Formation, diagenetic facies, image processing, machine learning, logging